linux/drivers/acpi/nfit/core.c

3680 lines
97 KiB
C
Raw Normal View History

/*
* Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of version 2 of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*/
#include <linux/list_sort.h>
#include <linux/libnvdimm.h>
#include <linux/module.h>
#include <linux/mutex.h>
2015-06-08 18:27:06 +00:00
#include <linux/ndctl.h>
#include <linux/sysfs.h>
#include <linux/delay.h>
#include <linux/list.h>
#include <linux/acpi.h>
2015-05-01 17:11:27 +00:00
#include <linux/sort.h>
#include <linux/io.h>
#include <linux/nd.h>
x86, pmem: clarify that ARCH_HAS_PMEM_API implies PMEM mapped WB Given that a write-back (WB) mapping plus non-temporal stores is expected to be the most efficient way to access PMEM, update the definition of ARCH_HAS_PMEM_API to imply arch support for WB-mapped-PMEM. This is needed as a pre-requisite for adding PMEM to the direct map and mapping it with struct page. The above clarification for X86_64 means that memcpy_to_pmem() is permitted to use the non-temporal arch_memcpy_to_pmem() rather than needlessly fall back to default_memcpy_to_pmem() when the pcommit instruction is not available. When arch_memcpy_to_pmem() is not guaranteed to flush writes out of cache, i.e. on older X86_32 implementations where non-temporal stores may just dirty cache, ARCH_HAS_PMEM_API is simply disabled. The default fall back for persistent memory handling remains. Namely, map it with the WT (write-through) cache-type and hope for the best. arch_has_pmem_api() is updated to only indicate whether the arch provides the proper helpers to meet the minimum "writes are visible outside the cache hierarchy after memcpy_to_pmem() + wmb_pmem()". Code that cares whether wmb_pmem() actually flushes writes to pmem must now call arch_has_wmb_pmem() directly. Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com> [hch: set ARCH_HAS_PMEM_API=n on x86_32] Reviewed-by: Christoph Hellwig <hch@lst.de> [toshi: x86_32 compile fixes] Signed-off-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-24 22:29:38 +00:00
#include <asm/cacheflush.h>
#include <acpi/nfit.h>
#include "nfit.h"
#include "intel.h"
/*
* For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
* irrelevant.
*/
#include <linux/io-64-nonatomic-hi-lo.h>
static bool force_enable_dimms;
module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
static bool disable_vendor_specific;
module_param(disable_vendor_specific, bool, S_IRUGO);
MODULE_PARM_DESC(disable_vendor_specific,
"Limit commands to the publicly specified set");
static unsigned long override_dsm_mask;
module_param(override_dsm_mask, ulong, S_IRUGO);
MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
static int default_dsm_family = -1;
module_param(default_dsm_family, int, S_IRUGO);
MODULE_PARM_DESC(default_dsm_family,
"Try this DSM type first when identifying NVDIMM family");
static bool no_init_ars;
module_param(no_init_ars, bool, 0644);
MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
LIST_HEAD(acpi_descs);
DEFINE_MUTEX(acpi_desc_lock);
static struct workqueue_struct *nfit_wq;
struct nfit_table_prev {
struct list_head spas;
struct list_head memdevs;
struct list_head dcrs;
struct list_head bdws;
struct list_head idts;
struct list_head flushes;
};
static guid_t nfit_uuid[NFIT_UUID_MAX];
const guid_t *to_nfit_uuid(enum nfit_uuids id)
{
return &nfit_uuid[id];
}
tools/testing/nvdimm: libnvdimm unit test infrastructure 'libnvdimm' is the first driver sub-system in the kernel to implement mocking for unit test coverage. The nfit_test module gets built as an external module and arranges for external module replacements of nfit, libnvdimm, nd_pmem, and nd_blk. These replacements use the linker --wrap option to redirect calls to ioremap() + request_mem_region() to custom defined unit test resources. The end result is a fully functional nvdimm_bus, as far as userspace is concerned, but with the capability to perform otherwise destructive tests on emulated resources. Q: Why not use QEMU for this emulation? QEMU is not suitable for unit testing. QEMU's role is to faithfully emulate the platform. A unit test's role is to unfaithfully implement the platform with the goal of triggering bugs in the corners of the sub-system implementation. As bugs are discovered in platforms, or the sub-system itself, the unit tests are extended to backstop a fix with a reproducer unit test. Another problem with QEMU is that it would require coordination of 3 software projects instead of 2 (kernel + libndctl [1]) to maintain and execute the tests. The chances for bit rot and the difficulty of getting the tests running goes up non-linearly the more components involved. Q: Why submit this to the kernel tree instead of external modules in libndctl? Simple, to alleviate the same risk that out-of-tree external modules face. Updates to drivers/nvdimm/ can be immediately evaluated to see if they have any impact on tools/testing/nvdimm/. Q: What are the negative implications of merging this? It is a unique maintenance burden because the purpose of mocking an interface to enable a unit test is to purposefully short circuit the semantics of a routine to enable testing. For example __wrap_ioremap_cache() fakes the pmem driver into "ioremap()'ing" a test resource buffer allocated by dma_alloc_coherent(). The future maintenance burden hits when someone changes the semantics of ioremap_cache() and wonders what the implications are for the unit test. [1]: https://github.com/pmem/ndctl Cc: <linux-acpi@vger.kernel.org> Cc: Lv Zheng <lv.zheng@intel.com> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-17 21:23:32 +00:00
EXPORT_SYMBOL(to_nfit_uuid);
2015-06-08 18:27:06 +00:00
static struct acpi_nfit_desc *to_acpi_nfit_desc(
struct nvdimm_bus_descriptor *nd_desc)
{
return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
}
static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
{
struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
/*
* If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
* acpi_device.
*/
if (!nd_desc->provider_name
|| strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
return NULL;
return to_acpi_device(acpi_desc->dev);
}
static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
{
struct nd_cmd_clear_error *clear_err;
struct nd_cmd_ars_status *ars_status;
u16 flags;
switch (cmd) {
case ND_CMD_ARS_CAP:
if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
return -ENOTTY;
/* Command failed */
if (status & 0xffff)
return -EIO;
/* No supported scan types for this range */
flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
if ((status >> 16 & flags) == 0)
return -ENOTTY;
return 0;
case ND_CMD_ARS_START:
/* ARS is in progress */
if ((status & 0xffff) == NFIT_ARS_START_BUSY)
return -EBUSY;
/* Command failed */
if (status & 0xffff)
return -EIO;
return 0;
case ND_CMD_ARS_STATUS:
ars_status = buf;
/* Command failed */
if (status & 0xffff)
return -EIO;
/* Check extended status (Upper two bytes) */
if (status == NFIT_ARS_STATUS_DONE)
return 0;
/* ARS is in progress */
if (status == NFIT_ARS_STATUS_BUSY)
return -EBUSY;
/* No ARS performed for the current boot */
if (status == NFIT_ARS_STATUS_NONE)
return -EAGAIN;
/*
* ARS interrupted, either we overflowed or some other
* agent wants the scan to stop. If we didn't overflow
* then just continue with the returned results.
*/
if (status == NFIT_ARS_STATUS_INTR) {
if (ars_status->out_length >= 40 && (ars_status->flags
& NFIT_ARS_F_OVERFLOW))
return -ENOSPC;
return 0;
}
/* Unknown status */
if (status >> 16)
return -EIO;
return 0;
case ND_CMD_CLEAR_ERROR:
clear_err = buf;
if (status & 0xffff)
return -EIO;
if (!clear_err->cleared)
return -EIO;
if (clear_err->length > clear_err->cleared)
return clear_err->cleared;
return 0;
default:
break;
}
/* all other non-zero status results in an error */
if (status)
return -EIO;
return 0;
}
#define ACPI_LABELS_LOCKED 3
static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
u32 status)
{
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
switch (cmd) {
case ND_CMD_GET_CONFIG_SIZE:
/*
* In the _LSI, _LSR, _LSW case the locked status is
* communicated via the read/write commands
*/
if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
break;
if (status >> 16 & ND_CONFIG_LOCKED)
return -EACCES;
break;
case ND_CMD_GET_CONFIG_DATA:
if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
&& status == ACPI_LABELS_LOCKED)
return -EACCES;
break;
case ND_CMD_SET_CONFIG_DATA:
if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
&& status == ACPI_LABELS_LOCKED)
return -EACCES;
break;
default:
break;
}
/* all other non-zero status results in an error */
if (status)
return -EIO;
return 0;
}
static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
u32 status)
{
if (!nvdimm)
return xlat_bus_status(buf, cmd, status);
return xlat_nvdimm_status(nvdimm, buf, cmd, status);
}
/* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
static union acpi_object *pkg_to_buf(union acpi_object *pkg)
{
int i;
void *dst;
size_t size = 0;
union acpi_object *buf = NULL;
if (pkg->type != ACPI_TYPE_PACKAGE) {
WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
pkg->type);
goto err;
}
for (i = 0; i < pkg->package.count; i++) {
union acpi_object *obj = &pkg->package.elements[i];
if (obj->type == ACPI_TYPE_INTEGER)
size += 4;
else if (obj->type == ACPI_TYPE_BUFFER)
size += obj->buffer.length;
else {
WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
obj->type);
goto err;
}
}
buf = ACPI_ALLOCATE(sizeof(*buf) + size);
if (!buf)
goto err;
dst = buf + 1;
buf->type = ACPI_TYPE_BUFFER;
buf->buffer.length = size;
buf->buffer.pointer = dst;
for (i = 0; i < pkg->package.count; i++) {
union acpi_object *obj = &pkg->package.elements[i];
if (obj->type == ACPI_TYPE_INTEGER) {
memcpy(dst, &obj->integer.value, 4);
dst += 4;
} else if (obj->type == ACPI_TYPE_BUFFER) {
memcpy(dst, obj->buffer.pointer, obj->buffer.length);
dst += obj->buffer.length;
}
}
err:
ACPI_FREE(pkg);
return buf;
}
static union acpi_object *int_to_buf(union acpi_object *integer)
{
union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
void *dst = NULL;
if (!buf)
goto err;
if (integer->type != ACPI_TYPE_INTEGER) {
WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
integer->type);
goto err;
}
dst = buf + 1;
buf->type = ACPI_TYPE_BUFFER;
buf->buffer.length = 4;
buf->buffer.pointer = dst;
memcpy(dst, &integer->integer.value, 4);
err:
ACPI_FREE(integer);
return buf;
}
static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
u32 len, void *data)
{
acpi_status rc;
struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
struct acpi_object_list input = {
.count = 3,
.pointer = (union acpi_object []) {
[0] = {
.integer.type = ACPI_TYPE_INTEGER,
.integer.value = offset,
},
[1] = {
.integer.type = ACPI_TYPE_INTEGER,
.integer.value = len,
},
[2] = {
.buffer.type = ACPI_TYPE_BUFFER,
.buffer.pointer = data,
.buffer.length = len,
},
},
};
rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
if (ACPI_FAILURE(rc))
return NULL;
return int_to_buf(buf.pointer);
}
static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
u32 len)
{
acpi_status rc;
struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
struct acpi_object_list input = {
.count = 2,
.pointer = (union acpi_object []) {
[0] = {
.integer.type = ACPI_TYPE_INTEGER,
.integer.value = offset,
},
[1] = {
.integer.type = ACPI_TYPE_INTEGER,
.integer.value = len,
},
},
};
rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
if (ACPI_FAILURE(rc))
return NULL;
return pkg_to_buf(buf.pointer);
}
static union acpi_object *acpi_label_info(acpi_handle handle)
{
acpi_status rc;
struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
if (ACPI_FAILURE(rc))
return NULL;
return pkg_to_buf(buf.pointer);
}
static u8 nfit_dsm_revid(unsigned family, unsigned func)
{
static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
[NVDIMM_FAMILY_INTEL] = {
[NVDIMM_INTEL_GET_MODES] = 2,
[NVDIMM_INTEL_GET_FWINFO] = 2,
[NVDIMM_INTEL_START_FWUPDATE] = 2,
[NVDIMM_INTEL_SEND_FWUPDATE] = 2,
[NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
[NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
[NVDIMM_INTEL_SET_THRESHOLD] = 2,
[NVDIMM_INTEL_INJECT_ERROR] = 2,
},
};
u8 id;
if (family > NVDIMM_FAMILY_MAX)
return 0;
if (func > 31)
return 0;
id = revid_table[family][func];
if (id == 0)
return 1; /* default */
return id;
}
int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
{
2015-06-08 18:27:06 +00:00
struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2015-06-08 18:27:06 +00:00
union acpi_object in_obj, in_buf, *out_obj;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
const struct nd_cmd_desc *desc = NULL;
2015-06-08 18:27:06 +00:00
struct device *dev = acpi_desc->dev;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
struct nd_cmd_pkg *call_pkg = NULL;
2015-06-08 18:27:06 +00:00
const char *cmd_name, *dimm_name;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
unsigned long cmd_mask, dsm_mask;
u32 offset, fw_status = 0;
2015-06-08 18:27:06 +00:00
acpi_handle handle;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
unsigned int func;
const guid_t *guid;
2015-06-08 18:27:06 +00:00
int rc, i;
if (cmd_rc)
*cmd_rc = -EINVAL;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
func = cmd;
if (cmd == ND_CMD_CALL) {
call_pkg = buf;
func = call_pkg->nd_command;
libnvdimm, nfit: move the check on nd_reserved2 to the endpoint Delay the check of nd_reserved2 to the actual endpoint (acpi_nfit_ctl) that uses it, as a prevention of a potential double-fetch bug. While examining the kernel source code, I found a dangerous operation that could turn into a double-fetch situation (a race condition bug) where the same userspace memory region are fetched twice into kernel with sanity checks after the first fetch while missing checks after the second fetch. In the case of _IOC_NR(ioctl_cmd) == ND_CMD_CALL: 1. The first fetch happens in line 935 copy_from_user(&pkg, p, sizeof(pkg) 2. subsequently `pkg.nd_reserved2` is asserted to be all zeroes (line 984 to 986). 3. The second fetch happens in line 1022 copy_from_user(buf, p, buf_len) 4. Given that `p` can be fully controlled in userspace, an attacker can race condition to override the header part of `p`, say, `((struct nd_cmd_pkg *)p)->nd_reserved2` to arbitrary value (say nine 0xFFFFFFFF for `nd_reserved2`) after the first fetch but before the second fetch. The changed value will be copied to `buf`. 5. There is no checks on the second fetches until the use of it in line 1034: nd_cmd_clear_to_send(nvdimm_bus, nvdimm, cmd, buf) and line 1038: nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc) which means that the assumed relation, `p->nd_reserved2` are all zeroes might not hold after the second fetch. And once the control goes to these functions we lose the context to assert the assumed relation. 6. Based on my manual analysis, `p->nd_reserved2` is not used in function `nd_cmd_clear_to_send` and potential implementations of `nd_desc->ndctl` so there is no working exploit against it right now. However, this could easily turns to an exploitable one if careless developers start to use `p->nd_reserved2` later and assume that they are all zeroes. Move the validation of the nd_reserved2 field to the ->ndctl() implementation where it has a stable buffer to evaluate. Signed-off-by: Meng Xu <mengxu.gatech@gmail.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2017-09-04 15:34:33 +00:00
for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
if (call_pkg->nd_reserved2[i])
return -EINVAL;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
}
2015-06-08 18:27:06 +00:00
if (nvdimm) {
struct acpi_device *adev = nfit_mem->adev;
if (!adev)
return -ENOTTY;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
if (call_pkg && nfit_mem->family != call_pkg->nd_family)
return -ENOTTY;
dimm_name = nvdimm_name(nvdimm);
2015-06-08 18:27:06 +00:00
cmd_name = nvdimm_cmd_name(cmd);
cmd_mask = nvdimm_cmd_mask(nvdimm);
2015-06-08 18:27:06 +00:00
dsm_mask = nfit_mem->dsm_mask;
desc = nd_cmd_dimm_desc(cmd);
guid = to_nfit_uuid(nfit_mem->family);
2015-06-08 18:27:06 +00:00
handle = adev->handle;
} else {
struct acpi_device *adev = to_acpi_dev(acpi_desc);
cmd_name = nvdimm_bus_cmd_name(cmd);
cmd_mask = nd_desc->cmd_mask;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
dsm_mask = cmd_mask;
if (cmd == ND_CMD_CALL)
dsm_mask = nd_desc->bus_dsm_mask;
2015-06-08 18:27:06 +00:00
desc = nd_cmd_bus_desc(cmd);
guid = to_nfit_uuid(NFIT_DEV_BUS);
2015-06-08 18:27:06 +00:00
handle = adev->handle;
dimm_name = "bus";
}
if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
return -ENOTTY;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
2015-06-08 18:27:06 +00:00
return -ENOTTY;
in_obj.type = ACPI_TYPE_PACKAGE;
in_obj.package.count = 1;
in_obj.package.elements = &in_buf;
in_buf.type = ACPI_TYPE_BUFFER;
in_buf.buffer.pointer = buf;
in_buf.buffer.length = 0;
/* libnvdimm has already validated the input envelope */
for (i = 0; i < desc->in_num; i++)
in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
i, buf);
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
if (call_pkg) {
/* skip over package wrapper */
in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
in_buf.buffer.length = call_pkg->nd_size_in;
}
dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
dimm_name, cmd, func, in_buf.buffer.length);
print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
in_buf.buffer.pointer,
min_t(u32, 256, in_buf.buffer.length), true);
2015-06-08 18:27:06 +00:00
/* call the BIOS, prefer the named methods over _DSM if available */
if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
&& test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
out_obj = acpi_label_info(handle);
else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
&& test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
struct nd_cmd_get_config_data_hdr *p = buf;
out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
} else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
&& test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
struct nd_cmd_set_config_hdr *p = buf;
out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
p->in_buf);
} else {
u8 revid;
if (nvdimm)
revid = nfit_dsm_revid(nfit_mem->family, func);
else
revid = 1;
out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
}
2015-06-08 18:27:06 +00:00
if (!out_obj) {
dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
2015-06-08 18:27:06 +00:00
return -EINVAL;
}
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
if (call_pkg) {
call_pkg->nd_fw_size = out_obj->buffer.length;
memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
out_obj->buffer.pointer,
min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
ACPI_FREE(out_obj);
/*
* Need to support FW function w/o known size in advance.
