Thomas Gleixner | 457c899 | 2019-05-19 13:08:55 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 3 | * linux/kernel/profile.c |
| 4 | * Simple profiling. Manages a direct-mapped profile hit count buffer, |
| 5 | * with configurable resolution, support for restricting the cpus on |
| 6 | * which profiling is done, and switching between cpu time and |
| 7 | * schedule() calls via kernel command line parameters passed at boot. |
| 8 | * |
| 9 | * Scheduler profiling support, Arjan van de Ven and Ingo Molnar, |
| 10 | * Red Hat, July 2004 |
| 11 | * Consolidation of architecture support code for profiling, |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 12 | * Nadia Yvette Chambers, Oracle, July 2004 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * Amortized hit count accounting via per-cpu open-addressed hashtables |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 14 | * to resolve timer interrupt livelocks, Nadia Yvette Chambers, |
| 15 | * Oracle, 2004 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | */ |
| 17 | |
Paul Gortmaker | 9984de1 | 2011-05-23 14:51:41 -0400 | [diff] [blame] | 18 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | #include <linux/profile.h> |
Mike Rapoport | 57c8a66 | 2018-10-30 15:09:49 -0700 | [diff] [blame] | 20 | #include <linux/memblock.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include <linux/notifier.h> |
| 22 | #include <linux/mm.h> |
| 23 | #include <linux/cpumask.h> |
| 24 | #include <linux/cpu.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/highmem.h> |
Arjan van de Ven | 97d1f15 | 2006-03-23 03:00:24 -0800 | [diff] [blame] | 26 | #include <linux/mutex.h> |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 27 | #include <linux/slab.h> |
| 28 | #include <linux/vmalloc.h> |
Ingo Molnar | 3905f9a | 2017-02-05 12:07:04 +0100 | [diff] [blame] | 29 | #include <linux/sched/stat.h> |
| 30 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include <asm/sections.h> |
David Howells | 7d12e78 | 2006-10-05 14:55:46 +0100 | [diff] [blame] | 32 | #include <asm/irq_regs.h> |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 33 | #include <asm/ptrace.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | |
| 35 | struct profile_hit { |
| 36 | u32 pc, hits; |
| 37 | }; |
| 38 | #define PROFILE_GRPSHIFT 3 |
| 39 | #define PROFILE_GRPSZ (1 << PROFILE_GRPSHIFT) |
| 40 | #define NR_PROFILE_HIT (PAGE_SIZE/sizeof(struct profile_hit)) |
| 41 | #define NR_PROFILE_GRP (NR_PROFILE_HIT/PROFILE_GRPSZ) |
| 42 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | static atomic_t *prof_buffer; |
Pavel Skripkin | 2d186af | 2021-09-07 19:58:21 -0700 | [diff] [blame] | 44 | static unsigned long prof_len; |
| 45 | static unsigned short int prof_shift; |
Ingo Molnar | 07031e1 | 2007-01-10 23:15:38 -0800 | [diff] [blame] | 46 | |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 47 | int prof_on __read_mostly; |
Ingo Molnar | 07031e1 | 2007-01-10 23:15:38 -0800 | [diff] [blame] | 48 | EXPORT_SYMBOL_GPL(prof_on); |
| 49 | |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 50 | static cpumask_var_t prof_cpu_mask; |
Arnd Bergmann | ade356b | 2016-03-22 14:27:26 -0700 | [diff] [blame] | 51 | #if defined(CONFIG_SMP) && defined(CONFIG_PROC_FS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | static DEFINE_PER_CPU(struct profile_hit *[2], cpu_profile_hits); |
| 53 | static DEFINE_PER_CPU(int, cpu_profile_flip); |
Arjan van de Ven | 97d1f15 | 2006-03-23 03:00:24 -0800 | [diff] [blame] | 54 | static DEFINE_MUTEX(profile_flip_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | #endif /* CONFIG_SMP */ |
| 56 | |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 57 | int profile_setup(char *str) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | { |
Fabian Frederick | f3da64d | 2014-06-06 14:37:30 -0700 | [diff] [blame] | 59 | static const char schedstr[] = "schedule"; |
| 60 | static const char sleepstr[] = "sleep"; |
| 61 | static const char kvmstr[] = "kvm"; |
wuchi | 35783cc | 2022-09-01 08:31:21 +0800 | [diff] [blame] | 62 | const char *select = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | int par; |
| 64 | |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 65 | if (!strncmp(str, sleepstr, strlen(sleepstr))) { |
Mel Gorman | b3da2a7 | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 66 | #ifdef CONFIG_SCHEDSTATS |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 67 | force_schedstat_enabled(); |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 68 | prof_on = SLEEP_PROFILING; |
wuchi | 35783cc | 2022-09-01 08:31:21 +0800 | [diff] [blame] | 69 | select = sleepstr; |
Mel Gorman | b3da2a7 | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 70 | #else |
Fabian Frederick | aba871f | 2014-06-06 14:37:29 -0700 | [diff] [blame] | 71 | pr_warn("kernel sleep profiling requires CONFIG_SCHEDSTATS\n"); |
Mel Gorman | b3da2a7 | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 72 | #endif /* CONFIG_SCHEDSTATS */ |
Ingo Molnar | a75acf8 | 2007-01-05 16:36:29 -0800 | [diff] [blame] | 73 | } else if (!strncmp(str, schedstr, strlen(schedstr))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | prof_on = SCHED_PROFILING; |
wuchi | 35783cc | 2022-09-01 08:31:21 +0800 | [diff] [blame] | 75 | select = schedstr; |
Ingo Molnar | 07031e1 | 2007-01-10 23:15:38 -0800 | [diff] [blame] | 76 | } else if (!strncmp(str, kvmstr, strlen(kvmstr))) { |
| 77 | prof_on = KVM_PROFILING; |
wuchi | 35783cc | 2022-09-01 08:31:21 +0800 | [diff] [blame] | 78 | select = kvmstr; |
William Lee Irwin III | dfaa9c9 | 2005-05-16 21:53:58 -0700 | [diff] [blame] | 79 | } else if (get_option(&str, &par)) { |
Pavel Skripkin | 2d186af | 2021-09-07 19:58:21 -0700 | [diff] [blame] | 80 | prof_shift = clamp(par, 0, BITS_PER_LONG - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | prof_on = CPU_PROFILING; |
Pavel Skripkin | 2d186af | 2021-09-07 19:58:21 -0700 | [diff] [blame] | 82 | pr_info("kernel profiling enabled (shift: %u)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | prof_shift); |
| 84 | } |
wuchi | 35783cc | 2022-09-01 08:31:21 +0800 | [diff] [blame] | 85 | |
| 86 | if (select) { |
| 87 | if (str[strlen(select)] == ',') |
| 88 | str += strlen(select) + 1; |
| 89 | if (get_option(&str, &par)) |
| 90 | prof_shift = clamp(par, 0, BITS_PER_LONG - 1); |
| 91 | pr_info("kernel %s profiling enabled (shift: %u)\n", |
| 92 | select, prof_shift); |
| 93 | } |
| 94 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | return 1; |
| 96 | } |
| 97 | __setup("profile=", profile_setup); |
| 98 | |
| 99 | |
Paul Mundt | ce05fcc | 2008-10-29 14:01:07 -0700 | [diff] [blame] | 100 | int __ref profile_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | { |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 102 | int buffer_bytes; |
Paolo Ciarrocchi | 1ad82fd | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 103 | if (!prof_on) |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 104 | return 0; |
Paolo Ciarrocchi | 1ad82fd | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 105 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | /* only text is profiled */ |
| 107 | prof_len = (_etext - _stext) >> prof_shift; |
Chen Zhongjin | 0fe6ee8f | 2022-05-31 09:28:54 +0800 | [diff] [blame] | 108 | |
| 109 | if (!prof_len) { |
| 110 | pr_warn("profiling shift: %u too large\n", prof_shift); |
| 111 | prof_on = 0; |
| 112 | return -EINVAL; |
| 113 | } |
| 114 | |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 115 | buffer_bytes = prof_len*sizeof(atomic_t); |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 116 | |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 117 | if (!alloc_cpumask_var(&prof_cpu_mask, GFP_KERNEL)) |
| 118 | return -ENOMEM; |
| 119 | |
Hugh Dickins | acd8957 | 2009-02-09 19:20:50 +0000 | [diff] [blame] | 120 | cpumask_copy(prof_cpu_mask, cpu_possible_mask); |
| 121 | |
Mel Gorman | b62f495 | 2009-07-29 15:04:09 -0700 | [diff] [blame] | 122 | prof_buffer = kzalloc(buffer_bytes, GFP_KERNEL|__GFP_NOWARN); |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 123 | if (prof_buffer) |
| 124 | return 0; |
| 125 | |
Mel Gorman | b62f495 | 2009-07-29 15:04:09 -0700 | [diff] [blame] | 126 | prof_buffer = alloc_pages_exact(buffer_bytes, |
| 127 | GFP_KERNEL|__GFP_ZERO|__GFP_NOWARN); |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 128 | if (prof_buffer) |
| 129 | return 0; |
| 130 | |
Jesper Juhl | 559fa6e | 2010-10-30 21:56:26 +0200 | [diff] [blame] | 131 | prof_buffer = vzalloc(buffer_bytes); |
| 132 | if (prof_buffer) |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 133 | return 0; |
| 134 | |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 135 | free_cpumask_var(prof_cpu_mask); |
Dave Hansen | 22b8ce9 | 2008-10-15 22:01:46 -0700 | [diff] [blame] | 136 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | } |
| 138 | |
Arnd Bergmann | ade356b | 2016-03-22 14:27:26 -0700 | [diff] [blame] | 139 | #if defined(CONFIG_SMP) && defined(CONFIG_PROC_FS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | /* |
| 141 | * Each cpu has a pair of open-addressed hashtables for pending |
| 142 | * profile hits. read_profile() IPI's all cpus to request them |
| 143 | * to flip buffers and flushes their contents to prof_buffer itself. |
| 144 | * Flip requests are serialized by the profile_flip_mutex. The sole |
| 145 | * use of having a second hashtable is for avoiding cacheline |
| 146 | * contention that would otherwise happen during flushes of pending |
| 147 | * profile hits required for the accuracy of reported profile hits |
| 148 | * and so resurrect the interrupt livelock issue. |
| 149 | * |
| 150 | * The open-addressed hashtables are indexed by profile buffer slot |
| 151 | * and hold the number of pending hits to that profile buffer slot on |
| 152 | * a cpu in an entry. When the hashtable overflows, all pending hits |
| 153 | * are accounted to their corresponding profile buffer slots with |
| 154 | * atomic_add() and the hashtable emptied. As numerous pending hits |
| 155 | * may be accounted to a profile buffer slot in a hashtable entry, |
| 156 | * this amortizes a number of atomic profile buffer increments likely |
| 157 | * to be far larger than the number of entries in the hashtable, |
| 158 | * particularly given that the number of distinct profile buffer |
| 159 | * positions to which hits are accounted during short intervals (e.g. |
| 160 | * several seconds) is usually very small. Exclusion from buffer |
| 161 | * flipping is provided by interrupt disablement (note that for |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 162 | * SCHED_PROFILING or SLEEP_PROFILING profile_hit() may be called from |
| 163 | * process context). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | * The hash function is meant to be lightweight as opposed to strong, |
| 165 | * and was vaguely inspired by ppc64 firmware-supported inverted |
| 166 | * pagetable hash functions, but uses a full hashtable full of finite |
| 167 | * collision chains, not just pairs of them. |
| 168 | * |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 169 | * -- nyc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | */ |
| 171 | static void __profile_flip_buffers(void *unused) |
| 172 | { |
| 173 | int cpu = smp_processor_id(); |
| 174 | |
| 175 | per_cpu(cpu_profile_flip, cpu) = !per_cpu(cpu_profile_flip, cpu); |
| 176 | } |
| 177 | |
| 178 | static void profile_flip_buffers(void) |
| 179 | { |
| 180 | int i, j, cpu; |
| 181 | |
Arjan van de Ven | 97d1f15 | 2006-03-23 03:00:24 -0800 | [diff] [blame] | 182 | mutex_lock(&profile_flip_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | j = per_cpu(cpu_profile_flip, get_cpu()); |
| 184 | put_cpu(); |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 185 | on_each_cpu(__profile_flip_buffers, NULL, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | for_each_online_cpu(cpu) { |
| 187 | struct profile_hit *hits = per_cpu(cpu_profile_hits, cpu)[j]; |
| 188 | for (i = 0; i < NR_PROFILE_HIT; ++i) { |
| 189 | if (!hits[i].hits) { |
| 190 | if (hits[i].pc) |
| 191 | hits[i].pc = 0; |
| 192 | continue; |
| 193 | } |
| 194 | atomic_add(hits[i].hits, &prof_buffer[hits[i].pc]); |
| 195 | hits[i].hits = hits[i].pc = 0; |
| 196 | } |
| 197 | } |
Arjan van de Ven | 97d1f15 | 2006-03-23 03:00:24 -0800 | [diff] [blame] | 198 | mutex_unlock(&profile_flip_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | } |
| 200 | |
| 201 | static void profile_discard_flip_buffers(void) |
| 202 | { |
| 203 | int i, cpu; |
| 204 | |
Arjan van de Ven | 97d1f15 | 2006-03-23 03:00:24 -0800 | [diff] [blame] | 205 | mutex_lock(&profile_flip_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | i = per_cpu(cpu_profile_flip, get_cpu()); |
| 207 | put_cpu(); |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 208 | on_each_cpu(__profile_flip_buffers, NULL, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | for_each_online_cpu(cpu) { |
| 210 | struct profile_hit *hits = per_cpu(cpu_profile_hits, cpu)[i]; |
| 211 | memset(hits, 0, NR_PROFILE_HIT*sizeof(struct profile_hit)); |
| 212 | } |
Arjan van de Ven | 97d1f15 | 2006-03-23 03:00:24 -0800 | [diff] [blame] | 213 | mutex_unlock(&profile_flip_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | } |
| 215 | |
Rakib Mullick | 6f7bd76 | 2011-05-26 16:26:00 -0700 | [diff] [blame] | 216 | static void do_profile_hits(int type, void *__pc, unsigned int nr_hits) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | { |
| 218 | unsigned long primary, secondary, flags, pc = (unsigned long)__pc; |
| 219 | int i, j, cpu; |
| 220 | struct profile_hit *hits; |
| 221 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | pc = min((pc - (unsigned long)_stext) >> prof_shift, prof_len - 1); |
| 223 | i = primary = (pc & (NR_PROFILE_GRP - 1)) << PROFILE_GRPSHIFT; |
| 224 | secondary = (~(pc << 1) & (NR_PROFILE_GRP - 1)) << PROFILE_GRPSHIFT; |
| 225 | cpu = get_cpu(); |
| 226 | hits = per_cpu(cpu_profile_hits, cpu)[per_cpu(cpu_profile_flip, cpu)]; |
| 227 | if (!hits) { |
| 228 | put_cpu(); |
| 229 | return; |
| 230 | } |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 231 | /* |
| 232 | * We buffer the global profiler buffer into a per-CPU |
| 233 | * queue and thus reduce the number of global (and possibly |
| 234 | * NUMA-alien) accesses. The write-queue is self-coalescing: |
| 235 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | local_irq_save(flags); |
| 237 | do { |
| 238 | for (j = 0; j < PROFILE_GRPSZ; ++j) { |
| 239 | if (hits[i + j].pc == pc) { |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 240 | hits[i + j].hits += nr_hits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | goto out; |
| 242 | } else if (!hits[i + j].hits) { |
| 243 | hits[i + j].pc = pc; |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 244 | hits[i + j].hits = nr_hits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | goto out; |
| 246 | } |
| 247 | } |
| 248 | i = (i + secondary) & (NR_PROFILE_HIT - 1); |
| 249 | } while (i != primary); |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 250 | |
| 251 | /* |
| 252 | * Add the current hit(s) and flush the write-queue out |
| 253 | * to the global buffer: |
| 254 | */ |
| 255 | atomic_add(nr_hits, &prof_buffer[pc]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | for (i = 0; i < NR_PROFILE_HIT; ++i) { |
| 257 | atomic_add(hits[i].hits, &prof_buffer[hits[i].pc]); |
| 258 | hits[i].pc = hits[i].hits = 0; |
| 259 | } |
| 260 | out: |
| 261 | local_irq_restore(flags); |
| 262 | put_cpu(); |
| 263 | } |
| 264 | |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 265 | static int profile_dead_cpu(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | { |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 267 | struct page *page; |
| 268 | int i; |
| 269 | |
Nathan Chancellor | ef70eff | 2019-12-04 16:50:50 -0800 | [diff] [blame] | 270 | if (cpumask_available(prof_cpu_mask)) |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 271 | cpumask_clear_cpu(cpu, prof_cpu_mask); |
| 272 | |
| 273 | for (i = 0; i < 2; i++) { |
| 274 | if (per_cpu(cpu_profile_hits, cpu)[i]) { |
| 275 | page = virt_to_page(per_cpu(cpu_profile_hits, cpu)[i]); |
| 276 | per_cpu(cpu_profile_hits, cpu)[i] = NULL; |
| 277 | __free_page(page); |
| 278 | } |
| 279 | } |
| 280 | return 0; |
| 281 | } |
| 282 | |
| 283 | static int profile_prepare_cpu(unsigned int cpu) |
| 284 | { |
| 285 | int i, node = cpu_to_mem(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | struct page *page; |
| 287 | |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 288 | per_cpu(cpu_profile_flip, cpu) = 0; |
| 289 | |
| 290 | for (i = 0; i < 2; i++) { |
| 291 | if (per_cpu(cpu_profile_hits, cpu)[i]) |
| 292 | continue; |
| 293 | |
| 294 | page = __alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0); |
| 295 | if (!page) { |
| 296 | profile_dead_cpu(cpu); |
| 297 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | } |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 299 | per_cpu(cpu_profile_hits, cpu)[i] = page_address(page); |
| 300 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | } |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 302 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | } |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 304 | |
| 305 | static int profile_online_cpu(unsigned int cpu) |
| 306 | { |
Nathan Chancellor | ef70eff | 2019-12-04 16:50:50 -0800 | [diff] [blame] | 307 | if (cpumask_available(prof_cpu_mask)) |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 308 | cpumask_set_cpu(cpu, prof_cpu_mask); |
| 309 | |
| 310 | return 0; |
| 311 | } |
| 312 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 313 | #else /* !CONFIG_SMP */ |
| 314 | #define profile_flip_buffers() do { } while (0) |
| 315 | #define profile_discard_flip_buffers() do { } while (0) |
| 316 | |
Rakib Mullick | 6f7bd76 | 2011-05-26 16:26:00 -0700 | [diff] [blame] | 317 | static void do_profile_hits(int type, void *__pc, unsigned int nr_hits) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | { |
| 319 | unsigned long pc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | pc = ((unsigned long)__pc - (unsigned long)_stext) >> prof_shift; |
Ingo Molnar | ece8a68 | 2006-12-06 20:37:24 -0800 | [diff] [blame] | 321 | atomic_add(nr_hits, &prof_buffer[min(pc, prof_len - 1)]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | } |
| 323 | #endif /* !CONFIG_SMP */ |
Rakib Mullick | 6f7bd76 | 2011-05-26 16:26:00 -0700 | [diff] [blame] | 324 | |
| 325 | void profile_hits(int type, void *__pc, unsigned int nr_hits) |
| 326 | { |
| 327 | if (prof_on != type || !prof_buffer) |
| 328 | return; |
| 329 | do_profile_hits(type, __pc, nr_hits); |
| 330 | } |
Andrew Morton | bbe1a59b | 2007-01-22 20:40:33 -0800 | [diff] [blame] | 331 | EXPORT_SYMBOL_GPL(profile_hits); |
| 332 | |
David Howells | 7d12e78 | 2006-10-05 14:55:46 +0100 | [diff] [blame] | 333 | void profile_tick(int type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | { |
David Howells | 7d12e78 | 2006-10-05 14:55:46 +0100 | [diff] [blame] | 335 | struct pt_regs *regs = get_irq_regs(); |
| 336 | |
Nathan Chancellor | ef70eff | 2019-12-04 16:50:50 -0800 | [diff] [blame] | 337 | if (!user_mode(regs) && cpumask_available(prof_cpu_mask) && |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 338 | cpumask_test_cpu(smp_processor_id(), prof_cpu_mask)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | profile_hit(type, (void *)profile_pc(regs)); |
| 340 | } |
| 341 | |
| 342 | #ifdef CONFIG_PROC_FS |
| 343 | #include <linux/proc_fs.h> |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 344 | #include <linux/seq_file.h> |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 345 | #include <linux/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 347 | static int prof_cpu_mask_proc_show(struct seq_file *m, void *v) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | { |
Tejun Heo | ccbd59c | 2015-02-13 14:38:13 -0800 | [diff] [blame] | 349 | seq_printf(m, "%*pb\n", cpumask_pr_args(prof_cpu_mask)); |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 350 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | } |
| 352 | |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 353 | static int prof_cpu_mask_proc_open(struct inode *inode, struct file *file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | { |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 355 | return single_open(file, prof_cpu_mask_proc_show, NULL); |
| 356 | } |
| 357 | |
| 358 | static ssize_t prof_cpu_mask_proc_write(struct file *file, |
| 359 | const char __user *buffer, size_t count, loff_t *pos) |
| 360 | { |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 361 | cpumask_var_t new_value; |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 362 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | |
Tetsuo Handa | c5e3a41 | 2021-04-01 14:58:23 +0900 | [diff] [blame] | 364 | if (!zalloc_cpumask_var(&new_value, GFP_KERNEL)) |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 365 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 367 | err = cpumask_parse_user(buffer, count, new_value); |
| 368 | if (!err) { |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 369 | cpumask_copy(prof_cpu_mask, new_value); |
| 370 | err = count; |
Rusty Russell | c309b91 | 2009-01-01 10:12:27 +1030 | [diff] [blame] | 371 | } |
| 372 | free_cpumask_var(new_value); |
| 373 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | } |
| 375 | |
Alexey Dobriyan | 97a3253 | 2020-02-03 17:37:17 -0800 | [diff] [blame] | 376 | static const struct proc_ops prof_cpu_mask_proc_ops = { |
| 377 | .proc_open = prof_cpu_mask_proc_open, |
| 378 | .proc_read = seq_read, |
| 379 | .proc_lseek = seq_lseek, |
| 380 | .proc_release = single_release, |
| 381 | .proc_write = prof_cpu_mask_proc_write, |
Alexey Dobriyan | 583a22e | 2009-09-18 12:57:09 -0700 | [diff] [blame] | 382 | }; |
| 383 | |
Al Viro | fbd387a | 2013-04-01 20:48:34 -0400 | [diff] [blame] | 384 | void create_prof_cpu_mask(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | /* create /proc/irq/prof_cpu_mask */ |
Alexey Dobriyan | 97a3253 | 2020-02-03 17:37:17 -0800 | [diff] [blame] | 387 | proc_create("irq/prof_cpu_mask", 0600, NULL, &prof_cpu_mask_proc_ops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | } |
| 389 | |
| 390 | /* |
| 391 | * This function accesses profiling information. The returned data is |
| 392 | * binary: the sampling step and the actual contents of the profile |
| 393 | * buffer. Use of the program readprofile is recommended in order to |
| 394 | * get meaningful info out of these data. |
| 395 | */ |
| 396 | static ssize_t |
| 397 | read_profile(struct file *file, char __user *buf, size_t count, loff_t *ppos) |
| 398 | { |
| 399 | unsigned long p = *ppos; |
| 400 | ssize_t read; |
Paolo Ciarrocchi | 1ad82fd | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 401 | char *pnt; |
Pavel Skripkin | 2d186af | 2021-09-07 19:58:21 -0700 | [diff] [blame] | 402 | unsigned long sample_step = 1UL << prof_shift; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 403 | |
| 404 | profile_flip_buffers(); |
| 405 | if (p >= (prof_len+1)*sizeof(unsigned int)) |
| 406 | return 0; |
| 407 | if (count > (prof_len+1)*sizeof(unsigned int) - p) |
| 408 | count = (prof_len+1)*sizeof(unsigned int) - p; |
| 409 | read = 0; |
| 410 | |
| 411 | while (p < sizeof(unsigned int) && count > 0) { |
Paolo Ciarrocchi | 1ad82fd | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 412 | if (put_user(*((char *)(&sample_step)+p), buf)) |
Heiko Carstens | 064b022 | 2006-12-06 20:36:37 -0800 | [diff] [blame] | 413 | return -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | buf++; p++; count--; read++; |
| 415 | } |
| 416 | pnt = (char *)prof_buffer + p - sizeof(atomic_t); |
Paolo Ciarrocchi | 1ad82fd | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 417 | if (copy_to_user(buf, (void *)pnt, count)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | return -EFAULT; |
| 419 | read += count; |
| 420 | *ppos += read; |
| 421 | return read; |
| 422 | } |
| 423 | |
Ben Dooks | 787dbea | 2022-07-21 20:55:09 +0100 | [diff] [blame] | 424 | /* default is to not implement this call */ |
| 425 | int __weak setup_profiling_timer(unsigned mult) |
| 426 | { |
| 427 | return -EINVAL; |
| 428 | } |
| 429 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | /* |
| 431 | * Writing to /proc/profile resets the counters |
| 432 | * |
| 433 | * Writing a 'profiling multiplier' value into it also re-sets the profiling |
| 434 | * interrupt frequency, on architectures that support this. |
| 435 | */ |
| 436 | static ssize_t write_profile(struct file *file, const char __user *buf, |
| 437 | size_t count, loff_t *ppos) |
| 438 | { |
| 439 | #ifdef CONFIG_SMP |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 440 | if (count == sizeof(int)) { |
| 441 | unsigned int multiplier; |
| 442 | |
| 443 | if (copy_from_user(&multiplier, buf, sizeof(int))) |
| 444 | return -EFAULT; |
| 445 | |
| 446 | if (setup_profiling_timer(multiplier)) |
| 447 | return -EINVAL; |
| 448 | } |
| 449 | #endif |
| 450 | profile_discard_flip_buffers(); |
| 451 | memset(prof_buffer, 0, prof_len * sizeof(atomic_t)); |
| 452 | return count; |
| 453 | } |
| 454 | |
Alexey Dobriyan | 97a3253 | 2020-02-03 17:37:17 -0800 | [diff] [blame] | 455 | static const struct proc_ops profile_proc_ops = { |
| 456 | .proc_read = read_profile, |
| 457 | .proc_write = write_profile, |
| 458 | .proc_lseek = default_llseek, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | }; |
| 460 | |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 461 | int __ref create_proc_profile(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | { |
| 463 | struct proc_dir_entry *entry; |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 464 | #ifdef CONFIG_SMP |
| 465 | enum cpuhp_state online_state; |
| 466 | #endif |
| 467 | |
Srivatsa S. Bhat | c270a81 | 2014-03-11 02:12:08 +0530 | [diff] [blame] | 468 | int err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | |
| 470 | if (!prof_on) |
| 471 | return 0; |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 472 | #ifdef CONFIG_SMP |
| 473 | err = cpuhp_setup_state(CPUHP_PROFILE_PREPARE, "PROFILE_PREPARE", |
| 474 | profile_prepare_cpu, profile_dead_cpu); |
| 475 | if (err) |
| 476 | return err; |
Srivatsa S. Bhat | c270a81 | 2014-03-11 02:12:08 +0530 | [diff] [blame] | 477 | |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 478 | err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "AP_PROFILE_ONLINE", |
| 479 | profile_online_cpu, NULL); |
| 480 | if (err < 0) |
| 481 | goto err_state_prep; |
| 482 | online_state = err; |
| 483 | err = 0; |
| 484 | #endif |
Denis V. Lunev | c33fff0 | 2008-04-29 01:02:31 -0700 | [diff] [blame] | 485 | entry = proc_create("profile", S_IWUSR | S_IRUGO, |
Alexey Dobriyan | 97a3253 | 2020-02-03 17:37:17 -0800 | [diff] [blame] | 486 | NULL, &profile_proc_ops); |
Paolo Ciarrocchi | 1ad82fd | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 487 | if (!entry) |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 488 | goto err_state_onl; |
David Howells | 271a15e | 2013-04-12 00:38:51 +0100 | [diff] [blame] | 489 | proc_set_size(entry, (1 + prof_len) * sizeof(atomic_t)); |
Srivatsa S. Bhat | c270a81 | 2014-03-11 02:12:08 +0530 | [diff] [blame] | 490 | |
Sebastian Andrzej Siewior | e722d8d | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 491 | return err; |
| 492 | err_state_onl: |
| 493 | #ifdef CONFIG_SMP |
| 494 | cpuhp_remove_state(online_state); |
| 495 | err_state_prep: |
| 496 | cpuhp_remove_state(CPUHP_PROFILE_PREPARE); |
| 497 | #endif |
Srivatsa S. Bhat | c270a81 | 2014-03-11 02:12:08 +0530 | [diff] [blame] | 498 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | } |
Paul Gortmaker | c96d666 | 2014-04-03 14:48:35 -0700 | [diff] [blame] | 500 | subsys_initcall(create_proc_profile); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | #endif /* CONFIG_PROC_FS */ |