perf/x86/intel/lbr: Zero the xstate buffer on allocation
Subsystems (2)
PATCH 1/1 perf/x86/intel/lbr: Zero the xstate buffer on allocation
XRSTORS requires a valid xstate buffer to work correctly. XSAVES does not guarantee to write a fully valid buffer according to the SDM:
"XSAVES does not write to any parts of the XSAVE header other than the
XSTATE_BV and XCOMP_BV fields."
XRSTORS triggers a #GP:
"If bytes 63:16 of the XSAVE header are not all zero."
It's dubious at best how this can work at all when the buffer is not zeroed before use.
Allocate the buffers with __GFP_ZERO to prevent XRSTORS failure.
--- arch/x86/events/intel/lbr.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
arch/x86/events/intel/lbr.c +2 -1
@@ -731,7 +731,8 @@ void reserve_lbr_buffers(void)if (!kmem_cache || cpuc->lbr_xsave)continue ; - cpuc->lbr_xsave =kmem_cache_alloc_node (kmem_cache,GFP_KERNEL , + cpuc->lbr_xsave =kmem_cache_alloc_node (kmem_cache, +GFP_KERNEL | __GFP_ZERO,cpu_to_node (cpu)); } }
Quoted 6 lines · Thomas Gleixner
XRSTORS requires a valid xstate buffer to work correctly. XSAVES does not guarantee to write a fully valid buffer according to the SDM:
"XSAVES does not write to any parts of the XSAVE header other than the
XSTATE_BVandXCOMP_BVfields."
Quoted 7 lines
XRSTORS triggers a #GP:
"If bytes 63:16 of the XSAVE header are not all zero."
It's dubious at best how this can work at all when the buffer is not zeroed before use.
I didn't run into any issues when I did the test. I guess the reserved bits in the buffer may always happen to be 0. That's why it didn't set off the alarm to me. Thank you very much for the fix.
Thanks,
Kan
Quoted 24 lines
Allocate the buffers with
__GFP_ZEROto prevent XRSTORS failure.Fixes ce711ea3cab9 ("perf/x86/intel/lbr: Support XSAVES/XRSTORS for LBR context switch")Signed-off-by Thomas Gleixner <[email protected]>Cc Kan Liang <[email protected]>Cc Peter Zijlstra <[email protected]>--- arch/x86/events/intel/lbr.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)arch/x86/events/intel/lbr.c +2
-1@@ -731,7 +731,8 @@ void reserve_lbr_buffers(void)if (!kmem_cache || cpuc->lbr_xsave)continue ; - cpuc->lbr_xsave =kmem_cache_alloc_node (kmem_cache,GFP_KERNEL , + cpuc->lbr_xsave =kmem_cache_alloc_node (kmem_cache, +GFP_KERNEL | __GFP_ZERO,cpu_to_node (cpu)); } }
The following commit has been merged into the perf/urgent branch of tip:
The following commit has been merged into the perf/urgent branch of tip:
Commit-ID: 7f049fbdd57f6ea71dc741d903c19c73b2f70950 Gitweb: https://git.kernel.org/tip/7f049fbdd57f6ea71dc741d903c19c73b2f70950 Author: Thomas Gleixner <[email protected]> AuthorDate: Fri, 11 Jun 2021 15:03:16 +02:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Thu, 24 Jun 2021 08:49:03 +02:00
perf/x86/intel/lbr: Zero the xstate buffer on allocation
XRSTORS requires a valid xstate buffer to work correctly. XSAVES does not guarantee to write a fully valid buffer according to the SDM:
"XSAVES does not write to any parts of the XSAVE header other than the
XSTATE_BV and XCOMP_BV fields."
XRSTORS triggers a #GP:
"If bytes 63:16 of the XSAVE header are not all zero."
It's dubious at best how this can work at all when the buffer is not zeroed before use.
Allocate the buffers with __GFP_ZERO to prevent XRSTORS failure.
--- arch/x86/events/intel/lbr.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
arch/x86/events/intel/lbr.c +2 -1
@@ -731,7 +731,8 @@ void reserve_lbr_buffers(void)if (!kmem_cache || cpuc->lbr_xsave)continue ; - cpuc->lbr_xsave =kmem_cache_alloc_node (kmem_cache,GFP_KERNEL , + cpuc->lbr_xsave =kmem_cache_alloc_node (kmem_cache, +GFP_KERNEL | __GFP_ZERO,cpu_to_node (cpu)); } }