* Caller can determine required size based upon nd_fw_size.
* If we return an error (like elsewhere) then caller wouldn't
* be able to rely upon data returned to make calculation.
*/
if (cmd_rc)
*cmd_rc = 0;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
return 0;
}
2015-06-08 18:27:06 +00:00
if (out_obj->package.type != ACPI_TYPE_BUFFER) {
dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
dimm_name, cmd_name, out_obj->type);
2015-06-08 18:27:06 +00:00
rc = -EINVAL;
goto out;
}
dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
cmd_name, out_obj->buffer.length);
print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
out_obj->buffer.pointer,
min_t(u32, 128, out_obj->buffer.length), true);
2015-06-08 18:27:06 +00:00
for (i = 0, offset = 0; i < desc->out_num; i++) {
u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
acpi, nfit, libnvdimm: fix / harden ars_status output length handling Given ambiguities in the ACPI 6.1 definition of the "Output (Size)" field of the ARS (Address Range Scrub) Status command, a firmware implementation may in practice return 0, 4, or 8 to indicate that there is no output payload to process. The specification states "Size of Output Buffer in bytes, including this field.". However, 'Output Buffer' is also the name of the entire payload, and earlier in the specification it states "Max Query ARS Status Output Buffer Size: Maximum size of buffer (including the Status and Extended Status fields)". Without this fix if the BIOS happens to return 0 it causes memory corruption as evidenced by this result from the acpi_nfit_ctl() unit test. ars_status00000000: 00020000 00000000 ........ BUG: stack guard page was hit at ffffc90001750000 (stack is ffffc9000174c000..ffffc9000174ffff) kernel stack overflow (page fault): 0000 [#1] SMP DEBUG_PAGEALLOC task: ffff8803332d2ec0 task.stack: ffffc9000174c000 RIP: 0010:[<ffffffff814cfe72>] [<ffffffff814cfe72>] __memcpy+0x12/0x20 RSP: 0018:ffffc9000174f9a8 EFLAGS: 00010246 RAX: ffffc9000174fab8 RBX: 0000000000000000 RCX: 000000001fffff56 RDX: 0000000000000000 RSI: ffff8803231f5a08 RDI: ffffc90001750000 RBP: ffffc9000174fa88 R08: ffffc9000174fab0 R09: ffff8803231f54b8 R10: 0000000000000008 R11: 0000000000000001 R12: 0000000000000000 R13: 0000000000000000 R14: 0000000000000003 R15: ffff8803231f54a0 FS: 00007f3a611af640(0000) GS:ffff88033ed00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: ffffc90001750000 CR3: 0000000325b20000 CR4: 00000000000406e0 Stack: ffffffffa00bc60d 0000000000000008 ffffc90000000001 ffffc9000174faac 0000000000000292 ffffffffa00c24e4 ffffffffa00c2914 0000000000000000 0000000000000000 ffffffff00000003 ffff880331ae8ad0 0000000800000246 Call Trace: [<ffffffffa00bc60d>] ? acpi_nfit_ctl+0x49d/0x750 [nfit] [<ffffffffa01f4fe0>] nfit_test_probe+0x670/0xb1b [nfit_test] Cc: <stable@vger.kernel.org> Fixes: 747ffe11b440 ("libnvdimm, tools/testing/nvdimm: fix 'ars_status' output buffer sizing") Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-12-06 17:10:12 +00:00
(u32 *) out_obj->buffer.pointer,
out_obj->buffer.length - offset);
2015-06-08 18:27:06 +00:00
if (offset + out_size > out_obj->buffer.length) {
dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
dimm_name, cmd_name, i);
2015-06-08 18:27:06 +00:00
break;
}
if (in_buf.buffer.length + offset + out_size > buf_len) {
dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
dimm_name, cmd_name, i);
2015-06-08 18:27:06 +00:00
rc = -ENXIO;
goto out;
}
memcpy(buf + in_buf.buffer.length + offset,
out_obj->buffer.pointer + offset, out_size);
offset += out_size;
}
/*
* Set fw_status for all the commands with a known format to be
* later interpreted by xlat_status().
*/
if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
&& cmd <= ND_CMD_CLEAR_ERROR)
|| (nvdimm && cmd >= ND_CMD_SMART
&& cmd <= ND_CMD_VENDOR)))
fw_status = *(u32 *) out_obj->buffer.pointer;
2015-06-08 18:27:06 +00:00
if (offset + in_buf.buffer.length < buf_len) {
if (i >= 1) {
/*
* status valid, return the number of bytes left
* unfilled in the output buffer
*/
rc = buf_len - offset - in_buf.buffer.length;
if (cmd_rc)
*cmd_rc = xlat_status(nvdimm, buf, cmd,
fw_status);
2015-06-08 18:27:06 +00:00
} else {
dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
__func__, dimm_name, cmd_name, buf_len,
offset);
rc = -ENXIO;
}
} else {
2015-06-08 18:27:06 +00:00
rc = 0;
if (cmd_rc)
*cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
}
2015-06-08 18:27:06 +00:00
out:
ACPI_FREE(out_obj);
return rc;
}
EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
static const char *spa_type_name(u16 type)
{
static const char *to_name[] = {
[NFIT_SPA_VOLATILE] = "volatile",
[NFIT_SPA_PM] = "pmem",
[NFIT_SPA_DCR] = "dimm-control-region",
[NFIT_SPA_BDW] = "block-data-window",
[NFIT_SPA_VDISK] = "volatile-disk",
[NFIT_SPA_VCD] = "volatile-cd",
[NFIT_SPA_PDISK] = "persistent-disk",
[NFIT_SPA_PCD] = "persistent-cd",
};
if (type > NFIT_SPA_PCD)
return "unknown";
return to_name[type];
}
int nfit_spa_type(struct acpi_nfit_system_address *spa)
{
int i;
for (i = 0; i < NFIT_UUID_MAX; i++)
if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
return i;
return -1;
}
static bool add_spa(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev,
struct acpi_nfit_system_address *spa)
{
struct device *dev = acpi_desc->dev;
struct nfit_spa *nfit_spa;
if (spa->header.length != sizeof(*spa))
return false;
list_for_each_entry(nfit_spa, &prev->spas, list) {
if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
list_move_tail(&nfit_spa->list, &acpi_desc->spas);
return true;
}
}
nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
GFP_KERNEL);
if (!nfit_spa)
return false;
INIT_LIST_HEAD(&nfit_spa->list);
memcpy(nfit_spa->spa, spa, sizeof(*spa));
list_add_tail(&nfit_spa->list, &acpi_desc->spas);
dev_dbg(dev, "spa index: %d type: %s\n",
spa->range_index,
spa_type_name(nfit_spa_type(spa)));
return true;
}
static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev,
struct acpi_nfit_memory_map *memdev)
{
struct device *dev = acpi_desc->dev;
struct nfit_memdev *nfit_memdev;
if (memdev->header.length != sizeof(*memdev))
return false;
list_for_each_entry(nfit_memdev, &prev->memdevs, list)
if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
return true;
}
nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
GFP_KERNEL);
if (!nfit_memdev)
return false;
INIT_LIST_HEAD(&nfit_memdev->list);
memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
memdev->device_handle, memdev->range_index,
memdev->region_index, memdev->flags);
return true;
}
int nfit_get_smbios_id(u32 device_handle, u16 *flags)
{
struct acpi_nfit_memory_map *memdev;
struct acpi_nfit_desc *acpi_desc;
struct nfit_mem *nfit_mem;
mutex_lock(&acpi_desc_lock);
list_for_each_entry(acpi_desc, &acpi_descs, list) {
mutex_lock(&acpi_desc->init_mutex);
list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
memdev = __to_nfit_memdev(nfit_mem);
if (memdev->device_handle == device_handle) {
mutex_unlock(&acpi_desc->init_mutex);
mutex_unlock(&acpi_desc_lock);
*flags = memdev->flags;
return memdev->physical_id;
}
}
mutex_unlock(&acpi_desc->init_mutex);
}
mutex_unlock(&acpi_desc_lock);
return -ENODEV;
}
EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
/*
* An implementation may provide a truncated control region if no block windows
* are defined.
*/
static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
{
if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
window_size))
return 0;
if (dcr->windows)
return sizeof(*dcr);
return offsetof(struct acpi_nfit_control_region, window_size);
}
static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev,
struct acpi_nfit_control_region *dcr)
{
struct device *dev = acpi_desc->dev;
struct nfit_dcr *nfit_dcr;
if (!sizeof_dcr(dcr))
return false;
list_for_each_entry(nfit_dcr, &prev->dcrs, list)
if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
return true;
}
nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
GFP_KERNEL);
if (!nfit_dcr)
return false;
INIT_LIST_HEAD(&nfit_dcr->list);
memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
dev_dbg(dev, "dcr index: %d windows: %d\n",
dcr->region_index, dcr->windows);
return true;
}
static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev,
struct acpi_nfit_data_region *bdw)
{
struct device *dev = acpi_desc->dev;
struct nfit_bdw *nfit_bdw;
if (bdw->header.length != sizeof(*bdw))
return false;
list_for_each_entry(nfit_bdw, &prev->bdws, list)
if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
return true;
}
nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
GFP_KERNEL);
if (!nfit_bdw)
return false;
INIT_LIST_HEAD(&nfit_bdw->list);
memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
dev_dbg(dev, "bdw dcr: %d windows: %d\n",
bdw->region_index, bdw->windows);
return true;
}
static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
{
if (idt->header.length < sizeof(*idt))
return 0;
return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
}
static bool add_idt(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev,
struct acpi_nfit_interleave *idt)
{
struct device *dev = acpi_desc->dev;
struct nfit_idt *nfit_idt;
if (!sizeof_idt(idt))
return false;
list_for_each_entry(nfit_idt, &prev->idts, list) {
if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
continue;
if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
list_move_tail(&nfit_idt->list, &acpi_desc->idts);
return true;
}
}
nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
GFP_KERNEL);
if (!nfit_idt)
return false;
INIT_LIST_HEAD(&nfit_idt->list);
memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
list_add_tail(&nfit_idt->list, &acpi_desc->idts);
dev_dbg(dev, "idt index: %d num_lines: %d\n",
idt->interleave_index, idt->line_count);
return true;
}
static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
{
if (flush->header.length < sizeof(*flush))
return 0;
return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
}
static bool add_flush(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev,
struct acpi_nfit_flush_address *flush)
{
struct device *dev = acpi_desc->dev;
struct nfit_flush *nfit_flush;
if (!sizeof_flush(flush))
return false;
list_for_each_entry(nfit_flush, &prev->flushes, list) {
if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
continue;
if (memcmp(nfit_flush->flush, flush,
sizeof_flush(flush)) == 0) {
list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
return true;
}
}
nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
+ sizeof_flush(flush), GFP_KERNEL);
if (!nfit_flush)
return false;
INIT_LIST_HEAD(&nfit_flush->list);
memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
flush->device_handle, flush->hint_count);
return true;
}
static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
struct acpi_nfit_capabilities *pcap)
{
struct device *dev = acpi_desc->dev;
u32 mask;
mask = (1 << (pcap->highest_capability + 1)) - 1;
acpi_desc->platform_cap = pcap->capabilities & mask;
dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
return true;
}
static void *add_table(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev, void *table, const void *end)
{
struct device *dev = acpi_desc->dev;
struct acpi_nfit_header *hdr;
void *err = ERR_PTR(-ENOMEM);
if (table >= end)
return NULL;
hdr = table;
if (!hdr->length) {
dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
hdr->type);
return NULL;
}
switch (hdr->type) {
case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
if (!add_spa(acpi_desc, prev, table))
return err;
break;
case ACPI_NFIT_TYPE_MEMORY_MAP:
if (!add_memdev(acpi_desc, prev, table))
return err;
break;
case ACPI_NFIT_TYPE_CONTROL_REGION:
if (!add_dcr(acpi_desc, prev, table))
return err;
break;
case ACPI_NFIT_TYPE_DATA_REGION:
if (!add_bdw(acpi_desc, prev, table))
return err;
break;
case ACPI_NFIT_TYPE_INTERLEAVE:
if (!add_idt(acpi_desc, prev, table))
return err;
break;
case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
if (!add_flush(acpi_desc, prev, table))
return err;
break;
case ACPI_NFIT_TYPE_SMBIOS:
dev_dbg(dev, "smbios\n");
break;
case ACPI_NFIT_TYPE_CAPABILITIES:
if (!add_platform_cap(acpi_desc, table))
return err;
break;
default:
dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
break;
}
return table + hdr->length;
}
static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
struct nfit_mem *nfit_mem)
{
u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
u16 dcr = nfit_mem->dcr->region_index;
struct nfit_spa *nfit_spa;
list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
u16 range_index = nfit_spa->spa->range_index;
int type = nfit_spa_type(nfit_spa->spa);
struct nfit_memdev *nfit_memdev;
if (type != NFIT_SPA_BDW)
continue;
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
if (nfit_memdev->memdev->range_index != range_index)
continue;
if (nfit_memdev->memdev->device_handle != device_handle)
continue;
if (nfit_memdev->memdev->region_index != dcr)
continue;
nfit_mem->spa_bdw = nfit_spa->spa;
return;
}
}
dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
nfit_mem->spa_dcr->range_index);
nfit_mem->bdw = NULL;
}
static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
{
u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
struct nfit_memdev *nfit_memdev;
struct nfit_bdw *nfit_bdw;
struct nfit_idt *nfit_idt;
u16 idt_idx, range_index;
list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
if (nfit_bdw->bdw->region_index != dcr)
continue;
nfit_mem->bdw = nfit_bdw->bdw;
break;
}
if (!nfit_mem->bdw)
return;
nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
if (!nfit_mem->spa_bdw)
return;
range_index = nfit_mem->spa_bdw->range_index;
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
if (nfit_memdev->memdev->range_index != range_index ||
nfit_memdev->memdev->region_index != dcr)
continue;
nfit_mem->memdev_bdw = nfit_memdev->memdev;
idt_idx = nfit_memdev->memdev->interleave_index;
list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
if (nfit_idt->idt->interleave_index != idt_idx)
continue;
nfit_mem->idt_bdw = nfit_idt->idt;
break;
}
break;
}
}
static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
struct acpi_nfit_system_address *spa)
{
struct nfit_mem *nfit_mem, *found;
struct nfit_memdev *nfit_memdev;
int type = spa ? nfit_spa_type(spa) : 0;
switch (type) {
case NFIT_SPA_DCR:
case NFIT_SPA_PM:
break;
default:
if (spa)
return 0;
}
/*
* This loop runs in two modes, when a dimm is mapped the loop
* adds memdev associations to an existing dimm, or creates a
* dimm. In the unmapped dimm case this loop sweeps for memdev
* instances with an invalid / zero range_index and adds those
* dimms without spa associations.
