linux/drivers/thunderbolt/ctl.c

1017 lines
24 KiB
C
Raw Normal View History

/*
* Thunderbolt Cactus Ridge driver - control channel and configuration commands
*
* Copyright (c) 2014 Andreas Noever <andreas.noever@gmail.com>
*/
#include <linux/crc32.h>
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
#include <linux/delay.h>
#include <linux/slab.h>
#include <linux/pci.h>
#include <linux/dmapool.h>
#include <linux/workqueue.h>
#include "ctl.h"
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
#define TB_CTL_RX_PKG_COUNT 10
#define TB_CTL_RETRIES 4
/**
* struct tb_cfg - thunderbolt control channel
*/
struct tb_ctl {
struct tb_nhi *nhi;
struct tb_ring *tx;
struct tb_ring *rx;
struct dma_pool *frame_pool;
struct ctl_pkg *rx_packets[TB_CTL_RX_PKG_COUNT];
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
struct mutex request_queue_lock;
struct list_head request_queue;
bool running;
event_cb callback;
void *callback_data;
};
#define tb_ctl_WARN(ctl, format, arg...) \
dev_WARN(&(ctl)->nhi->pdev->dev, format, ## arg)
#define tb_ctl_err(ctl, format, arg...) \
dev_err(&(ctl)->nhi->pdev->dev, format, ## arg)
#define tb_ctl_warn(ctl, format, arg...) \
dev_warn(&(ctl)->nhi->pdev->dev, format, ## arg)
#define tb_ctl_info(ctl, format, arg...) \
dev_info(&(ctl)->nhi->pdev->dev, format, ## arg)
#define tb_ctl_dbg(ctl, format, arg...) \
dev_dbg(&(ctl)->nhi->pdev->dev, format, ## arg)
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static DECLARE_WAIT_QUEUE_HEAD(tb_cfg_request_cancel_queue);
/* Serializes access to request kref_get/put */
static DEFINE_MUTEX(tb_cfg_request_lock);
/**
* tb_cfg_request_alloc() - Allocates a new config request
*
* This is refcounted object so when you are done with this, call
* tb_cfg_request_put() to it.
*/
struct tb_cfg_request *tb_cfg_request_alloc(void)
{
struct tb_cfg_request *req;
req = kzalloc(sizeof(*req), GFP_KERNEL);
if (!req)
return NULL;
kref_init(&req->kref);
return req;
}
/**
* tb_cfg_request_get() - Increase refcount of a request
* @req: Request whose refcount is increased
*/
void tb_cfg_request_get(struct tb_cfg_request *req)
{
mutex_lock(&tb_cfg_request_lock);
kref_get(&req->kref);
mutex_unlock(&tb_cfg_request_lock);
}
static void tb_cfg_request_destroy(struct kref *kref)
{
struct tb_cfg_request *req = container_of(kref, typeof(*req), kref);
kfree(req);
}
/**
* tb_cfg_request_put() - Decrease refcount and possibly release the request
* @req: Request whose refcount is decreased
*
* Call this function when you are done with the request. When refcount
* goes to %0 the object is released.
