The eeh_ops->probe() function is called from two different contexts:
1. On pseries, where we set EEH_PROBE_MODE_DEVTREE, it's called in
eeh_add_device_early() which is supposed to run before we create
a pci_dev.
2. On PowerNV, where we set EEH_PROBE_MODE_DEV, it's called in
eeh_device_add_late() which is supposed to run *after* the
pci_dev is created.
The "early" probe is required because PAPR requires that we perform an RTAS
call to enable EEH support on a device before we start interacting with it
via config space or MMIO. This requirement doesn't exist on PowerNV and
shoehorning two completely separate initialisation paths into a common
interface just results in a convoluted code everywhere.
Additionally the early probe requires the probe function to take an pci_dn
rather than a pci_dev argument. We'd like to make pci_dn a pseries specific
data structure since there's no real requirement for them on PowerNV. To
help both goals move the early probe into the pseries containment zone
so the platform depedence is more explicit.
Reviewed-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200306073904.4737-5-oohall@gmail.com
This check for a missing PHB has existing in various forms since the
initial PPC64 port was upstreamed in 2002. The idea seems to be that we
need to guard against creating pci-specific data structures for the non-pci
children of a PCI device tree node (e.g. USB devices). However, we only
create pci_dn structures for DT nodes that correspond to PCI devices so
there's not much point in doing this check in the eeh_probe path.
Reviewed-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200306073904.4737-4-oohall@gmail.com
The pci hotplug helper (pci_hp_add_devices()) calls
eeh_add_device_tree_early() to scan the device-tree for new PCI devices and
do the early EEH probe before the device is scanned. This early probe is a
no-op in a lot of cases because:
a) The early init is only required to satisfy a PAPR requirement that EEH
be configured before we start doing config accesses. On PowerNV it is
a no-op.
b) It's a no-op for devices that have already had their eeh_dev
initialised.
There are four callers of pci_hp_add_devices():
1. arch/powerpc/kernel/eeh_driver.c
Here the hotplug helper is called when re-scanning pci_devs that
were removed during an EEH recovery pass. The EEH stat for each
removed device (the eeh_dev) is retained across a recovery pass
so the early init is a no-op in this case.
2. drivers/pci/hotplug/pnv_php.c
This is also a no-op since the PowerNV hotplug driver is, suprisingly,
PowerNV specific.
3. drivers/pci/hotplug/rpaphp_core.c
4. drivers/pci/hotplug/rpaphp_pci.c
In these two cases new devices have been hotplugged and FW has
provided new DT nodes for each. These are the only two cases where
the EEH we might have new PCI device nodes in the DT so these are
the only two cases where the early EEH probe needs to be done.
We can move the calls to eeh_add_device_tree_early() to the locations where
it's needed and remove it from the generic path. This is preparation for
making the early EEH probe pseries specific.
Reviewed-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200306073904.4737-3-oohall@gmail.com
On pseries and PowerNV pcibios_bus_add_device() calls eeh_add_device_late()
so there's no need to do a separate tree traversal to bind the eeh_dev and
pci_dev together setting up the PHB at boot. As a result we can remove
eeh_add_device_tree_late().
Reviewed-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200306073904.4737-2-oohall@gmail.com
Move creating the EEH specific sysfs files into eeh_add_device_late()
rather than being open-coded all over the place. Calling the function is
generally done immediately after calling eeh_add_device_late() anyway. This
is also a correctness fix since currently the sysfs files will be added
even if the EEH probe happens to fail.
Similarly, on pseries we currently add the sysfs files before calling
eeh_add_device_late(). This is flat-out broken since the sysfs files
require the pci_dev->dev.archdata.edev pointer to be set, and that is done
in eeh_add_device_late().
Reviewed-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200306073904.4737-1-oohall@gmail.com
The previous commit reduced the amount of code that is run before we
setup a paca. However there are still a few remaining functions that
run with no paca, or worse, with an arbitrary value in r13 that will
be used as a paca pointer.
In particular the stack protector canary is stored in the paca, so if
stack protector is activated for any of these functions we will read
the stack canary from wherever r13 points. If r13 happens to point
outside of memory we will get a machine check / checkstop.
