]> git.baikalelectronics.ru Git - kernel.git/commit
kernel/events/uprobes.c: only do FOLL_SPLIT_PMD for uprobe register
authorSong Liu <songliubraving@fb.com>
Sat, 19 Oct 2019 03:20:40 +0000 (20:20 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 19 Oct 2019 10:32:33 +0000 (06:32 -0400)
commit3f0d49701b3a5b23866553157be200f1dd012822
tree6ce098dda02fa906d14be6e8bd338d34a105945d
parentdfee40e2ff0868a983b360e5ea6eb10b0a1ba850
kernel/events/uprobes.c: only do FOLL_SPLIT_PMD for uprobe register

Attaching uprobe to text section in THP splits the PMD mapped page table
into PTE mapped entries.  On uprobe detach, we would like to regroup PMD
mapped page table entry to regain performance benefit of THP.

However, the regroup is broken For perf_event based trace_uprobe.  This
is because perf_event based trace_uprobe calls uprobe_unregister twice
on close: first in TRACE_REG_PERF_CLOSE, then in
TRACE_REG_PERF_UNREGISTER.  The second call will split the PMD mapped
page table entry, which is not the desired behavior.

Fix this by only use FOLL_SPLIT_PMD for uprobe register case.

Add a WARN() to confirm uprobe unregister never work on huge pages, and
abort the operation when this WARN() triggers.

Link: http://lkml.kernel.org/r/20191017164223.2762148-6-songliubraving@fb.com
Fixes: 45cd613d4542 ("uprobe: use FOLL_SPLIT_PMD instead of FOLL_SPLIT")
Signed-off-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/events/uprobes.c