[LTP] hugemmap34 trigger oom if compile with CFLAGS='-fstack-clash-protection'

[LTP] hugemmap34 trigger oom if compile with CFLAGS='-fstack-clash-protection'

[Wei Gao via ltp]

Hi Jan

Compiling hugemmap34 with CFLAGS='-fstack-clash-protection' causes gcc to insert additional stack address validation checks during alloca calls.  This results in kernel start allocate memory really , eventually triggering an OOM.

In our normal openqa env we do not have such big memory(1024GB) to support this test. So we
 have to disable CFLAGS='-fstack-clash-protection' if we still need this case running in our env.

===log with CFLAGS='-fstack-clash-protection'===
[71073.562094] [T13546] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0-1,global_oom,task_memcg=/,task=hugemmap34,pid=13542,uid=0
[71073.562109] [T13546] Out of memory: Killed process 13542 (hugemmap34) total-vm:4764864kB, anon-rss:2816000kB, file-rss:0kB, shmem-rss:0kB, UID:0 pgtables:636kB oom_score_adj:0

tst_hugepage.c:84: TINFO: 1 hugepage(s) reserved
tst_tmpdir.c:317: TINFO: Using /tmp/LTP_hug2vxfh1 as tmpdir (btrfs filesystem)
tst_test.c:1097: TINFO: Mounting none to /tmp/LTP_hug2vxfh1/hugetlbfs fstyp=hugetlbfs flags=0
tst_test.c:1900: TINFO: LTP version: 20240930
tst_test.c:1904: TINFO: Tested kernel: 6.4.0-150700.40-default #1 SMP Thu Jan 9 12:43:31 UTC 2025 (f78fd5e) ppc64le
tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
tst_kconfig.c:669: TINFO: CONFIG_LATENCYTOP kernel option detected which might slow the execution
tst_test.c:1724: TINFO: Overall timeout per run is 0h 02m 00s
hugemmap34.c:152: TINFO: searching for huge page and child stack placement
hugemmap34.c:175: TINFO: potential stack at address 0x7effff000000
hugemmap34.c:192: TINFO: stack = 0x7effff000000-0x7f0000000000, hugepage = 0x7dffff000000-0x7e0000000000
tst_coredump.c:32: TINFO: Avoid dumping corefile for process(pid=20714)
hugemmap34.c:56: TINFO: Child process starting with top of stack at 0x7effffffff80

tst_test.c:1785: TINFO: Killed the leftover descendant processes
tst_test.c:1792: TINFO: If you are running on slow machine, try exporting LTP_TIMEOUT_MUL > 1
tst_test.c:1793: TBROK: Test killed! (timeout?)


===log without CFLAGS='-fstack-clash-protection'===
tst_hugepage.c:84: TINFO: 1 hugepage(s) reserved
tst_tmpdir.c:317: TINFO: Using /tmp/LTP_hugjztvoX as tmpdir (btrfs filesystem)
tst_test.c:1097: TINFO: Mounting none to /tmp/LTP_hugjztvoX/hugetlbfs fstyp=hugetlbfs flags=0
tst_test.c:1900: TINFO: LTP version: 20240930
tst_test.c:1904: TINFO: Tested kernel: 6.4.0-150700.40-default #1 SMP Thu Jan 9 12:43:31 UTC 2025 (f78fd5e) ppc64le
tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
tst_kconfig.c:669: TINFO: CONFIG_LATENCYTOP kernel option detected which might slow the execution
tst_test.c:1724: TINFO: Overall timeout per run is 0h 02m 00s
hugemmap34.c:152: TINFO: searching for huge page and child stack placement
hugemmap34.c:175: TINFO: potential stack at address 0x7effff000000
hugemmap34.c:192: TINFO: stack = 0x7effff000000-0x7f0000000000, hugepage = 0x7dffff000000-0x7e0000000000
tst_coredump.c:32: TINFO: Avoid dumping corefile for process(pid=20704)
hugemmap34.c:56: TINFO: Child process starting with top of stack at 0x7effffffff80
hugemmap34.c:77: TINFO: Child process extended stack up to 0x7dfffffeff80, hasn't reached 0x7dffeffeff80
hugemmap34.c:79: TPASS: Child killed by SIGSEGV as expected

