Skip to contentSkip to conversation

perf/core: Bail out early if the request AUX area is out of bound

PATCH In review Shuai Xue 1 participant 4 messages Sep 7, 2023 – Oct 9, 2023

Subsystems (2)
lore
Share & tools
lore

Shuai Xue

author
#0 lore raw

v5 cover perf/core: Bail out early if the request AUX area is out of bound

changes since v4 (per Leo):
- refine comments as Leo asked
- correct aux limitation in doc by adding a formula

changes since v3 (per James):
- change to return with existing value -ENOMEM
- split doc part to a separate patch since they are going to be merged through separate trees.

changes since v2:
- remove unnecessary overflow check (per Peter)

changes since v1:
- drop out patch2 because it has been fixed on upstream (Thanks James for reminding)
- move sanity check into rb_alloc_aux (per Leo)
- add overflow check (per James)

Shuai Xue (2):
perf/core: Bail out early if the request AUX area is out of bound perf record: Update docs regarding the maximum limitation of AUX area

 kernel/events/ring_buffer.c              |  6 ++++++
 tools/perf/Documentation/perf-record.txt | 11 +++++++++++
 2 files changed, 17 insertions(+)
Signature · 1 line
2.39.3
author
#1 lore raw

PATCH v5 1/2 perf/core: Bail out early if the request AUX area is out of bound

When perf-record with a large AUX area, e.g 4GB, it fails with:

#perf record -C 0 -m ,4G -e arm_spe_0// -- sleep 1 failed to mmap with 12 (Cannot allocate memory)

and it reveals a WARNING with __alloc_pages():

[   66.595604] ------------[ cut here ]------------
[   66.600206] WARNING: CPU: 44 PID: 17573 at mm/page_alloc.c:5568 __alloc_pages+0x1ec/0x248
[   66.608375] Modules linked in: ip6table_filter(E) ip6_tables(E) iptable_filter(E) ebtable_nat(E) ebtables(E) aes_ce_blk(E) vfat(E) fat(E) aes_ce_cipher(E) crct10dif_ce(E) ghash_ce(E) sm4_ce_cipher(E) sm4(E) sha2_ce(E) sha256_arm64(E) sha1_ce(E) acpi_ipmi(E) sbsa_gwdt(E) sg(E) ipmi_si(E) ipmi_devintf(E) ipmi_msghandler(E) ip_tables(E) sd_mod(E) ast(E) drm_kms_helper(E) syscopyarea(E) sysfillrect(E) nvme(E) sysimgblt(E) i2c_algo_bit(E) nvme_core(E) drm_shmem_helper(E) ahci(E) t10_pi(E) libahci(E) drm(E) crc64_rocksoft(E) i40e(E) crc64(E) libata(E) i2c_core(E)
[   66.657719] CPU: 44 PID: 17573 Comm: perf Kdump: loaded Tainted: G            E      6.3.0-rc4+ #58
[   66.666749] Hardware name: Default Default/Default, BIOS 1.2.M1.AL.P.139.00 03/22/2023
[   66.674650] pstate: 23400009 (nzCv daif +PAN -UAO +TCO +DIT -SSBS BTYPE=--)
[   66.681597] pc : __alloc_pages+0x1ec/0x248
[   66.685680] lr : __kmalloc_large_node+0xc0/0x1f8
[   66.690285] sp : ffff800020523980
[   66.693585] pmr_save: 000000e0
[   66.696624] x29: ffff800020523980 x28: ffff000832975800 x27: 0000000000000000
[   66.703746] x26: 0000000000100000 x25: 0000000000100000 x24: ffff8000083615d0
[   66.710866] x23: 0000000000040dc0 x22: ffff000823d6d140 x21: 000000000000000b
[   66.717987] x20: 000000000000000b x19: 0000000000000000 x18: 0000000000000030
[   66.725108] x17: 0000000000000000 x16: ffff800008f05be8 x15: ffff000823d6d6d0
[   66.732229] x14: 0000000000000000 x13: 343373656761705f x12: 726e202c30206574
[   66.739350] x11: 00000000ffff7fff x10: 00000000ffff7fff x9 : ffff8000083af570
[   66.746471] x8 : 00000000000bffe8 x7 : c0000000ffff7fff x6 : 000000000005fff4
[   66.753592] x5 : 0000000000000000 x4 : ffff000823d6d8d8 x3 : 0000000000000000
[   66.760713] x2 : 0000000000000000 x1 : 0000000000000001 x0 : 0000000000040dc0
[   66.767834] Call trace:
[   66.770267]  __alloc_pages+0x1ec/0x248
[   66.774003]  __kmalloc_large_node+0xc0/0x1f8
[   66.778259]  __kmalloc_node+0x134/0x1e8
[   66.782081]  rb_alloc_aux+0xe0/0x298
[   66.785643]  perf_mmap+0x440/0x660
[   66.789031]  mmap_region+0x308/0x8a8
[   66.792593]  do_mmap+0x3c0/0x528
[   66.795807]  vm_mmap_pgoff+0xf4/0x1b8
[   66.799456]  ksys_mmap_pgoff+0x18c/0x218
[   66.803365]  __arm64_sys_mmap+0x38/0x58
[   66.807187]  invoke_syscall+0x50/0x128
[   66.810922]  el0_svc_common.constprop.0+0x58/0x188
[   66.815698]  do_el0_svc+0x34/0x50
[   66.818999]  el0_svc+0x34/0x108
[   66.822127]  el0t_64_sync_handler+0xb8/0xc0
[   66.826296]  el0t_64_sync+0x1a4/0x1a8
[   66.829946] ---[ end trace 0000000000000000 ]---