For example if we modify initialise_paca() to trigger stack
protection, and then boot in the mambo simulator with r13 poisoned in
skiboot before calling the kernel:
DEBUG: 19952232: (19952232): INSTRUCTION: PC=0xC0000000191FC1E8: [0x3C4C006D]: addis r2,r12,0x6D [fetch]
DEBUG: 19952236: (19952236): INSTRUCTION: PC=0xC00000001807EAD8: [0x7D8802A6]: mflr r12 [fetch]
FATAL ERROR: 19952276: (19952276): Check Stop for 0:0: Machine Check with ME bit of MSR off
DEBUG: 19952276: (19952276): INSTRUCTION: PC=0xC0000000191FCA7C: [0xE90D0CF8]: ld r8,0xCF8(r13) [Instruction Failed]
INFO: 19952276: (19952277): ** Execution stopped: Mambo Error, Machine Check Stop, **
systemsim % bt
pc: 0xC0000000191FCA7C initialise_paca+0x54
lr: 0xC0000000191FC22C early_setup+0x44
stack:0x00000000198CBED0 0x0 +0x0
stack:0x00000000198CBF00 0xC0000000191FC22C early_setup+0x44
stack:0x00000000198CBF90 0x1801C968 +0x1801C968
So annotate the relevant functions to ensure stack protection is never
enabled for them.
Fixes: 06ec27aea9 ("powerpc/64: add stack protector support")
Cc: stable@vger.kernel.org # v4.20+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200320032116.1024773-2-mpe@ellerman.id.au
Currently we set up the paca after parsing the device tree for CPU
features. Prior to that, r13 contains random data, which means there
is random data in r13 while we're running the generic dt parsing code.
This random data varies depending on whether we boot through a vmlinux
or a zImage: for the vmlinux case it's usually around zero, but for
zImages we see random values like 912a72603d420015.
This is poor practice, and can also lead to difficult-to-debug
crashes. For example, when kcov is enabled, the kcov instrumentation
attempts to read preempt_count out of the current task, which goes via
the paca. This then crashes in the zImage case.
Similarly stack protector can cause crashes if r13 is bogus, by
reading from the stack canary in the paca.
To resolve this:
- move the paca setup to before the CPU feature parsing.
- because we no longer have access to CPU feature flags in paca
setup, change the HV feature test in the paca setup path to consider
the actual value of the MSR rather than the CPU feature.
Translations get switched on once we leave early_setup, so I think
we'd already catch any other cases where the paca or task aren't set
up.
Boot tested on a P9 guest and host.
Fixes: fb0b0a73b2 ("powerpc: Enable kcov")
Fixes: 06ec27aea9 ("powerpc/64: add stack protector support")
Cc: stable@vger.kernel.org # v4.20+
Reviewed-by: Andrew Donnellan <ajd@linux.ibm.com>
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Daniel Axtens <dja@axtens.net>
[mpe: Reword comments & change log a bit to mention stack protector]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200320032116.1024773-1-mpe@ellerman.id.au
H_PAGE_THP_HUGE is used to differentiate between a THP hugepage and
hugetlb hugepage entries. The difference is WRT how we handle hash
fault on these address. THP address enables MPSS in segments. We want
to manage devmap hugepage entries similar to THP pt entries. Hence use
H_PAGE_THP_HUGE for devmap huge PTE entries.
With current code while handling hash PTE fault, we do set is_thp =
true when finding devmap PTE huge PTE entries.
Current code also does the below sequence we setting up huge devmap
entries.
entry = pmd_mkhuge(pfn_t_pmd(pfn, prot));
if (pfn_t_devmap(pfn))
entry = pmd_mkdevmap(entry);
In that case we would find both H_PAGE_THP_HUGE and PAGE_DEVMAP set
for huge devmap PTE entries. This results in false positive error like
below.
kernel BUG at /home/kvaneesh/src/linux/mm/memory.c:4321!
Oops: Exception in kernel mode, sig: 5 [#1]
LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA pSeries
Modules linked in:
CPU: 56 PID: 67996 Comm: t_mmap_dio Not tainted 5.6.0-rc4-59640-g371c804dedbc #128
....
