| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Third time's the charm, I hope?
Fixes: d3116756a710 ("drm/ttm: rename bo->mem and make it a pointer")
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3837
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 695c2c745e5dff201b75da8a1d237ce403600d04)
Cc: stable@vger.kernel.org
|
|
The VM pointer might already be outdated when that function is called.
Use the PASID instead to gather the information instead.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 57f812d171af4ba233d3ed7c94dfa5b8e92dcc04)
Cc: stable@vger.kernel.org
|
|
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8f2cd1067afe68372a1723e05e19b68ed187676a)
Cc: stable@vger.kernel.org
|
|
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit f1fd1d0f40272948aa6ab82a3a82ecbbc76dff53)
Cc: stable@vger.kernel.org
|
|
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 63bfd24088b42c6f55c2096bfc41b50213d419b2)
Cc: stable@vger.kernel.org
|
|
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 2c8eeaaa0fe5841ccf07a0eb51b1426f34ef39f7)
Cc: stable@vger.kernel.org
|
|
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 22b9555bc90df22b585bdd1f161b61584b13af51)
Cc: stable@vger.kernel.org
|
|
Since 2320c9e6a768 ("drm/sched: memset() 'job' in drm_sched_job_init()")
accessing job->base.sched can produce unexpected results as the initialisation
of (*job)->base.sched done in amdgpu_job_alloc is overwritten by the
memset.
This commit fixes an issue when a CS would fail validation and would
be rejected after job->num_ibs is incremented. In this case,
amdgpu_ib_free(ring->adev, ...) will be called, which would crash the
machine because the ring value is bogus.
To fix this, pass a NULL pointer to amdgpu_ib_free(): we can do this
because the device is actually not used in this function.
The next commit will remove the ring argument completely.
Fixes: 2320c9e6a768 ("drm/sched: memset() 'job' in drm_sched_job_init()")
Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 2ae520cb12831d264ceb97c61f72c59d33c0dbd7)
|
|
If the kernel hasn't been compiled with PCIe hotplug support this
can lead to problems with dGPUs that use BOCO because they effectively
drop off the bus.
To prevent issues, disable BOCO support when compiled without PCIe hotplug.
Reported-by: Gabriel Marcano <gabemarcano@yahoo.com>
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/1707#note_2696862
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://lore.kernel.org/r/20241211155601.3585256-1-superm1@kernel.org
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 1ad5bdc28bafa66db0f041cc6cdd278a80426aae)
|
|
That pool implementation doesn't really work: if the krealloc happens to
move the memory and return another address, the entries in the xarray
become invalid, leading to use-after-free later:
BUG: KASAN: slab-use-after-free in xe_reg_sr_apply_mmio+0x570/0x760 [xe]
Read of size 4 at addr ffff8881244b2590 by task modprobe/2753
Allocated by task 2753:
kasan_save_stack+0x39/0x70
kasan_save_track+0x14/0x40
kasan_save_alloc_info+0x37/0x60
__kasan_kmalloc+0xc3/0xd0
__kmalloc_node_track_caller_noprof+0x200/0x6d0
krealloc_noprof+0x229/0x380
Simplify the code to fix the bug. A better pooling strategy may be added
back later if needed.
Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241209232739.147417-2-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
(cherry picked from commit e5283bd4dfecbd3335f43b62a68e24dae23f59e4)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
|
|
Invalidation_fence_init takes a PM reference, which is released in its
_fini counterpart, so we need to make sure that the latter is called,
even if the fence is in an error state.
Since we already have a function that calls _fini() and signals the
fence in the tlb inval code, we can expose that and call it from the PT
code.
Fixes: f002702290fc ("drm/xe: Hold a PM ref when GT TLB invalidations are inflight")
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: <stable@vger.kernel.org> # v6.11+
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241206015022.1567113-1-daniele.ceraolospurio@intel.com
(cherry picked from commit 65338639b79ce88aef5263cd518cde570a3c7c8e)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
|
|
Running coccinelle spatch gave the following warning:
./drivers/gpu/drm/xe/tests/xe_migrate.c:226:5-11: inconsistent IS_ERR
and PTR_ERR on line 228.
The code reports PTR_ERR(pt) when IS_ERR(tiny) is checked:
→ 211 pt = xe_bo_create_pin_map(xe, tile, m->q->vm, XE_PAGE_SIZE,
212 ttm_bo_type_kernel,
213 XE_BO_FLAG_VRAM_IF_DGFX(tile) |
214 XE_BO_FLAG_PINNED);
215 if (IS_ERR(pt)) {
216 KUNIT_FAIL(test, "Failed to allocate fake pt: %li\n",
217 PTR_ERR(pt));
218 goto free_big;
219 }
220
221 tiny = xe_bo_create_pin_map(xe, tile, m->q->vm,
→ 222 2 * SZ_4K,
223 ttm_bo_type_kernel,
224 XE_BO_FLAG_VRAM_IF_DGFX(tile) |
225 XE_BO_FLAG_PINNED);
→ 226 if (IS_ERR(tiny)) {
→ 227 KUNIT_FAIL(test, "Failed to allocate fake pt: %li\n",
→ 228 PTR_ERR(pt));
229 goto free_pt;
230 }
Now, the IS_ERR(tiny) and the corresponding PTR_ERR(pt) do not match.