*/
void tb_cfg_request_put(struct tb_cfg_request *req)
{
mutex_lock(&tb_cfg_request_lock);
kref_put(&req->kref, tb_cfg_request_destroy);
mutex_unlock(&tb_cfg_request_lock);
}
static int tb_cfg_request_enqueue(struct tb_ctl *ctl,
struct tb_cfg_request *req)
{
WARN_ON(test_bit(TB_CFG_REQUEST_ACTIVE, &req->flags));
WARN_ON(req->ctl);
mutex_lock(&ctl->request_queue_lock);
if (!ctl->running) {
mutex_unlock(&ctl->request_queue_lock);
return -ENOTCONN;
}
req->ctl = ctl;
list_add_tail(&req->list, &ctl->request_queue);
set_bit(TB_CFG_REQUEST_ACTIVE, &req->flags);
mutex_unlock(&ctl->request_queue_lock);
return 0;
}
static void tb_cfg_request_dequeue(struct tb_cfg_request *req)
{
struct tb_ctl *ctl = req->ctl;
mutex_lock(&ctl->request_queue_lock);
list_del(&req->list);
clear_bit(TB_CFG_REQUEST_ACTIVE, &req->flags);
if (test_bit(TB_CFG_REQUEST_CANCELED, &req->flags))
wake_up(&tb_cfg_request_cancel_queue);
mutex_unlock(&ctl->request_queue_lock);
}
static bool tb_cfg_request_is_active(struct tb_cfg_request *req)
{
return test_bit(TB_CFG_REQUEST_ACTIVE, &req->flags);
}
static struct tb_cfg_request *
tb_cfg_request_find(struct tb_ctl *ctl, struct ctl_pkg *pkg)
{
struct tb_cfg_request *req;
bool found = false;
mutex_lock(&pkg->ctl->request_queue_lock);
list_for_each_entry(req, &pkg->ctl->request_queue, list) {
tb_cfg_request_get(req);
if (req->match(req, pkg)) {
found = true;
break;
}
tb_cfg_request_put(req);
}
mutex_unlock(&pkg->ctl->request_queue_lock);
return found ? req : NULL;
}
/* utility functions */
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static int check_header(const struct ctl_pkg *pkg, u32 len,
enum tb_cfg_pkg_type type, u64 route)
{
struct tb_cfg_header *header = pkg->buffer;
/* check frame, TODO: frame flags */
if (WARN(len != pkg->frame.size,
"wrong framesize (expected %#x, got %#x)\n",
len, pkg->frame.size))
return -EIO;
if (WARN(type != pkg->frame.eof, "wrong eof (expected %#x, got %#x)\n",
type, pkg->frame.eof))
return -EIO;
if (WARN(pkg->frame.sof, "wrong sof (expected 0x0, got %#x)\n",
pkg->frame.sof))
return -EIO;
/* check header */
if (WARN(header->unknown != 1 << 9,
"header->unknown is %#x\n", header->unknown))
return -EIO;
if (WARN(route != tb_cfg_get_route(header),
"wrong route (expected %llx, got %llx)",
route, tb_cfg_get_route(header)))
return -EIO;
return 0;
}
static int check_config_address(struct tb_cfg_address addr,
enum tb_cfg_space space, u32 offset,
u32 length)
{
if (WARN(addr.zero, "addr.zero is %#x\n", addr.zero))
return -EIO;
if (WARN(space != addr.space, "wrong space (expected %x, got %x\n)",
space, addr.space))
return -EIO;
if (WARN(offset != addr.offset, "wrong offset (expected %x, got %x\n)",
offset, addr.offset))
return -EIO;
if (WARN(length != addr.length, "wrong space (expected %x, got %x\n)",
length, addr.length))
return -EIO;
/*
* We cannot check addr->port as it is set to the upstream port of the
* sender.
*/
return 0;
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static struct tb_cfg_result decode_error(const struct ctl_pkg *response)
{
struct cfg_error_pkg *pkg = response->buffer;
struct tb_cfg_result res = { 0 };
res.response_route = tb_cfg_get_route(&pkg->header);
res.response_port = 0;
res.err = check_header(response, sizeof(*pkg), TB_CFG_PKG_ERROR,
tb_cfg_get_route(&pkg->header));
if (res.err)
return res;
WARN(pkg->zero1, "pkg->zero1 is %#x\n", pkg->zero1);
WARN(pkg->zero2, "pkg->zero1 is %#x\n", pkg->zero1);
WARN(pkg->zero3, "pkg->zero1 is %#x\n", pkg->zero1);
res.err = 1;
res.tb_error = pkg->error;
res.response_port = pkg->port;
return res;
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static struct tb_cfg_result parse_header(const struct ctl_pkg *pkg, u32 len,
enum tb_cfg_pkg_type type, u64 route)
{
struct tb_cfg_header *header = pkg->buffer;
struct tb_cfg_result res = { 0 };
if (pkg->frame.eof == TB_CFG_PKG_ERROR)
return decode_error(pkg);
res.response_port = 0; /* will be updated later for cfg_read/write */
res.response_route = tb_cfg_get_route(header);
res.err = check_header(pkg, len, type, route);
return res;
}
static void tb_cfg_print_error(struct tb_ctl *ctl,
const struct tb_cfg_result *res)
{
WARN_ON(res->err != 1);
switch (res->tb_error) {
case TB_CFG_ERROR_PORT_NOT_CONNECTED:
/* Port is not connected. This can happen during surprise
* removal. Do not warn. */
return;
case TB_CFG_ERROR_INVALID_CONFIG_SPACE:
/*
* Invalid cfg_space/offset/length combination in
* cfg_read/cfg_write.