NIP [c00000000044c9e4] __follow_pte_pmd+0x264/0x900
LR [c0000000005d45f8] dax_writeback_one+0x1a8/0x740
Call Trace:
str_spec.74809+0x22ffb4/0x2d116c (unreliable)
dax_writeback_one+0x1a8/0x740
dax_writeback_mapping_range+0x26c/0x700
ext4_dax_writepages+0x150/0x5a0
do_writepages+0x68/0x180
__filemap_fdatawrite_range+0x138/0x180
file_write_and_wait_range+0xa4/0x110
ext4_sync_file+0x370/0x6e0
vfs_fsync_range+0x70/0xf0
sys_msync+0x220/0x2e0
system_call+0x5c/0x68
This is because our pmd_trans_huge check doesn't exclude _PAGE_DEVMAP.
To make this all consistent, update pmd_mkdevmap to set
H_PAGE_THP_HUGE and pmd_trans_huge check now excludes _PAGE_DEVMAP
correctly.
Fixes: ebd3119793 ("powerpc/mm: Add devmap support for ppc64")
Cc: stable@vger.kernel.org # v4.13+
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200313094842.351830-1-aneesh.kumar@linux.ibm.com
Reorder Linux PTE bits to (almost) match Hash PTE bits.
RW Kernel : PP = 00
RO Kernel : PP = 00
RW User : PP = 01
RO User : PP = 11
So naturally, we should have
_PAGE_USER = 0x001
_PAGE_RW = 0x002
Today 0x001 and 0x002 and _PAGE_PRESENT and _PAGE_HASHPTE which
both are software only bits.
Switch _PAGE_USER and _PAGE_PRESET
Switch _PAGE_RW and _PAGE_HASHPTE
This allows to remove a few insns.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/c4d6c18a7f8d9d3b899bc492f55fbc40ef38896a.1583861325.git.christophe.leroy@c-s.fr
At the time being we have something like
if (something) {
p = get();
if (p) {
if (something_wrong)
goto out;
...
return;
} else if (a != b) {
if (some_error)
goto out;
...
}
goto out;
}
p = get();
if (!p) {
if (a != b) {
if (some_error)
goto out;
...
}
goto out;
}
This is similar to
p = get();
if (!p) {
if (a != b) {
if (some_error)
goto out;
...
}
goto out;
}
if (something) {
if (something_wrong)
goto out;
...
return;
}
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Reflow the comment that was moved]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/07a17425743600460ce35fa9432d42487a825583.1582099499.git.christophe.leroy@c-s.fr
We currently have two section mismatch warnings:
The function __boot_from_prom() references
the function __init prom_init().
The function start_here_common() references
the function __init start_kernel().
The warnings are correct, we do have branches from non-init code into
init code, which is freed after boot. But we don't expect to ever
execute any of that early boot code after boot, if we did that would
be a bug. In particular calling into OF after boot would be fatal
because OF is no longer resident.
So for now fix the warnings by marking the relevant functions as
__REF, which puts them in the ".ref.text" section.
This causes some reordering of the functions in the final link:
@@ -217,10 +217,9 @@
c00000000000b088 t generic_secondary_common_init
c00000000000b124 t __mmu_off
c00000000000b14c t __start_initialization_multiplatform
-c00000000000b1ac t __boot_from_prom
-c00000000000b1ec t __after_prom_start
-c00000000000b260 t p_end
-c00000000000b27c T copy_and_flush
+c00000000000b1ac t __after_prom_start
+c00000000000b220 t p_end
+c00000000000b23c T copy_and_flush
c00000000000b300 T __secondary_start
c00000000000b300 t copy_to_here
c00000000000b344 t start_secondary_prolog
@@ -228,8 +227,9 @@
c00000000000b36c t enable_64b_mode
c00000000000b388 T relative_toc
c00000000000b3a8 t p_toc
-c00000000000b3b0 t start_here_common
-c00000000000b3d0 t start_here_multiplatform
+c00000000000b3b0 t __boot_from_prom
+c00000000000b3f0 t start_here_multiplatform
+c00000000000b480 t start_here_common
c00000000000b880 T system_call_common
c00000000000b974 t system_call
c00000000000b9dc t system_call_exit
In particular __boot_from_prom moves after copy_to_here, which means
it's not copied to zero in the first stage of copy of the kernel to
zero.