Returning PTR_ERR(tiny), as the last failed function call, seems logical.
Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Mirsad Todorovac <mtodorovac69@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241121212057.1526634-2-mtodorovac69@gmail.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit cb57c75098c1c449a007ba301f9073f96febaaa9)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
|
|
When using MES creating a pdd will require talking to the GPU to
setup the relevant context. The code here forgot to wake up the GPU
in case it was in suspend, this causes KVM to EFAULT for passthrough
GPU for example. This issue can be masked if the GPU was woken up by
other things (e.g. opening the KMS node) first and have not yet gone to sleep.
v4: do the allocation of proc_ctx_bo in a lazy fashion
when the first queue is created in a process (Felix)
Signed-off-by: Jesse Zhang <jesse.zhang@amd.com>
Reviewed-by: Yunxiang Li <Yunxiang.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
|
|
Emitting the cleaner shader must come after the check if a VM switch is
necessary or not.
Otherwise we will emit the cleaner shader every time and not just when it is
necessary because we switched between applications.
This can otherwise crash on gang submit and probably decreases performance
quite a bit.
v2: squash in fix from Srini (Alex)
Signed-off-by: Christian König <christian.koenig@amd.com>
Fixes: ee7a846ea27b ("drm/amdgpu: Emit cleaner shader at end of IB submission")
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
|
|
ISP hw_init is not called with the recent changes related
to hw init levels. AMDGPU_INIT_LEVEL_DEFAULT is ignoring
the ISP IP block as AMDGPU_IP_BLK_MASK_ALL is derived using
incorrect max number of IP blocks.
Update AMDGPU_IP_BLK_MASK_ALL to use AMD_IP_BLOCK_TYPE_NUM
instead of AMDGPU_MAX_IP_NUM to fix the issue.
Fixes: 14f2fe34f5c6 ("drm/amdgpu: Add init levels")
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Pratap Nirujogi <pratap.nirujogi@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
This information is not available in ip discovery table.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Reviewed-by: David Belanger <david.belanger@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
|
|
This information is not available in ip discovery table.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Reviewed-by: David Belanger <david.belanger@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
|
|
In the function pqm_uninit there is a call-assignment of "pdd =
kfd_get_process_device_data" which could be null, and this value was
later dereferenced without checking.
Fixes: fb91065851cd ("drm/amdkfd: Refactor queue wptr_bo GART mapping")
Signed-off-by: Andrew Martin <Andrew.Martin@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
|
|
Replace "slab_priorities" with "slab_dependencies" in the error handler
to avoid memory leak.
Fixes: 32eb6bcfdda9 ("drm/i915: Make request allocation caches global")
Cc: <stable@vger.kernel.org> # v5.2+
Signed-off-by: Jiasheng Jiang <jiashengjiangcool@outlook.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241127201042.29620-1-jiashengjiangcool@gmail.com
(cherry picked from commit 9bc5e7dc694d3112bbf0fa4c46ef0fa0f114937a)
Signed-off-by: Tvrtko Ursulin <tursulin@ursulin.net>
|
|
When the intel_context structure contains NULL,
it raises a NULL pointer dereference error in drm_info().
Fixes: e8a3319c31a1 ("drm/i915: Allow error capture without a request")
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/12309
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: <stable@vger.kernel.org> # v6.3+
Signed-off-by: Eugene Kobyak <eugene.kobyak@intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/xmsgfynkhycw3cf56akp4he2ffg44vuratocsysaowbsnhutzi@augnqbm777at
(cherry picked from commit 754302a5bc1bd8fd3b7d85c168b0a1af6d4bba4d)
Signed-off-by: Tvrtko Ursulin <tursulin@ursulin.net>
|
|
DSB LUT register writes vs. palette anti-collision logic
appear to interact in interesting ways:
- posted DSB writes simply vanish into thin air while
anti-collision is active
- non-posted DSB writes actually get blocked by the anti-collision
logic, but unfortunately this ends up hogging the bus for
long enough that unrelated parallel CPU MMIO accesses start
to disappear instead
Even though we are updating the LUT during vblank we aren't
immune to the anti-collision logic because it kicks in briefly
for pipe prefill (initiated at frame start). The safe time
window for performing the LUT update is thus between the
undelayed vblank and frame start. Turns out that with low
enough CDCLK frequency (DSB execution speed depends on CDCLK)
we can exceed that.
As we are currently using non-posted writes for the legacy LUT
updates, in which case we can hit the far more severe failure
mode. The problem is exacerbated by the fact that non-posted
writes are much slower than posted writes (~4x it seems).
To mititage the problem let's switch to using posted DSB
writes for legacy LUT updates (which will involve using the
double write approach to avoid other problems with DSB
vs. legacy LUT writes). Despite writing each register twice
this will in fact make the legacy LUT update faster when
compared to the non-posted write approach, making the
problem less likely to appear. The failure mode is also
less severe.