*/
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
struct nfit_flush *nfit_flush;
struct nfit_dcr *nfit_dcr;
u32 device_handle;
u16 dcr;
if (spa && nfit_memdev->memdev->range_index != spa->range_index)
continue;
if (!spa && nfit_memdev->memdev->range_index)
continue;
found = NULL;
dcr = nfit_memdev->memdev->region_index;
device_handle = nfit_memdev->memdev->device_handle;
list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
if (__to_nfit_memdev(nfit_mem)->device_handle
== device_handle) {
found = nfit_mem;
break;
}
if (found)
nfit_mem = found;
else {
nfit_mem = devm_kzalloc(acpi_desc->dev,
sizeof(*nfit_mem), GFP_KERNEL);
if (!nfit_mem)
return -ENOMEM;
INIT_LIST_HEAD(&nfit_mem->list);
nfit_mem->acpi_desc = acpi_desc;
list_add(&nfit_mem->list, &acpi_desc->dimms);
}
list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
if (nfit_dcr->dcr->region_index != dcr)
continue;
/*
* Record the control region for the dimm. For
* the ACPI 6.1 case, where there are separate
* control regions for the pmem vs blk
* interfaces, be sure to record the extended
* blk details.
*/
if (!nfit_mem->dcr)
nfit_mem->dcr = nfit_dcr->dcr;
else if (nfit_mem->dcr->windows == 0
&& nfit_dcr->dcr->windows)
nfit_mem->dcr = nfit_dcr->dcr;
break;
}
list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
struct acpi_nfit_flush_address *flush;
u16 i;
if (nfit_flush->flush->device_handle != device_handle)
continue;
nfit_mem->nfit_flush = nfit_flush;
flush = nfit_flush->flush;
treewide: devm_kzalloc() -> devm_kcalloc() The devm_kzalloc() function has a 2-factor argument form, devm_kcalloc(). This patch replaces cases of: devm_kzalloc(handle, a * b, gfp) with: devm_kcalloc(handle, a * b, gfp) as well as handling cases of: devm_kzalloc(handle, a * b * c, gfp) with: devm_kzalloc(handle, array3_size(a, b, c), gfp) as it's slightly less ugly than: devm_kcalloc(handle, array_size(a, b), c, gfp) This does, however, attempt to ignore constant size factors like: devm_kzalloc(handle, 4 * 1024, gfp) though any constants defined via macros get caught up in the conversion. Any factors with a sizeof() of "unsigned char", "char", and "u8" were dropped, since they're redundant. Some manual whitespace fixes were needed in this patch, as Coccinelle really liked to write "=devm_kcalloc..." instead of "= devm_kcalloc...". The Coccinelle script used for this was: // Fix redundant parens around sizeof(). @@ expression HANDLE; type TYPE; expression THING, E; @@ ( devm_kzalloc(HANDLE, - (sizeof(TYPE)) * E + sizeof(TYPE) * E , ...) | devm_kzalloc(HANDLE, - (sizeof(THING)) * E + sizeof(THING) * E , ...) ) // Drop single-byte sizes and redundant parens. @@ expression HANDLE; expression COUNT; typedef u8; typedef __u8; @@ ( devm_kzalloc(HANDLE, - sizeof(u8) * (COUNT) + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(__u8) * (COUNT) + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(char) * (COUNT) + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(unsigned char) * (COUNT) + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(u8) * COUNT + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(__u8) * COUNT + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(char) * COUNT + COUNT , ...) | devm_kzalloc(HANDLE, - sizeof(unsigned char) * COUNT + COUNT , ...) ) // 2-factor product with sizeof(type/expression) and identifier or constant. @@ expression HANDLE; type TYPE; expression THING; identifier COUNT_ID; constant COUNT_CONST; @@ ( - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(TYPE) * (COUNT_ID) + COUNT_ID, sizeof(TYPE) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(TYPE) * COUNT_ID + COUNT_ID, sizeof(TYPE) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(TYPE) * (COUNT_CONST) + COUNT_CONST, sizeof(TYPE) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(TYPE) * COUNT_CONST + COUNT_CONST, sizeof(TYPE) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(THING) * (COUNT_ID) + COUNT_ID, sizeof(THING) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(THING) * COUNT_ID + COUNT_ID, sizeof(THING) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(THING) * (COUNT_CONST) + COUNT_CONST, sizeof(THING) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(THING) * COUNT_CONST + COUNT_CONST, sizeof(THING) , ...) ) // 2-factor product, only identifiers. @@ expression HANDLE; identifier SIZE, COUNT; @@ - devm_kzalloc + devm_kcalloc (HANDLE, - SIZE * COUNT + COUNT, SIZE , ...) // 3-factor product with 1 sizeof(type) or sizeof(expression), with // redundant parens removed. @@ expression HANDLE; expression THING; identifier STRIDE, COUNT; type TYPE; @@ ( devm_kzalloc(HANDLE, - sizeof(TYPE) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | devm_kzalloc(HANDLE, - sizeof(TYPE) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | devm_kzalloc(HANDLE, - sizeof(TYPE) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | devm_kzalloc(HANDLE, - sizeof(TYPE) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | devm_kzalloc(HANDLE, - sizeof(THING) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | devm_kzalloc(HANDLE, - sizeof(THING) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | devm_kzalloc(HANDLE, - sizeof(THING) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | devm_kzalloc(HANDLE, - sizeof(THING) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) ) // 3-factor product with 2 sizeof(variable), with redundant parens removed. @@ expression HANDLE; expression THING1, THING2; identifier COUNT; type TYPE1, TYPE2; @@ ( devm_kzalloc(HANDLE, - sizeof(TYPE1) * sizeof(TYPE2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | devm_kzalloc(HANDLE, - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | devm_kzalloc(HANDLE, - sizeof(THING1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | devm_kzalloc(HANDLE, - sizeof(THING1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | devm_kzalloc(HANDLE, - sizeof(TYPE1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) | devm_kzalloc(HANDLE, - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) ) // 3-factor product, only identifiers, with redundant parens removed. @@ expression HANDLE; identifier STRIDE, SIZE, COUNT; @@ ( devm_kzalloc(HANDLE, - (COUNT) * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - COUNT * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - COUNT * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - (COUNT) * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - COUNT * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - (COUNT) * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - (COUNT) * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | devm_kzalloc(HANDLE, - COUNT * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) ) // Any remaining multi-factor products, first at least 3-factor products, // when they're not all constants... @@ expression HANDLE; expression E1, E2, E3; constant C1, C2, C3; @@ ( devm_kzalloc(HANDLE, C1 * C2 * C3, ...) | devm_kzalloc(HANDLE, - (E1) * E2 * E3 + array3_size(E1, E2, E3) , ...) | devm_kzalloc(HANDLE, - (E1) * (E2) * E3 + array3_size(E1, E2, E3) , ...) | devm_kzalloc(HANDLE, - (E1) * (E2) * (E3) + array3_size(E1, E2, E3) , ...) | devm_kzalloc(HANDLE, - E1 * E2 * E3 + array3_size(E1, E2, E3) , ...) ) // And then all remaining 2 factors products when they're not all constants, // keeping sizeof() as the second factor argument. @@ expression HANDLE; expression THING, E1, E2; type TYPE; constant C1, C2, C3; @@ ( devm_kzalloc(HANDLE, sizeof(THING) * C2, ...) | devm_kzalloc(HANDLE, sizeof(TYPE) * C2, ...) | devm_kzalloc(HANDLE, C1 * C2 * C3, ...) | devm_kzalloc(HANDLE, C1 * C2, ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(TYPE) * (E2) + E2, sizeof(TYPE) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(TYPE) * E2 + E2, sizeof(TYPE) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(THING) * (E2) + E2, sizeof(THING) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - sizeof(THING) * E2 + E2, sizeof(THING) , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - (E1) * E2 + E1, E2 , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - (E1) * (E2) + E1, E2 , ...) | - devm_kzalloc + devm_kcalloc (HANDLE, - E1 * E2 + E1, E2 , ...) ) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-12 21:07:58 +00:00
nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
flush->hint_count,
sizeof(struct resource),
GFP_KERNEL);
if (!nfit_mem->flush_wpq)
return -ENOMEM;
for (i = 0; i < flush->hint_count; i++) {
struct resource *res = &nfit_mem->flush_wpq[i];
res->start = flush->hint_address[i];
res->end = res->start + 8 - 1;
}
break;
}
if (dcr && !nfit_mem->dcr) {
dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
spa->range_index, dcr);
return -ENODEV;
}
if (type == NFIT_SPA_DCR) {
struct nfit_idt *nfit_idt;
u16 idt_idx;
/* multiple dimms may share a SPA when interleaved */
nfit_mem->spa_dcr = spa;
nfit_mem->memdev_dcr = nfit_memdev->memdev;
idt_idx = nfit_memdev->memdev->interleave_index;
list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
if (nfit_idt->idt->interleave_index != idt_idx)
continue;
nfit_mem->idt_dcr = nfit_idt->idt;
break;
}
nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
} else if (type == NFIT_SPA_PM) {
/*
* A single dimm may belong to multiple SPA-PM
* ranges, record at least one in addition to
* any SPA-DCR range.
*/
nfit_mem->memdev_pmem = nfit_memdev->memdev;
} else
nfit_mem->memdev_dcr = nfit_memdev->memdev;
}
return 0;
}
static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
{
struct nfit_mem *a = container_of(_a, typeof(*a), list);
struct nfit_mem *b = container_of(_b, typeof(*b), list);
u32 handleA, handleB;
handleA = __to_nfit_memdev(a)->device_handle;
handleB = __to_nfit_memdev(b)->device_handle;
if (handleA < handleB)
return -1;
else if (handleA > handleB)
return 1;
return 0;
}
static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
{
struct nfit_spa *nfit_spa;
int rc;
/*
* For each SPA-DCR or SPA-PMEM address range find its
* corresponding MEMDEV(s). From each MEMDEV find the
* corresponding DCR. Then, if we're operating on a SPA-DCR,
* try to find a SPA-BDW and a corresponding BDW that references
* the DCR. Throw it all into an nfit_mem object. Note, that
* BDWs are optional.
*/
list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
if (rc)
return rc;
}
/*
* If a DIMM has failed to be mapped into SPA there will be no
* SPA entries above. Find and register all the unmapped DIMMs
* for reporting and recovery purposes.
*/
rc = __nfit_mem_init(acpi_desc, NULL);
if (rc)
return rc;
list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
return 0;
}
static ssize_t bus_dsm_mask_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
}
static struct device_attribute dev_attr_bus_dsm_mask =
__ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
static ssize_t revision_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
}
static DEVICE_ATTR_RO(revision);
static ssize_t hw_error_scrub_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
}
/*
* The 'hw_error_scrub' attribute can have the following values written to it:
* '0': Switch to the default mode where an exception will only insert
* the address of the memory error into the poison and badblocks lists.
* '1': Enable a full scrub to happen if an exception for a memory error is
* received.
*/
static ssize_t hw_error_scrub_store(struct device *dev,
struct device_attribute *attr, const char *buf, size_t size)
{
struct nvdimm_bus_descriptor *nd_desc;
ssize_t rc;
long val;
rc = kstrtol(buf, 0, &val);
if (rc)
return rc;
device_lock(dev);
nd_desc = dev_get_drvdata(dev);
if (nd_desc) {
struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
switch (val) {
case HW_ERROR_SCRUB_ON:
acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
break;
case HW_ERROR_SCRUB_OFF:
acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
break;
default:
rc = -EINVAL;
break;
}
}
device_unlock(dev);
if (rc)
return rc;
return size;
}
static DEVICE_ATTR_RW(hw_error_scrub);
/*
* This shows the number of full Address Range Scrubs that have been
* completed since driver load time. Userspace can wait on this using
* select/poll etc. A '+' at the end indicates an ARS is in progress
*/
static ssize_t scrub_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm_bus_descriptor *nd_desc;
ssize_t rc = -ENXIO;
device_lock(dev);
nd_desc = dev_get_drvdata(dev);
if (nd_desc) {
struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
mutex_lock(&acpi_desc->init_mutex);
rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
acpi_desc->scrub_busy
&& !acpi_desc->cancel ? "+\n" : "\n");
mutex_unlock(&acpi_desc->init_mutex);
}
device_unlock(dev);
return rc;
}
static ssize_t scrub_store(struct device *dev,
struct device_attribute *attr, const char *buf, size_t size)
{
struct nvdimm_bus_descriptor *nd_desc;
ssize_t rc;
long val;
rc = kstrtol(buf, 0, &val);
if (rc)
return rc;
if (val != 1)
return -EINVAL;
device_lock(dev);
nd_desc = dev_get_drvdata(dev);
if (nd_desc) {
struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
rc = acpi_nfit_ars_rescan(acpi_desc, 0);
}
device_unlock(dev);
if (rc)
return rc;
return size;
}
static DEVICE_ATTR_RW(scrub);
static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
{
struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
| 1 << ND_CMD_ARS_STATUS;
return (nd_desc->cmd_mask & mask) == mask;
}
static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
{
struct device *dev = container_of(kobj, struct device, kobj);
struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
return 0;
return a->mode;
}
static struct attribute *acpi_nfit_attributes[] = {
&dev_attr_revision.attr,
&dev_attr_scrub.attr,
&dev_attr_hw_error_scrub.attr,
&dev_attr_bus_dsm_mask.attr,
NULL,
};
static const struct attribute_group acpi_nfit_attribute_group = {
.name = "nfit",
.attrs = acpi_nfit_attributes,
.is_visible = nfit_visible,
};
static const struct attribute_group *acpi_nfit_attribute_groups[] = {
&nvdimm_bus_attribute_group,
&acpi_nfit_attribute_group,
NULL,
};
static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
return __to_nfit_memdev(nfit_mem);
}
static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
return nfit_mem->dcr;
}
static ssize_t handle_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
return sprintf(buf, "%#x\n", memdev->device_handle);
}
static DEVICE_ATTR_RO(handle);
static ssize_t phys_id_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
return sprintf(buf, "%#x\n", memdev->physical_id);
}
static DEVICE_ATTR_RO(phys_id);
static ssize_t vendor_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
}
static DEVICE_ATTR_RO(vendor);
static ssize_t rev_id_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
}
static DEVICE_ATTR_RO(rev_id);
static ssize_t device_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
}
static DEVICE_ATTR_RO(device);
static ssize_t subsystem_vendor_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
}
static DEVICE_ATTR_RO(subsystem_vendor);
static ssize_t subsystem_rev_id_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n",
be16_to_cpu(dcr->subsystem_revision_id));
}
static DEVICE_ATTR_RO(subsystem_rev_id);
static ssize_t subsystem_device_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
}
static DEVICE_ATTR_RO(subsystem_device);
static int num_nvdimm_formats(struct nvdimm *nvdimm)
{
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
int formats = 0;
if (nfit_mem->memdev_pmem)
formats++;
if (nfit_mem->memdev_bdw)
formats++;
return formats;
}
static ssize_t format_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
}
static DEVICE_ATTR_RO(format);
static ssize_t format1_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
u32 handle;
ssize_t rc = -ENXIO;
struct nfit_mem *nfit_mem;
struct nfit_memdev *nfit_memdev;
struct acpi_nfit_desc *acpi_desc;
struct nvdimm *nvdimm = to_nvdimm(dev);
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
nfit_mem = nvdimm_provider_data(nvdimm);
acpi_desc = nfit_mem->acpi_desc;
handle = to_nfit_memdev(dev)->device_handle;
/* assumes DIMMs have at most 2 published interface codes */
mutex_lock(&acpi_desc->init_mutex);
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
struct nfit_dcr *nfit_dcr;
if (memdev->device_handle != handle)
continue;
list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
if (nfit_dcr->dcr->region_index != memdev->region_index)
continue;
if (nfit_dcr->dcr->code == dcr->code)
continue;
rc = sprintf(buf, "0x%04x\n",
le16_to_cpu(nfit_dcr->dcr->code));
break;
}
if (rc != ENXIO)
break;
}
mutex_unlock(&acpi_desc->init_mutex);
return rc;
}
static DEVICE_ATTR_RO(format1);
static ssize_t formats_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
}
static DEVICE_ATTR_RO(formats);
static ssize_t serial_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
}
static DEVICE_ATTR_RO(serial);
static ssize_t family_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
if (nfit_mem->family < 0)
return -ENXIO;
return sprintf(buf, "%d\n", nfit_mem->family);
}
static DEVICE_ATTR_RO(family);
static ssize_t dsm_mask_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
if (nfit_mem->family < 0)
return -ENXIO;
return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
}
static DEVICE_ATTR_RO(dsm_mask);
static ssize_t flags_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
u16 flags = __to_nfit_memdev(nfit_mem)->flags;
if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
return sprintf(buf, "%s%s%s%s%s%s%s\n",
2015-08-26 16:20:23 +00:00
flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
}
static DEVICE_ATTR_RO(flags);
static ssize_t id_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
return sprintf(buf, "%04x-%02x-%04x-%08x\n",
be16_to_cpu(dcr->vendor_id),
dcr->manufacturing_location,
be16_to_cpu(dcr->manufacturing_date),
be32_to_cpu(dcr->serial_number));
else
return sprintf(buf, "%04x-%08x\n",
be16_to_cpu(dcr->vendor_id),
be32_to_cpu(dcr->serial_number));
}
static DEVICE_ATTR_RO(id);
static ssize_t dirty_shutdown_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
}
static DEVICE_ATTR_RO(dirty_shutdown);
static struct attribute *acpi_nfit_dimm_attributes[] = {
&dev_attr_handle.attr,
&dev_attr_phys_id.attr,
&dev_attr_vendor.attr,
&dev_attr_device.attr,
&dev_attr_rev_id.attr,
&dev_attr_subsystem_vendor.attr,
&dev_attr_subsystem_device.attr,
&dev_attr_subsystem_rev_id.attr,
&dev_attr_format.attr,
&dev_attr_formats.attr,
&dev_attr_format1.attr,
&dev_attr_serial.attr,
&dev_attr_flags.attr,
&dev_attr_id.attr,
&dev_attr_family.attr,
&dev_attr_dsm_mask.attr,
&dev_attr_dirty_shutdown.attr,
NULL,
};
static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
struct attribute *a, int n)
{
struct device *dev = container_of(kobj, struct device, kobj);
struct nvdimm *nvdimm = to_nvdimm(dev);
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
if (!to_nfit_dcr(dev)) {
/* Without a dcr only the memdev attributes can be surfaced */
if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
|| a == &dev_attr_flags.attr
|| a == &dev_attr_family.attr
|| a == &dev_attr_dsm_mask.attr)
return a->mode;
return 0;
}
if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
return 0;
if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
&& a == &dev_attr_dirty_shutdown.attr)
return 0;
return a->mode;
}
static const struct attribute_group acpi_nfit_dimm_attribute_group = {
.name = "nfit",
.attrs = acpi_nfit_dimm_attributes,
.is_visible = acpi_nfit_dimm_attr_visible,
};
static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
2015-06-08 18:27:06 +00:00
&nvdimm_attribute_group,
&nd_device_attribute_group,
&acpi_nfit_dimm_attribute_group,
NULL,
};
static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
u32 device_handle)
{
struct nfit_mem *nfit_mem;
list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
return nfit_mem->nvdimm;
return NULL;
}
void __acpi_nvdimm_notify(struct device *dev, u32 event)
{
struct nfit_mem *nfit_mem;
struct acpi_nfit_desc *acpi_desc;
dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
event);
if (event != NFIT_NOTIFY_DIMM_HEALTH) {
dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
event);
return;
}
acpi_desc = dev_get_drvdata(dev->parent);
if (!acpi_desc)
return;
/*
* If we successfully retrieved acpi_desc, then we know nfit_mem data
* is still valid.