But that's OK, because we only call __boot_from_prom before we do the
copy, so it makes no difference when it's copied. The call sequence
is:
__start
-> __start_initialization_multiplatform
-> __boot_from_prom
-> __start
-> __start_initialization_multiplatform
-> __after_prom_start
-> copy_and_flush
-> copy_and_flush (relocated to 0)
-> start_here_multiplatform
-> early_setup
Reported-by: Mauricio Faria de Oliveira <mauricfo@linux.ibm.com>
Reported-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200225031328.14676-1-mpe@ellerman.id.au
In xmon we have two variables that are used by the dump commands.
There's ndump which is the number of bytes to dump using 'd', and
nidump which is the number of instructions to dump using 'di'.
ndump starts as 64 and nidump starts as 16, but both can be set by the
user.
It's fairly common to be pasting addresses into xmon when trying to
debug something, and if you inadvertently double paste an address like
so:
0:mon> di c000000002101f6c c000000002101f6c
The second value is interpreted as the number of instructions to dump.
Luckily it doesn't dump 13 quintrillion instructions, the value is
limited to MAX_DUMP (128K). But as each instruction is dumped on a
single line, that's still a lot of output. If you're on a slow console
that can take multiple minutes to print. If you were "just popping in
and out of xmon quickly before the RCU/hardlockup detector fires" you
are now having a bad day.
Things are not as bad with 'd' because we print 16 bytes per line, so
it's fewer lines. But it's still quite a lot.
So shrink the maximum for 'd' to 64K (one page), which is 4096 lines.
For 'di' add a new limit which is the above / 4 - because instructions
are 4 bytes, meaning again we can dump one page.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200219110007.31195-1-mpe@ellerman.id.au
The "os-term" RTAS calls has one argument with a message address of OS
termination cause. rtas_os_term() already passes it but the recently
added prom_init's version of that missed it; it also does not fill
args correctly.
This passes the message address and initializes the number of arguments.
Fixes: 6a9c930bd7 ("powerpc/prom_init: Add the ESM call to prom_init")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200312074404.87293-1-aik@ozlabs.ru
ld instruction should have 14 bit immediate field (DS) concatenated
with 0b00 on the right, encode it accordingly. Introduce macro
`IMM_DS()` to encode DS form instructions with 14 bit immediate field.
Fixes: 4ceae137bd ("powerpc: emulate_step() tests for load/store instructions")
Reviewed-by: Sandipan Das <sandipan@linux.ibm.com>
Signed-off-by: Balamuruhan S <bala24@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200311102405.392263-1-bala24@linux.ibm.com
The expectation is that when calling of_read_drc_info_cell()
repeatedly to parse multiple drc-info records that the in/out curval
parameter points at the start of the next record on return. However,
the current behavior has curval still pointing at the final value of
the record just parsed. The result of which is that if the
ibm,drc-info property contains multiple properties the parsed value
of the drc_type for any record after the first has the power_domain
value of the previous record appended to the type string.
eg: observed the following 0xffffffff prepended to PHB
drc-info: type: \xff\xff\xff\xffPHB, prefix: PHB , index_start: 0x20000001
drc-info: suffix_start: 1, sequential_elems: 3072, sequential_inc: 1
drc-info: power-domain: 0xffffffff, last_index: 0x20000c00
In practice PHBs are the only type of connector in the ibm,drc-info
property that has multiple records. So, it breaks PHB hotplug, but by
chance not PCI, CPU, slot, or memory because they happen to only ever
be a single record.
Fix by incrementing curval past the power_domain value to point at
drc_type string of next record.
Fixes: e83636ac33 ("pseries/drc-info: Search DRC properties for CPU indexes")
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Acked-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200307024547.5748-1-tyreld@linux.ibm.com
Currently you can enable PPC_KUAP_DEBUG when PPC_KUAP is disabled,
even though the former has not effect without the latter.
Fix it so that PPC_KUAP_DEBUG can only be enabled when PPC_KUAP is
enabled, not when the platform could support KUAP (PPC_HAVE_KUAP).
Fixes: 890274c2dc ("powerpc/64s: Implement KUAP for Radix MMU")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200301111738.22497-1-mpe@ellerman.id.au
We should be checking that the instruction was stepped *and* that the
target register has the right value.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
[mpe: Write change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200226055302.1577954-1-npiggin@gmail.com
When building with ppc64_defconfig, the compiler reports
that these 2 variables are not used:
warning: unused variable 'core99_l2_cache' [-Wunused-variable]
warning: unused variable 'core99_l3_cache' [-Wunused-variable]
They are only used when CONFIG_PPC64 is not defined. Move
them into a section which does the same macro check.
Reported-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Ilie Halip <ilie.halip@gmail.com>
[mpe: Move them into core99_init_caches() which is their only user]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190920153951.25762-1-ilie.halip@gmail.com
In the current implementation, the call to loadcam_multi() is wrapped
between switch_to_as1() and restore_to_as0() calls so, when it tries
to create its own temporary AS=1 TLB1 entry, it ends up duplicating
the existing one created by switch_to_as1(). Add a check to skip
creating the temporary entry if already running in AS=1.
Fixes: d9e1831a42 ("powerpc/85xx: Load all early TLB entries at once")
Cc: stable@vger.kernel.org # v4.4+
Signed-off-by: Laurentiu Tudor <laurentiu.tudor@nxp.com>
Acked-by: Scott Wood <oss@buserror.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200123111914.2565-1-laurentiu.tudor@nxp.com
The original 2005 patch that introduced the powerpc vdso, pre-git
("ppc64: Implement a vDSO and use it for signal trampoline") notes that:
... symbols exposed by the vDSO aren't "normal" function symbols, apps
can't be expected to link against them directly, the vDSO's are both
seen as if they were linked at 0 and the symbols just contain offsets
to the various functions. This is done on purpose to avoid a
relocation step (ppc64 functions normally have descriptors with abs
addresses in them). When glibc uses those functions, it's expected to
use it's own trampolines that know how to reach them.
Despite that explanation, there remains dead #ifdef
VDS64_HAS_DESCRIPTORS code-blocks that provide alternate function
definitions that setup function descriptors.
Since VDS64_HAS_DESCRIPTORS has been unused for all these years, we
might as well finally remove it from the codebase.
Signed-off-by: Joe Lawrence <joe.lawrence@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200224211848.26087-1-joe.lawrence@redhat.com
Commit 2efc7c085f ("powerpc/32: drop get_pteptr()"),
replaced get_pteptr() by virt_to_kpte(). But virt_to_kpte() lacks a
NULL pmd check and returns an invalid non NULL pointer when there
is no page table.
Reported-by: Nick Desaulniers <ndesaulniers@google.com>
Fixes: 2efc7c085f ("powerpc/32: drop get_pteptr()")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Tested-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b1177cdfc6af74a3e277bba5d9e708c4b3315ebe.1583575707.git.christophe.leroy@c-s.fr
Stefan reported a strange kernel fault which turned out to be due to a
missing KUAP disable in flush_coherent_icache() called from
flush_icache_range().