Thanks.
Regards
Gao Wei

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] hugemmap34 trigger oom if compile with CFLAGS='-fstack-clash-protection'

[Jan Stancek]

On Wed, Feb 5, 2025 at 3:54 AM Wei Gao <wegao@suse.com> wrote:
>
> Hi Jan
>
> Compiling hugemmap34 with CFLAGS='-fstack-clash-protection' causes gcc to insert additional stack address validation checks during alloca calls.  This results in kernel start allocate memory really , eventually triggering an OOM.
>
> In our normal openqa env we do not have such big memory(1024GB) to support this test. So we
>  have to disable CFLAGS='-fstack-clash-protection' if we still need this case running in our env.

Hi,

It sounds like you already found the cause. We seem to have couple choices:

1) strip the option from the CFLAGS for this testcase
diff --git a/testcases/kernel/mem/hugetlb/hugemmap/Makefile
b/testcases/kernel/mem/hugetlb/hugemmap/Makefile
index 2d651b4aa..1154ab3de 100644
--- a/testcases/kernel/mem/hugetlb/hugemmap/Makefile
+++ b/testcases/kernel/mem/hugetlb/hugemmap/Makefile
@@ -8,4 +8,7 @@ include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../Makefile.inc
 include $(top_srcdir)/include/mk/generic_leaf_target.mk

+CFLAGS_no_stack_prot := $(filter-out -fstack-clash-protection, $(CFLAGS))
+
+hugemmap34: CFLAGS=$(CFLAGS_no_stack_prot)

2) increase the memory requirements for the test.

I'd probably go with 1) since we want kernel to catch any issues anyway. And it
allows test to run on larger pool of systems.

>
> ===log with CFLAGS='-fstack-clash-protection'===
> [71073.562094] [T13546] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0-1,global_oom,task_memcg=/,task=hugemmap34,pid=13542,uid=0
> [71073.562109] [T13546] Out of memory: Killed process 13542 (hugemmap34) total-vm:4764864kB, anon-rss:2816000kB, file-rss:0kB, shmem-rss:0kB, UID:0 pgtables:636kB oom_score_adj:0
>
> tst_hugepage.c:84: TINFO: 1 hugepage(s) reserved
> tst_tmpdir.c:317: TINFO: Using /tmp/LTP_hug2vxfh1 as tmpdir (btrfs filesystem)
> tst_test.c:1097: TINFO: Mounting none to /tmp/LTP_hug2vxfh1/hugetlbfs fstyp=hugetlbfs flags=0
> tst_test.c:1900: TINFO: LTP version: 20240930
> tst_test.c:1904: TINFO: Tested kernel: 6.4.0-150700.40-default #1 SMP Thu Jan 9 12:43:31 UTC 2025 (f78fd5e) ppc64le
> tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
> tst_kconfig.c:669: TINFO: CONFIG_LATENCYTOP kernel option detected which might slow the execution
> tst_test.c:1724: TINFO: Overall timeout per run is 0h 02m 00s
> hugemmap34.c:152: TINFO: searching for huge page and child stack placement
> hugemmap34.c:175: TINFO: potential stack at address 0x7effff000000
> hugemmap34.c:192: TINFO: stack = 0x7effff000000-0x7f0000000000, hugepage = 0x7dffff000000-0x7e0000000000
> tst_coredump.c:32: TINFO: Avoid dumping corefile for process(pid=20714)
> hugemmap34.c:56: TINFO: Child process starting with top of stack at 0x7effffffff80
>
> tst_test.c:1785: TINFO: Killed the leftover descendant processes
> tst_test.c:1792: TINFO: If you are running on slow machine, try exporting LTP_TIMEOUT_MUL > 1
> tst_test.c:1793: TBROK: Test killed! (timeout?)
>
>
> ===log without CFLAGS='-fstack-clash-protection'===
> tst_hugepage.c:84: TINFO: 1 hugepage(s) reserved
> tst_tmpdir.c:317: TINFO: Using /tmp/LTP_hugjztvoX as tmpdir (btrfs filesystem)
> tst_test.c:1097: TINFO: Mounting none to /tmp/LTP_hugjztvoX/hugetlbfs fstyp=hugetlbfs flags=0
> tst_test.c:1900: TINFO: LTP version: 20240930
> tst_test.c:1904: TINFO: Tested kernel: 6.4.0-150700.40-default #1 SMP Thu Jan 9 12:43:31 UTC 2025 (f78fd5e) ppc64le
> tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
> tst_kconfig.c:669: TINFO: CONFIG_LATENCYTOP kernel option detected which might slow the execution
> tst_test.c:1724: TINFO: Overall timeout per run is 0h 02m 00s
> hugemmap34.c:152: TINFO: searching for huge page and child stack placement
> hugemmap34.c:175: TINFO: potential stack at address 0x7effff000000
> hugemmap34.c:192: TINFO: stack = 0x7effff000000-0x7f0000000000, hugepage = 0x7dffff000000-0x7e0000000000
> tst_coredump.c:32: TINFO: Avoid dumping corefile for process(pid=20704)
> hugemmap34.c:56: TINFO: Child process starting with top of stack at 0x7effffffff80
> hugemmap34.c:77: TINFO: Child process extended stack up to 0x7dfffffeff80, hasn't reached 0x7dffeffeff80
> hugemmap34.c:79: TPASS: Child killed by SIGSEGV as expected
>
> Thanks.
> Regards
> Gao Wei
>


