As noted in the "Deprecated Interfaces, Language Features, Attributes,
and Conventions" documentation [1], size calculations (especially
multiplication) should not be performed in memory allocator (or similar)
function arguments due to the risk of them overflowing.
This could lead to values wrapping around and a smaller allocation being
made than the caller was expecting. Using those allocations could lead
to linear overflows of heap memory and other misbehaviors.
So, use the struct_size() helper to do the arithmetic instead of the
argument "size + count * size" in the kvmalloc() functions.
Also, take the opportunity to refactor the memcpy() call to use the
flex_array_size() helper.
This code was detected with the help of Coccinelle and audited and fixed
manually.
[1] https://www.kernel.org/doc/html/latest/process/deprecated.html#open-coded-arithmetic-in-allocator-arguments
Link: https://lkml.kernel.org/r/20211017105929.9284-1-len.baker@gmx.com
Signed-off-by: Len Baker <len.baker@gmx.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The memcg stats can be flushed in multiple context and potentially in
parallel too. For example multiple parallel user space readers for
memcg stats will contend on the rstat locks with each other. There is
no need for that. We just need one flusher and everyone else can
benefit.
In addition after aa48e47e39 ("memcg: infrastructure to flush memcg
stats") the kernel periodically flush the memcg stats from the root, so,
the other flushers will potentially have much less work to do.
Link: https://lkml.kernel.org/r/20211001190040.48086-2-shakeelb@google.com
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: "Michal Koutný" <mkoutny@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
At the moment, the kernel flushes the memcg stats on every refault and
also on every reclaim iteration. Although rstat maintains per-cpu
update tree but on the flush the kernel still has to go through all the
cpu rstat update tree to check if there is anything to flush. This
patch adds the tracking on the stats update side to make flush side more
clever by skipping the flush if there is no update.
The stats update codepath is very sensitive performance wise for many
workloads and benchmarks. So, we can not follow what the commit
aa48e47e39 ("memcg: infrastructure to flush memcg stats") did which
was triggering async flush through queue_work() and caused a lot
performance regression reports. That got reverted by the commit
1f828223b7 ("memcg: flush lruvec stats in the refault").
In this patch we kept the stats update codepath very minimal and let the
stats reader side to flush the stats only when the updates are over a
specific threshold. For now the threshold is (nr_cpus * CHARGE_BATCH).
To evaluate the impact of this patch, an 8 GiB tmpfs file is created on
a system with swap-on-zram and the file was pushed to swap through
memory.force_empty interface. On reading the whole file, the memcg stat
flush in the refault code path is triggered. With this patch, we
observed 63% reduction in the read time of 8 GiB file.
Link: https://lkml.kernel.org/r/20211001190040.48086-1-shakeelb@google.com
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Reviewed-by: "Michal Koutný" <mkoutny@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This one is just a minor nuisance for people going through /proc/swaps
if any of their swapareas is bigger than, or equal to 1073741824 pages
(4TB).
seq_printf() format string casts as uint the conversion from pages to
KB, and that will overflow in the aforementioned case.
Albeit being almost unthinkable that someone would actually set up such
big of a single swaparea, there is a ticket recently filed against RHEL:
https://bugzilla.redhat.com/show_bug.cgi?id=2008812
Given that all other codesites that use format strings for the same swap
pages-to-KB conversion do cast it as ulong, this patch just follows
suit.
Link: https://lkml.kernel.org/r/20211006184011.2579054-1-aquini@redhat.com
Signed-off-by: Rafael Aquini <aquini@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
GCC and Clang can use the "alloc_size" attribute to better inform the
results of __builtin_object_size() (for compile-time constant values).
Clang can additionally use alloc_size to inform the results of
__builtin_dynamic_object_size() (for run-time values).
Because GCC sees the frequent use of struct_size() as an allocator size
argument, and notices it can return SIZE_MAX (the overflow indication),
it complains about these call sites overflowing (since SIZE_MAX is
greater than the default -Walloc-size-larger-than=PTRDIFF_MAX). This
isn't helpful since we already know a SIZE_MAX will be caught at
run-time (this was an intentional design). To deal with this, we must
disable this check as it is both a false positive and redundant. (Clang
does not have this warning option.)
Unfortunately, just checking the -Wno-alloc-size-larger-than is not
sufficient to make the __alloc_size attribute behave correctly under
older GCC versions. The attribute itself must be disabled in those
situations too, as there appears to be no way to reliably silence the
SIZE_MAX constant expression cases for GCC versions less than 9.1:
In file included from ./include/linux/resource_ext.h:11,
from ./include/linux/pci.h:40,
from drivers/net/ethernet/intel/ixgbe/ixgbe.h:9,
from drivers/net/ethernet/intel/ixgbe/ixgbe_lib.c:4:
In function 'kmalloc_node',
inlined from 'ixgbe_alloc_q_vector' at ./include/linux/slab.h:743:9:
./include/linux/slab.h:618:9: error: argument 1 value '18446744073709551615' exceeds maximum object size 9223372036854775807 [-Werror=alloc-size-larger-than=]
return __kmalloc_node(size, flags, node);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
./include/linux/slab.h: In function 'ixgbe_alloc_q_vector':
./include/linux/slab.h:455:7: note: in a call to allocation function '__kmalloc_node' declared here
void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_slab_alignment __malloc;
^~~~~~~~~~~~~~
Specifically:
'-Wno-alloc-size-larger-than' is not correctly handled by GCC < 9.1
https://godbolt.org/z/hqsfG7q84 (doesn't disable)
https://godbolt.org/z/P9jdrPTYh (doesn't admit to not knowing about option)
https://godbolt.org/z/465TPMWKb (only warns when other warnings appear)
'-Walloc-size-larger-than=18446744073709551615' is not handled by GCC < 8.2
https://godbolt.org/z/73hh1EPxz (ignores numeric value)
Since anything marked with __alloc_size would also qualify for marking
with __malloc, just include __malloc along with it to avoid redundant
markings. (Suggested by Linus Torvalds.)
Finally, make sure checkpatch.pl doesn't get confused about finding the
__alloc_size attribute on functions. (Thanks to Joe Perches.)
Link: https://lkml.kernel.org/r/20210930222704.2631604-3-keescook@chromium.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Tested-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Dwaipayan Ray <dwaipayanray1@gmail.com>
Cc: Joe Perches <joe@perches.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Alexandre Bounine <alex.bou9@gmail.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Jing Xiangfeng <jingxiangfeng@huawei.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: kernel test robot <lkp@intel.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Miguel Ojeda <ojeda@kernel.org>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "Add __alloc_size()", v3.
GCC and Clang both use the "alloc_size" attribute to assist with bounds
checking around the use of allocation functions. Add the attribute,
adjust the Makefile to silence needless warnings, and add the hints to
the allocators where possible. These changes have been in use for a
while now in GrapheneOS.
This patch (of 8):
After adding __alloc_size attributes to the allocators, GCC 9.3 (but not
later) may incorrectly evaluate the arguments to check_copy_size(),
getting seemingly confused by the size being returned from array_size().
Instead, perform the calculation once, which both makes the code more
readable and avoids the bug in GCC.
In file included from arch/x86/include/asm/preempt.h:7,
from include/linux/preempt.h:78,
from include/linux/spinlock.h:55,
from include/linux/mm_types.h:9,
from include/linux/buildid.h:5,
from include/linux/module.h:14,
from drivers/rapidio/devices/rio_mport_cdev.c:13:
In function 'check_copy_size',
inlined from 'copy_from_user' at include/linux/uaccess.h:191:6,
inlined from 'rio_mport_transfer_ioctl' at drivers/rapidio/devices/rio_mport_cdev.c:983:6:
include/linux/thread_info.h:213:4: error: call to '__bad_copy_to' declared with attribute error: copy destination size is too small
213 | __bad_copy_to();
| ^~~~~~~~~~~~~~~
But the allocation size and the copy size are identical:
transfer = vmalloc(array_size(sizeof(*transfer), transaction.count));
if (!transfer)
return -ENOMEM;
if (unlikely(copy_from_user(transfer,
(void __user *)(uintptr_t)transaction.block,
array_size(sizeof(*transfer), transaction.count)))) {
Link: https://lkml.kernel.org/r/20210930222704.2631604-1-keescook@chromium.org
Link: https://lkml.kernel.org/r/20210930222704.2631604-2-keescook@chromium.org
Link: https://lore.kernel.org/linux-mm/202109091134.FHnRmRxu-lkp@intel.com/
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: John Hubbard <jhubbard@nvidia.com>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Alexandre Bounine <alex.bou9@gmail.com>
Cc: Jing Xiangfeng <jingxiangfeng@huawei.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Gustavo A. R. Silva <gustavoars@kernel.org>
Cc: Andy Whitcroft <apw@canonical.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Dwaipayan Ray <dwaipayanray1@gmail.com>
Cc: Joe Perches <joe@perches.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Cc: Miguel Ojeda <ojeda@kernel.org>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Firstly, check_shmem_swap variable is actually not necessary, because
it's always set with pte_hole hook; checking each would work.
Meanwhile, the check within smaps_pte_entry is not easy to follow.
E.g., pte_none() check is not needed as "!pte_present && !is_swap_pte"
is the same. Since at it, use the pte_hole() helper rather than dup the
page cache lookup.
Still keep the CONFIG_SHMEM part so the code can be optimized to nop for
!SHMEM.
There will be a very slight functional change in smaps_pte_entry(), that
for !SHMEM we'll return early for pte_none (before checking page==NULL),
but that's even nicer.
Link: https://lkml.kernel.org/r/20210917164756.8586-4-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "mm/smaps: Fixes and optimizations on shmem swap handling".
This patch (of 3):
The shmem swap calculation on the privately writable mappings are using
wrong parameters as spotted by Vlastimil. Fix them. This was
introduced in commit 48131e03ca ("mm, proc: reduce cost of
/proc/pid/smaps for unpopulated shmem mappings"), when shmem_swap_usage
was reworked to shmem_partial_swap_usage.
Test program:
void main(void)
{
char *buffer, *p;
int i, fd;
fd = memfd_create("test", 0);
assert(fd > 0);
/* isize==2M*3, fill in pages, swap them out */
ftruncate(fd, SIZE_2M * 3);
buffer = mmap(NULL, SIZE_2M * 3, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
assert(buffer);
for (i = 0, p = buffer; i < SIZE_2M * 3 / 4096; i++) {
*p = 1;
p += 4096;
}
madvise(buffer, SIZE_2M * 3, MADV_PAGEOUT);
munmap(buffer, SIZE_2M * 3);
/*
* Remap with private+writtable mappings on partial of the inode (<= 2M*3),
* while the size must also be >= 2M*2 to make sure there's a none pmd so
* smaps_pte_hole will be triggered.
*/
buffer = mmap(NULL, SIZE_2M * 2, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
printf("pid=%d, buffer=%p\n", getpid(), buffer);
/* Check /proc/$PID/smap_rollup, should see 4MB swap */
sleep(1000000);
}
Before the patch, smaps_rollup shows <4MB swap and the number will be
random depending on the alignment of the buffer of mmap() allocated.
After this patch, it'll show 4MB.
Link: https://lkml.kernel.org/r/20210917164756.8586-1-peterx@redhat.com
Link: https://lkml.kernel.org/r/20210917164756.8586-2-peterx@redhat.com
Fixes: 48131e03ca ("mm, proc: reduce cost of /proc/pid/smaps for unpopulated shmem mappings")
Signed-off-by: Peter Xu <peterx@redhat.com>
Reported-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
With HW tag-based KASAN, error checks are performed implicitly by the
load and store instructions in the memcpy implementation. A failed
check results in tag checks being disabled and execution will keep
going. As a result, under HW tag-based KASAN, prior to commit
1b0668be62 ("kasan: test: disable kmalloc_memmove_invalid_size for
HW_TAGS"), this memcpy would end up corrupting memory until it hits an
inaccessible page and causes a kernel panic.
This is a pre-existing issue that was revealed by commit 285133040e
("arm64: Import latest memcpy()/memmove() implementation") which changed
the memcpy implementation from using signed comparisons (incorrectly,
resulting in the memcpy being terminated early for negative sizes) to
using unsigned comparisons.
It is unclear how this could be handled by memcpy itself in a reasonable
way. One possibility would be to add an exception handler that would
force memcpy to return if a tag check fault is detected -- this would
make the behavior roughly similar to generic and SW tag-based KASAN.
However, this wouldn't solve the problem for asynchronous mode and also
makes memcpy behavior inconsistent with manually copying data.