The fault looks like:
Kernel attempted to access user page (7fffc30d9c00) - exploit attempt? (uid: 1009)
BUG: Unable to handle kernel data access on read at 0x7fffc30d9c00
Faulting instruction address: 0xc00000000007232c
Oops: Kernel access of bad area, sig: 11 [#1]
LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV
CPU: 35 PID: 5886 Comm: sigtramp Not tainted 5.6.0-rc2-gcc-8.2.0-00003-gfc37a1632d40 #79
NIP: c00000000007232c LR: c00000000003b7fc CTR: 0000000000000000
REGS: c000001e11093940 TRAP: 0300 Not tainted (5.6.0-rc2-gcc-8.2.0-00003-gfc37a1632d40)
MSR: 900000000280b033 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 28000884 XER: 00000000
CFAR: c0000000000722fc DAR: 00007fffc30d9c00 DSISR: 08000000 IRQMASK: 0
GPR00: c00000000003b7fc c000001e11093bd0 c0000000023ac200 00007fffc30d9c00
GPR04: 00007fffc30d9c18 0000000000000000 c000001e11093bd4 0000000000000000
GPR08: 0000000000000000 0000000000000001 0000000000000000 c000001e1104ed80
GPR12: 0000000000000000 c000001fff6ab380 c0000000016be2d0 4000000000000000
GPR16: c000000000000000 bfffffffffffffff 0000000000000000 0000000000000000
GPR20: 00007fffc30d9c00 00007fffc30d8f58 00007fffc30d9c18 00007fffc30d9c20
GPR24: 00007fffc30d9c18 0000000000000000 c000001e11093d90 c000001e1104ed80
GPR28: c000001e11093e90 0000000000000000 c0000000023d9d18 00007fffc30d9c00
NIP flush_icache_range+0x5c/0x80
LR handle_rt_signal64+0x95c/0xc2c
Call Trace:
0xc000001e11093d90 (unreliable)
handle_rt_signal64+0x93c/0xc2c
do_notify_resume+0x310/0x430
ret_from_except_lite+0x70/0x74
Instruction dump:
409e002c 7c0802a6 3c62ff31 3863f6a0 f8010080 48195fed 60000000 48fe4c8d
60000000 e8010080 7c0803a6 7c0004ac <7c00ffac> 7c0004ac 4c00012c 38210070
This path through handle_rt_signal64() to setup_trampoline() and
flush_icache_range() is only triggered by 64-bit processes that have
unmapped their VDSO, which is rare.
flush_icache_range() takes a range of addresses to flush. In
flush_coherent_icache() we implement an optimisation for CPUs where we
know we don't actually have to flush the whole range, we just need to
do a single icbi.
However we still execute the icbi on the user address of the start of
the range we're flushing. On CPUs that also implement KUAP (Power9)
that leads to the spurious fault above.
We should be able to pass any address, including a kernel address, to
the icbi on these CPUs, which would avoid any interaction with KUAP.
But I don't want to make that change in a bug fix, just in case it
surfaces some strange behaviour on some CPU.
So for now just disable KUAP around the icbi. Note the icbi is treated
as a load, so we allow read access, not write as you'd expect.
Fixes: 890274c2dc ("powerpc/64s: Implement KUAP for Radix MMU")
Cc: stable@vger.kernel.org # v5.2+
Reported-by: Stefan Berger <stefanb@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200303235708.26004-1-mpe@ellerman.id.au
With commit ("powerpc/numa: Early request for home node associativity"),
commit 2ea6263068 ("powerpc/topology: Get topology for shared
processors at boot") which was requesting home node associativity
becomes redundant.
Hence remove the late request for home node associativity.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reported-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Reviewed-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200129135301.24739-6-srikar@linux.vnet.ibm.com
Currently the kernel detects if its running on a shared lpar platform
and requests home node associativity before the scheduler sched_domains
are setup. However between the time NUMA setup is initialized and the
request for home node associativity, workqueue initializes its per node
cpumask. The per node workqueue possible cpumask may turn invalid
after home node associativity resulting in weird situations like
workqueue possible cpumask being a subset of workqueue online cpumask.
This can be fixed by requesting home node associativity earlier just
before NUMA setup. However at the NUMA setup time, kernel may not be in
a position to detect if its running on a shared lpar platform. So
request for home node associativity and if the request fails, fallback
on the device tree property.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reported-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200129135301.24739-5-srikar@linux.vnet.ibm.com
All the sibling threads of a core have to be part of the same node.
To ensure that all the sibling threads map to the same node, always
lookup/update the cpu-to-node map of the first thread in the core.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reported-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Reviewed-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200129135301.24739-4-srikar@linux.vnet.ibm.com
Currently code handles H_FUNCTION, H_SUCCESS, H_HARDWARE return codes.