*/
nfit_mem = dev_get_drvdata(dev);
if (nfit_mem && nfit_mem->flags_attr)
sysfs_notify_dirent(nfit_mem->flags_attr);
}
EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
{
struct acpi_device *adev = data;
struct device *dev = &adev->dev;
device_lock(dev->parent);
__acpi_nvdimm_notify(dev, event);
device_unlock(dev->parent);
}
static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
{
acpi_handle handle;
acpi_status status;
status = acpi_get_handle(adev->handle, method, &handle);
if (ACPI_SUCCESS(status))
return true;
return false;
}
__weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
{
struct nd_intel_smart smart = { 0 };
union acpi_object in_buf = {
.type = ACPI_TYPE_BUFFER,
.buffer.pointer = (char *) &smart,
.buffer.length = sizeof(smart),
};
union acpi_object in_obj = {
.type = ACPI_TYPE_PACKAGE,
.package.count = 1,
.package.elements = &in_buf,
};
const u8 func = ND_INTEL_SMART;
const guid_t *guid = to_nfit_uuid(nfit_mem->family);
u8 revid = nfit_dsm_revid(nfit_mem->family, func);
struct acpi_device *adev = nfit_mem->adev;
acpi_handle handle = adev->handle;
union acpi_object *out_obj;
if ((nfit_mem->dsm_mask & (1 << func)) == 0)
return;
out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
if (!out_obj)
return;
if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
if (smart.shutdown_state)
set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
}
if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
nfit_mem->dirty_shutdown = smart.shutdown_count;
}
ACPI_FREE(out_obj);
}
static void populate_shutdown_status(struct nfit_mem *nfit_mem)
{
/*
* For DIMMs that provide a dynamic facility to retrieve a
* dirty-shutdown status and/or a dirty-shutdown count, cache
* these values in nfit_mem.
*/
if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
nfit_intel_shutdown_status(nfit_mem);
}
2015-06-08 18:27:06 +00:00
static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
struct nfit_mem *nfit_mem, u32 device_handle)
{
struct acpi_device *adev, *adev_dimm;
struct device *dev = acpi_desc->dev;
unsigned long dsm_mask, label_mask;
const guid_t *guid;
int i;
int family = -1;
2015-06-08 18:27:06 +00:00
/* nfit test assumes 1:1 relationship between commands and dsms */
nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
nfit_mem->family = NVDIMM_FAMILY_INTEL;
2015-06-08 18:27:06 +00:00
adev = to_acpi_dev(acpi_desc);
if (!adev) {
/* unit test case */
populate_shutdown_status(nfit_mem);
2015-06-08 18:27:06 +00:00
return 0;
}
2015-06-08 18:27:06 +00:00
adev_dimm = acpi_find_child_device(adev, device_handle, false);
nfit_mem->adev = adev_dimm;
if (!adev_dimm) {
dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
device_handle);
return force_enable_dimms ? 0 : -ENODEV;
2015-06-08 18:27:06 +00:00
}
if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
dev_err(dev, "%s: notification registration failed\n",
dev_name(&adev_dimm->dev));
return -ENXIO;
}
/*
* Record nfit_mem for the notification path to track back to
* the nfit sysfs attributes for this dimm device object.
*/
dev_set_drvdata(&adev_dimm->dev, nfit_mem);
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
/*
* Until standardization materializes we need to consider 4
* different command sets. Note, that checking for function0 (bit0)
* tells us if any commands are reachable through this GUID.
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
*/
for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
if (family < 0 || i == default_dsm_family)
family = i;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
/* limit the supported commands to those that are publicly documented */
nfit_mem->family = family;
if (override_dsm_mask && !disable_vendor_specific)
dsm_mask = override_dsm_mask;
else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
dsm_mask = NVDIMM_INTEL_CMDMASK;
if (disable_vendor_specific)
dsm_mask &= ~(1 << ND_CMD_VENDOR);
} else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
dsm_mask = 0x1c3c76;
} else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
dsm_mask = 0x1fe;
if (disable_vendor_specific)
dsm_mask &= ~(1 << 8);
} else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
dsm_mask = 0xffffffff;
} else {
dev_dbg(dev, "unknown dimm command family\n");
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
nfit_mem->family = -1;
/* DSMs are optional, continue loading the driver... */
return 0;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
}
guid = to_nfit_uuid(nfit_mem->family);
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
if (acpi_check_dsm(adev_dimm->handle, guid,
nfit_dsm_revid(nfit_mem->family, i),
1ULL << i))
2015-06-08 18:27:06 +00:00
set_bit(i, &nfit_mem->dsm_mask);
/*
* Prefer the NVDIMM_FAMILY_INTEL label read commands if present
* due to their better semantics handling locked capacity.
*/
label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
| 1 << ND_CMD_SET_CONFIG_DATA;
if (family == NVDIMM_FAMILY_INTEL
&& (dsm_mask & label_mask) == label_mask)
return 0;
if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
&& acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
}
if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
&& acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
}
populate_shutdown_status(nfit_mem);
return 0;
2015-06-08 18:27:06 +00:00
}
static void shutdown_dimm_notify(void *data)
{
struct acpi_nfit_desc *acpi_desc = data;
struct nfit_mem *nfit_mem;
mutex_lock(&acpi_desc->init_mutex);
/*
* Clear out the nfit_mem->flags_attr and shut down dimm event
* notifications.
*/
list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
struct acpi_device *adev_dimm = nfit_mem->adev;
if (nfit_mem->flags_attr) {
sysfs_put(nfit_mem->flags_attr);
nfit_mem->flags_attr = NULL;
}
if (adev_dimm) {
acpi_remove_notify_handler(adev_dimm->handle,
ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
dev_set_drvdata(&adev_dimm->dev, NULL);
}
}
mutex_unlock(&acpi_desc->init_mutex);
}
static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
{
struct nfit_mem *nfit_mem;
int dimm_count = 0, rc;
struct nvdimm *nvdimm;
list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
struct acpi_nfit_flush_address *flush;
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
unsigned long flags = 0, cmd_mask;
struct nfit_memdev *nfit_memdev;
u32 device_handle;
u16 mem_flags;
device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
if (nvdimm) {
dimm_count++;
continue;
}
if (nfit_mem->bdw && nfit_mem->memdev_pmem)
set_bit(NDD_ALIASING, &flags);
/* collate flags across all memdevs for this dimm */
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
struct acpi_nfit_memory_map *dimm_memdev;
dimm_memdev = __to_nfit_memdev(nfit_mem);
if (dimm_memdev->device_handle
!= nfit_memdev->memdev->device_handle)
continue;
dimm_memdev->flags |= nfit_memdev->memdev->flags;
}
mem_flags = __to_nfit_memdev(nfit_mem)->flags;
if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
set_bit(NDD_UNARMED, &flags);
2015-06-08 18:27:06 +00:00
rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
if (rc)
continue;
/*
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
* TODO: provide translation for non-NVDIMM_FAMILY_INTEL
* devices (i.e. from nd_cmd to acpi_dsm) to standardize the
* userspace interface.
*/
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
cmd_mask = 1UL << ND_CMD_CALL;
if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
/*
* These commands have a 1:1 correspondence
* between DSM payload and libnvdimm ioctl
* payload format.
*/
cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
}
nfit, libnvdimm: limited/whitelisted dimm command marshaling mechanism There are currently 4 known similar but incompatible definitions of the command sets that can be sent to an NVDIMM through ACPI. It is also clear that future platform generations (ACPI or not) will continue to revise and extend the DIMM command set as new devices and use cases arrive. It is obviously untenable to continue to proliferate divergence of these command definitions, and to that end a standardization process has begun to provide for a unified specification. However, that leaves a problem about what to do with this first generation where vendors are already shipping divergence. The Linux kernel can support these initial diverged platforms without giving platform-firmware free reign to continue to diverge and compound kernel maintenance overhead. The kernel implementation can encourage standardization in two ways: 1/ Require that any function code that userspace wants to send be explicitly white-listed in the implementation. For ACPI this means function codes marked as supported by acpi_check_dsm() may only be invoked if they appear in the white-list. A function must be publicly documented before it is added to the white-list. 2/ The above restrictions can be trivially bypassed by using the "vendor-specific" payload command. However, since vendor-specific commands are by definition not publicly documented and have the potential to corrupt the kernel's view of the dimm state, we provide a toggle to disable vendor-specific operations. Enabling undefined behavior is a policy decision that can be made by the platform owner and encourages firmware implementations to choose public over private command implementations. Based on an initial patch from Jerry Hoemann Cc: Jerry Hoemann <jerry.hoemann@hpe.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-04-28 23:23:43 +00:00
if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
}
if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
: NULL;
nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2015-06-08 18:27:06 +00:00
acpi_nfit_dimm_attribute_groups,
flags, cmd_mask, flush ? flush->hint_count : 0,
nfit_mem->flush_wpq);
if (!nvdimm)
return -ENOMEM;
nfit_mem->nvdimm = nvdimm;
dimm_count++;
if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
continue;
dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
nvdimm_name(nvdimm),
2015-08-26 16:20:23 +00:00
mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
}
rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
if (rc)
return rc;
/*
* Now that dimms are successfully registered, and async registration
* is flushed, attempt to enable event notification.
*/
list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
struct kernfs_node *nfit_kernfs;
nvdimm = nfit_mem->nvdimm;
if (!nvdimm)
continue;
nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
if (nfit_kernfs)
nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
"flags");
sysfs_put(nfit_kernfs);
if (!nfit_mem->flags_attr)
dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
nvdimm_name(nvdimm));
}
return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
acpi_desc);
}
/*
* These constants are private because there are no kernel consumers of
* these commands.