This isn't the 100% solution we need though. That will involve
estimating how long the LUT update will take, and pushing
frame start and/or delayed vblank forward to guarantee that
the update will have finished by the time the pipe prefill
starts...
Cc: stable@vger.kernel.org
Fixes: 34d8311f4a1c ("drm/i915/dsb: Re-instate DSB for LUT updates")
Fixes: 25ea3411bd23 ("drm/i915/dsb: Use non-posted register writes for legacy LUT")
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/12494
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241120164123.12706-3-ville.syrjala@linux.intel.com
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
(cherry picked from commit 2504a316b35d49522f39cf0dc01830d7c36a9be4)
Signed-off-by: Tvrtko Ursulin <tursulin@ursulin.net>
|
|
Turns out the DSB indexed register write command has
rather significant initial overhead compared to the normal
MMIO write command. Based on some quick experiments on TGL
you have to write the register at least ~5 times for the
indexed write command to come out ahead. If you write the
register less times than that the MMIO write is faster.
So it seems my automagic indexed write logic was a bit
misguided. Go back to the original approach only use
indexed writes for the cases we know will benefit from
it (indexed LUT register updates).
Currently we shouldn't have any cases where this truly
matters (just some rare double writes to the precision
LUT index registers), but we will need to switch the
legacy LUT updates to write each LUT register twice (to
avoid some palette anti-collision logic troubles).
This would be close to the worst case for using indexed
writes (two writes per register, and 256 separate registers).
Using the MMIO write command should shave off around 30%
of the execution time compared to using the indexed write
command.
Cc: stable@vger.kernel.org
Fixes: 34d8311f4a1c ("drm/i915/dsb: Re-instate DSB for LUT updates")
Fixes: 25ea3411bd23 ("drm/i915/dsb: Use non-posted register writes for legacy LUT")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241120164123.12706-2-ville.syrjala@linux.intel.com
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
(cherry picked from commit ecba559a88ab8399a41893d7828caf4dccbeab6c)
Signed-off-by: Tvrtko Ursulin <tursulin@ursulin.net>
|
|
|
|
Since commit 13b25489b6f8 ("kbuild: change working directory to external
module directory with M="), the Debian package build fails if a relative
path is specified with the O= option.
$ make O=build bindeb-pkg
[ snip ]
dpkg-deb: building package 'linux-image-6.13.0-rc1' in '../linux-image-6.13.0-rc1_6.13.0-rc1-6_amd64.deb'.
Rebuilding host programs with x86_64-linux-gnu-gcc...
make[6]: Entering directory '/home/masahiro/linux/build'
/home/masahiro/linux/Makefile:190: *** specified kernel directory "build" does not exist. Stop.
This occurs because the sub_make_done flag is cleared, even though the
working directory is already in the output directory.
Passing KBUILD_OUTPUT=. resolves the issue.
Fixes: 13b25489b6f8 ("kbuild: change working directory to external module directory with M=")
Reported-by: Charlie Jenkins <charlie@rivosinc.com>
Closes: https://lore.kernel.org/all/Z1DnP-GJcfseyrM3@ghost/
Tested-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
|
|
The compiler can fully inline the actual handler function of an interrupt
entry into the .irqentry.text entry point. If such a function contains an
access which has an exception table entry, modpost complains about a
section mismatch:
WARNING: vmlinux.o(__ex_table+0x447c): Section mismatch in reference ...
The relocation at __ex_table+0x447c references section ".irqentry.text"
which is not in the list of authorized sections.
Add .irqentry.text to OTHER_SECTIONS to cure the issue.
Reported-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org # needed for linux-5.4-y
Link: https://lore.kernel.org/all/20241128111844.GE10431@google.com/
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
|
|
When ensuring EFER.AUTOIBRS is set, WARN only on a negative return code
from msr_set_bit(), as '1' is used to indicate the WRMSR was successful
('0' indicates the MSR bit was already set).
Fixes: 8cc68c9c9e92 ("x86/CPU/AMD: Make sure EFER[AIBRSE] is set")
Reported-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/Z1MkNofJjt7Oq0G6@google.com
Closes: https://lore.kernel.org/all/20241205220604.GA2054199@thelio-3990X
|
|
Add more test cases for LPM trie in test_maps:
1) test_lpm_trie_update_flags
It constructs various use cases for BPF_EXIST and BPF_NOEXIST and check
whether the return value of update operation is expected.
2) test_lpm_trie_update_full_maps
It tests the update operations on a full LPM trie map. Adding new node
will fail and overwriting the value of existed node will succeed.
3) test_lpm_trie_iterate_strs and test_lpm_trie_iterate_ints
There two test cases test whether the iteration through get_next_key is
sorted and expected. These two test cases delete the minimal key after
each iteration and check whether next iteration returns the second
minimal key. The only difference between these two test cases is the
former one saves strings in the LPM trie and the latter saves integers.
Without the fix of get_next_key, these two cases will fail as shown
below:
test_lpm_trie_iterate_strs(1091):FAIL:iterate #2 got abc exp abS
test_lpm_trie_iterate_ints(1142):FAIL:iterate #1 got 0x2 exp 0x1
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-10-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Move test_lpm_map.c to map_tests/ to include LPM trie test cases in
regular test_maps run. Most code remains unchanged, including the use of
assert(). Only reduce n_lookups from 64K to 512, which decreases
test_lpm_map runtime from 37s to 0.7s.
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-9-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
After switching from kmalloc() to the bpf memory allocator, there will be
no blocking operation during the update of LPM trie. Therefore, change
trie->lock from spinlock_t to raw_spinlock_t to make LPM trie usable in
atomic context, even on RT kernels.
The max value of prefixlen is 2048. Therefore, update or deletion
operations will find the target after at most 2048 comparisons.
Constructing a test case which updates an element after 2048 comparisons
under a 8 CPU VM, and the average time and the maximal time for such
update operation is about 210us and 900us.
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-8-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Multiple syzbot warnings have been reported. These warnings are mainly
about the lock order between trie->lock and kmalloc()'s internal lock.
See report [1] as an example:
======================================================
WARNING: possible circular locking dependency detected
6.10.0-rc7-syzkaller-00003-g4376e966ecb7 #0 Not tainted
------------------------------------------------------
syz.3.2069/15008 is trying to acquire lock:
ffff88801544e6d8 (&n->list_lock){-.-.}-{2:2}, at: get_partial_node ...
but task is already holding lock:
ffff88802dcc89f8 (&trie->lock){-.-.}-{2:2}, at: trie_update_elem ...
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (&trie->lock){-.-.}-{2:2}:
__raw_spin_lock_irqsave
_raw_spin_lock_irqsave+0x3a/0x60
trie_delete_elem+0xb0/0x820
___bpf_prog_run+0x3e51/0xabd0
__bpf_prog_run32+0xc1/0x100
bpf_dispatcher_nop_func
......
bpf_trace_run2+0x231/0x590
__bpf_trace_contention_end+0xca/0x110
trace_contention_end.constprop.0+0xea/0x170
__pv_queued_spin_lock_slowpath+0x28e/0xcc0
pv_queued_spin_lock_slowpath
queued_spin_lock_slowpath
queued_spin_lock
do_raw_spin_lock+0x210/0x2c0
__raw_spin_lock_irqsave
_raw_spin_lock_irqsave+0x42/0x60
__put_partials+0xc3/0x170
qlink_free
qlist_free_all+0x4e/0x140
kasan_quarantine_reduce+0x192/0x1e0
__kasan_slab_alloc+0x69/0x90
kasan_slab_alloc
slab_post_alloc_hook
slab_alloc_node
kmem_cache_alloc_node_noprof+0x153/0x310
__alloc_skb+0x2b1/0x380
......
-> #0 (&n->list_lock){-.-.}-{2:2}:
check_prev_add
check_prevs_add
validate_chain
__lock_acquire+0x2478/0x3b30
lock_acquire
lock_acquire+0x1b1/0x560
__raw_spin_lock_irqsave
_raw_spin_lock_irqsave+0x3a/0x60
get_partial_node.part.0+0x20/0x350
get_partial_node
get_partial
___slab_alloc+0x65b/0x1870
__slab_alloc.constprop.0+0x56/0xb0
__slab_alloc_node
slab_alloc_node
__do_kmalloc_node
__kmalloc_node_noprof+0x35c/0x440
kmalloc_node_noprof
bpf_map_kmalloc_node+0x98/0x4a0
lpm_trie_node_alloc
trie_update_elem+0x1ef/0xe00
bpf_map_update_value+0x2c1/0x6c0
map_update_elem+0x623/0x910
__sys_bpf+0x90c/0x49a0
...
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&trie->lock);
lock(&n->list_lock);
lock(&trie->lock);
lock(&n->list_lock);
*** DEADLOCK ***
[1]: https://syzkaller.appspot.com/bug?extid=9045c0a3d5a7f1b119f7
A bpf program attached to trace_contention_end() triggers after
acquiring &n->list_lock. The program invokes trie_delete_elem(), which
then acquires trie->lock. However, it is possible that another
process is invoking trie_update_elem(). trie_update_elem() will acquire
trie->lock first, then invoke kmalloc_node(). kmalloc_node() may invoke
get_partial_node() and try to acquire &n->list_lock (not necessarily the
same lock object). Therefore, lockdep warns about the circular locking
dependency.
Invoking kmalloc() before acquiring trie->lock could fix the warning.
However, since BPF programs call be invoked from any context (e.g.,
through kprobe/tracepoint/fentry), there may still be lock ordering
problems for internal locks in kmalloc() or trie->lock itself.
To eliminate these potential lock ordering problems with kmalloc()'s
internal locks, replacing kmalloc()/kfree()/kfree_rcu() with equivalent
BPF memory allocator APIs that can be invoked in any context. The lock
ordering problems with trie->lock (e.g., reentrance) will be handled
separately.
Three aspects of this change require explanation:
1. Intermediate and leaf nodes are allocated from the same allocator.
Since the value size of LPM trie is usually small, using a single
alocator reduces the memory overhead of the BPF memory allocator.