This test was added as a part of a series that taught KASAN to detect
negative sizes in memory operations, see commit 8cceeff48f ("kasan:
detect negative size in memory operation function"). Therefore we
should keep testing for negative sizes with generic and SW tag-based
KASAN. But there is some value in testing small memcpy overflows, so
let's add another test with memcpy that does not destabilize the kernel
by performing out-of-bounds writes, and run it in all modes.
Link: https://linux-review.googlesource.com/id/I048d1e6a9aff766c4a53f989fb0c83de68923882
Link: https://lkml.kernel.org/r/20210910211356.3603758-1-pcc@google.com
Signed-off-by: Peter Collingbourne <pcc@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Acked-by: Marco Elver <elver@google.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Evgenii Stepanov <eugenis@google.com>
Cc: Alexander Potapenko <glider@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Shuah Khan reported:
| When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
| kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
| it tries to allocate memory attempting to acquire spinlock in page
| allocation code while holding workqueue pool raw_spinlock.
|
| There are several instances of this problem when block layer tries
| to __queue_work(). Call trace from one of these instances is below:
|
| kblockd_mod_delayed_work_on()
| mod_delayed_work_on()
| __queue_delayed_work()
| __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
| insert_work()
| kasan_record_aux_stack()
| kasan_save_stack()
| stack_depot_save()
| alloc_pages()
| __alloc_pages()
| get_page_from_freelist()
| rm_queue()
| rm_queue_pcplist()
| local_lock_irqsave(&pagesets.lock, flags);
| [ BUG: Invalid wait context triggered ]
The default kasan_record_aux_stack() calls stack_depot_save() with
GFP_NOWAIT, which in turn can then call alloc_pages(GFP_NOWAIT, ...).
In general, however, it is not even possible to use either GFP_ATOMIC
nor GFP_NOWAIT in certain non-preemptive contexts, including
raw_spin_locks (see gfp.h and commmit ab00db216c).
Fix it by instructing stackdepot to not expand stack storage via
alloc_pages() in case it runs out by using
kasan_record_aux_stack_noalloc().
While there is an increased risk of failing to insert the stack trace,
this is typically unlikely, especially if the same insertion had already
succeeded previously (stack depot hit).
For frequent calls from the same location, it therefore becomes
extremely unlikely that kasan_record_aux_stack_noalloc() fails.
Link: https://lkml.kernel.org/r/20210902200134.25603-1-skhan@linuxfoundation.org
Link: https://lkml.kernel.org/r/20210913112609.2651084-7-elver@google.com
Signed-off-by: Marco Elver <elver@google.com>
Reported-by: Shuah Khan <skhan@linuxfoundation.org>
Tested-by: Shuah Khan <skhan@linuxfoundation.org>
Acked-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Taras Madan <tarasmadan@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "stackdepot, kasan, workqueue: Avoid expanding stackdepot
slabs when holding raw_spin_lock", v2.
Shuah Khan reported [1]:
| When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
| kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
| it tries to allocate memory attempting to acquire spinlock in page
| allocation code while holding workqueue pool raw_spinlock.
|
| There are several instances of this problem when block layer tries
| to __queue_work(). Call trace from one of these instances is below:
|
| kblockd_mod_delayed_work_on()
| mod_delayed_work_on()
| __queue_delayed_work()
| __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
| insert_work()
| kasan_record_aux_stack()
| kasan_save_stack()
| stack_depot_save()
| alloc_pages()
| __alloc_pages()
| get_page_from_freelist()
| rm_queue()
| rm_queue_pcplist()
| local_lock_irqsave(&pagesets.lock, flags);
| [ BUG: Invalid wait context triggered ]
PROVE_RAW_LOCK_NESTING is pointing out that (on RT kernels) the locking
rules are being violated. More generally, memory is being allocated
from a non-preemptive context (raw_spin_lock'd c-s) where it is not
allowed.
To properly fix this, we must prevent stackdepot from replenishing its
"stack slab" pool if memory allocations cannot be done in the current
context: it's a bug to use either GFP_ATOMIC nor GFP_NOWAIT in certain
non-preemptive contexts, including raw_spin_locks (see gfp.h and commit
ab00db216c).
The only downside is that saving a stack trace may fail if: stackdepot
runs out of space AND the same stack trace has not been recorded before.