However hcall_vphn can return other return codes. Now it also handles
H_PARAMETER return code. Also the rest return codes are handled under the
default case.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reported-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Reviewed-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200129135301.24739-3-srikar@linux.vnet.ibm.com
There is no value in unpacking associativity, if
H_HOME_NODE_ASSOCIATIVITY hcall has returned an error.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reported-by: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Reviewed-by: Nathan Lynch <nathanl@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200129135301.24739-2-srikar@linux.vnet.ibm.com
package_id is to match cores that are part of the same chip. On
PowerNV machines, package_id defaults to chip_id. However ibm,chip_id
property is not present in device-tree of PowerVM LPARs. Hence lscpu
output shows one core per socket and multiple cores.
To overcome this, use nid as the package_id on PowerVM LPARs.
Before the patch:
Architecture: ppc64le
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Thread(s) per core: 8
Core(s) per socket: 1 <----------------------
Socket(s): 16 <----------------------
NUMA node(s): 2
Model: 2.2 (pvr 004e 0202)
Model name: POWER9 (architected), altivec supported
Hypervisor vendor: pHyp
Virtualization type: para
L1d cache: 32K
L1i cache: 32K
L2 cache: 512K
L3 cache: 10240K
NUMA node0 CPU(s): 0-63
NUMA node1 CPU(s): 64-127
#
# cat /sys/devices/system/cpu/cpu0/topology/physical_package_id
-1
After the patch:
Architecture: ppc64le
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Thread(s) per core: 8 <---------------------
Core(s) per socket: 8 <---------------------
Socket(s): 2
NUMA node(s): 2
Model: 2.2 (pvr 004e 0202)
Model name: POWER9 (architected), altivec supported
Hypervisor vendor: pHyp
Virtualization type: para
L1d cache: 32K
L1i cache: 32K
L2 cache: 512K
L3 cache: 10240K
NUMA node0 CPU(s): 0-63
NUMA node1 CPU(s): 64-127
#
# cat /sys/devices/system/cpu/cpu0/topology/physical_package_id
0
Now lscpu output is more in line with the system configuration.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
[mpe: Use pkg_id instead of ppid, tweak change log and comment]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200129135121.24617-1-srikar@linux.vnet.ibm.com
Until commit 7306e83ccf ("powerpc: Don't use CURRENT_THREAD_INFO to
find the stack"), the current stack base address was obtained by
calling current_thread_info(). That inline function was simply masking
out the value of r1.
In that commit, it was changed to using current_stack_pointer() (since
renamed current_stack_frame()), which is a heavier function as it is
an outline assembly function which cannot be inlined and which reads
the content of the stack at 0(r1).
Convert to using current_stack_pointer for geting r1 and masking out
its value to obtain the base address of the stack pointer as before.
Fixes: 7306e83ccf ("powerpc: Don't use CURRENT_THREAD_INFO to find the stack")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200220115141.2707-5-mpe@ellerman.id.au
Instead of #ifdef, use IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW).
This enable GCC to check for code validity even when the option
is not selected.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200220115141.2707-4-mpe@ellerman.id.au
The purpose of check_stack_overflow() is to verify that the stack has
not overflowed.
To really know whether the stack pointer is still within boundaries,
the check must be done directly on the value of r1.
So use current_stack_pointer, which returns the current value of r1,
rather than current_stack_frame() which causes a frame to be created
and then returns that value.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200220115141.2707-3-mpe@ellerman.id.au
current_stack_frame() doesn't return the stack pointer, but the
caller's stack frame. See commit bfe9a2cfe9 ("powerpc: Reimplement
__get_SP() as a function not a define") and commit
acf620ecf5 ("powerpc: Rename __get_SP() to current_stack_pointer()")
for details.
In some cases this is overkill or incorrect, as it doesn't return the
current value of r1.
So add a current_stack_pointer register global to get the value of r1
directly.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: Split out of other patch, tweak change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200220115141.2707-2-mpe@ellerman.id.au
current_stack_pointer(), which was called __get_SP(), used to just
return the value in r1.
But that caused problems in some cases, so it was turned into a
function in commit bfe9a2cfe9 ("powerpc: Reimplement __get_SP() as a
function not a define").
Because it's a function in a separate compilation unit to all its
callers, it has the effect of causing a stack frame to be created, and
then returns the address of that frame. This is good in some cases
like those described in the above commit, but in other cases it's
overkill, we just need to know what stack page we're on.