2. Leaf nodes are allocated before disabling IRQs. This handles cases
where leaf_size is large (e.g., > 4KB - 8) and updates require
intermediate node allocation. If leaf nodes were allocated in
IRQ-disabled region, the free objects in BPF memory allocator would not
be refilled timely and the intermediate node allocation may fail.
3. Paired migrate_{disable|enable}() calls for node alloc and free. The
BPF memory allocator uses per-CPU struct internally, these paired calls
are necessary to guarantee correctness.
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-7-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
trie_get_next_key() uses node->prefixlen == key->prefixlen to identify
an exact match, However, it is incorrect because when the target key
doesn't fully match the found node (e.g., node->prefixlen != matchlen),
these two nodes may also have the same prefixlen. It will return
expected result when the passed key exist in the trie. However when a
recently-deleted key or nonexistent key is passed to
trie_get_next_key(), it may skip keys and return incorrect result.
Fix it by using node->prefixlen == matchlen to identify exact matches.
When the condition is true after the search, it also implies
node->prefixlen equals key->prefixlen, otherwise, the search would
return NULL instead.
Fixes: b471f2f1de8b ("bpf: implement MAP_GET_NEXT_KEY command for LPM_TRIE map")
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-6-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
When a LPM trie is full, in-place updates of existing elements
incorrectly return -ENOSPC.
Fix this by deferring the check of trie->n_entries. For new insertions,
n_entries must not exceed max_entries. However, in-place updates are
allowed even when the trie is full.
Fixes: b95a5c4db09b ("bpf: add a longest prefix match trie map implementation")
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-5-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Add the currently missing handling for the BPF_EXIST and BPF_NOEXIST
flags. These flags can be specified by users and are relevant since LPM
trie supports exact matches during update.
Fixes: b95a5c4db09b ("bpf: add a longest prefix match trie map implementation")
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-4-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
There is no need to call kfree(im_node) when updating element fails,
because im_node must be NULL. Remove the unnecessary kfree() for
im_node.
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-3-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
When "node->prefixlen == matchlen" is true, it means that the node is
fully matched. If "node->prefixlen == key->prefixlen" is false, it means
the prefix length of key is greater than the prefix length of node,
otherwise, matchlen will not be equal with node->prefixlen. However, it
also implies that the prefix length of node must be less than
max_prefixlen.
Therefore, "node->prefixlen == trie->max_prefixlen" will always be false
when the check of "node->prefixlen == key->prefixlen" returns false.
Remove this unnecessary comparison.
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20241206110622.1161752-2-houtao@huaweicloud.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Registering and unregistering cpuhp callback requires global cpu hotplug lock,
which is used everywhere. Meantime q->sysfs_lock is used in block layer
almost everywhere.
It is easy to trigger lockdep warning[1] by connecting the two locks.
Fix the warning by moving blk-mq's cpuhp callback registering out of
q->sysfs_lock. Add one dedicated global lock for covering registering &
unregistering hctx's cpuhp, and it is safe to do so because hctx is
guaranteed to be live if our request_queue is live.
[1] https://lore.kernel.org/lkml/Z04pz3AlvI4o0Mr8@agluck-desk3/
Cc: Reinette Chatre <reinette.chatre@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Peter Newman <peternewman@google.com>
Cc: Babu Moger <babu.moger@amd.com>
Reported-by: Luck Tony <tony.luck@intel.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/r/20241206111611.978870-3-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
We need to retrieve 'hctx' from xarray table in the cpuhp callback, so the
callback should be registered after this 'hctx' is added to xarray table.
Cc: Reinette Chatre <reinette.chatre@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Peter Newman <peternewman@google.com>
Cc: Babu Moger <babu.moger@amd.com>
Cc: Luck Tony <tony.luck@intel.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: https://lore.kernel.org/r/20241206111611.978870-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
dfs_cache_refresh() delayed worker could race with cifs_put_tcon(), so
make sure to call list_replace_init() on @tcon->dfs_ses_list after
kworker is cancelled or finished.
Fixes: 4f42a8b54b5c ("smb: client: fix DFS interlink failover")
Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
|
|
Some servers which implement the SMB3.1.1 POSIX extensions did not
set the file type in the mode in the infolevel 100 response.
With the recent changes for checking the file type via the mode field,
this can cause the root directory to be reported incorrectly and
mounts (e.g. to ksmbd) to fail.
Fixes: 6a832bc8bbb2 ("fs/smb/client: Implement new SMB3 POSIX type")
Cc: stable@vger.kernel.org
Acked-by: Paulo Alcantara (Red Hat) <pc@manguebit.com>
Cc: Ralph Boehme <slow@samba.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
|
|
Linux remembers cpu_cachinfo::num_leaves per CPU, but x86 initializes all
CPUs from the same global "num_cache_leaves".
This is erroneous on systems such as Meteor Lake, where each CPU has a
distinct num_leaves value. Delete the global "num_cache_leaves" and
initialize num_leaves on each CPU.
init_cache_level() no longer needs to set num_leaves. Also, it never had to
set num_levels as it is unnecessary in x86. Keep checking for zero cache
leaves. Such condition indicates a bug.
[ bp: Cleanup. ]
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Cc: stable@vger.kernel.org # 6.3+
Link: https://lore.kernel.org/r/20241128002247.26726-3-ricardo.neri-calderon@linux.intel.com
|
|
Commit
5944ce092b97 ("arch_topology: Build cacheinfo from primary CPU")
adds functionality that architectures can use to optionally allocate and
build cacheinfo early during boot. Commit
6539cffa9495 ("cacheinfo: Add arch specific early level initializer")
lets secondary CPUs correct (and reallocate memory) cacheinfo data if
needed.
If the early build functionality is not used and cacheinfo does not need
correction, memory for cacheinfo is never allocated. x86 does not use
the early build functionality. Consequently, during the cacheinfo CPU
hotplug callback, last_level_cache_is_valid() attempts to dereference
a NULL pointer:
BUG: kernel NULL pointer dereference, address: 0000000000000100
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not present page
PGD 0 P4D 0
Oops: 0000 [#1] PREEPMT SMP NOPTI
CPU: 0 PID 19 Comm: cpuhp/0 Not tainted 6.4.0-rc2 #1
RIP: 0010: last_level_cache_is_valid+0x95/0xe0a
Allocate memory for cacheinfo during the cacheinfo CPU hotplug callback
if not done earlier.
Moreover, before determining the validity of the last-level cache info,
ensure that it has been allocated. Simply checking for non-zero
cache_leaves() is not sufficient, as some architectures (e.g., Intel
processors) have non-zero cache_leaves() before allocation.
Dereferencing NULL cacheinfo can occur in update_per_cpu_data_slice_size().
This function iterates over all online CPUs. However, a CPU may have come
online recently, but its cacheinfo may not have been allocated yet.
While here, remove an unnecessary indentation in allocate_cache_info().
[ bp: Massage. ]
Fixes: 6539cffa9495 ("cacheinfo: Add arch specific early level initializer")
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Radu Rendec <rrendec@redhat.com>
Reviewed-by: Nikolay Borisov <nik.borisov@suse.com>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Cc: stable@vger.kernel.org # 6.3+
Link: https://lore.kernel.org/r/20241128002247.26726-2-ricardo.neri-calderon@linux.intel.com
|
|
The restore_processor_state() function explicitly states that "the asm code
that gets us here will have restored a usable GDT". That wasn't true in the
case of returning from a ::preserve_context kexec. Make it so.
Without this, the kernel was depending on the called function to reload a
GDT which is appropriate for the kernel before returning.
Test program:
#include <unistd.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <linux/kexec.h>
#include <linux/reboot.h>
#include <sys/reboot.h>
#include <sys/syscall.h>
int main (void)
{
struct kexec_segment segment = {};
unsigned char purgatory[] = {
0x66, 0xba, 0xf8, 0x03, // mov $0x3f8, %dx
0xb0, 0x42, // mov $0x42, %al
0xee, // outb %al, (%dx)
0xc3, // ret
};
int ret;
segment.buf = &purgatory;
segment.bufsz = sizeof(purgatory);
segment.mem = (void *)0x400000;
segment.memsz = 0x1000;
ret = syscall(__NR_kexec_load, 0x400000, 1, &segment, KEXEC_PRESERVE_CONTEXT);
if (ret) {
perror("kexec_load");
exit(1);
}
ret = syscall(__NR_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2, LINUX_REBOOT_CMD_KEXEC);
if (ret) {
perror("kexec reboot");
exit(1);
}
printf("Success\n");
return 0;
}
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20241205153343.3275139-2-dwmw2@infradead.org
|
|
In the function yas537_measure() there is a clamp_val() with limits of
-BIT(13) and BIT(13) - 1. The input clamp value h[] is of type s32. The
BIT() is of type unsigned long integer due to its define in
include/vdso/bits.h. The lower limit -BIT(13) is recognized as -8192 but
expressed as an unsigned long integer. The size of an unsigned long
integer differs between 32-bit and 64-bit architectures. Converting this
to type s32 may lead to undesired behavior.
Additionally, in the calculation lines h[0], h[1] and h[2] the unsigned
long integer divisor BIT(13) causes an unsigned division, shifting the
left-hand side of the equation back and forth, possibly ending up in large
positive values instead of negative values on 32-bit architectures.
To solve those two issues, declare a signed integer with a value of
BIT(13).
There is another omission in the clamp line: clamp_val() returns a value
and it's going nowhere here. Self-assign it to h[i] to make use of the
clamp macro.
Finally, replace clamp_val() macro by clamp() because after changing the
limits from type unsigned long integer to signed integer it's fine that
way.
Link: https://lkml.kernel.org/r/11609b2243c295d65ab4d47e78c239d61ad6be75.1732914810.git.jahau@rocketmail.com
Fixes: 65f79b501030 ("iio: magnetometer: yas530: Add YAS537 variant")
Signed-off-by: Jakob Hauser <jahau@rocketmail.com>
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202411230458.dhZwh3TT-lkp@intel.com/
Closes: https://lore.kernel.org/oe-kbuild-all/202411282222.oF0B4110-lkp@intel.com/
Reviewed-by: David Laight <david.laight@aculab.com>
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: Lars-Peter Clausen <lars@metafoo.de>
Cc: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
[Problem Description]
When running the hackbench program of LTP, the following memory leak is
reported by kmemleak.
# /opt/ltp/testcases/bin/hackbench 20 thread 1000
Running with 20*40 (== 800) tasks.
# dmesg | grep kmemleak
...
kmemleak: 480 new suspected memory leaks (see /sys/kernel/debug/kmemleak)
kmemleak: 665 new suspected memory leaks (see /sys/kernel/debug/kmemleak)
# cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff888cd8ca2c40 (size 64):
comm "hackbench", pid 17142, jiffies 4299780315
hex dump (first 32 bytes):
ac 74 49 00 01 00 00 00 4c 84 49 00 01 00 00 00 .tI.....L.I.....
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace (crc bff18fd4):
[<ffffffff81419a89>] __kmalloc_cache_noprof+0x2f9/0x3f0
[<ffffffff8113f715>] task_numa_work+0x725/0xa00
[<ffffffff8110f878>] task_work_run+0x58/0x90
[<ffffffff81ddd9f8>] syscall_exit_to_user_mode+0x1c8/0x1e0
[<ffffffff81dd78d5>] do_syscall_64+0x85/0x150
[<ffffffff81e0012b>] entry_SYSCALL_64_after_hwframe+0x76/0x7e
...
This issue can be consistently reproduced on three different servers:
* a 448-core server
* a 256-core server
* a 192-core server
[Root Cause]
Since multiple threads are created by the hackbench program (along with
the command argument 'thread'), a shared vma might be accessed by two or
more cores simultaneously. When two or more cores observe that
vma->numab_state is NULL at the same time, vma->numab_state will be
overwritten.
Although current code ensures that only one thread scans the VMAs in a
single 'numa_scan_period', there might be a chance for another thread
to enter in the next 'numa_scan_period' while we have not gotten till
numab_state allocation [1].
Note that the command `/opt/ltp/testcases/bin/hackbench 50 process 1000`
cannot the reproduce the issue. It is verified with 200+ test runs.
[Solution]
Use the cmpxchg atomic operation to ensure that only one thread executes
the vma->numab_state assignment.
[1] https://lore.kernel.org/lkml/1794be3c-358c-4cdc-a43d-a1f841d91ef7@amd.com/
Link: https://lkml.kernel.org/r/20241113102146.2384-1-ahuang12@lenovo.com
Fixes: ef6a22b70f6d ("sched/numa: apply the scan delay to every new vma")
Signed-off-by: Adrian Huang <ahuang12@lenovo.com>
Reported-by: Jiwei Sun <sunjw10@lenovo.com>
Reviewed-by: Raghavendra K T <raghavendra.kt@amd.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Ben Segall <bsegall@google.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Valentin Schneider <vschneid@redhat.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
Since the order of the scheme_idx and target_idx arguments in TP_ARGS is
reversed, they are stored in the trace record in reverse.
Link: https://lkml.kernel.org/r/20241115182023.43118-1-sj@kernel.org
Link: https://patch.msgid.link/20241112154828.40307-1-akinobu.mita@gmail.com
Fixes: c603c630b509 ("mm/damon/core: add a tracepoint for damos apply target regions")
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: SeongJae Park <sj@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
The never-taken branch leads to an invalid bounds condition, which is by
design. To avoid the unwanted warning from the compiler, hide the
variable from the optimizer.
../lib/stackinit_kunit.c: In function 'do_nothing_u16_zero':
../lib/stackinit_kunit.c:51:49: error: array subscript 1 is outside array bounds of 'u16[0]' {aka 'short unsigned int[]'} [-Werror=array-bounds=]
51 | #define DO_NOTHING_RETURN_SCALAR(ptr) *(ptr)
| ^~~~~~
../lib/stackinit_kunit.c:219:24: note: in expansion of macro 'DO_NOTHING_RETURN_SCALAR'
219 | return DO_NOTHING_RETURN_ ## which(ptr + 1); \
| ^~~~~~~~~~~~~~~~~~
Link: https://lkml.kernel.org/r/20241117113813.work.735-kees@kernel.org
Signed-off-by: Kees Cook <kees@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
The folio can get freed + buddy-merged + reallocated in the meantime,
resulting in us calling folio_test_locked() possibly on a tail page.
This makes const_folio_flags VM_BUG_ON_PGFLAGS() when stumbling over the
tail page.
Could this result in other issues? Doesn't look like it. False positives
and false negatives don't really matter, because this folio would get
skipped either way when detecting that they have been reallocated in the
meantime.
Fix it by performing the folio_test_locked() checked after grabbing a
reference. If this ever becomes a real problem, we could add a special
helper that racily checks if the bit is set even on tail pages ... but
let's hope that's not required so we can just handle it cleaner: work on
the folio after we hold a reference.
Do we really need the folio_test_locked() check if we are going to trylock
briefly after? Well, we can at least avoid a xas_reload().
It's a bit unclear which exact change introduced that issue. Likely, ever
since we made PG_locked obey to the PF_NO_TAIL policy it could have been
triggered in some way.
Link: https://lkml.kernel.org/r/20241129125303.4033164-1-david@redhat.com
Fixes: 48c935ad88f5 ("page-flags: define PG_locked behavior on compound pages")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reported-by: syzbot+9f9a7f73fb079b2387a6@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/lkml/674184c9.050a0220.1cc393.0001.GAE@google.com/
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Hillf Danton <hdanton@sina.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
Fix a kernel-doc warning by making the kernel-doc function description
match the function name:
include/linux/scatterlist.h:323: warning: expecting prototype for sg_unmark_bus_address(). Prototype was for sg_dma_unmark_bus_address() instead
Link: https://lkml.kernel.org/r/20241130022406.537973-1-rdunlap@infradead.org
Fixes: 42399301203e ("lib/scatterlist: add flag for indicating P2PDMA segments in an SGL")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
We mistakenly refer to len rather than len_ here. The only existing
caller passes len to the len_ parameter so this has no impact on the code,
but it is obviously incorrect to do this, so fix it.
Link: https://lkml.kernel.org/r/20241118175414.390827-1-lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Cc: Jann Horn <jannh@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
|
Commit efa7df3e3bb5 ("mm: align larger anonymous mappings on THP
boundaries") updated __get_unmapped_area() to align the start address for
the VMA to a PMD boundary if CONFIG_TRANSPARENT_HUGEPAGE=y.
It does this by effectively looking up a region that is of size,
request_size + PMD_SIZE, and aligning up the start to a PMD boundary.
Commit 4ef9ad19e176 ("mm: huge_memory: don't force huge page alignment on
32 bit") opted out of this for 32bit due to regressions in mmap base
randomization.
Commit d4148aeab412 ("mm, mmap: limit THP alignment of anonymous mappings
to PMD-aligned sizes") restricted this to only mmap sizes that are
multiples of the PMD_SIZE due to reported regressions in some performance
benchmarks -- which seemed mostly due to the reduced spatial locality of
related mappings due to the forced PMD-alignment.
Another unintended side effect has emerged: When a user specifies an mmap
hint address, the THP alignment logic modifies the behavior, potentially
ignoring the hint even if a sufficiently large gap exists at the requested
hint location.
Example Scenario:
Consider the following simplified virtual address (VA) space:
...
0x200000-0x400000 --- VMA A
0x400000-0x600000 --- Hole
0x600000-0x800000 --- VMA B
...
A call to mmap() with hint=0x400000 and len=0x200000 behaves differently:
- Before THP alignment: The requested region (size 0x200000) fits into
the gap at 0x400000, so the hint is respected.
- After alignment: The logic searches for a region of size
0x400000 (len + PMD_SIZE) starting at 0x400000.
This search fails due to the mapping at 0x600000 (VMA B), and the hint
is ignored, falling back to arch_get_unmapped_area[_topdown]().
In general the hint is effectively ignored, if there is any existing
mapping in the below range:
[mmap_hint + mmap_size, mmap_hint + mmap_size + PMD_SIZE)
This changes the semantics of mmap hint; from ""Respect the hint if a
sufficiently large gap exists at the requested location" to "Respect the
hint only if an additional PMD-sized gap exists beyond the requested
size".
This has performance implications for allocators that allocate their heap
using mmap but try to keep it "as contiguous as possible" by using the end
of the exisiting heap as the address hint. With the new behavior it's
more likely to get a much less contiguous heap, adding extra fragmentation
and performance overhead.
To restore the expected behavior; don't use
thp_get_unmapped_area_vmflags() when the user provided a hint address, for
anonymous mappings.
Note: As Yang Shi pointed out: the issue still remains for filesystems
which are using thp_get_unmapped_area() for their get_unmapped_area() op.
It is unclear what worklaods will regress for if we ignore THP alignment
when the hint address is provided for such file backed mappings -- so this
fix will be handled separately.
Link: https://lkml.kernel.org/r/20241118214650.3667577-1-kaleshsingh@google.com
Fixes: efa7df3e3bb5 ("mm: align larger anonymous mappings on THP boundaries")
Signed-off-by: Kalesh Singh <kaleshsingh@google.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <yang@os.amperecomputing.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hans Boehm <hboehm@google.com>
Cc: Lokesh Gidra <lokeshgidra@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
|
Michel Dänzer
Christian König
Alex Deucher
Pierre-Eric Pelloux-Prayer
Mario Limonciello
Lucas De Marchi
Daniele Ceraolo Spurio
Mirsad Todorovac
Jesse.zhang@amd.com
Pratap Nirujogi
Harish Kasiviswanathan
Andrew Martin
Jiasheng Jiang
Eugene Kobyak
Ville Syrjälä
Linus Torvalds
Masahiro Yamada
Thomas Gleixner
Sean Christopherson
Hou Tao
Ming Lei
Paulo Alcantara
Steve French
Ricardo Neri
David Woodhouse
Jakob Hauser
Adrian Huang
Akinobu Mita
Kees Cook
David Hildenbrand
Randy Dunlap
Lorenzo Stoakes
Kalesh Singh