I expect this to be unlikely, and a simple experiment (boot the kernel)
didn't result in any failure to record stack trace from insert_work().
The series includes a few minor fixes to stackdepot that I noticed in
preparing the series. It then introduces __stack_depot_save(), which
exposes the option to force stackdepot to not allocate any memory.
Finally, KASAN is changed to use the new stackdepot interface and
provide kasan_record_aux_stack_noalloc(), which is then used by
workqueue code.
[1] https://lkml.kernel.org/r/20210902200134.25603-1-skhan@linuxfoundation.org
This patch (of 6):
<linux/stackdepot.h> refers to gfp_t, but doesn't include gfp.h.
Fix it by including <linux/gfp.h>.
Link: https://lkml.kernel.org/r/20210913112609.2651084-1-elver@google.com
Link: https://lkml.kernel.org/r/20210913112609.2651084-2-elver@google.com
Signed-off-by: Marco Elver <elver@google.com>
Tested-by: Shuah Khan <skhan@linuxfoundation.org>
Acked-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Walter Wu <walter-zh.wu@mediatek.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Cc: Taras Madan <tarasmadan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 0ad9500e16 ("slub: prefetch next freelist pointer in
slab_alloc()") introduced prefetch_freepointer() because when other
cpu(s) freed objects into a page that current cpu owns, the freelist
link is hot on cpu(s) which freed objects and possibly very cold on
current cpu.
But if freelist link chain is hot on cpu(s) which freed objects, it's
better to invalidate that chain because they're not going to access
again within a short time.
So use prefetchw instead of prefetch. On supported architectures like
x86 and arm, it invalidates other copied instances of a cache line when
prefetching it.
Before:
Time: 91.677
Performance counter stats for 'hackbench -g 100 -l 10000':
1462938.07 msec cpu-clock # 15.908 CPUs utilized
18072550 context-switches # 12.354 K/sec
1018814 cpu-migrations # 696.416 /sec
104558 page-faults # 71.471 /sec
1580035699271 cycles # 1.080 GHz (54.51%)
2003670016013 instructions # 1.27 insn per cycle (54.31%)
5702204863 branch-misses (54.28%)
643368500985 cache-references # 439.778 M/sec (54.26%)
18475582235 cache-misses # 2.872 % of all cache refs (54.28%)
642206796636 L1-dcache-loads # 438.984 M/sec (46.87%)
18215813147 L1-dcache-load-misses # 2.84% of all L1-dcache accesses (46.83%)
653842996501 dTLB-loads # 446.938 M/sec (46.63%)
3227179675 dTLB-load-misses # 0.49% of all dTLB cache accesses (46.85%)
537531951350 iTLB-loads # 367.433 M/sec (54.33%)
114750630 iTLB-load-misses # 0.02% of all iTLB cache accesses (54.37%)
630135543177 L1-icache-loads # 430.733 M/sec (46.80%)
22923237620 L1-icache-load-misses # 3.64% of all L1-icache accesses (46.76%)
91.964452802 seconds time elapsed
43.416742000 seconds user
1422.441123000 seconds sys
After:
Time: 90.220
Performance counter stats for 'hackbench -g 100 -l 10000':
1437418.48 msec cpu-clock # 15.880 CPUs utilized
17694068 context-switches # 12.310 K/sec
958257 cpu-migrations # 666.651 /sec
100604 page-faults # 69.989 /sec
1583259429428 cycles # 1.101 GHz (54.57%)
2004002484935 instructions # 1.27 insn per cycle (54.37%)
5594202389 branch-misses (54.36%)
643113574524 cache-references # 447.409 M/sec (54.39%)
18233791870 cache-misses # 2.835 % of all cache refs (54.37%)
640205852062 L1-dcache-loads # 445.386 M/sec (46.75%)
17968160377 L1-dcache-load-misses # 2.81% of all L1-dcache accesses (46.79%)
651747432274 dTLB-loads # 453.415 M/sec (46.59%)
3127124271 dTLB-load-misses # 0.48% of all dTLB cache accesses (46.75%)
535395273064 iTLB-loads # 372.470 M/sec (54.38%)
113500056 iTLB-load-misses # 0.02% of all iTLB cache accesses (54.35%)
628871845924 L1-icache-loads # 437.501 M/sec (46.80%)
22585641203 L1-icache-load-misses # 3.59% of all L1-icache accesses (46.79%)
90.514819303 seconds time elapsed
43.877656000 seconds user
1397.176001000 seconds sys
Link: https://lkml.org/lkml/2021/10/8/598=20
Link: https://lkml.kernel.org/r/20211011144331.70084-1-42.hyeyoo@gmail.com
Signed-off-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The defaults are determined based on object size and can go up to 30 for
objects smaller than 256 bytes. Before the previous patch changed the
accounting, this could have made cpu partial list contain up to 30
pages. After that patch, only up to 2 pages with default allocation
order.