On some other arches current_stack_pointer is just a register global
giving the stack pointer, and we'd like to do that too. So rename our
current_stack_pointer() to current_stack_frame() to make that
possible.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Christophe Leroy <christophe.leroy@c-s.fr>
Link: https://lore.kernel.org/r/20200220115141.2707-1-mpe@ellerman.id.au
Many of the performance monitoring unit (PMU) SPRs are
exposed in the sysfs. This may not be a desirable since
"perf" API is the primary interface to program PMU and
collect counter data in the system. But that said, we
cant remove these sysfs files since we dont whether
anyone/anything is using them.
So the patch adds a new CONFIG option 'CONFIG_PMU_SYSFS'
(user selectable) to be used in sysfs file creation for
PMU SPRs. New option by default is disabled, but can be
enabled if user needs it.
Tested this patch behaviour in powernv and pseries machines.
Patch is also tested for pmac32_defconfig.
Signed-off-by: Kajol Jain <kjain@linux.ibm.com>
Tested-by: Nageswara R Sastry <nasastry@in.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200214080606.26872-2-kjain@linux.ibm.com
An attempt to refactor the current sysfs.c file.
To start with a big chuck of macro #defines and dscr
functions are moved to start of the file. Secondly,
HAS_ #define macros are cleanup based on CONFIG_ options
Finally new HAS_ macro added:
1. HAS_PPC_PA6T (for PA6T) to separate out non-PMU SPRs.
2. HAS_PPC_PMC56 to separate out PMC SPR's from HAS_PPC_PMC_CLASSIC
which come under CONFIG_PPC64.
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200214080606.26872-1-kjain@linux.ibm.com
Add a way to manually invoke a fast-reboot rather than setting the NVRAM
flag. The idea is to allow userspace to invoke a fast-reboot using the
optional string argument to the reboot() system call, or using the xmon
zr command so we don't need to leave around a persistent changes on
a system to use the feature.
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200217024833.30580-2-oohall@gmail.com
Treat an empty reboot cmd string the same as a NULL string. This squashes a
spurious unsupported reboot message that sometimes gets out when using
xmon.
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200217024833.30580-1-oohall@gmail.com
Some of our phony targets are not marked as such. This can lead to
confusing errors, eg:
$ make clean
$ touch install
$ make install
make: 'install' is up to date.
$
Fix it by adding them to the PHONY variable which is marked phony in
the top-level Makefile, or in scripts/Makefile.build for the boot
Makefile.
Suggested-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200219000434.15872-1-mpe@ellerman.id.au
On PPC32, pte_offset_map() does a kmap_atomic() in order to support
page tables allocated in high memory, just like ARM and x86/32.
But since at least 2008 and commit 8054a3428f ("powerpc: Remove dead
CONFIG_HIGHPTE"), page tables are never allocated in high memory.
When the page is in low mem, kmap_atomic() just returns the page
address but still disable preemption and pagefault. And it is
not an inlined function, so we suffer function call for no reason.
Make pte_offset_map() the same as pte_offset_kernel() and make
pte_unmap() void, in the same way as PPC64 which doesn't have HIGHMEM.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/03c97f0f6b3790d164822563be80f2fd4713a955.1581932480.git.christophe.leroy@c-s.fr
The last jump to free_exit in mm_iommu_do_alloc() happens after page
pointers in struct mm_iommu_table_group_mem_t were already converted to
physical addresses. Thus calling put_page() on these physical addresses
will likely crash.
This moves the loop which calculates the pageshift and converts page
struct pointers to physical addresses later after the point when
we cannot fail; thus eliminating the need to convert pointers back.
Fixes: eb9d7a62c3 ("powerpc/mm_iommu: Fix potential deadlock")
Reported-by: Jan Kara <jack@suse.cz>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20191223060351.26359-1-aik@ozlabs.ru
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200209105901.1620958-6-gregkh@linuxfoundation.org
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200209105901.1620958-5-gregkh@linuxfoundation.org
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200209105901.1620958-4-gregkh@linuxfoundation.org
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200209105901.1620958-3-gregkh@linuxfoundation.org
