cpumask: make "nr_cpumask_bits" unsigned
authorAlexey Dobriyan <adobriyan@gmail.com>
Mon, 8 May 2017 22:56:15 +0000 (15:56 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 9 May 2017 00:15:11 +0000 (17:15 -0700)
Bit searching functions accept "unsigned long" indices but
"nr_cpumask_bits" is "int" which is signed, so inevitable sign
extensions occur on x86_64.  Those MOVSX are #1 MOVSX bloat by number of
uses across whole kernel.

Change "nr_cpumask_bits" to unsigned, this number can't be negative
after all.  It allows to do implicit zero-extension on x86_64 without
MOVSX.

Change signed comparisons into unsigned comparisons where necessary.

Other uses looks fine because it is either argument passed to a function
or comparison is already unsigned.

Net win on allyesconfig type of kernel: ~2.8 KB (!)

add/remove: 0/0 grow/shrink: 8/725 up/down: 93/-2926 (-2833)
function                                     old     new   delta
xen_exit_mmap                                691     735     +44
qstat_read                                   426     440     +14
__cpufreq_cooling_register                  1678    1687      +9
trace_rb_cpu_prepare                         447     455      +8
vermagic                                      54      60      +6
nfp_driver_version                            54      60      +6
rcu_torture_stats_print                     1147    1151      +4
find_next_push_cpu                           267     269      +2
xen_irq_resume                               961     960      -1
...
init_vp_index                                946     906     -40
od_set_powersave_bias                        328     281     -47
power_cpu_exit                               193     139     -54
arch_show_interrupts                        3538    3484     -54
select_idle_sibling                         1558    1471     -87
Total: Before=158358910, After=158356077, chg -0.00%

Same arguments apply to "nr_cpu_ids" but I haven't yet found enough
courage to delve into this issue (and proper fix may require new type
"cpu_t" which is whole separate story).

Link: http://lkml.kernel.org/r/20170309205322.GA1728@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/mips/kernel/perf_event_mipsxx.c
arch/s390/kernel/perf_cpum_sf.c
include/linux/cpumask.h

index 9452b02ce0797e7ac890d1785f50907d9e747b5d..313a88b2973f673f0fac0998c36517fd2139504b 100644 (file)
@@ -618,7 +618,7 @@ static int mipspmu_event_init(struct perf_event *event)
                return -ENOENT;
        }
 
-       if (event->cpu >= nr_cpumask_bits ||
+       if ((unsigned int)event->cpu >= nr_cpumask_bits ||
            (event->cpu >= 0 && !cpu_online(event->cpu)))
                return -ENODEV;
 
index 9a4f279d25ca9675a711d604f0a9ec7c633cae7b..ca960d0370d56517f7cf10ec86f20c7ea31c3746 100644 (file)
@@ -823,7 +823,7 @@ static int cpumsf_pmu_event_init(struct perf_event *event)
        }
 
        /* Check online status of the CPU to which the event is pinned */
-       if (event->cpu >= nr_cpumask_bits ||
+       if ((unsigned int)event->cpu >= nr_cpumask_bits ||
            (event->cpu >= 0 && !cpu_online(event->cpu)))
                return -ENODEV;
 
index 1a675604b17dfc5cc0357136e1b1eb93e53a5f1d..2404ad238c0b8c0309e8c1480339a19967968edf 100644 (file)
@@ -40,9 +40,9 @@ extern int nr_cpu_ids;
 #ifdef CONFIG_CPUMASK_OFFSTACK
 /* Assuming NR_CPUS is huge, a runtime limit is more efficient.  Also,
  * not all bits may be allocated. */
-#define nr_cpumask_bits        nr_cpu_ids
+#define nr_cpumask_bits        ((unsigned int)nr_cpu_ids)
 #else
-#define nr_cpumask_bits        NR_CPUS
+#define nr_cpumask_bits        ((unsigned int)NR_CPUS)
 #endif
 
 /*