*/
tb_ctl_WARN(ctl,
"CFG_ERROR(%llx:%x): Invalid config space or offset\n",
res->response_route, res->response_port);
return;
case TB_CFG_ERROR_NO_SUCH_PORT:
/*
* - The route contains a non-existent port.
* - The route contains a non-PHY port (e.g. PCIe).
* - The port in cfg_read/cfg_write does not exist.
*/
tb_ctl_WARN(ctl, "CFG_ERROR(%llx:%x): Invalid port\n",
res->response_route, res->response_port);
return;
case TB_CFG_ERROR_LOOP:
tb_ctl_WARN(ctl, "CFG_ERROR(%llx:%x): Route contains a loop\n",
res->response_route, res->response_port);
return;
default:
/* 5,6,7,9 and 11 are also valid error codes */
tb_ctl_WARN(ctl, "CFG_ERROR(%llx:%x): Unknown error\n",
res->response_route, res->response_port);
return;
}
}
static void cpu_to_be32_array(__be32 *dst, const u32 *src, size_t len)
{
int i;
for (i = 0; i < len; i++)
dst[i] = cpu_to_be32(src[i]);
}
static void be32_to_cpu_array(u32 *dst, __be32 *src, size_t len)
{
int i;
for (i = 0; i < len; i++)
dst[i] = be32_to_cpu(src[i]);
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static __be32 tb_crc(const void *data, size_t len)
{
return cpu_to_be32(~__crc32c_le(~0, data, len));
}
static void tb_ctl_pkg_free(struct ctl_pkg *pkg)
{
if (pkg) {
dma_pool_free(pkg->ctl->frame_pool,
pkg->buffer, pkg->frame.buffer_phy);
kfree(pkg);
}
}
static struct ctl_pkg *tb_ctl_pkg_alloc(struct tb_ctl *ctl)
{
struct ctl_pkg *pkg = kzalloc(sizeof(*pkg), GFP_KERNEL);
if (!pkg)
return NULL;
pkg->ctl = ctl;
pkg->buffer = dma_pool_alloc(ctl->frame_pool, GFP_KERNEL,
&pkg->frame.buffer_phy);
if (!pkg->buffer) {
kfree(pkg);
return NULL;
}
return pkg;
}
/* RX/TX handling */
static void tb_ctl_tx_callback(struct tb_ring *ring, struct ring_frame *frame,
bool canceled)
{
struct ctl_pkg *pkg = container_of(frame, typeof(*pkg), frame);
tb_ctl_pkg_free(pkg);
}
/**
* tb_cfg_tx() - transmit a packet on the control channel
*
* len must be a multiple of four.
*
* Return: Returns 0 on success or an error code on failure.
*/
static int tb_ctl_tx(struct tb_ctl *ctl, const void *data, size_t len,
enum tb_cfg_pkg_type type)
{
int res;
struct ctl_pkg *pkg;
if (len % 4 != 0) { /* required for le->be conversion */
tb_ctl_WARN(ctl, "TX: invalid size: %zu\n", len);
return -EINVAL;
}
if (len > TB_FRAME_SIZE - 4) { /* checksum is 4 bytes */
tb_ctl_WARN(ctl, "TX: packet too large: %zu/%d\n",
len, TB_FRAME_SIZE - 4);
return -EINVAL;
}
pkg = tb_ctl_pkg_alloc(ctl);
if (!pkg)
return -ENOMEM;
pkg->frame.callback = tb_ctl_tx_callback;
pkg->frame.size = len + 4;
pkg->frame.sof = type;
pkg->frame.eof = type;
cpu_to_be32_array(pkg->buffer, data, len / 4);
*(__be32 *) (pkg->buffer + len) = tb_crc(pkg->buffer, len);
res = ring_tx(ctl->tx, &pkg->frame);
if (res) /* ring is stopped */
tb_ctl_pkg_free(pkg);
return res;
}
/**
* tb_ctl_handle_event() - acknowledge a plug event, invoke ctl->callback
*/
static void tb_ctl_handle_event(struct tb_ctl *ctl, enum tb_cfg_pkg_type type,
struct ctl_pkg *pkg, size_t size)
{
ctl->callback(ctl->callback_data, type, pkg->buffer, size);
}
static void tb_ctl_rx_submit(struct ctl_pkg *pkg)
{
ring_rx(pkg->ctl->rx, &pkg->frame); /*
* We ignore failures during stop.