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] hugemmap34 trigger oom if compile with CFLAGS='-fstack-clash-protection'

[Wei Gao via ltp]

On Wed, Feb 05, 2025 at 11:33:08AM +0100, Jan Stancek wrote:
> On Wed, Feb 5, 2025 at 3:54 AM Wei Gao <wegao@suse.com> wrote:
> >
> > Hi Jan
> >
> > Compiling hugemmap34 with CFLAGS='-fstack-clash-protection' causes gcc to insert additional stack address validation checks during alloca calls.  This results in kernel start allocate memory really , eventually triggering an OOM.
> >
> > In our normal openqa env we do not have such big memory(1024GB) to support this test. So we
> >  have to disable CFLAGS='-fstack-clash-protection' if we still need this case running in our env.
> 
> Hi,
> 
> It sounds like you already found the cause. We seem to have couple choices:
> 
> 1) strip the option from the CFLAGS for this testcase
> diff --git a/testcases/kernel/mem/hugetlb/hugemmap/Makefile
> b/testcases/kernel/mem/hugetlb/hugemmap/Makefile
> index 2d651b4aa..1154ab3de 100644
> --- a/testcases/kernel/mem/hugetlb/hugemmap/Makefile
> +++ b/testcases/kernel/mem/hugetlb/hugemmap/Makefile
> @@ -8,4 +8,7 @@ include $(top_srcdir)/include/mk/testcases.mk
>  include $(abs_srcdir)/../Makefile.inc
>  include $(top_srcdir)/include/mk/generic_leaf_target.mk
> 
> +CFLAGS_no_stack_prot := $(filter-out -fstack-clash-protection, $(CFLAGS))
> +
> +hugemmap34: CFLAGS=$(CFLAGS_no_stack_prot)
> 
> 2) increase the memory requirements for the test.
> 
> I'd probably go with 1) since we want kernel to catch any issues anyway. And it
> allows test to run on larger pool of systems.
> 

Agree go with 1), i send new patch 
https://patchwork.ozlabs.org/project/ltp/patch/20250206022327.11821-1-wegao@suse.com/

Thanks for your quick support.

> >
> > Thanks.
> > Regards
> > Gao Wei
> >
> 

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp