>From 49f8827eff8ce94e7e1fb62ebfa0add419238850 Mon Sep 17 00:00:00 2001 From: Maria Guseva Date: Wed, 13 Jan 2016 19:58:24 +0300 Subject: [PATCH 2/2] Generic time implementation. * elf/dl-support.c (_dl_non_dynamic_init) [HP_TIMING_AVAIL]: Time _dl_cpuclock_offset. * elf/rtld.c (relocate_time, load_time, start_time): Declare when (HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL) is defined instead of HP_TIMING_AVAIL. (dl_start_final_info): Check (HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL) instead of HP_TIMING_AVAIL. (DL_STATISTICS_TIMING_NOW): New macro to define timing in case of DL_DEBUG_STATISTICS set. (DL_STATISTICS_TIMING_DIFF): Likewise. (DL_STATISTICS_TIMING_ACCUM_NT): Likewise. (_dl_start_final) [HP_TIMING_AVAIL]: Time dl_cpuclock_offset. (_dl_start_final): Check (HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL) instead of HP_TIMING_AVAIL. (dl_main): Likewise. Use DL_STATISTICS_TIMING_NOW, DL_STATISTICS_TIMING_DIFF, DL_STATISTICS_TIMING_ACCUM_NT instead of HP_TIMING_NOW, HP_TIMING_DIFF, HP_TIMING_ACCUM_NT. (dl_main) [HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL]: Initialize start with 0 to prevent warnings. (print_statistics): Check (HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL) instead of HP_TIMING_AVAIL. * sysdeps/generic/hp-timing-common.h (HP_TIMING_UNIT): New. Define to "cycles". (HP_TIMING_PRINT): Print HP_TIMING_UNIT instead of "cycles". * sysdeps/generic/hp-timing.h: Add includes , , . (HP_SMALL_TIMING_AVAIL): Define to 1. (HP_GENERIC_TIMING_AVAIL): New. Define to 1. (hp_timing_t): Add typedef. [!HAVE_CLOCK_GETTIME_VSYSCALL](__INTERNAL_SYSCALL): New. Define to INTERNAL_SYSCALL. [HAVE_CLOCK_GETTIME_VSYSCALL]: Add include . [HAVE_CLOCK_GETTIME_VSYSCALL](__INTERNAL_SYSCALL): New. Define to INTERNAL_VSYSCALL. [__NR_clock_gettime] (HP_TIMING_NOW): Define using clock_gettime. [__NR_clock_gettime] (HP_TIMING_UNIT): Define to "nsec". [!__NR_clock_gettime]: Add include . [!__NR_clock_gettime] (HP_TIMING_NOW): Define using __gettimeofday. [!__NR_clock_gettime] (HP_TIMING_UNIT): Define to "usec". --- elf/dl-support.c | 2 + elf/rtld.c | 115 ++++++++++++++++++++++++------------- sysdeps/generic/hp-timing-common.h | 9 ++- sysdeps/generic/hp-timing.h | 64 +++++++++++++++++---- 4 files changed, 137 insertions(+), 53 deletions(-) diff --git a/elf/dl-support.c b/elf/dl-support.c index 303e5d0..f2b4788 100644 --- a/elf/dl-support.c +++ b/elf/dl-support.c @@ -311,8 +311,10 @@ _dl_non_dynamic_init (void) _dl_main_map.l_phdr = GL(dl_phdr); _dl_main_map.l_phnum = GL(dl_phnum); +#if HP_TIMING_AVAIL if (HP_SMALL_TIMING_AVAIL) HP_TIMING_NOW (_dl_cpuclock_offset); +#endif _dl_verbose = *(getenv ("LD_WARN") ?: "") == '\0' ? 0 : 1; diff --git a/elf/rtld.c b/elf/rtld.c index b1125d9..34c4ab5 100644 --- a/elf/rtld.c +++ b/elf/rtld.c @@ -195,7 +195,7 @@ static struct libname_list _dl_rtld_libname; static struct libname_list _dl_rtld_libname2; /* Variable for statistics. */ -#if HP_TIMING_AVAIL +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL static hp_timing_t relocate_time; static hp_timing_t load_time attribute_relro; static hp_timing_t start_time attribute_relro; @@ -227,7 +227,7 @@ static ElfW(Addr) _dl_start_final (void *arg); struct dl_start_final_info { struct link_map l; -#if HP_TIMING_AVAIL && HP_TIMING_INLINE +#if (HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL) && HP_TIMING_INLINE hp_timing_t start_time; #endif }; @@ -240,6 +240,35 @@ extern char _begin[] attribute_hidden; extern char _etext[] attribute_hidden; extern char _end[] attribute_hidden; +/* Macros to define timing in case of DL_DEBUG_STATISTICS set. */ +# if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL +#define DL_STATISTICS_TIMING_NOW(tval) \ +do { \ + if (__glibc_unlikely (HP_SMALL_TIMING_AVAIL && \ + (GLRO(dl_debug_mask) & DL_DEBUG_STATISTICS))) \ + HP_TIMING_NOW (tval); \ +} while (0) + +#define DL_STATISTICS_TIMING_DIFF(tdiff, tval1, tval2) \ +do { \ +if (__glibc_unlikely (HP_SMALL_TIMING_AVAIL && \ + (GLRO(dl_debug_mask) & DL_DEBUG_STATISTICS))) \ + HP_TIMING_DIFF (tdiff, tval1, tval2); \ +} while (0) + +#define DL_STATISTICS_TIMING_ACCUM_NT(tsum, tval) \ +do { \ +if (__glibc_unlikely (HP_SMALL_TIMING_AVAIL && \ + (GLRO(dl_debug_mask) & DL_DEBUG_STATISTICS))) \ + HP_TIMING_ACCUM_NT (tsum, tval); \ +} while (0) + +# else +#define DL_STATISTICS_TIMING_NOW(tval) +#define DL_STATISTICS_TIMING_DIFF(tdiff, tval1, tval2) +#define DL_STATISTICS_TIMING_ACCUM(tsum, tval) +# endif + #ifdef RTLD_START RTLD_START @@ -295,7 +324,9 @@ _dl_start_final (void *arg, struct dl_start_final_info *info) # endif #endif +#if HP_TIMING_AVAIL HP_TIMING_NOW (GL(dl_cpuclock_offset)); +#endif /* Initialize the stack end variable. */ __libc_stack_end = __builtin_frame_address (0); @@ -306,23 +337,23 @@ _dl_start_final (void *arg, struct dl_start_final_info *info) entry point on the same stack we entered on. */ start_addr = _dl_sysdep_start (arg, &dl_main); -#if HP_TIMING_AVAIL +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL hp_timing_t rtld_total_time; if (HP_SMALL_TIMING_AVAIL) { hp_timing_t end_time; /* Get the current time. */ - HP_TIMING_NOW (end_time); + DL_STATISTICS_TIMING_NOW (end_time); /* Compute the difference. */ - HP_TIMING_DIFF (rtld_total_time, start_time, end_time); + DL_STATISTICS_TIMING_DIFF (rtld_total_time, start_time, end_time); } #endif if (__glibc_unlikely (GLRO(dl_debug_mask) & DL_DEBUG_STATISTICS)) { -#if HP_TIMING_AVAIL +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL print_statistics (&rtld_total_time); #else print_statistics (NULL); @@ -350,12 +381,14 @@ _dl_start (void *arg) #define RESOLVE_MAP(sym, version, flags) (&bootstrap_map) #include "dynamic-link.h" +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL if (HP_TIMING_INLINE && HP_SMALL_TIMING_AVAIL) #ifdef DONT_USE_BOOTSTRAP_MAP HP_TIMING_NOW (start_time); #else HP_TIMING_NOW (info.start_time); #endif +#endif /* Partly clean the `bootstrap_map' structure up. Don't use `memset' since it might not be built in or inlined and we cannot @@ -745,8 +778,8 @@ dl_main (const ElfW(Phdr) *phdr, unsigned int i; bool prelinked = false; bool rtld_is_main = false; -#if HP_TIMING_AVAIL - hp_timing_t start; +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL + hp_timing_t start = 0; hp_timing_t stop; hp_timing_t diff; #endif @@ -922,12 +955,12 @@ of this helper program; chances are you did not intend to run this program.\n\ } else { - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW (start); _dl_map_object (NULL, rtld_progname, lt_executable, 0, __RTLD_OPENEXEC, LM_ID_BASE); - HP_TIMING_NOW (stop); + DL_STATISTICS_TIMING_NOW (stop); - HP_TIMING_DIFF (load_time, start, stop); + DL_STATISTICS_TIMING_DIFF (load_time, start, stop); } /* Now the map for the main executable is available. */ @@ -1489,7 +1522,7 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", char *list = strdupa (preloadlist); char *p; - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW (start); /* Prevent optimizing strsep. Speed is not important here. */ while ((p = (strsep) (&list, " :")) != NULL) @@ -1498,9 +1531,9 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", || strchr (p, '/') == NULL)) npreloads += do_preload (p, main_map, "LD_PRELOAD"); - HP_TIMING_NOW (stop); - HP_TIMING_DIFF (diff, start, stop); - HP_TIMING_ACCUM_NT (load_time, diff); + DL_STATISTICS_TIMING_NOW (stop); + DL_STATISTICS_TIMING_DIFF (diff, start, stop); + DL_STATISTICS_TIMING_ACCUM_NT (load_time, diff); } /* There usually is no ld.so.preload file, it should only be used @@ -1560,7 +1593,7 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", file[file_size - 