* All rx packets are referenced
* from ctl->rx_packets, so we do
* not loose them.
*/
}
static int tb_async_error(const struct ctl_pkg *pkg)
{
const struct cfg_error_pkg *error = (const struct cfg_error_pkg *)pkg;
if (pkg->frame.eof != TB_CFG_PKG_ERROR)
return false;
switch (error->error) {
case TB_CFG_ERROR_LINK_ERROR:
case TB_CFG_ERROR_HEC_ERROR_DETECTED:
case TB_CFG_ERROR_FLOW_CONTROL_ERROR:
return true;
default:
return false;
}
}
static void tb_ctl_rx_callback(struct tb_ring *ring, struct ring_frame *frame,
bool canceled)
{
struct ctl_pkg *pkg = container_of(frame, typeof(*pkg), frame);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
struct tb_cfg_request *req;
__be32 crc32;
if (canceled)
return; /*
* ring is stopped, packet is referenced from
* ctl->rx_packets.
*/
if (frame->size < 4 || frame->size % 4 != 0) {
tb_ctl_err(pkg->ctl, "RX: invalid size %#x, dropping packet\n",
frame->size);
goto rx;
}
frame->size -= 4; /* remove checksum */
crc32 = tb_crc(pkg->buffer, frame->size);
be32_to_cpu_array(pkg->buffer, pkg->buffer, frame->size / 4);
switch (frame->eof) {
case TB_CFG_PKG_READ:
case TB_CFG_PKG_WRITE:
case TB_CFG_PKG_ERROR:
case TB_CFG_PKG_OVERRIDE:
case TB_CFG_PKG_RESET:
if (*(__be32 *)(pkg->buffer + frame->size) != crc32) {
tb_ctl_err(pkg->ctl,
"RX: checksum mismatch, dropping packet\n");
goto rx;
}
if (tb_async_error(pkg)) {
tb_ctl_handle_event(pkg->ctl, frame->eof,
pkg, frame->size);
goto rx;
}
break;
case TB_CFG_PKG_EVENT:
if (*(__be32 *)(pkg->buffer + frame->size) != crc32) {
tb_ctl_err(pkg->ctl,
"RX: checksum mismatch, dropping packet\n");
goto rx;
}
thunderbolt: Add support for Internal Connection Manager (ICM) Starting from Intel Falcon Ridge the internal connection manager running on the Thunderbolt host controller has been supporting 4 security levels. One reason for this is to prevent DMA attacks and only allow connecting devices the user trusts. The internal connection manager (ICM) is the preferred way of connecting Thunderbolt devices over software only implementation typically used on Macs. The driver communicates with ICM using special Thunderbolt ring 0 (control channel) messages. In order to handle these messages we add support for the ICM messages to the control channel. The security levels are as follows: none - No security, all tunnels are created automatically user - User needs to approve the device before tunnels are created secure - User need to approve the device before tunnels are created. The device is sent a challenge on future connects to be able to verify it is actually the approved device. dponly - Only Display Port and USB tunnels can be created and those are created automatically. The security levels are typically configurable from the system BIOS and by default it is set to "user" on many systems. In this patch each Thunderbolt device will have either one or two new sysfs attributes: authorized and key. The latter appears for devices that support secure connect. In order to identify the device the user can read identication information, including UUID and name of the device from sysfs and based on that make a decision to authorize the device. The device is authorized by simply writing 1 to the "authorized" sysfs attribute. This is following the USB bus device authorization mechanism. The secure connect requires an additional challenge step (writing 2 to the "authorized" attribute) in future connects when the key has already been stored to the NVM of the device. Non-ICM systems (before Alpine Ridge) continue to use the existing functionality and the security level is set to none. For systems with Alpine Ridge, even on Apple hardware, we will use ICM. This code is based on the work done by Amir Levy and Michael Jamet. Signed-off-by: Michael Jamet <michael.jamet@intel.com> Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:16 +00:00
/* Fall through */
case TB_CFG_PKG_ICM_EVENT:
tb_ctl_handle_event(pkg->ctl, frame->eof, pkg, frame->size);
goto rx;
default:
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
break;
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
/*
* The received packet will be processed only if there is an
* active request and that the packet is what is expected. This
* prevents packets such as replies coming after timeout has
* triggered from messing with the active requests.
