releases/3.18.101/powerpc-avoid-taking-a-data-miss-on-every-userspace-instruction-miss.patch

   1 From foo@baz Mon Mar 19 10:11:52 CET 2018
   2 From: Anton Blanchard <anton@samba.org>
   3 Date: Mon, 3 Apr 2017 16:41:02 +1000
   4 Subject: powerpc: Avoid taking a data miss on every userspace instruction miss
   5
   6 From: Anton Blanchard <anton@samba.org>
   7
   8
   9 [ Upstream commit a7a9dcd882a67b68568868b988289fce5ffd8419 ]
  10
  11 Early on in do_page_fault() we call store_updates_sp(), regardless of
  12 the type of exception. For an instruction miss this doesn't make
  13 sense, because we only use this information to detect if a data miss
  14 is the result of a stack expansion instruction or not.
  15
  16 Worse still, it results in a data miss within every userspace
  17 instruction miss handler, because we try and load the very instruction
  18 we are about to install a pte for!
  19
  20 A simple exec microbenchmark runs 6% faster on POWER8 with this fix:
  21
  22  #include <stdlib.h>
  23  #include <stdio.h>
  24  #include <unistd.h>
  25
  26 int main(int argc, char *argv[])
  27 {
  28         unsigned long left = atol(argv[1]);
  29         char leftstr[16];
  30
  31         if (left-- == 0)
  32                 return 0;
  33
  34         sprintf(leftstr, "%ld", left);
  35         execlp(argv[0], argv[0], leftstr, NULL);
  36         perror("exec failed\n");
  37
  38         return 0;
  39 }
  40
  41 Pass the number of iterations on the command line (eg 10000) and time
  42 how long it takes to execute.
  43
  44 Signed-off-by: Anton Blanchard <anton@samba.org>
  45 Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
  46 Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
  47 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  48 ---
  49  arch/powerpc/mm/fault.c |    2 +-
  50  1 file changed, 1 insertion(+), 1 deletion(-)
  51
  52 --- a/arch/powerpc/mm/fault.c
  53 +++ b/arch/powerpc/mm/fault.c
  54 @@ -294,7 +294,7 @@ int __kprobes do_page_fault(struct pt_re
  55          * can result in fault, which will cause a deadlock when called with
  56          * mmap_sem held
  57          */
  58 -       if (user_mode(regs))
  59 +       if (!is_exec && user_mode(regs))
  60                 store_update_sp = store_updates_sp(regs);
  61
  62         if (user_mode(regs))