*/
enum nfit_aux_cmds {
NFIT_CMD_TRANSLATE_SPA = 5,
NFIT_CMD_ARS_INJECT_SET = 7,
NFIT_CMD_ARS_INJECT_CLEAR = 8,
NFIT_CMD_ARS_INJECT_GET = 9,
};
2015-06-08 18:27:06 +00:00
static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
{
struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2015-06-08 18:27:06 +00:00
struct acpi_device *adev;
unsigned long dsm_mask;
2015-06-08 18:27:06 +00:00
int i;
nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2015-06-08 18:27:06 +00:00
adev = to_acpi_dev(acpi_desc);
if (!adev)
return;
for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
set_bit(i, &nd_desc->cmd_mask);
set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
dsm_mask =
(1 << ND_CMD_ARS_CAP) |
(1 << ND_CMD_ARS_START) |
(1 << ND_CMD_ARS_STATUS) |
(1 << ND_CMD_CLEAR_ERROR) |
(1 << NFIT_CMD_TRANSLATE_SPA) |
(1 << NFIT_CMD_ARS_INJECT_SET) |
(1 << NFIT_CMD_ARS_INJECT_CLEAR) |
(1 << NFIT_CMD_ARS_INJECT_GET);
for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
set_bit(i, &nd_desc->bus_dsm_mask);
2015-06-08 18:27:06 +00:00
}
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
static ssize_t range_index_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct nd_region *nd_region = to_nd_region(dev);
struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
}
static DEVICE_ATTR_RO(range_index);
static struct attribute *acpi_nfit_region_attributes[] = {
&dev_attr_range_index.attr,
NULL,
};
static const struct attribute_group acpi_nfit_region_attribute_group = {
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
.name = "nfit",
.attrs = acpi_nfit_region_attributes,
};
static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
&nd_region_attribute_group,
&nd_mapping_attribute_group,
&nd_device_attribute_group,
&nd_numa_attribute_group,
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
&acpi_nfit_region_attribute_group,
NULL,
};
2015-05-01 17:11:27 +00:00
/* enough info to uniquely specify an interleave set */
struct nfit_set_info {
struct nfit_set_info_map {
u64 region_offset;
u32 serial_number;
u32 pad;
} mapping[0];
};
struct nfit_set_info2 {
struct nfit_set_info_map2 {
u64 region_offset;
u32 serial_number;
u16 vendor_id;
u16 manufacturing_date;
u8 manufacturing_location;
u8 reserved[31];
} mapping[0];
};
2015-05-01 17:11:27 +00:00
static size_t sizeof_nfit_set_info(int num_mappings)
{
return sizeof(struct nfit_set_info)
+ num_mappings * sizeof(struct nfit_set_info_map);
}
static size_t sizeof_nfit_set_info2(int num_mappings)
{
return sizeof(struct nfit_set_info2)
+ num_mappings * sizeof(struct nfit_set_info_map2);
}
2017-03-01 02:32:48 +00:00
static int cmp_map_compat(const void *m0, const void *m1)
2015-05-01 17:11:27 +00:00
{
const struct nfit_set_info_map *map0 = m0;
const struct nfit_set_info_map *map1 = m1;
return memcmp(&map0->region_offset, &map1->region_offset,
sizeof(u64));
}
2017-03-01 02:32:48 +00:00
static int cmp_map(const void *m0, const void *m1)
{
const struct nfit_set_info_map *map0 = m0;
const struct nfit_set_info_map *map1 = m1;
if (map0->region_offset < map1->region_offset)
return -1;
else if (map0->region_offset > map1->region_offset)
return 1;
return 0;
2017-03-01 02:32:48 +00:00
}
static int cmp_map2(const void *m0, const void *m1)
{
const struct nfit_set_info_map2 *map0 = m0;
const struct nfit_set_info_map2 *map1 = m1;
if (map0->region_offset < map1->region_offset)
return -1;
else if (map0->region_offset > map1->region_offset)
return 1;
return 0;
}
2015-05-01 17:11:27 +00:00
/* Retrieve the nth entry referencing this spa */
static struct acpi_nfit_memory_map *memdev_from_spa(
struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
{
struct nfit_memdev *nfit_memdev;
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
if (nfit_memdev->memdev->range_index == range_index)
if (n-- == 0)
return nfit_memdev->memdev;
return NULL;
}
static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
struct nd_region_desc *ndr_desc,
struct acpi_nfit_system_address *spa)
{
struct device *dev = acpi_desc->dev;
struct nd_interleave_set *nd_set;
u16 nr = ndr_desc->num_mappings;
struct nfit_set_info2 *info2;
2015-05-01 17:11:27 +00:00
struct nfit_set_info *info;
int i;
2015-05-01 17:11:27 +00:00
nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
if (!nd_set)
return -ENOMEM;
ndr_desc->nd_set = nd_set;
guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2015-05-01 17:11:27 +00:00
info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
if (!info)
return -ENOMEM;
info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
if (!info2)
return -ENOMEM;
2015-05-01 17:11:27 +00:00
for (i = 0; i < nr; i++) {
struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2015-05-01 17:11:27 +00:00
struct nfit_set_info_map *map = &info->mapping[i];
struct nfit_set_info_map2 *map2 = &info2->mapping[i];
struct nvdimm *nvdimm = mapping->nvdimm;
2015-05-01 17:11:27 +00:00
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
spa->range_index, i);
struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2015-05-01 17:11:27 +00:00
if (!memdev || !nfit_mem->dcr) {
dev_err(dev, "%s: failed to find DCR\n", __func__);
return -ENODEV;
}
map->region_offset = memdev->region_offset;
map->serial_number = dcr->serial_number;
map2->region_offset = memdev->region_offset;
map2->serial_number = dcr->serial_number;
map2->vendor_id = dcr->vendor_id;
map2->manufacturing_date = dcr->manufacturing_date;
map2->manufacturing_location = dcr->manufacturing_location;
2015-05-01 17:11:27 +00:00
}
/* v1.1 namespaces */
2015-05-01 17:11:27 +00:00
sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
cmp_map, NULL);
nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
/* v1.2 namespaces */
sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
cmp_map2, NULL);
nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2017-03-01 02:32:48 +00:00
/* support v1.1 namespaces created with the wrong sort order */
2017-03-01 02:32:48 +00:00
sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
cmp_map_compat, NULL);
nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
/* record the result of the sort for the mapping position */
for (i = 0; i < nr; i++) {
struct nfit_set_info_map2 *map2 = &info2->mapping[i];
int j;
for (j = 0; j < nr; j++) {
struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
struct nvdimm *nvdimm = mapping->nvdimm;
struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
if (map2->serial_number == dcr->serial_number &&
map2->vendor_id == dcr->vendor_id &&
map2->manufacturing_date == dcr->manufacturing_date &&
map2->manufacturing_location
== dcr->manufacturing_location) {
mapping->position = i;
break;
}
}
}
2015-05-01 17:11:27 +00:00
ndr_desc->nd_set = nd_set;
devm_kfree(dev, info);
devm_kfree(dev, info2);
2015-05-01 17:11:27 +00:00
return 0;
}
static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
{
struct acpi_nfit_interleave *idt = mmio->idt;
u32 sub_line_offset, line_index, line_offset;
u64 line_no, table_skip_count, table_offset;
line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
line_offset = idt->line_offset[line_index]
* mmio->line_size;
table_offset = table_skip_count * mmio->table_size;
return mmio->base_offset + line_offset + table_offset + sub_line_offset;
}
static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
{
struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
u64 offset = nfit_blk->stat_offset + mmio->size * bw;
const u32 STATUS_MASK = 0x80000037;
if (mmio->num_lines)
offset = to_interleave_offset(offset, mmio);
return readl(mmio->addr.base + offset) & STATUS_MASK;
}
static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
resource_size_t dpa, unsigned int len, unsigned int write)
{
u64 cmd, offset;
struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
enum {
BCW_OFFSET_MASK = (1ULL << 48)-1,
BCW_LEN_SHIFT = 48,
BCW_LEN_MASK = (1ULL << 8) - 1,
BCW_CMD_SHIFT = 56,
};
cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
len = len >> L1_CACHE_SHIFT;
cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
cmd |= ((u64) write) << BCW_CMD_SHIFT;
offset = nfit_blk->cmd_offset + mmio->size * bw;
if (mmio->num_lines)
offset = to_interleave_offset(offset, mmio);
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
writeq(cmd, mmio->addr.base + offset);
libnvdimm: introduce nvdimm_flush() and nvdimm_has_flush() nvdimm_flush() is a replacement for the x86 'pcommit' instruction. It is an optional write flushing mechanism that an nvdimm bus can provide for the pmem driver to consume. In the case of the NFIT nvdimm-bus-provider nvdimm_flush() is implemented as a series of flush-hint-address [1] writes to each dimm in the interleave set (region) that backs the namespace. The nvdimm_has_flush() routine relies on platform firmware to describe the flushing capabilities of a platform. It uses the heuristic of whether an nvdimm bus provider provides flush address data to return a ternary result: 1: flush addresses defined 0: dimm topology described without flush addresses (assume ADR) -errno: no topology information, unable to determine flush mechanism The pmem driver is expected to take the following actions on this ternary result: 1: nvdimm_flush() in response to REQ_FUA / REQ_FLUSH and shutdown 0: do not set, WC or FUA on the queue, take no further action -errno: warn and then operate as if nvdimm_has_flush() returned '0' The caveat of this heuristic is that it can not distinguish the "dimm does not have flush address" case from the "platform firmware is broken and failed to describe a flush address". Given we are already explicitly trusting the NFIT there's not much more we can do beyond blacklisting broken firmwares if they are ever encountered. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-07-08 02:44:50 +00:00
nvdimm_flush(nfit_blk->nd_region);
if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
readq(mmio->addr.base + offset);
}
static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
resource_size_t dpa, void *iobuf, size_t len, int rw,
unsigned int lane)
{
struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
unsigned int copied = 0;
u64 base_offset;
int rc;
base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
+ lane * mmio->size;
write_blk_ctl(nfit_blk, lane, dpa, len, rw);
while (len) {
unsigned int c;
u64 offset;
if (mmio->num_lines) {
u32 line_offset;
offset = to_interleave_offset(base_offset + copied,
mmio);
div_u64_rem(offset, mmio->line_size, &line_offset);
c = min_t(size_t, len, mmio->line_size - line_offset);
} else {
offset = base_offset + nfit_blk->bdw_offset;
c = len;
}
if (rw)
x86, uaccess: introduce copy_from_iter_flushcache for pmem / cache-bypass operations The pmem driver has a need to transfer data with a persistent memory destination and be able to rely on the fact that the destination writes are not cached. It is sufficient for the writes to be flushed to a cpu-store-buffer (non-temporal / "movnt" in x86 terms), as we expect userspace to call fsync() to ensure data-writes have reached a power-fail-safe zone in the platform. The fsync() triggers a REQ_FUA or REQ_FLUSH to the pmem driver which will turn around and fence previous writes with an "sfence". Implement a __copy_from_user_inatomic_flushcache, memcpy_page_flushcache, and memcpy_flushcache, that guarantee that the destination buffer is not dirty in the cpu cache on completion. The new copy_from_iter_flushcache and sub-routines will be used to replace the "pmem api" (include/linux/pmem.h + arch/x86/include/asm/pmem.h). The availability of copy_from_iter_flushcache() and memcpy_flushcache() are gated by the CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE config symbol, and fallback to copy_from_iter_nocache() and plain memcpy() otherwise. This is meant to satisfy the concern from Linus that if a driver wants to do something beyond the normal nocache semantics it should be something private to that driver [1], and Al's concern that anything uaccess related belongs with the rest of the uaccess code [2]. The first consumer of this interface is a new 'copy_from_iter' dax operation so that pmem can inject cache maintenance operations without imposing this overhead on other dax-capable drivers. [1]: https://lists.01.org/pipermail/linux-nvdimm/2017-January/008364.html [2]: https://lists.01.org/pipermail/linux-nvdimm/2017-April/009942.html Cc: <x86@kernel.org> Cc: Jan Kara <jack@suse.cz> Cc: Jeff Moyer <jmoyer@redhat.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Toshi Kani <toshi.kani@hpe.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Matthew Wilcox <mawilcox@microsoft.com> Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2017-05-29 19:22:50 +00:00
memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
else {
if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
arch_invalidate_pmem((void __force *)
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
mmio->addr.aperture + offset, c);
memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
}
copied += c;
len -= c;
}
if (rw)
libnvdimm: introduce nvdimm_flush() and nvdimm_has_flush() nvdimm_flush() is a replacement for the x86 'pcommit' instruction. It is an optional write flushing mechanism that an nvdimm bus can provide for the pmem driver to consume. In the case of the NFIT nvdimm-bus-provider nvdimm_flush() is implemented as a series of flush-hint-address [1] writes to each dimm in the interleave set (region) that backs the namespace. The nvdimm_has_flush() routine relies on platform firmware to describe the flushing capabilities of a platform. It uses the heuristic of whether an nvdimm bus provider provides flush address data to return a ternary result: 1: flush addresses defined 0: dimm topology described without flush addresses (assume ADR) -errno: no topology information, unable to determine flush mechanism The pmem driver is expected to take the following actions on this ternary result: 1: nvdimm_flush() in response to REQ_FUA / REQ_FLUSH and shutdown 0: do not set, WC or FUA on the queue, take no further action -errno: warn and then operate as if nvdimm_has_flush() returned '0' The caveat of this heuristic is that it can not distinguish the "dimm does not have flush address" case from the "platform firmware is broken and failed to describe a flush address". Given we are already explicitly trusting the NFIT there's not much more we can do beyond blacklisting broken firmwares if they are ever encountered. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-07-08 02:44:50 +00:00
nvdimm_flush(nfit_blk->nd_region);
rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
return rc;
}
static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
resource_size_t dpa, void *iobuf, u64 len, int rw)
{
struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
struct nd_region *nd_region = nfit_blk->nd_region;
unsigned int lane, copied = 0;
int rc = 0;
lane = nd_region_acquire_lane(nd_region);
while (len) {
u64 c = min(len, mmio->size);
rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
iobuf + copied, c, rw, lane);
if (rc)
break;
copied += c;
len -= c;
}
nd_region_release_lane(nd_region, lane);
return rc;
}
static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
struct acpi_nfit_interleave *idt, u16 interleave_ways)
{
if (idt) {
mmio->num_lines = idt->line_count;
mmio->line_size = idt->line_size;
if (interleave_ways == 0)
return -ENXIO;
mmio->table_size = mmio->num_lines * interleave_ways
* mmio->line_size;
}
return 0;
}
static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
{
struct nd_cmd_dimm_flags flags;
int rc;
memset(&flags, 0, sizeof(flags));
rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
sizeof(flags), NULL);
if (rc >= 0 && flags.status == 0)
nfit_blk->dimm_flags = flags.flags;
else if (rc == -ENOTTY) {
/* fall back to a conservative default */
nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
rc = 0;
} else
rc = -ENXIO;
return rc;
}
static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
struct device *dev)
{
struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
struct nd_blk_region *ndbr = to_nd_blk_region(dev);
struct nfit_blk_mmio *mmio;
struct nfit_blk *nfit_blk;
struct nfit_mem *nfit_mem;
struct nvdimm *nvdimm;
int rc;
nvdimm = nd_blk_region_to_dimm(ndbr);
nfit_mem = nvdimm_provider_data(nvdimm);
if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
dev_dbg(dev, "missing%s%s%s\n",
nfit_mem ? "" : " nfit_mem",
(nfit_mem && nfit_mem->dcr) ? "" : " dcr",
(nfit_mem && nfit_mem->bdw) ? "" : " bdw");
return -ENXIO;
}
nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
if (!nfit_blk)
return -ENOMEM;
nd_blk_region_set_provider_data(ndbr, nfit_blk);
nfit_blk->nd_region = to_nd_region(dev);
/* map block aperture memory */
nfit_blk->bdw_offset = nfit_mem->bdw->offset;
mmio = &nfit_blk->mmio[BDW];
mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
if (!mmio->addr.base) {
dev_dbg(dev, "%s failed to map bdw\n",
nvdimm_name(nvdimm));
return -ENOMEM;
}
mmio->size = nfit_mem->bdw->size;
mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
mmio->idt = nfit_mem->idt_bdw;
mmio->spa = nfit_mem->spa_bdw;
rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
nfit_mem->memdev_bdw->interleave_ways);
if (rc) {
dev_dbg(dev, "%s failed to init bdw interleave\n",
nvdimm_name(nvdimm));
return rc;
}
/* map block control memory */
nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
mmio = &nfit_blk->mmio[DCR];
mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
nfit_mem->spa_dcr->length);
nd_blk: change aperture mapping from WC to WB This should result in a pretty sizeable performance gain for reads. For rough comparison I did some simple read testing using PMEM to compare reads of write combining (WC) mappings vs write-back (WB). This was done on a random lab machine. PMEM reads from a write combining mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=100000 100000+0 records in 100000+0 records out 409600000 bytes (410 MB) copied, 9.2855 s, 44.1 MB/s PMEM reads from a write-back mapping: # dd of=/dev/null if=/dev/pmem0 bs=4096 count=1000000 1000000+0 records in 1000000+0 records out 4096000000 bytes (4.1 GB) copied, 3.44034 s, 1.2 GB/s To be able to safely support a write-back aperture I needed to add support for the "read flush" _DSM flag, as outlined in the DSM spec: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf This flag tells the ND BLK driver that it needs to flush the cache lines associated with the aperture after the aperture is moved but before any new data is read. This ensures that any stale cache lines from the previous contents of the aperture will be discarded from the processor cache, and the new data will be read properly from the DIMM. We know that the cache lines are clean and will be discarded without any writeback because either a) the previous aperture operation was a read, and we never modified the contents of the aperture, or b) the previous aperture operation was a write and we must have written back the dirtied contents of the aperture to the DIMM before the I/O was completed. In order to add support for the "read flush" flag I needed to add a generic routine to invalidate cache lines, mmio_flush_range(). This is protected by the ARCH_HAS_MMIO_FLUSH Kconfig variable, and is currently only supported on x86. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:14:20 +00:00
if (!mmio->addr.base) {
dev_dbg(dev, "%s failed to map dcr\n",
nvdimm_name(nvdimm));
return -ENOMEM;
}
mmio->size = nfit_mem->dcr->window_size;
mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
mmio->idt = nfit_mem->idt_dcr;
mmio->spa = nfit_mem->spa_dcr;
rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
nfit_mem->memdev_dcr->interleave_ways);
if (rc) {
dev_dbg(dev, "%s failed to init dcr interleave\n",
nvdimm_name(nvdimm));
return rc;
}
rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
if (rc < 0) {
dev_dbg(dev, "%s failed get DIMM flags\n",
nvdimm_name(nvdimm));
return rc;
}
libnvdimm: introduce nvdimm_flush() and nvdimm_has_flush() nvdimm_flush() is a replacement for the x86 'pcommit' instruction. It is an optional write flushing mechanism that an nvdimm bus can provide for the pmem driver to consume. In the case of the NFIT nvdimm-bus-provider nvdimm_flush() is implemented as a series of flush-hint-address [1] writes to each dimm in the interleave set (region) that backs the namespace. The nvdimm_has_flush() routine relies on platform firmware to describe the flushing capabilities of a platform. It uses the heuristic of whether an nvdimm bus provider provides flush address data to return a ternary result: 1: flush addresses defined 0: dimm topology described without flush addresses (assume ADR) -errno: no topology information, unable to determine flush mechanism The pmem driver is expected to take the following actions on this ternary result: 1: nvdimm_flush() in response to REQ_FUA / REQ_FLUSH and shutdown 0: do not set, WC or FUA on the queue, take no further action -errno: warn and then operate as if nvdimm_has_flush() returned '0' The caveat of this heuristic is that it can not distinguish the "dimm does not have flush address" case from the "platform firmware is broken and failed to describe a flush address". Given we are already explicitly trusting the NFIT there's not much more we can do beyond blacklisting broken firmwares if they are ever encountered. Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-07-08 02:44:50 +00:00
if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
dev_warn(dev, "unable to guarantee persistence of writes\n");
if (mmio->line_size == 0)
return 0;
if ((u32) nfit_blk->cmd_offset % mmio->line_size
+ 8 > mmio->line_size) {
dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
return -ENXIO;
} else if ((u32) nfit_blk->stat_offset % mmio->line_size
+ 8 > mmio->line_size) {
dev_dbg(dev, "stat_offset crosses interleave boundary\n");
return -ENXIO;
}
return 0;
}
static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
{
struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
struct acpi_nfit_system_address *spa = nfit_spa->spa;
int cmd_rc, rc;
cmd->address = spa->address;
cmd->length = spa->length;
rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
sizeof(*cmd), &cmd_rc);
if (rc < 0)
return rc;
return cmd_rc;
}
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
static int ars_start(struct acpi_nfit_desc *acpi_desc,
struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
{
int rc;
int cmd_rc;
struct nd_cmd_ars_start ars_start;
struct acpi_nfit_system_address *spa = nfit_spa->spa;
struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
memset(&ars_start, 0, sizeof(ars_start));
ars_start.address = spa->address;
ars_start.length = spa->length;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
if (req_type == ARS_REQ_SHORT)
ars_start.flags = ND_ARS_RETURN_PREV_DATA;
if (nfit_spa_type(spa) == NFIT_SPA_PM)
ars_start.type = ND_ARS_PERSISTENT;
else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
ars_start.type = ND_ARS_VOLATILE;
else
return -ENOTTY;
rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
sizeof(ars_start), &cmd_rc);
if (rc < 0)
return rc;
return cmd_rc;
}
static int ars_continue(struct acpi_nfit_desc *acpi_desc)
{
int rc, cmd_rc;
struct nd_cmd_ars_start ars_start;
struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
memset(&ars_start, 0, sizeof(ars_start));
ars_start.address = ars_status->restart_address;
ars_start.length = ars_status->restart_length;
ars_start.type = ars_status->type;
ars_start.flags = acpi_desc->ars_start_flags;
rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
sizeof(ars_start), &cmd_rc);
if (rc < 0)
return rc;
return cmd_rc;
}
static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
{
struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
int rc, cmd_rc;
rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
acpi_desc->max_ars, &cmd_rc);
if (rc < 0)
return rc;
return cmd_rc;
}
static void ars_complete(struct acpi_nfit_desc *acpi_desc,
struct nfit_spa *nfit_spa)
{
struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
struct acpi_nfit_system_address *spa = nfit_spa->spa;
struct nd_region *nd_region = nfit_spa->nd_region;
struct device *dev;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
lockdep_assert_held(&acpi_desc->init_mutex);
/*
* Only advance the ARS state for ARS runs initiated by the
* kernel, ignore ARS results from BIOS initiated runs for scrub
* completion tracking.