'rb->aux_pages' allocated by kcalloc() is a pointer array which is used to maintains AUX trace pages. The allocated page for this array is physically contiguous (and virtually contiguous) with an order of 0..MAX_ORDER. If the size of pointer array crosses the limitation set by MAX_ORDER, it reveals a WARNING.

So bail out early with -ENOMEM if the request AUX area is out of bound, e.g.:

#perf record -C 0 -m ,4G -e arm_spe_0// -- sleep 1 failed to mmap with 12 (Cannot allocate memory)

---
 kernel/events/ring_buffer.c | 6 ++++++
 1 file changed, 6 insertions(+)
kernel/events/ring_buffer.c +6 -0
index fb1e180b5f0a..e8d82c2f07d0 100644
@@ -700,6 +700,12 @@ int rb_alloc_aux(struct perf_buffer *rb, struct perf_event *event,
 		watermark = 0;
 	}
 
+	/*
+	 * kcalloc_node() is unable to allocate buffer if the size is larger
+	 * than: PAGE_SIZE << MAX_ORDER; directly bail out in this case.
+	 */
+	if (get_order((unsigned long)nr_pages * sizeof(void *)) > MAX_ORDER)
+		return -ENOMEM;
 	rb->aux_pages = kcalloc_node(nr_pages, sizeof(void *), GFP_KERNEL,
 				     node);
 	if (!rb->aux_pages)
Signature · 1 line
2.39.3
author
#2 lore raw

PATCH v5 2/2 perf record: Update docs regarding the maximum limitation of AUX area

In reply to Shuai Xue

The maximum AUX area is limited by the page size of the system. Update the documentation to reflect this.

---
 tools/perf/Documentation/perf-record.txt | 11 +++++++++++
 1 file changed, 11 insertions(+)
tools/perf/Documentation/perf-record.txt +11 -0
index 680396c56bd1..dfc322d6f1f1 100644
@@ -292,6 +292,17 @@ OPTIONS
 	Also, by adding a comma, the number of mmap pages for AUX
 	area tracing can be specified.
 
+	The maximum AUX area is limited by the maximum physically contiguous
+	memory allocated from slab/slub. It can be calculated with following
+	formula:
+
+	  PAGE_SIZE << MAX_ORDER
+	( ---------------------- ) * PAGE_SIZE
+	   sizeof(page_pointer)
+
+	For example with 4K pages and MAX_ORDER=10 configured, the maximum AUX
+	area is 2GiB.
+
 -g::
 	Enables call-graph (stack chain/backtrace) recording for both
 	kernel space and user space.
Signature · 1 line
2.39.3
author
#3 lore raw

Re: [PATCH v5 2/2] perf record: Update docs regarding the maximum limitation of AUX area

Quoted 20 lines · Shuai Xue

The maximum AUX area is limited by the page size of the system. Update the documentation to reflect this.

---
 tools/perf/Documentation/perf-record.txt | 11 +++++++++++
 1 file changed, 11 insertions(+)
tools/perf/Documentation/perf-record.txt +4 -0
index 680396c56bd1..dfc322d6f1f1 100644
@@ -292,6 +292,17 @@ OPTIONS
 	Also, by adding a comma, the number of mmap pages for AUX
 	area tracing can be specified.
 
+	The maximum AUX area is limited by the maximum physically contiguous
+	memory allocated from slab/slub. It can be calculated with following
+	formula:
+

tools/perf/Documentation/perf-record.txt OPTIONS · line 308

Quoted 10 lines
tools/perf/Documentation/perf-record.txt +7 -0
+	  PAGE_SIZE << MAX_ORDER
+	( ---------------------- ) * PAGE_SIZE
+	   sizeof(page_pointer)
+
+	For example with 4K pages and MAX_ORDER=10 configured, the maximum AUX
+	area is 2GiB.
+
 -g::
 	Enables call-graph (stack chain/backtrace) recording for both
 	kernel space and user space.

Hi, Namhyung,

Since PATCH [1/2] of this series has been merge into perf/core[1], are you happy to pick this part [2/2] into perf-tools-next?

Thank you.

Best Regards,
Shuai

[1] https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/commit/?h=perf/core&id=54aee5f15b83437f23b2b2469bcf21bdd9823916