*/
req = tb_cfg_request_find(pkg->ctl, pkg);
if (req) {
if (req->copy(req, pkg))
schedule_work(&req->work);
tb_cfg_request_put(req);
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
rx:
tb_ctl_rx_submit(pkg);
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static void tb_cfg_request_work(struct work_struct *work)
{
struct tb_cfg_request *req = container_of(work, typeof(*req), work);
if (!test_bit(TB_CFG_REQUEST_CANCELED, &req->flags))
req->callback(req->callback_data);
tb_cfg_request_dequeue(req);
tb_cfg_request_put(req);
}
/**
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
* tb_cfg_request() - Start control request not waiting for it to complete
* @ctl: Control channel to use
* @req: Request to start
* @callback: Callback called when the request is completed
* @callback_data: Data to be passed to @callback
*
* This queues @req on the given control channel without waiting for it
* to complete. When the request completes @callback is called.
*/
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
int tb_cfg_request(struct tb_ctl *ctl, struct tb_cfg_request *req,
void (*callback)(void *), void *callback_data)
{
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
int ret;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
req->flags = 0;
req->callback = callback;
req->callback_data = callback_data;
INIT_WORK(&req->work, tb_cfg_request_work);
INIT_LIST_HEAD(&req->list);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
tb_cfg_request_get(req);
ret = tb_cfg_request_enqueue(ctl, req);
if (ret)
goto err_put;
ret = tb_ctl_tx(ctl, req->request, req->request_size,
req->request_type);
if (ret)
goto err_dequeue;
if (!req->response)
schedule_work(&req->work);
return 0;
err_dequeue:
tb_cfg_request_dequeue(req);
err_put:
tb_cfg_request_put(req);
return ret;
}
/**
* tb_cfg_request_cancel() - Cancel a control request
* @req: Request to cancel
* @err: Error to assign to the request
*
* This function can be used to cancel ongoing request. It will wait
* until the request is not active anymore.
*/
void tb_cfg_request_cancel(struct tb_cfg_request *req, int err)
{
set_bit(TB_CFG_REQUEST_CANCELED, &req->flags);
schedule_work(&req->work);
wait_event(tb_cfg_request_cancel_queue, !tb_cfg_request_is_active(req));
req->result.err = err;
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static void tb_cfg_request_complete(void *data)
{
complete(data);
}
/**
* tb_cfg_request_sync() - Start control request and wait until it completes
* @ctl: Control channel to use
* @req: Request to start
* @timeout_msec: Timeout how long to wait @req to complete
*
* Starts a control request and waits until it completes. If timeout
* triggers the request is canceled before function returns. Note the
* caller needs to make sure only one message for given switch is active
* at a time.
*/
struct tb_cfg_result tb_cfg_request_sync(struct tb_ctl *ctl,
struct tb_cfg_request *req,
int timeout_msec)
{
unsigned long timeout = msecs_to_jiffies(timeout_msec);
struct tb_cfg_result res = { 0 };
DECLARE_COMPLETION_ONSTACK(done);
int ret;
ret = tb_cfg_request(ctl, req, tb_cfg_request_complete, &done);
if (ret) {
res.err = ret;
return res;
}
if (!wait_for_completion_timeout(&done, timeout))
tb_cfg_request_cancel(req, -ETIMEDOUT);
flush_work(&req->work);
return req->result;
}
/* public interface, alloc/start/stop/free */
/**
* tb_ctl_alloc() - allocate a control channel
*
* cb will be invoked once for every hot plug event.
*
* Return: Returns a pointer on success or NULL on failure.