*/
if (acpi_desc->scrub_spa != nfit_spa)
return;
if ((ars_status->address >= spa->address && ars_status->address
< spa->address + spa->length)
|| (ars_status->address < spa->address)) {
/*
* Assume that if a scrub starts at an offset from the
* start of nfit_spa that we are in the continuation
* case.
*
* Otherwise, if the scrub covers the spa range, mark
* any pending request complete.
*/
if (ars_status->address + ars_status->length
>= spa->address + spa->length)
/* complete */;
else
return;
} else
return;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
acpi_desc->scrub_spa = NULL;
if (nd_region) {
dev = nd_region_dev(nd_region);
nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
} else
dev = acpi_desc->dev;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
}
static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
{
struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
int rc;
u32 i;
/*
* First record starts at 44 byte offset from the start of the
* payload.
*/
if (ars_status->out_length < 44)
return 0;
for (i = 0; i < ars_status->num_records; i++) {
/* only process full records */
if (ars_status->out_length
< 44 + sizeof(struct nd_ars_record) * (i + 1))
break;
rc = nvdimm_bus_add_badrange(nvdimm_bus,
ars_status->records[i].err_address,
ars_status->records[i].length);
if (rc)
return rc;
}
if (i < ars_status->num_records)
dev_warn(acpi_desc->dev, "detected truncated ars results\n");
return 0;
}
ACPI: Change NFIT driver to insert new resource ACPI 6 defines persistent memory (PMEM) ranges in multiple firmware interfaces, e820, EFI, and ACPI NFIT table. This EFI change, however, leads to hit a bug in the grub bootloader, which treats EFI_PERSISTENT_MEMORY type as regular memory and corrupts stored user data [1]. Therefore, BIOS may set generic reserved type in e820 and EFI to cover PMEM ranges. The kernel can initialize PMEM ranges from ACPI NFIT table alone. This scheme causes a problem in the iomem table, though. On x86, for instance, e820_reserve_resources() initializes top-level entries (iomem_resource.child) from the e820 table at early boot-time. This creates "reserved" entry for a PMEM range, which does not allow region_intersects() to check with PMEM type. Change acpi_nfit_register_region() to call acpi_nfit_insert_resource(), which calls insert_resource() to insert a PMEM entry from NFIT when the iomem table does not have a PMEM entry already. That is, when a PMEM range is marked as reserved type in e820, it inserts "Persistent Memory" entry, which results as follows. + "Persistent Memory" + "reserved" This allows the EINJ driver, which calls region_intersects() to check PMEM ranges, to work continuously even if BIOS sets reserved type (or sets nothing) to PMEM ranges in e820 and EFI. [1]: https://lists.gnu.org/archive/html/grub-devel/2015-11/msg00209.html Signed-off-by: Toshi Kani <toshi.kani@hpe.com> Cc: Rafael J. Wysocki <rjw@rjwysocki.net> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-03-09 19:47:06 +00:00
static void acpi_nfit_remove_resource(void *data)
{
struct resource *res = data;
remove_resource(res);
}
static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
struct nd_region_desc *ndr_desc)
{
struct resource *res, *nd_res = ndr_desc->res;
int is_pmem, ret;
/* No operation if the region is already registered as PMEM */
is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
if (is_pmem == REGION_INTERSECTS)
return 0;
res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
if (!res)
return -ENOMEM;
res->name = "Persistent Memory";
res->start = nd_res->start;
res->end = nd_res->end;
res->flags = IORESOURCE_MEM;
res->desc = IORES_DESC_PERSISTENT_MEMORY;
ret = insert_resource(&iomem_resource, res);
if (ret)
return ret;
ret = devm_add_action_or_reset(acpi_desc->dev,
acpi_nfit_remove_resource,
res);
if (ret)
ACPI: Change NFIT driver to insert new resource ACPI 6 defines persistent memory (PMEM) ranges in multiple firmware interfaces, e820, EFI, and ACPI NFIT table. This EFI change, however, leads to hit a bug in the grub bootloader, which treats EFI_PERSISTENT_MEMORY type as regular memory and corrupts stored user data [1]. Therefore, BIOS may set generic reserved type in e820 and EFI to cover PMEM ranges. The kernel can initialize PMEM ranges from ACPI NFIT table alone. This scheme causes a problem in the iomem table, though. On x86, for instance, e820_reserve_resources() initializes top-level entries (iomem_resource.child) from the e820 table at early boot-time. This creates "reserved" entry for a PMEM range, which does not allow region_intersects() to check with PMEM type. Change acpi_nfit_register_region() to call acpi_nfit_insert_resource(), which calls insert_resource() to insert a PMEM entry from NFIT when the iomem table does not have a PMEM entry already. That is, when a PMEM range is marked as reserved type in e820, it inserts "Persistent Memory" entry, which results as follows. + "Persistent Memory" + "reserved" This allows the EINJ driver, which calls region_intersects() to check PMEM ranges, to work continuously even if BIOS sets reserved type (or sets nothing) to PMEM ranges in e820 and EFI. [1]: https://lists.gnu.org/archive/html/grub-devel/2015-11/msg00209.html Signed-off-by: Toshi Kani <toshi.kani@hpe.com> Cc: Rafael J. Wysocki <rjw@rjwysocki.net> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-03-09 19:47:06 +00:00
return ret;
return 0;
}
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
struct acpi_nfit_memory_map *memdev,
struct nfit_spa *nfit_spa)
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
{
struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
memdev->device_handle);
struct acpi_nfit_system_address *spa = nfit_spa->spa;
struct nd_blk_region_desc *ndbr_desc;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
struct nfit_mem *nfit_mem;
int rc;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
if (!nvdimm) {
dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
spa->range_index, memdev->device_handle);
return -ENODEV;
}
mapping->nvdimm = nvdimm;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
switch (nfit_spa_type(spa)) {
case NFIT_SPA_PM:
case NFIT_SPA_VOLATILE:
mapping->start = memdev->address;
mapping->size = memdev->region_size;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
break;
case NFIT_SPA_DCR:
nfit_mem = nvdimm_provider_data(nvdimm);
if (!nfit_mem || !nfit_mem->bdw) {
dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
spa->range_index, nvdimm_name(nvdimm));
break;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
}
mapping->size = nfit_mem->bdw->capacity;
mapping->start = nfit_mem->bdw->start_address;
ndr_desc->num_lanes = nfit_mem->bdw->windows;
ndr_desc->mapping = mapping;
ndr_desc->num_mappings = 1;
ndbr_desc = to_blk_region_desc(ndr_desc);
ndbr_desc->enable = acpi_nfit_blk_region_enable;
tools/testing/nvdimm: libnvdimm unit test infrastructure 'libnvdimm' is the first driver sub-system in the kernel to implement mocking for unit test coverage. The nfit_test module gets built as an external module and arranges for external module replacements of nfit, libnvdimm, nd_pmem, and nd_blk. These replacements use the linker --wrap option to redirect calls to ioremap() + request_mem_region() to custom defined unit test resources. The end result is a fully functional nvdimm_bus, as far as userspace is concerned, but with the capability to perform otherwise destructive tests on emulated resources. Q: Why not use QEMU for this emulation? QEMU is not suitable for unit testing. QEMU's role is to faithfully emulate the platform. A unit test's role is to unfaithfully implement the platform with the goal of triggering bugs in the corners of the sub-system implementation. As bugs are discovered in platforms, or the sub-system itself, the unit tests are extended to backstop a fix with a reproducer unit test. Another problem with QEMU is that it would require coordination of 3 software projects instead of 2 (kernel + libndctl [1]) to maintain and execute the tests. The chances for bit rot and the difficulty of getting the tests running goes up non-linearly the more components involved. Q: Why submit this to the kernel tree instead of external modules in libndctl? Simple, to alleviate the same risk that out-of-tree external modules face. Updates to drivers/nvdimm/ can be immediately evaluated to see if they have any impact on tools/testing/nvdimm/. Q: What are the negative implications of merging this? It is a unique maintenance burden because the purpose of mocking an interface to enable a unit test is to purposefully short circuit the semantics of a routine to enable testing. For example __wrap_ioremap_cache() fakes the pmem driver into "ioremap()'ing" a test resource buffer allocated by dma_alloc_coherent(). The future maintenance burden hits when someone changes the semantics of ioremap_cache() and wonders what the implications are for the unit test. [1]: https://github.com/pmem/ndctl Cc: <linux-acpi@vger.kernel.org> Cc: Lv Zheng <lv.zheng@intel.com> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-17 21:23:32 +00:00
ndbr_desc->do_io = acpi_desc->blk_do_io;
rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
if (rc)
return rc;
nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
ndr_desc);
if (!nfit_spa->nd_region)
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
return -ENOMEM;
break;
}
return 0;
}
acpi, nfit: treat virtual ramdisk SPA as pmem region This patch adds logic to treat virtual ramdisk SPA as pmem region, then ramdisk's /dev/pmem* device can be mounted with iso9660. It's useful to work with the httpboot in EFI firmware to pull a remote ISO file to the local memory region for booting and installation. Wiki page of UEFI HTTPBoot with OVMF: https://en.opensuse.org/UEFI_HTTPBoot_with_OVMF The ramdisk function in EDK2/OVMF generates a ACPI0012 root device that it contains empty _STA but without _DSM: DefinitionBlock ("ssdt2.aml", "SSDT", 2, "INTEL ", "RamDisk ", 0x00001000) { Scope (\_SB) { Device (NVDR) { Name (_HID, "ACPI0012") // _HID: Hardware ID Name (_STR, Unicode ("NVDIMM Root Device")) // _STR: Description String Method (_STA, 0, NotSerialized) // _STA: Status { Return (0x0F) } } } } In section 5.2.25.2 of ACPI 6.1 spec, it mentions that the "SPA Range Structure Index" of virtual SPA shall be set to zero. That means virtual SPA will not be associated by any NVDIMM region mapping. The VCD's SPA Range Structure in NFIT is similar to virtual disk region as following: [028h 0040 2] Subtable Type : 0000 [System Physical Address Range] [02Ah 0042 2] Length : 0038 [02Ch 0044 2] Range Index : 0000 [02Eh 0046 2] Flags (decoded below) : 0000 Add/Online Operation Only : 0 Proximity Domain Valid : 0 [030h 0048 4] Reserved : 00000000 [034h 0052 4] Proximity Domain : 00000000 [038h 0056 16] Address Range GUID : 77AB535A-45FC-624B-5560-F7B281D1F96E [048h 0072 8] Address Range Base : 00000000B6ABD018 [050h 0080 8] Address Range Length : 0000000005500000 [058h 0088 8] Memory Map Attribute : 0000000000000000 The way to not associate a SPA range is to never reference it from a "flush hint", "interleave", or "control region" table. After testing on OVMF, pmem driver can support the region that it doesn't assoicate to any NVDIMM mapping. So, treat VCD like pmem is a idea to get a pmem block device that it contains iso. v4: Instoduce nfit_spa_is_virtual() to check virtual ramdisk SPA and create pmem region. v3: To simplify patch, removed useless VCD region in libnvdimm. v2: Removed the code for setting VCD to a read-only region. Cc: Gary Lin <GLin@suse.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Linda Knippers <linda.knippers@hpe.com> Signed-off-by: Lee, Chun-Yi <jlee@suse.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-07-15 04:05:35 +00:00
static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
{
return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
nfit_spa_type(spa) == NFIT_SPA_VCD ||
nfit_spa_type(spa) == NFIT_SPA_PDISK ||
nfit_spa_type(spa) == NFIT_SPA_PCD);
}
static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
{
return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
nfit_spa_type(spa) == NFIT_SPA_VCD ||
nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
}
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
struct nfit_spa *nfit_spa)
{
static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
struct acpi_nfit_system_address *spa = nfit_spa->spa;
struct nd_blk_region_desc ndbr_desc;
struct nd_region_desc *ndr_desc;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
struct nfit_memdev *nfit_memdev;
struct nvdimm_bus *nvdimm_bus;
struct resource res;
2015-05-01 17:11:27 +00:00
int count = 0, rc;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
if (nfit_spa->nd_region)
return 0;
acpi, nfit: treat virtual ramdisk SPA as pmem region This patch adds logic to treat virtual ramdisk SPA as pmem region, then ramdisk's /dev/pmem* device can be mounted with iso9660. It's useful to work with the httpboot in EFI firmware to pull a remote ISO file to the local memory region for booting and installation. Wiki page of UEFI HTTPBoot with OVMF: https://en.opensuse.org/UEFI_HTTPBoot_with_OVMF The ramdisk function in EDK2/OVMF generates a ACPI0012 root device that it contains empty _STA but without _DSM: DefinitionBlock ("ssdt2.aml", "SSDT", 2, "INTEL ", "RamDisk ", 0x00001000) { Scope (\_SB) { Device (NVDR) { Name (_HID, "ACPI0012") // _HID: Hardware ID Name (_STR, Unicode ("NVDIMM Root Device")) // _STR: Description String Method (_STA, 0, NotSerialized) // _STA: Status { Return (0x0F) } } } } In section 5.2.25.2 of ACPI 6.1 spec, it mentions that the "SPA Range Structure Index" of virtual SPA shall be set to zero. That means virtual SPA will not be associated by any NVDIMM region mapping. The VCD's SPA Range Structure in NFIT is similar to virtual disk region as following: [028h 0040 2] Subtable Type : 0000 [System Physical Address Range] [02Ah 0042 2] Length : 0038 [02Ch 0044 2] Range Index : 0000 [02Eh 0046 2] Flags (decoded below) : 0000 Add/Online Operation Only : 0 Proximity Domain Valid : 0 [030h 0048 4] Reserved : 00000000 [034h 0052 4] Proximity Domain : 00000000 [038h 0056 16] Address Range GUID : 77AB535A-45FC-624B-5560-F7B281D1F96E [048h 0072 8] Address Range Base : 00000000B6ABD018 [050h 0080 8] Address Range Length : 0000000005500000 [058h 0088 8] Memory Map Attribute : 0000000000000000 The way to not associate a SPA range is to never reference it from a "flush hint", "interleave", or "control region" table. After testing on OVMF, pmem driver can support the region that it doesn't assoicate to any NVDIMM mapping. So, treat VCD like pmem is a idea to get a pmem block device that it contains iso. v4: Instoduce nfit_spa_is_virtual() to check virtual ramdisk SPA and create pmem region. v3: To simplify patch, removed useless VCD region in libnvdimm. v2: Removed the code for setting VCD to a read-only region. Cc: Gary Lin <GLin@suse.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Linda Knippers <linda.knippers@hpe.com> Signed-off-by: Lee, Chun-Yi <jlee@suse.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-07-15 04:05:35 +00:00
if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
return 0;
}
memset(&res, 0, sizeof(res));
memset(&mappings, 0, sizeof(mappings));
memset(&ndbr_desc, 0, sizeof(ndbr_desc));
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
res.start = spa->address;
res.end = res.start + spa->length - 1;
ndr_desc = &ndbr_desc.ndr_desc;
ndr_desc->res = &res;
ndr_desc->provider_data = nfit_spa;
ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
ndr_desc->numa_node = acpi_map_pxm_to_online_node(
spa->proximity_domain);
else
ndr_desc->numa_node = NUMA_NO_NODE;
/*
* Persistence domain bits are hierarchical, if
* ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
* ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
*/
if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
struct nd_mapping_desc *mapping;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
if (memdev->range_index != spa->range_index)
continue;
if (count >= ND_MAX_MAPPINGS) {
dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
spa->range_index, ND_MAX_MAPPINGS);
return -ENXIO;
}
mapping = &mappings[count++];
rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
memdev, nfit_spa);
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
if (rc)
goto out;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
}
ndr_desc->mapping = mappings;
ndr_desc->num_mappings = count;
rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2015-05-01 17:11:27 +00:00
if (rc)
goto out;
2015-05-01 17:11:27 +00:00
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
nvdimm_bus = acpi_desc->nvdimm_bus;
if (nfit_spa_type(spa) == NFIT_SPA_PM) {
ACPI: Change NFIT driver to insert new resource ACPI 6 defines persistent memory (PMEM) ranges in multiple firmware interfaces, e820, EFI, and ACPI NFIT table. This EFI change, however, leads to hit a bug in the grub bootloader, which treats EFI_PERSISTENT_MEMORY type as regular memory and corrupts stored user data [1]. Therefore, BIOS may set generic reserved type in e820 and EFI to cover PMEM ranges. The kernel can initialize PMEM ranges from ACPI NFIT table alone. This scheme causes a problem in the iomem table, though. On x86, for instance, e820_reserve_resources() initializes top-level entries (iomem_resource.child) from the e820 table at early boot-time. This creates "reserved" entry for a PMEM range, which does not allow region_intersects() to check with PMEM type. Change acpi_nfit_register_region() to call acpi_nfit_insert_resource(), which calls insert_resource() to insert a PMEM entry from NFIT when the iomem table does not have a PMEM entry already. That is, when a PMEM range is marked as reserved type in e820, it inserts "Persistent Memory" entry, which results as follows. + "Persistent Memory" + "reserved" This allows the EINJ driver, which calls region_intersects() to check PMEM ranges, to work continuously even if BIOS sets reserved type (or sets nothing) to PMEM ranges in e820 and EFI. [1]: https://lists.gnu.org/archive/html/grub-devel/2015-11/msg00209.html Signed-off-by: Toshi Kani <toshi.kani@hpe.com> Cc: Rafael J. Wysocki <rjw@rjwysocki.net> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-03-09 19:47:06 +00:00
rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
if (rc) {
ACPI: Change NFIT driver to insert new resource ACPI 6 defines persistent memory (PMEM) ranges in multiple firmware interfaces, e820, EFI, and ACPI NFIT table. This EFI change, however, leads to hit a bug in the grub bootloader, which treats EFI_PERSISTENT_MEMORY type as regular memory and corrupts stored user data [1]. Therefore, BIOS may set generic reserved type in e820 and EFI to cover PMEM ranges. The kernel can initialize PMEM ranges from ACPI NFIT table alone. This scheme causes a problem in the iomem table, though. On x86, for instance, e820_reserve_resources() initializes top-level entries (iomem_resource.child) from the e820 table at early boot-time. This creates "reserved" entry for a PMEM range, which does not allow region_intersects() to check with PMEM type. Change acpi_nfit_register_region() to call acpi_nfit_insert_resource(), which calls insert_resource() to insert a PMEM entry from NFIT when the iomem table does not have a PMEM entry already. That is, when a PMEM range is marked as reserved type in e820, it inserts "Persistent Memory" entry, which results as follows. + "Persistent Memory" + "reserved" This allows the EINJ driver, which calls region_intersects() to check PMEM ranges, to work continuously even if BIOS sets reserved type (or sets nothing) to PMEM ranges in e820 and EFI. [1]: https://lists.gnu.org/archive/html/grub-devel/2015-11/msg00209.html Signed-off-by: Toshi Kani <toshi.kani@hpe.com> Cc: Rafael J. Wysocki <rjw@rjwysocki.net> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-03-09 19:47:06 +00:00
