releases/4.4.157/perf-tools-allow-overriding-max_nr_cpus-at-compile-time.patch

   1 From foo@baz Mon Sep 17 12:15:09 CEST 2018
   2 From: Christophe Leroy <christophe.leroy@c-s.fr>
   3 Date: Fri, 22 Sep 2017 13:20:43 +0200
   4 Subject: perf tools: Allow overriding MAX_NR_CPUS at compile time
   5
   6 From: Christophe Leroy <christophe.leroy@c-s.fr>
   7
   8 [ Upstream commit 21b8732eb4479b579bda9ee38e62b2c312c2a0e5 ]
   9
  10 After update of kernel, the perf tool doesn't run anymore on my 32MB RAM
  11 powerpc board, but still runs on a 128MB RAM board:
  12
  13   ~# strace perf
  14   execve("/usr/sbin/perf", ["perf"], [/* 12 vars */]) = -1 ENOMEM (Cannot allocate memory)
  15   --- SIGSEGV {si_signo=SIGSEGV, si_code=SI_KERNEL, si_addr=0} ---
  16   +++ killed by SIGSEGV +++
  17   Segmentation fault
  18
  19 objdump -x shows that .bss section has a huge size of 24Mbytes:
  20
  21  27 .bss          016baca8  101cebb8  101cebb8  001cd988  2**3
  22
  23 With especially the following objects having quite big size:
  24
  25   10205f80 l     O .bss 00140000     runtime_cycles_stats
  26   10345f80 l     O .bss 00140000     runtime_stalled_cycles_front_stats
  27   10485f80 l     O .bss 00140000     runtime_stalled_cycles_back_stats
  28   105c5f80 l     O .bss 00140000     runtime_branches_stats
  29   10705f80 l     O .bss 00140000     runtime_cacherefs_stats
  30   10845f80 l     O .bss 00140000     runtime_l1_dcache_stats
  31   10985f80 l     O .bss 00140000     runtime_l1_icache_stats
  32   10ac5f80 l     O .bss 00140000     runtime_ll_cache_stats
  33   10c05f80 l     O .bss 00140000     runtime_itlb_cache_stats
  34   10d45f80 l     O .bss 00140000     runtime_dtlb_cache_stats
  35   10e85f80 l     O .bss 00140000     runtime_cycles_in_tx_stats
  36   10fc5f80 l     O .bss 00140000     runtime_transaction_stats
  37   11105f80 l     O .bss 00140000     runtime_elision_stats
  38   11245f80 l     O .bss 00140000     runtime_topdown_total_slots
  39   11385f80 l     O .bss 00140000     runtime_topdown_slots_retired
  40   114c5f80 l     O .bss 00140000     runtime_topdown_slots_issued
  41   11605f80 l     O .bss 00140000     runtime_topdown_fetch_bubbles
  42   11745f80 l     O .bss 00140000     runtime_topdown_recovery_bubbles
  43
  44 This is due to commit 4d255766d28b1 ("perf: Bump max number of cpus
  45 to 1024"), because many tables are sized with MAX_NR_CPUS
  46
  47 This patch gives the opportunity to redefine MAX_NR_CPUS via
  48
  49   $ make EXTRA_CFLAGS=-DMAX_NR_CPUS=1
  50
  51 Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
  52 Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
  53 Cc: Peter Zijlstra <peterz@infradead.org>
  54 Cc: linuxppc-dev@lists.ozlabs.org
  55 Link: http://lkml.kernel.org/r/20170922112043.8349468C57@po15668-vm-win7.idsi0.si.c-s.fr
  56 Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
  57 Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
  58 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  59 ---
  60  tools/perf/perf.h |    2 ++
  61  1 file changed, 2 insertions(+)
  62
  63 --- a/tools/perf/perf.h
  64 +++ b/tools/perf/perf.h
  65 @@ -29,7 +29,9 @@ static inline unsigned long long rdclock
  66         return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
  67  }
  68
  69 +#ifndef MAX_NR_CPUS
  70  #define MAX_NR_CPUS                    1024
  71 +#endif
  72
  73  extern const char *input_name;
  74  extern bool perf_host, perf_guest;