Very short lists limit the usefulness of the whole concept of cpu
partial lists, so this patch aims at a more reasonable default under the
new accounting. The defaults are quadrupled, except for object size >=
PAGE_SIZE where it's doubled. This makes the lists grow up to 10 pages
in practice.
A quick test of booting a kernel under virtme with 4GB RAM and 8 vcpus
shows the following slab memory usage after boot:
Before previous patch (using page->pobjects):
Slab: 36732 kB
SReclaimable: 14836 kB
SUnreclaim: 21896 kB
After previous patch (using page->pages):
Slab: 34720 kB
SReclaimable: 13716 kB
SUnreclaim: 21004 kB
After this patch (using page->pages, higher defaults):
Slab: 35252 kB
SReclaimable: 13944 kB
SUnreclaim: 21308 kB
In the same setup, I also ran 5 times:
hackbench -l 16000 -g 16
Differences in time were in the noise, we can compare slub stats as
given by slabinfo -r skbuff_head_cache (the other cache heavily used by
hackbench, kmalloc-cg-512 looks similar). Negligible stats left out for
brevity.
Before previous patch (using page->pobjects):
Objects: 1408, Memory Total: 401408 Used : 304128
Slab Perf Counter Alloc Free %Al %Fr
--------------------------------------------------
Fastpath 469952498 5946606 91 1
Slowpath 42053573 506059465 8 98
Page Alloc 41093 41044 0 0
Add partial 18 21229327 0 4
Remove partial 20039522 36051 3 0
Cpu partial list 4686640 24767229 0 4
RemoteObj/SlabFrozen 16 124027841 0 24
Total 512006071 512006071
Flushes 18
Slab Deactivation Occurrences %
-------------------------------------------------
Slab empty 4993 0%
Deactivation bypass 24767229 99%
Refilled from foreign frees 21972674 88%
After previous patch (using page->pages):
Objects: 480, Memory Total: 131072 Used : 103680
Slab Perf Counter Alloc Free %Al %Fr
--------------------------------------------------
Fastpath 473016294 5405653 92 1
Slowpath 38989777 506600418 7 98
Page Alloc 32717 32701 0 0
Add partial 3 22749164 0 4
Remove partial 11371127 32474 2 0
Cpu partial list 11686226 23090059 2 4
RemoteObj/SlabFrozen 2 67541803 0 13
Total 512006071 512006071
Flushes 3
Slab Deactivation Occurrences %
-------------------------------------------------
Slab empty 227 0%
Deactivation bypass 23090059 99%
Refilled from foreign frees 27585695 119%
After this patch (using page->pages, higher defaults):
Objects: 896, Memory Total: 229376 Used : 193536
Slab Perf Counter Alloc Free %Al %Fr
--------------------------------------------------
Fastpath 473799295 4980278 92 0
Slowpath 38206776 507025793 7 99
Page Alloc 32295 32267 0 0
Add partial 11 23291143 0 4
Remove partial 5815764 31278 1 0
Cpu partial list 18119280 23967320 3 4
RemoteObj/SlabFrozen 10 76974794 0 15
Total 512006071 512006071
Flushes 11
Slab Deactivation Occurrences %
-------------------------------------------------
Slab empty 989 0%
Deactivation bypass 23967320 99%
Refilled from foreign frees 32358473 135%
As expected, memory usage dropped significantly with change of
accounting, increasing the defaults increased it, but not as much. The
number of page allocation/frees dropped significantly with the new
accounting, but didn't increase with the higher defaults.
Interestingly, the number of fasthpath allocations increased, as well as
allocations from the cpu partial list, even though it's shorter.
Link: https://lkml.kernel.org/r/20211012134651.11258-2-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