dev_warn(acpi_desc->dev,
"failed to insert pmem resource to iomem: %d\n",
rc);
goto out;
}
nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
ndr_desc);
if (!nfit_spa->nd_region)
rc = -ENOMEM;
} else if (nfit_spa_is_volatile(spa)) {
nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
ndr_desc);
if (!nfit_spa->nd_region)
rc = -ENOMEM;
acpi, nfit: treat virtual ramdisk SPA as pmem region This patch adds logic to treat virtual ramdisk SPA as pmem region, then ramdisk's /dev/pmem* device can be mounted with iso9660. It's useful to work with the httpboot in EFI firmware to pull a remote ISO file to the local memory region for booting and installation. Wiki page of UEFI HTTPBoot with OVMF: https://en.opensuse.org/UEFI_HTTPBoot_with_OVMF The ramdisk function in EDK2/OVMF generates a ACPI0012 root device that it contains empty _STA but without _DSM: DefinitionBlock ("ssdt2.aml", "SSDT", 2, "INTEL ", "RamDisk ", 0x00001000) { Scope (\_SB) { Device (NVDR) { Name (_HID, "ACPI0012") // _HID: Hardware ID Name (_STR, Unicode ("NVDIMM Root Device")) // _STR: Description String Method (_STA, 0, NotSerialized) // _STA: Status { Return (0x0F) } } } } In section 5.2.25.2 of ACPI 6.1 spec, it mentions that the "SPA Range Structure Index" of virtual SPA shall be set to zero. That means virtual SPA will not be associated by any NVDIMM region mapping. The VCD's SPA Range Structure in NFIT is similar to virtual disk region as following: [028h 0040 2] Subtable Type : 0000 [System Physical Address Range] [02Ah 0042 2] Length : 0038 [02Ch 0044 2] Range Index : 0000 [02Eh 0046 2] Flags (decoded below) : 0000 Add/Online Operation Only : 0 Proximity Domain Valid : 0 [030h 0048 4] Reserved : 00000000 [034h 0052 4] Proximity Domain : 00000000 [038h 0056 16] Address Range GUID : 77AB535A-45FC-624B-5560-F7B281D1F96E [048h 0072 8] Address Range Base : 00000000B6ABD018 [050h 0080 8] Address Range Length : 0000000005500000 [058h 0088 8] Memory Map Attribute : 0000000000000000 The way to not associate a SPA range is to never reference it from a "flush hint", "interleave", or "control region" table. After testing on OVMF, pmem driver can support the region that it doesn't assoicate to any NVDIMM mapping. So, treat VCD like pmem is a idea to get a pmem block device that it contains iso. v4: Instoduce nfit_spa_is_virtual() to check virtual ramdisk SPA and create pmem region. v3: To simplify patch, removed useless VCD region in libnvdimm. v2: Removed the code for setting VCD to a read-only region. Cc: Gary Lin <GLin@suse.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Ross Zwisler <ross.zwisler@linux.intel.com> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net> Cc: Linda Knippers <linda.knippers@hpe.com> Signed-off-by: Lee, Chun-Yi <jlee@suse.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2016-07-15 04:05:35 +00:00
} else if (nfit_spa_is_virtual(spa)) {
nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
ndr_desc);
if (!nfit_spa->nd_region)
rc = -ENOMEM;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
}
out:
if (rc)
dev_err(acpi_desc->dev, "failed to register spa range %d\n",
nfit_spa->spa->range_index);
return rc;
}
static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
{
struct device *dev = acpi_desc->dev;
struct nd_cmd_ars_status *ars_status;
if (acpi_desc->ars_status) {
memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
return 0;
}
ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
if (!ars_status)
return -ENOMEM;
acpi_desc->ars_status = ars_status;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
return 0;
}
static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
{
int rc;
if (ars_status_alloc(acpi_desc))
return -ENOMEM;
rc = ars_get_status(acpi_desc);
if (rc < 0 && rc != -ENOSPC)
return rc;
if (ars_status_process_records(acpi_desc))
dev_err(acpi_desc->dev, "Failed to process ARS records\n");
return rc;
}
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
static int ars_register(struct acpi_nfit_desc *acpi_desc,
struct nfit_spa *nfit_spa)
{
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
int rc;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
if (no_init_ars || test_bit(ARS_FAILED, &nfit_spa->ars_state))
return acpi_nfit_register_region(acpi_desc, nfit_spa);
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
switch (acpi_nfit_query_poison(acpi_desc)) {
case 0:
case -EAGAIN:
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
/* shouldn't happen, try again later */
if (rc == -EBUSY)
break;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
if (rc) {
set_bit(ARS_FAILED, &nfit_spa->ars_state);
break;
}
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
rc = acpi_nfit_query_poison(acpi_desc);
if (rc)
break;
acpi_desc->scrub_spa = nfit_spa;
ars_complete(acpi_desc, nfit_spa);
/*
* If ars_complete() says we didn't complete the
* short scrub, we'll try again with a long
* request.
*/
acpi_desc->scrub_spa = NULL;
break;
case -EBUSY:
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
case -ENOMEM:
case -ENOSPC:
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
/*
* BIOS was using ARS, wait for it to complete (or
* resources to become available) and then perform our
* own scrubs.
*/
break;
default:
set_bit(ARS_FAILED, &nfit_spa->ars_state);
break;
}
return acpi_nfit_register_region(acpi_desc, nfit_spa);
}
static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
{
struct nfit_spa *nfit_spa;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
continue;
ars_complete(acpi_desc, nfit_spa);
}
}
static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
int query_rc)
{
unsigned int tmo = acpi_desc->scrub_tmo;
struct device *dev = acpi_desc->dev;
struct nfit_spa *nfit_spa;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
lockdep_assert_held(&acpi_desc->init_mutex);
if (acpi_desc->cancel)
return 0;
if (query_rc == -EBUSY) {
dev_dbg(dev, "ARS: ARS busy\n");
return min(30U * 60U, tmo * 2);
}
if (query_rc == -ENOSPC) {
dev_dbg(dev, "ARS: ARS continue\n");
ars_continue(acpi_desc);
return 1;
}
if (query_rc && query_rc != -EAGAIN) {
unsigned long long addr, end;
addr = acpi_desc->ars_status->address;
end = addr + acpi_desc->ars_status->length;
dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
query_rc);
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
}
ars_complete_all(acpi_desc);
list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
enum nfit_ars_state req_type;
int rc;
if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
continue;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
/* prefer short ARS requests first */
if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
req_type = ARS_REQ_SHORT;
else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
req_type = ARS_REQ_LONG;
else
continue;
rc = ars_start(acpi_desc, nfit_spa, req_type);
dev = nd_region_dev(nfit_spa->nd_region);
dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
nfit_spa->spa->range_index,
req_type == ARS_REQ_SHORT ? "short" : "long",
rc);
/*
* Hmm, we raced someone else starting ARS? Try again in
* a bit.
*/
if (rc == -EBUSY)
return 1;
if (rc == 0) {
dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
"scrub start while range %d active\n",
acpi_desc->scrub_spa->spa->range_index);
clear_bit(req_type, &nfit_spa->ars_state);
acpi_desc->scrub_spa = nfit_spa;
/*
* Consider this spa last for future scrub
* requests
*/
list_move_tail(&nfit_spa->list, &acpi_desc->spas);
return 1;
}
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
dev_err(dev, "ARS: range %d ARS failed (%d)\n",
nfit_spa->spa->range_index, rc);
set_bit(ARS_FAILED, &nfit_spa->ars_state);
}
return 0;
}
static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
{
lockdep_assert_held(&acpi_desc->init_mutex);
acpi_desc->scrub_busy = 1;
/* note this should only be set from within the workqueue */
if (tmo)
acpi_desc->scrub_tmo = tmo;
queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
}
static void sched_ars(struct acpi_nfit_desc *acpi_desc)
{
__sched_ars(acpi_desc, 0);
}
static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
{
lockdep_assert_held(&acpi_desc->init_mutex);
acpi_desc->scrub_busy = 0;
acpi_desc->scrub_count++;
if (acpi_desc->scrub_count_state)
sysfs_notify_dirent(acpi_desc->scrub_count_state);
}
static void acpi_nfit_scrub(struct work_struct *work)
{
struct acpi_nfit_desc *acpi_desc;
unsigned int tmo;
int query_rc;
acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
mutex_lock(&acpi_desc->init_mutex);
query_rc = acpi_nfit_query_poison(acpi_desc);
tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
if (tmo)
__sched_ars(acpi_desc, tmo);
else
notify_ars_done(acpi_desc);
memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
mutex_unlock(&acpi_desc->init_mutex);
}
static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
struct nfit_spa *nfit_spa)
{
int type = nfit_spa_type(nfit_spa->spa);
struct nd_cmd_ars_cap ars_cap;
int rc;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
set_bit(ARS_FAILED, &nfit_spa->ars_state);
memset(&ars_cap, 0, sizeof(ars_cap));
rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
if (rc < 0)
return;
/* check that the supported scrub types match the spa type */
if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
& ND_ARS_VOLATILE) == 0)
return;
if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
& ND_ARS_PERSISTENT) == 0)
return;
nfit_spa->max_ars = ars_cap.max_ars_out;
nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
clear_bit(ARS_FAILED, &nfit_spa->ars_state);
}
static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
{
struct nfit_spa *nfit_spa;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
int rc;
list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
switch (nfit_spa_type(nfit_spa->spa)) {
case NFIT_SPA_VOLATILE:
case NFIT_SPA_PM:
acpi_nfit_init_ars(acpi_desc, nfit_spa);
break;
}
}
list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
switch (nfit_spa_type(nfit_spa->spa)) {
case NFIT_SPA_VOLATILE:
case NFIT_SPA_PM:
/* register regions and kick off initial ARS run */
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
rc = ars_register(acpi_desc, nfit_spa);
if (rc)
return rc;
break;
case NFIT_SPA_BDW:
/* nothing to register */
break;
case NFIT_SPA_DCR:
case NFIT_SPA_VDISK:
case NFIT_SPA_VCD:
case NFIT_SPA_PDISK:
case NFIT_SPA_PCD:
/* register known regions that don't support ARS */
rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
if (rc)
return rc;
break;
default:
/* don't register unknown regions */
break;
}
sched_ars(acpi_desc);
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
return 0;
}
static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
struct nfit_table_prev *prev)
{
struct device *dev = acpi_desc->dev;
if (!list_empty(&prev->spas) ||
!list_empty(&prev->memdevs) ||
!list_empty(&prev->dcrs) ||
!list_empty(&prev->bdws) ||
!list_empty(&prev->idts) ||
!list_empty(&prev->flushes)) {
dev_err(dev, "new nfit deletes entries (unsupported)\n");
return -ENXIO;
}
return 0;
}
static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
{
struct device *dev = acpi_desc->dev;
struct kernfs_node *nfit;
struct device *bus_dev;
if (!ars_supported(acpi_desc->nvdimm_bus))
return 0;
bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
if (!nfit) {
dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
return -ENODEV;
}
acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
sysfs_put(nfit);
if (!acpi_desc->scrub_count_state) {
dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
return -ENODEV;
}
return 0;
}
static void acpi_nfit_unregister(void *data)
{
struct acpi_nfit_desc *acpi_desc = data;
nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
}
int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
{
struct device *dev = acpi_desc->dev;
struct nfit_table_prev prev;
const void *end;
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
int rc;
if (!acpi_desc->nvdimm_bus) {
acpi_nfit_init_dsms(acpi_desc);
acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
&acpi_desc->nd_desc);
if (!acpi_desc->nvdimm_bus)
return -ENOMEM;
rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
acpi_desc);
if (rc)
return rc;
rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
if (rc)
return rc;
/* register this acpi_desc for mce notifications */
mutex_lock(&acpi_desc_lock);
list_add_tail(&acpi_desc->list, &acpi_descs);
mutex_unlock(&acpi_desc_lock);
}
mutex_lock(&acpi_desc->init_mutex);
INIT_LIST_HEAD(&prev.spas);
INIT_LIST_HEAD(&prev.memdevs);
INIT_LIST_HEAD(&prev.dcrs);
INIT_LIST_HEAD(&prev.bdws);
INIT_LIST_HEAD(&prev.idts);
INIT_LIST_HEAD(&prev.flushes);
list_cut_position(&prev.spas, &acpi_desc->spas,
acpi_desc->spas.prev);
list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
acpi_desc->memdevs.prev);
list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
acpi_desc->dcrs.prev);
list_cut_position(&prev.bdws, &acpi_desc->bdws,
acpi_desc->bdws.prev);
list_cut_position(&prev.idts, &acpi_desc->idts,
acpi_desc->idts.prev);
list_cut_position(&prev.flushes, &acpi_desc->flushes,
acpi_desc->flushes.prev);
end = data + sz;
while (!IS_ERR_OR_NULL(data))
data = add_table(acpi_desc, &prev, data, end);
if (IS_ERR(data)) {
dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
rc = PTR_ERR(data);
goto out_unlock;
}
rc = acpi_nfit_check_deletions(acpi_desc, &prev);
if (rc)
goto out_unlock;
rc = nfit_mem_init(acpi_desc);
if (rc)
goto out_unlock;
2015-06-08 18:27:06 +00:00
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
rc = acpi_nfit_register_dimms(acpi_desc);
if (rc)
goto out_unlock;
rc = acpi_nfit_register_regions(acpi_desc);
libnvdimm, nfit: regions (block-data-window, persistent memory, volatile memory) A "region" device represents the maximum capacity of a BLK range (mmio block-data-window(s)), or a PMEM range (DAX-capable persistent memory or volatile memory), without regard for aliasing. Aliasing, in the dimm-local address space (DPA), is resolved by metadata on a dimm to designate which exclusive interface will access the aliased DPA ranges. Support for the per-dimm metadata/label arrvies is in a subsequent patch. The name format of "region" devices is "regionN" where, like dimms, N is a global ida index assigned at discovery time. This id is not reliable across reboots nor in the presence of hotplug. Look to attributes of the region or static id-data of the sub-namespace to generate a persistent name. However, if the platform configuration does not change it is reasonable to expect the same region id to be assigned at the next boot. "region"s have 2 generic attributes "size", and "mapping"s where: - size: the BLK accessible capacity or the span of the system physical address range in the case of PMEM. - mappingN: a tuple describing a dimm's contribution to the region's capacity in the format (<nmemX>,<dpa>,<size>). For a PMEM-region there will be at least one mapping per dimm in the interleave set. For a BLK-region there is only "mapping0" listing the starting DPA of the BLK-region and the available DPA capacity of that space (matches "size" above). The max number of mappings per "region" is hard coded per the constraints of sysfs attribute groups. That said the number of mappings per region should never exceed the maximum number of possible dimms in the system. If the current number turns out to not be enough then the "mappings" attribute clarifies how many there are supposed to be. "32 should be enough for anybody...". Cc: Neil Brown <neilb@suse.de> Cc: <linux-acpi@vger.kernel.org> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Acked-by: Christoph Hellwig <hch@lst.de> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Tested-by: Toshi Kani <toshi.kani@hp.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-10 00:13:14 +00:00
out_unlock:
mutex_unlock(&acpi_desc->init_mutex);
return rc;
}
tools/testing/nvdimm: libnvdimm unit test infrastructure 'libnvdimm' is the first driver sub-system in the kernel to implement mocking for unit test coverage. The nfit_test module gets built as an external module and arranges for external module replacements of nfit, libnvdimm, nd_pmem, and nd_blk. These replacements use the linker --wrap option to redirect calls to ioremap() + request_mem_region() to custom defined unit test resources. The end result is a fully functional nvdimm_bus, as far as userspace is concerned, but with the capability to perform otherwise destructive tests on emulated resources. Q: Why not use QEMU for this emulation? QEMU is not suitable for unit testing. QEMU's role is to faithfully emulate the platform. A unit test's role is to unfaithfully implement the platform with the goal of triggering bugs in the corners of the sub-system implementation. As bugs are discovered in platforms, or the sub-system itself, the unit tests are extended to backstop a fix with a reproducer unit test. Another problem with QEMU is that it would require coordination of 3 software projects instead of 2 (kernel + libndctl [1]) to maintain and execute the tests. The chances for bit rot and the difficulty of getting the tests running goes up non-linearly the more components involved. Q: Why submit this to the kernel tree instead of external modules in libndctl? Simple, to alleviate the same risk that out-of-tree external modules face. Updates to drivers/nvdimm/ can be immediately evaluated to see if they have any impact on tools/testing/nvdimm/. Q: What are the negative implications of merging this? It is a unique maintenance burden because the purpose of mocking an interface to enable a unit test is to purposefully short circuit the semantics of a routine to enable testing. For example __wrap_ioremap_cache() fakes the pmem driver into "ioremap()'ing" a test resource buffer allocated by dma_alloc_coherent(). The future maintenance burden hits when someone changes the semantics of ioremap_cache() and wonders what the implications are for the unit test. [1]: https://github.com/pmem/ndctl Cc: <linux-acpi@vger.kernel.org> Cc: Lv Zheng <lv.zheng@intel.com> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-17 21:23:32 +00:00
EXPORT_SYMBOL_GPL(acpi_nfit_init);
static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
{
struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
struct device *dev = acpi_desc->dev;
/* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
device_lock(dev);
device_unlock(dev);
/* Bounce the init_mutex to complete initial registration */
mutex_lock(&acpi_desc->init_mutex);
mutex_unlock(&acpi_desc->init_mutex);
return 0;
}
static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
struct nvdimm *nvdimm, unsigned int cmd)
{
struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
struct nfit_spa *nfit_spa;
int rc = 0;
if (nvdimm)
return 0;
if (cmd != ND_CMD_ARS_START)
return 0;
/*
* The kernel and userspace may race to initiate a scrub, but
* the scrub thread is prepared to lose that initial race. It
* just needs guarantees that any ARS it initiates are not
* interrupted by any intervening start requests from userspace.
*/
mutex_lock(&acpi_desc->init_mutex);
list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
if (acpi_desc->scrub_spa
|| test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state)
|| test_bit(ARS_REQ_LONG, &nfit_spa->ars_state)) {
rc = -EBUSY;
break;
}
mutex_unlock(&acpi_desc->init_mutex);
return rc;
}
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
enum nfit_ars_state req_type)
{
struct device *dev = acpi_desc->dev;
int scheduled = 0, busy = 0;
struct nfit_spa *nfit_spa;
mutex_lock(&acpi_desc->init_mutex);
if (acpi_desc->cancel) {
mutex_unlock(&acpi_desc->init_mutex);
return 0;
}
list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
int type = nfit_spa_type(nfit_spa->spa);
if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
continue;
if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
continue;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
if (test_and_set_bit(req_type, &nfit_spa->ars_state))
busy++;
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
else
scheduled++;
}
if (scheduled) {
sched_ars(acpi_desc);
dev_dbg(dev, "ars_scan triggered\n");
}
mutex_unlock(&acpi_desc->init_mutex);
if (scheduled)
return 0;
if (busy)
return -EBUSY;
return -ENOTTY;
}
void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
{
struct nvdimm_bus_descriptor *nd_desc;
dev_set_drvdata(dev, acpi_desc);
acpi_desc->dev = dev;
tools/testing/nvdimm: libnvdimm unit test infrastructure 'libnvdimm' is the first driver sub-system in the kernel to implement mocking for unit test coverage. The nfit_test module gets built as an external module and arranges for external module replacements of nfit, libnvdimm, nd_pmem, and nd_blk. These replacements use the linker --wrap option to redirect calls to ioremap() + request_mem_region() to custom defined unit test resources. The end result is a fully functional nvdimm_bus, as far as userspace is concerned, but with the capability to perform otherwise destructive tests on emulated resources. Q: Why not use QEMU for this emulation? QEMU is not suitable for unit testing. QEMU's role is to faithfully emulate the platform. A unit test's role is to unfaithfully implement the platform with the goal of triggering bugs in the corners of the sub-system implementation. As bugs are discovered in platforms, or the sub-system itself, the unit tests are extended to backstop a fix with a reproducer unit test. Another problem with QEMU is that it would require coordination of 3 software projects instead of 2 (kernel + libndctl [1]) to maintain and execute the tests. The chances for bit rot and the difficulty of getting the tests running goes up non-linearly the more components involved. Q: Why submit this to the kernel tree instead of external modules in libndctl? Simple, to alleviate the same risk that out-of-tree external modules face. Updates to drivers/nvdimm/ can be immediately evaluated to see if they have any impact on tools/testing/nvdimm/. Q: What are the negative implications of merging this? It is a unique maintenance burden because the purpose of mocking an interface to enable a unit test is to purposefully short circuit the semantics of a routine to enable testing. For example __wrap_ioremap_cache() fakes the pmem driver into "ioremap()'ing" a test resource buffer allocated by dma_alloc_coherent(). The future maintenance burden hits when someone changes the semantics of ioremap_cache() and wonders what the implications are for the unit test. [1]: https://github.com/pmem/ndctl Cc: <linux-acpi@vger.kernel.org> Cc: Lv Zheng <lv.zheng@intel.com> Cc: Robert Moore <robert.moore@intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-17 21:23:32 +00:00
acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
nd_desc = &acpi_desc->nd_desc;
nd_desc->provider_name = "ACPI.NFIT";
nd_desc->module = THIS_MODULE;
nd_desc->ndctl = acpi_nfit_ctl;
nd_desc->flush_probe = acpi_nfit_flush_probe;
nd_desc->clear_to_send = acpi_nfit_clear_to_send;
nd_desc->attr_groups = acpi_nfit_attribute_groups;
INIT_LIST_HEAD(&acpi_desc->spas);
INIT_LIST_HEAD(&acpi_desc->dcrs);
INIT_LIST_HEAD(&acpi_desc->bdws);
INIT_LIST_HEAD(&acpi_desc->idts);
INIT_LIST_HEAD(&acpi_desc->flushes);
INIT_LIST_HEAD(&acpi_desc->memdevs);
INIT_LIST_HEAD(&acpi_desc->dimms);
INIT_LIST_HEAD(&acpi_desc->list);
mutex_init(&acpi_desc->init_mutex);
acpi_desc->scrub_tmo = 1;
INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
}
EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
static void acpi_nfit_put_table(void *table)
{
acpi_put_table(table);
}
void acpi_nfit_shutdown(void *data)
{
struct acpi_nfit_desc *acpi_desc = data;
struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
/*
* Destruct under acpi_desc_lock so that nfit_handle_mce does not
* race teardown
*/
mutex_lock(&acpi_desc_lock);
list_del(&acpi_desc->list);
mutex_unlock(&acpi_desc_lock);
mutex_lock(&acpi_desc->init_mutex);
acpi_desc->cancel = 1;
cancel_delayed_work_sync(&acpi_desc->dwork);
mutex_unlock(&acpi_desc->init_mutex);
/*
* Bounce the nvdimm bus lock to make sure any in-flight
* acpi_nfit_ars_rescan() submissions have had a chance to
* either submit or see ->cancel set.
*/
device_lock(bus_dev);
device_unlock(bus_dev);
flush_workqueue(nfit_wq);
}
EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
static int acpi_nfit_add(struct acpi_device *adev)
{
struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
struct acpi_nfit_desc *acpi_desc;
struct device *dev = &adev->dev;
struct acpi_table_header *tbl;
acpi_status status = AE_OK;
acpi_size sz;
int rc = 0;
status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
if (ACPI_FAILURE(status)) {
/* This is ok, we could have an nvdimm hotplugged later */
dev_dbg(dev, "failed to find NFIT at startup\n");
return 0;
}
rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
if (rc)
return rc;
sz = tbl->length;
acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
if (!acpi_desc)
return -ENOMEM;
acpi_nfit_desc_init(acpi_desc, &adev->dev);
/* Save the acpi header for exporting the revision via sysfs */
acpi_desc->acpi_header = *tbl;
/* Evaluate _FIT and override with that if present */
status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
if (ACPI_SUCCESS(status) && buf.length > 0) {
union acpi_object *obj = buf.pointer;
if (obj->type == ACPI_TYPE_BUFFER)
rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
obj->buffer.length);
else
dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
(int) obj->type);
kfree(buf.pointer);
} else
/* skip over the lead-in header table */
rc = acpi_nfit_init(acpi_desc, (void *) tbl
+ sizeof(struct acpi_table_nfit),
sz - sizeof(struct acpi_table_nfit));
if (rc)
return rc;
return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
}
static int acpi_nfit_remove(struct acpi_device *adev)
{
/* see acpi_nfit_unregister */
return 0;
}
static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
{
struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
union acpi_object *obj;
acpi_status status;
int ret;
if (!dev->driver) {
/* dev->driver may be null if we're being removed */
dev_dbg(dev, "no driver found for dev\n");
return;
}
if (!acpi_desc) {
acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
if (!acpi_desc)
return;
acpi_nfit_desc_init(acpi_desc, dev);
} else {
/*
* Finish previous registration before considering new
* regions.
*/
flush_workqueue(nfit_wq);
}
/* Evaluate _FIT */
status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
if (ACPI_FAILURE(status)) {
dev_err(dev, "failed to evaluate _FIT\n");
return;
}
obj = buf.pointer;
if (obj->type == ACPI_TYPE_BUFFER) {
ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
obj->buffer.length);
if (ret)
dev_err(dev, "failed to merge updated NFIT\n");
} else
dev_err(dev, "Invalid _FIT\n");
kfree(buf.pointer);
}
static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
{
struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
acpi, nfit: Fix Address Range Scrub completion tracking The Address Range Scrub implementation tried to skip running scrubs against ranges that were already scrubbed by the BIOS. Unfortunately that support also resulted in early scrub completions as evidenced by this debug output from nfit_test: nd_region region9: ARS: range 1 short complete nd_region region3: ARS: range 1 short complete nd_region region4: ARS: range 2 ARS start (0) nd_region region4: ARS: range 2 short complete ...i.e. completions without any indications that the scrub was started. This state of affairs was hard to see in the code due to the proliferation of state bits and mistakenly trying to track done state per-range when the completion is a global property of the bus. So, kill the four ARS state bits (ARS_REQ, ARS_REQ_REDO, ARS_DONE, and ARS_SHORT), and replace them with just 2 request flags ARS_REQ_SHORT and ARS_REQ_LONG. The implementation will still complete and reap the results of BIOS initiated ARS, but it will not attempt to use that information to affect the completion status of scrubbing the ranges from a Linux perspective. Instead, try to synchronously run a short ARS per range at init time and schedule a long scrub in the background. If ARS is busy with an ARS request, schedule both a short and a long scrub for when ARS returns to idle. This logic also satisfies the intent of what ARS_REQ_REDO was trying to achieve. The new rule is that the REQ flag stays set until the next successful ars_start() for that range. With the new policy that the REQ flags are not cleared until the next start, the implementation no longer loses requests as can be seen from the following log: nd_region region3: ARS: range 1 ARS start short (0) nd_region region9: ARS: range 1 ARS start short (0) nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start short (0) nd_region region9: ARS: range 1 complete nd_region region9: ARS: range 1 ARS start long (0) nd_region region4: ARS: range 2 complete nd_region region3: ARS: range 1 ARS start long (0) nd_region region9: ARS: range 1 complete nd_region region3: ARS: range 1 complete nd_region region4: ARS: range 2 ARS start long (0) nd_region region4: ARS: range 2 complete ...note that the nfit_test emulated driver provides 2 buses, that is why some of the range indices are duplicated. Notice that each range now successfully completes a short and long scrub. Cc: <stable@vger.kernel.org> Fixes: 14c73f997a5e ("nfit, address-range-scrub: introduce nfit_spa->ars_state") Fixes: cc3d3458d46f ("acpi/nfit: queue issuing of ars when an uc error...") Reported-by: Jacek Zloch <jacek.zloch@intel.com> Reported-by: Krzysztof Rusocki <krzysztof.rusocki@intel.com> Reviewed-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-10-14 03:32:17 +00:00
if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
else
acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
}
void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
{
dev_dbg(dev, "event: 0x%x\n", event);
switch (event) {
case NFIT_NOTIFY_UPDATE:
return acpi_nfit_update_notify(dev, handle);
case NFIT_NOTIFY_UC_MEMORY_ERROR:
return acpi_nfit_uc_error_notify(dev, handle);
default:
return;
}
}
EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
{
device_lock(&adev->dev);
__acpi_nfit_notify(&adev->dev, adev->handle, event);
device_unlock(&adev->dev);
}
static const struct acpi_device_id acpi_nfit_ids[] = {
{ "ACPI0012", 0 },
{ "", 0 },
};
MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
static struct acpi_driver acpi_nfit_driver = {
.name = KBUILD_MODNAME,
.ids = acpi_nfit_ids,
.ops = {
.add = acpi_nfit_add,
.remove = acpi_nfit_remove,
.notify = acpi_nfit_notify,
},
};
static __init int nfit_init(void)
{
int ret;
BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
nfit_wq = create_singlethread_workqueue("nfit");
if (!nfit_wq)
return -ENOMEM;
nfit_mce_register();
ret = acpi_bus_register_driver(&acpi_nfit_driver);
if (ret) {
nfit_mce_unregister();
destroy_workqueue(nfit_wq);
}
return ret;
}
static __exit void nfit_exit(void)
{
nfit_mce_unregister();
acpi_bus_unregister_driver(&acpi_nfit_driver);
destroy_workqueue(nfit_wq);
WARN_ON(!list_empty(&acpi_descs));
}
module_init(nfit_init);
module_exit(nfit_exit);
MODULE_LICENSE("GPL v2");
MODULE_AUTHOR("Intel Corporation");