1] = '\0'; } - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW (start); if (file != problem) { @@ -1578,9 +1611,9 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", npreloads += do_preload (p, main_map, preload_file); } - HP_TIMING_NOW (stop); - HP_TIMING_DIFF (diff, start, stop); - HP_TIMING_ACCUM_NT (load_time, diff); + DL_STATISTICS_TIMING_NOW (stop); + DL_STATISTICS_TIMING_DIFF (diff, start, stop); + DL_STATISTICS_TIMING_ACCUM_NT (load_time, diff); /* We don't need the file anymore. */ __munmap (file, file_size); @@ -1604,11 +1637,11 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", /* Load all the libraries specified by DT_NEEDED entries. If LD_PRELOAD specified some libraries to load, these are inserted before the actual dependencies in the executable's searchlist for symbol resolution. */ - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW (start); _dl_map_object_deps (main_map, preloads, npreloads, mode == trace, 0); - HP_TIMING_NOW (stop); - HP_TIMING_DIFF (diff, start, stop); - HP_TIMING_ACCUM_NT (load_time, diff); + DL_STATISTICS_TIMING_NOW (stop); + DL_STATISTICS_TIMING_DIFF (diff, start, stop); + DL_STATISTICS_TIMING_ACCUM_NT (load_time, diff); /* Mark all objects as being in the global scope. */ for (i = main_map->l_searchlist.r_nlist; i > 0; ) @@ -1997,12 +2030,12 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", if (main_map->l_info [ADDRIDX (DT_GNU_CONFLICT)] != NULL) { ElfW(Rela) *conflict, *conflictend; -#if HP_TIMING_AVAIL - hp_timing_t start; +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL + hp_timing_t start = 0; hp_timing_t stop; #endif - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW(start); assert (main_map->l_info [VALIDX (DT_GNU_CONFLICTSZ)] != NULL); conflict = (ElfW(Rela) *) main_map->l_info [ADDRIDX (DT_GNU_CONFLICT)]->d_un.d_ptr; @@ -2010,8 +2043,8 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", ((char *) conflict + main_map->l_info [VALIDX (DT_GNU_CONFLICTSZ)]->d_un.d_val); _dl_resolve_conflicts (main_map, conflict, conflictend); - HP_TIMING_NOW (stop); - HP_TIMING_DIFF (relocate_time, start, stop); + DL_STATISTICS_TIMING_NOW (stop); + DL_STATISTICS_TIMING_DIFF (relocate_time, start, stop); } @@ -2039,15 +2072,15 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", know that because it is self-contained). */ int consider_profiling = GLRO(dl_profile) != NULL; -#if HP_TIMING_AVAIL - hp_timing_t start; +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL + hp_timing_t start = 0; hp_timing_t stop; #endif /* If we are profiling we also must do lazy reloaction. */ GLRO(dl_lazy) |= consider_profiling; - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW (start); unsigned i = main_map->l_searchlist.r_nlist; while (i-- > 0) { @@ -2074,9 +2107,9 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", if (l->l_tls_blocksize != 0 && tls_init_tp_called) _dl_add_to_slotinfo (l); } - HP_TIMING_NOW (stop); + DL_STATISTICS_TIMING_NOW (stop); - HP_TIMING_DIFF (relocate_time, start, stop); + DL_STATISTICS_TIMING_DIFF (relocate_time, start, stop); /* Now enable profiling if needed. Like the previous call, this has to go here because the calls it makes should use the @@ -2117,19 +2150,19 @@ ERROR: ld.so: object '%s' cannot be loaded as audit interface: %s; ignored.\n", re-relocation, we might call a user-supplied function (e.g. calloc from _dl_relocate_object) that uses TLS data. */ -#if HP_TIMING_AVAIL - hp_timing_t start; +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL + hp_timing_t start = 0; hp_timing_t stop; hp_timing_t add; #endif - HP_TIMING_NOW (start); + DL_STATISTICS_TIMING_NOW (start); /* Mark the link map as not yet relocated again. */ GL(dl_rtld_map).l_relocated = 0; _dl_relocate_object (&GL(dl_rtld_map), main_map->l_scope, 0, 0); - HP_TIMING_NOW (stop); - HP_TIMING_DIFF (add, start, stop); - HP_TIMING_ACCUM_NT (relocate_time, add); + DL_STATISTICS_TIMING_NOW (stop); + DL_STATISTICS_TIMING_DIFF (add, start, stop); + DL_STATISTICS_TIMING_ACCUM_NT (relocate_time, add); } /* Do any necessary cleanups for the startup OS interface code. @@ -2568,7 +2601,7 @@ static void __attribute ((noinline)) print_statistics (hp_timing_t *rtld_total_timep) { -#if HP_TIMING_AVAIL +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL char buf[200]; char *cp; char *wp; @@ -2640,7 +2673,7 @@ print_statistics (hp_timing_t *rtld_total_timep) GL(dl_num_cache_relocations), num_relative_relocations); -#if HP_TIMING_AVAIL +#if HP_TIMING_AVAIL || HP_GENERIC_TIMING_AVAIL /* Time spend while loading the object and the dependencies. */ if (HP_SMALL_TIMING_AVAIL) { diff --git a/sysdeps/generic/hp-timing-common.h b/sysdeps/generic/hp-timing-common.h index 257cdd3..77cdeba 100644 --- a/sysdeps/generic/hp-timing-common.h +++ b/sysdeps/generic/hp-timing-common.h @@ -47,6 +47,11 @@ /* Accumulate ADD into SUM. No attempt is made to be thread-safe. */ #define HP_TIMING_ACCUM_NT(Sum, Diff) ((Sum) += (Diff)) +/* Define timing unit name to be printed, default is cycles. */ +#ifndef HP_TIMING_UNIT +#define HP_TIMING_UNIT "cycles" +#endif + /* Write a decimal representation of the timing value into the given string. */ #define HP_TIMING_PRINT(Dest, Len, Val) \ do { \ @@ -56,7 +61,7 @@ char *__cp = _itoa ((Val), __buf + sizeof (__buf), 10, 0); \ size_t __cp_len = MIN (__buf + sizeof (__buf) - __cp, __len); \ memcpy (__dest, __cp, __cp_len); \ - memcpy (__dest + __cp_len, " cycles", \ - MIN (__len - __cp_len, sizeof (" cycles"))); \ + memcpy (__dest + __cp_len, " " HP_TIMING_UNIT, \ + MIN (__len - __cp_len, sizeof (" " HP_TIMING_UNIT))); \ __dest[__len - 1] = '\0'; \ } while (0) diff --git a/sysdeps/generic/hp-timing.h b/sysdeps/generic/hp-timing.h index 5597378..a9905c0 100644 --- a/sysdeps/generic/hp-timing.h +++ b/sysdeps/generic/hp-timing.h @@ -20,18 +20,62 @@ #ifndef _HP_TIMING_H #define _HP_TIMING_H 1 -/* There are no generic definitions for the times. We could write something - using the `gettimeofday' system call where available but the overhead of - the system call might be too high. */ +/* Generic definitions for the times using the `clock_gettime' + system or vdso call where available + or gettimeofday function call in other cases. */ -/* Provide dummy definitions. */ #define HP_TIMING_AVAIL (0) -#define HP_SMALL_TIMING_AVAIL (0) +#define HP_GENERIC_TIMING_AVAIL (1) +#define HP_SMALL_TIMING_AVAIL (1) #define HP_TIMING_INLINE (0) -typedef int hp_timing_t; -#define HP_TIMING_NOW(var) -#define HP_TIMING_DIFF(Diff, Start, End) -#define HP_TIMING_ACCUM_NT(Sum, Diff) -#define HP_TIMING_PRINT(Buf, Len, Val) + +/* Use 64bit values for the times in nano- or microseconds. */ +#include +typedef uint64_t hp_timing_t; + +/* Use vdso syscall where available. */ +#include +#ifndef HAVE_CLOCK_GETTIME_VSYSCALL +# define __INTERNAL_SYSCALL INTERNAL_SYSCALL +#else +# include +# define __INTERNAL_SYSCALL INTERNAL_VSYSCALL +#endif + +# ifdef __NR_clock_gettime + +/* Use clockgettime where available. */ +#define HP_TIMING_NOW(var) \ +do \ + { \ + struct timespec __t; \ + INTERNAL_SYSCALL_DECL (__err); \ + __INTERNAL_SYSCALL (clock_gettime, __err, 2, CLOCK_MONOTONIC, &__t);\ + var = (uint64_t)__t.tv_nsec + (uint64_t)__t.tv_sec * 1000000000L; \ + } \ +while (0) + +/* Define timing unit name to be printed. */ +#define HP_TIMING_UNIT "nsec" + +# else + +/* Use gettimeofday function in other cases. */ +#include +#define HP_TIMING_NOW(var) \ +do \ + { \ + struct timeval __t; \ + __gettimeofday (&__t, NULL); \ + var = (uint64_t)__t.tv_usec + (uint64_t)__t.tv_sec * 1000000L; \ + } \ +while (0) + +/* Define timing unit name to be printed. */ +#define HP_TIMING_UNIT "usec" + +# endif /* __NR_clock_gettime */ + +#include #endif /* hp-timing.h */ -- 1.9.1