*/
struct tb_ctl *tb_ctl_alloc(struct tb_nhi *nhi, event_cb cb, void *cb_data)
{
int i;
struct tb_ctl *ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
if (!ctl)
return NULL;
ctl->nhi = nhi;
ctl->callback = cb;
ctl->callback_data = cb_data;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
mutex_init(&ctl->request_queue_lock);
INIT_LIST_HEAD(&ctl->request_queue);
ctl->frame_pool = dma_pool_create("thunderbolt_ctl", &nhi->pdev->dev,
TB_FRAME_SIZE, 4, 0);
if (!ctl->frame_pool)
goto err;
ctl->tx = ring_alloc_tx(nhi, 0, 10, RING_FLAG_NO_SUSPEND);
if (!ctl->tx)
goto err;
ctl->rx = ring_alloc_rx(nhi, 0, 10, RING_FLAG_NO_SUSPEND);
if (!ctl->rx)
goto err;
for (i = 0; i < TB_CTL_RX_PKG_COUNT; i++) {
ctl->rx_packets[i] = tb_ctl_pkg_alloc(ctl);
if (!ctl->rx_packets[i])
goto err;
ctl->rx_packets[i]->frame.callback = tb_ctl_rx_callback;
}
tb_ctl_info(ctl, "control channel created\n");
return ctl;
err:
tb_ctl_free(ctl);
return NULL;
}
/**
* tb_ctl_free() - free a control channel
*
* Must be called after tb_ctl_stop.
*
* Must NOT be called from ctl->callback.
*/
void tb_ctl_free(struct tb_ctl *ctl)
{
int i;
if (!ctl)
return;
if (ctl->rx)
ring_free(ctl->rx);
if (ctl->tx)
ring_free(ctl->tx);
/* free RX packets */
for (i = 0; i < TB_CTL_RX_PKG_COUNT; i++)
tb_ctl_pkg_free(ctl->rx_packets[i]);
if (ctl->frame_pool)
dma_pool_destroy(ctl->frame_pool);
kfree(ctl);
}
/**
* tb_cfg_start() - start/resume the control channel
*/
void tb_ctl_start(struct tb_ctl *ctl)
{
int i;
tb_ctl_info(ctl, "control channel starting...\n");
ring_start(ctl->tx); /* is used to ack hotplug packets, start first */
ring_start(ctl->rx);
for (i = 0; i < TB_CTL_RX_PKG_COUNT; i++)
tb_ctl_rx_submit(ctl->rx_packets[i]);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
ctl->running = true;
}
/**
* control() - pause the control channel
*
* All invocations of ctl->callback will have finished after this method
* returns.
*
* Must NOT be called from ctl->callback.
*/
void tb_ctl_stop(struct tb_ctl *ctl)
{
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
mutex_lock(&ctl->request_queue_lock);
ctl->running = false;
mutex_unlock(&ctl->request_queue_lock);
ring_stop(ctl->rx);
ring_stop(ctl->tx);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
if (!list_empty(&ctl->request_queue))
tb_ctl_WARN(ctl, "dangling request in request_queue\n");
INIT_LIST_HEAD(&ctl->request_queue);
tb_ctl_info(ctl, "control channel stopped\n");
}
/* public interface, commands */
/**
* tb_cfg_error() - send error packet
*
* Return: Returns 0 on success or an error code on failure.
*/
int tb_cfg_error(struct tb_ctl *ctl, u64 route, u32 port,
enum tb_cfg_error error)
{
struct cfg_error_pkg pkg = {
.header = tb_cfg_make_header(route),
.port = port,
.error = error,
};
tb_ctl_info(ctl, "resetting error on %llx:%x.\n", route, port);
return tb_ctl_tx(ctl, &pkg, sizeof(pkg), TB_CFG_PKG_ERROR);
}
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
static bool tb_cfg_match(const struct tb_cfg_request *req,
const struct ctl_pkg *pkg)
{
u64 route = tb_cfg_get_route(pkg->buffer) & ~BIT_ULL(63);
if (pkg->frame.eof == TB_CFG_PKG_ERROR)
return true;
if (pkg->frame.eof != req->response_type)
return false;
if (route != tb_cfg_get_route(req->request))
return false;
if (pkg->frame.size != req->response_size)
return false;
if (pkg->frame.eof == TB_CFG_PKG_READ ||
pkg->frame.eof == TB_CFG_PKG_WRITE) {
const struct cfg_read_pkg *req_hdr = req->request;
const struct cfg_read_pkg *res_hdr = pkg->buffer;
if (req_hdr->addr.seq != res_hdr->addr.seq)
return false;
}
return true;
}
static bool tb_cfg_copy(struct tb_cfg_request *req, const struct ctl_pkg *pkg)
{
struct tb_cfg_result res;
/* Now make sure it is in expected format */
res = parse_header(pkg, req->response_size, req->response_type,
tb_cfg_get_route(req->request));
if (!res.err)
memcpy(req->response, pkg->buffer, req->response_size);
req->result = res;
/* Always complete when first response is received */
return true;
}
/**
* tb_cfg_reset() - send a reset packet and wait for a response
*
* If the switch at route is incorrectly configured then we will not receive a
* reply (even though the switch will reset). The caller should check for
* -ETIMEDOUT and attempt to reconfigure the switch.
*/
struct tb_cfg_result tb_cfg_reset(struct tb_ctl *ctl, u64 route,
int timeout_msec)
{
struct cfg_reset_pkg request = { .header = tb_cfg_make_header(route) };
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
struct tb_cfg_result res = { 0 };
struct tb_cfg_header reply;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
struct tb_cfg_request *req;
req = tb_cfg_request_alloc();
if (!req) {
res.err = -ENOMEM;
return res;
}
req->match = tb_cfg_match;
req->copy = tb_cfg_copy;
req->request = &request;
req->request_size = sizeof(request);
req->request_type = TB_CFG_PKG_RESET;
req->response = &reply;
req->response_size = sizeof(reply);
req->response_type = sizeof(TB_CFG_PKG_RESET);
res = tb_cfg_request_sync(ctl, req, timeout_msec);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
tb_cfg_request_put(req);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
return res;
}
/**
* tb_cfg_read() - read from config space into buffer
*
* Offset and length are in dwords.
*/
struct tb_cfg_result tb_cfg_read_raw(struct tb_ctl *ctl, void *buffer,
u64 route, u32 port, enum tb_cfg_space space,
u32 offset, u32 length, int timeout_msec)
{
struct tb_cfg_result res = { 0 };
struct cfg_read_pkg request = {
.header = tb_cfg_make_header(route),
.addr = {
.port = port,
.space = space,
.offset = offset,
.length = length,
},
};
struct cfg_write_pkg reply;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
int retries = 0;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
while (retries < TB_CTL_RETRIES) {
struct tb_cfg_request *req;
req = tb_cfg_request_alloc();
if (!req) {
res.err = -ENOMEM;
return res;
}
request.addr.seq = retries++;
req->match = tb_cfg_match;
req->copy = tb_cfg_copy;
req->request = &request;
req->request_size = sizeof(request);
req->request_type = TB_CFG_PKG_READ;
req->response = &reply;
req->response_size = 12 + 4 * length;
req->response_type = TB_CFG_PKG_READ;
res = tb_cfg_request_sync(ctl, req, timeout_msec);
tb_cfg_request_put(req);
if (res.err != -ETIMEDOUT)
break;
/* Wait a bit (arbitrary time) until we send a retry */
usleep_range(10, 100);
}
if (res.err)
return res;
res.response_port = reply.addr.port;
res.err = check_config_address(reply.addr, space, offset, length);
if (!res.err)
memcpy(buffer, &reply.data, 4 * length);
return res;
}
/**
* tb_cfg_write() - write from buffer into config space
*
* Offset and length are in dwords.
*/
struct tb_cfg_result tb_cfg_write_raw(struct tb_ctl *ctl, const void *buffer,
u64 route, u32 port, enum tb_cfg_space space,
u32 offset, u32 length, int timeout_msec)
{
struct tb_cfg_result res = { 0 };
struct cfg_write_pkg request = {
.header = tb_cfg_make_header(route),
.addr = {
.port = port,
.space = space,
.offset = offset,
.length = length,
},
};
struct cfg_read_pkg reply;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
int retries = 0;
memcpy(&request.data, buffer, length * 4);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
while (retries < TB_CTL_RETRIES) {
struct tb_cfg_request *req;
req = tb_cfg_request_alloc();
if (!req) {
res.err = -ENOMEM;
return res;
}
request.addr.seq = retries++;
req->match = tb_cfg_match;
req->copy = tb_cfg_copy;
req->request = &request;
req->request_size = 12 + 4 * length;
req->request_type = TB_CFG_PKG_WRITE;
req->response = &reply;
req->response_size = sizeof(reply);
req->response_type = TB_CFG_PKG_WRITE;
res = tb_cfg_request_sync(ctl, req, timeout_msec);
tb_cfg_request_put(req);
if (res.err != -ETIMEDOUT)
break;
/* Wait a bit (arbitrary time) until we send a retry */
usleep_range(10, 100);
}
if (res.err)
return res;
res.response_port = reply.addr.port;
res.err = check_config_address(reply.addr, space, offset, length);
return res;
}
int tb_cfg_read(struct tb_ctl *ctl, void *buffer, u64 route, u32 port,
enum tb_cfg_space space, u32 offset, u32 length)
{
struct tb_cfg_result res = tb_cfg_read_raw(ctl, buffer, route, port,
space, offset, length, TB_CFG_DEFAULT_TIMEOUT);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
switch (res.err) {
case 0:
/* Success */
break;
case 1:
/* Thunderbolt error, tb_error holds the actual number */
tb_cfg_print_error(ctl, &res);
return -EIO;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
case -ETIMEDOUT:
tb_ctl_warn(ctl, "timeout reading config space %u from %#x\n",
space, offset);
break;
default:
WARN(1, "tb_cfg_read: %d\n", res.err);
break;
}
return res.err;
}
int tb_cfg_write(struct tb_ctl *ctl, const void *buffer, u64 route, u32 port,
enum tb_cfg_space space, u32 offset, u32 length)
{
struct tb_cfg_result res = tb_cfg_write_raw(ctl, buffer, route, port,
space, offset, length, TB_CFG_DEFAULT_TIMEOUT);
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
switch (res.err) {
case 0:
/* Success */
break;
case 1:
/* Thunderbolt error, tb_error holds the actual number */
tb_cfg_print_error(ctl, &res);
return -EIO;
thunderbolt: Rework control channel to be more reliable If a request times out the response might arrive right after the request is failed. This response is pushed to the kfifo and next request will read it instead. Since it most likely will not pass our validation checks in parse_header() the next request will fail as well, and response to that request will be pushed to the kfifo, ad infinitum. We end up in a situation where all requests fail and no devices can be added anymore until the driver is unloaded and reloaded again. To overcome this, rework the control channel so that we will have a queue of outstanding requests. Each request will be handled in turn and the response is validated against what is expected. Unexpected packets (for example responses for requests that have been timed out) are dropped. This model is copied from Greybus implementation with small changes here and there to get it cope with Thunderbolt control packets. In addition the configuration packets support sequence number which the switch is supposed to copy from the request to response. We use this to drop responses that are already timed out. Taking advantage of the sequence number, we automatically retry configuration read/write 4 times before giving up. Also timeout is not a programming error so there is no need to trigger a scary backtrace (WARN), instead we just log a warning. After all Thunderbolt devices are hot-pluggable by definition which means user can unplug a device any time and that is totally acceptable. With this change there is no need to take the global domain lock when sending configuration packets anymore. This is useful when we add support for cross-domain (XDomain) communication later on. Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com> Reviewed-by: Michael Jamet <michael.jamet@intel.com> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Andreas Noever <andreas.noever@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-06-06 12:25:10 +00:00
case -ETIMEDOUT:
tb_ctl_warn(ctl, "timeout writing config space %u to %#x\n",
space, offset);
break;
default:
WARN(1, "tb_cfg_write: %d\n", res.err);
break;
}
return res.err;
}
/**
* tb_cfg_get_upstream_port() - get upstream port number of switch at route
*
* Reads the first dword from the switches TB_CFG_SWITCH config area and
* returns the port number from which the reply originated.
*
* Return: Returns the upstream port number on success or an error code on
* failure.
*/
int tb_cfg_get_upstream_port(struct tb_ctl *ctl, u64 route)
{
u32 dummy;
struct tb_cfg_result res = tb_cfg_read_raw(ctl, &dummy, route, 0,
TB_CFG_SWITCH, 0, 1,
TB_CFG_DEFAULT_TIMEOUT);
if (res.err == 1)
return -EIO;
if (res.err)
return res.err;
return res.response_port;
}