From 1342798cc13e3b48d9b5738f0c8fa812ccea8101 Mon Sep 17 00:00:00 2001 From: David Ahern Date: Thu, 13 Sep 2012 14:59:13 -0600 Subject: perf tool: Precise mode requires exclude_guest Summary of events per Peter: "Intel PEBS in VT-x context uses the DS address as a guest linear address, even though its programmed by the host as a host linear address. This either results in guest memory corruption and or the hardware faulting and 'crashing' the virtual machine. Therefore we have to disable PEBS on VT-x enter and re-enable on VT-x exit, enforcing a strict exclude_guest. AMB IBS does work but doesn't currently support exclude_* at all, setting an exclude_* bit will make it fail." This patch handles userspace perf command, setting the exclude_guest attribute if precise mode is requested, but only if a user has not specified a request for guest or host only profiling (G or H attribute). Kernel side AMD currently ignores all exclude_* bits, so there is no impact to existing IBS code paths. Robert Richter has a patch where IBS code will return EINVAL if an exclude_* bit is set. When this goes in it means use of :p on AMD with IBS will first fail with EINVAL (because exclude_guest will be set). Then the existing fallback code within perf will unset exclude_guest and try again. The second attempt will succeed if the CPU supports IBS profiling. Signed-off-by: David Ahern Acked-by: Peter Zijlstra Acked-by: Robert Richter Tested-by: Robert Richter Reviewed-by: Robert Richter Cc: Avi Kivity Cc: Gleb Natapov Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Robert Richter Link: http://lkml.kernel.org/r/1347569955-54626-2-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/parse-events.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index aed38e4b9df..75c7b0fca6d 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -690,6 +690,9 @@ static int get_event_modifier(struct event_modifier *mod, char *str, eH = 0; } else if (*str == 'p') { precise++; + /* use of precise requires exclude_guest */ + if (!exclude_GH) + eG = 1; } else break; -- cgit v1.2.3 From 20b279ddb38ca42f8863cec07b4d45ec24589f13 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 13 Sep 2012 14:59:14 -0600 Subject: perf: Require exclude_guest to use PEBS - kernel side enforcement Intel PEBS in VT-x context uses the DS address as a guest linear address, even though its programmed by the host as a host linear address. This either results in guest memory corruption and or the hardware faulting and 'crashing' the virtual machine. Therefore we have to disable PEBS on VT-x enter and re-enable on VT-x exit, enforcing a strict exclude_guest. This patch enforces exclude_guest kernel side. Signed-off-by: Peter Zijlstra Cc: Avi Kivity Cc: David Ahern Cc: Gleb Natapov Cc: Ingo Molnar Cc: Robert Richter Link: http://lkml.kernel.org/r/1347569955-54626-3-git-send-email-dsahern@gmail.com Signed-off-by: David Ahern Signed-off-by: Arnaldo Carvalho de Melo --- arch/x86/kernel/cpu/perf_event.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c index 915b876edd1..3373f84d139 100644 --- a/arch/x86/kernel/cpu/perf_event.c +++ b/arch/x86/kernel/cpu/perf_event.c @@ -338,6 +338,9 @@ int x86_setup_perfctr(struct perf_event *event) /* BTS is currently only allowed for user-mode. */ if (!attr->exclude_kernel) return -EOPNOTSUPP; + + if (!attr->exclude_guest) + return -EOPNOTSUPP; } hwc->config |= config; @@ -380,6 +383,9 @@ int x86_pmu_hw_config(struct perf_event *event) if (event->attr.precise_ip) { int precise = 0; + if (!event->attr.exclude_guest) + return -EOPNOTSUPP; + /* Support for constant skid */ if (x86_pmu.pebs_active && !x86_pmu.pebs_broken) { precise++; -- cgit v1.2.3 From ffe10c6f95412da01695e659e967747333d5e812 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Mon, 15 Oct 2012 12:39:42 +0900 Subject: perf tools: Fix segfault when using srcline sort key The srcline sort key is for grouping samples based on their source file and line number. It use addr2line tool to get the information but it requires dso name. It caused a segfault when a sample does not have the name by dereferencing a NULL pointer. Fix it by using raw ip addresses for those samples. Signed-off-by: Namhyung Kim Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1350272383-7016-1-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/sort.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tools/perf/util/sort.c b/tools/perf/util/sort.c index b5b1b921196..dd68f115d39 100644 --- a/tools/perf/util/sort.c +++ b/tools/perf/util/sort.c @@ -260,6 +260,9 @@ static int hist_entry__srcline_snprintf(struct hist_entry *self, char *bf, if (path != NULL) goto out_path; + if (!self->ms.map) + goto out_ip; + snprintf(cmd, sizeof(cmd), "addr2line -e %s %016" PRIx64, self->ms.map->dso->long_name, self->ip); fp = popen(cmd, "r"); -- cgit v1.2.3 From 88481b6b33d6cb5edb57e5794abae4daeabd08c5 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Mon, 15 Oct 2012 12:39:43 +0900 Subject: perf tools: Remove warnings on JIT samples for srcline sort key When using the srcline sort key with perf report, I see many lines of warning related to JIT samples like below: addr2line: '/tmp/perf-1397.map': No such file Since it's not a ELF binary and doesn't provide such information, just use the raw ip address. Signed-off-by: Namhyung Kim Cc: David Ahern Cc: Ingo Molnar Cc: Irina Tirdea Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1350272383-7016-2-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/sort.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tools/perf/util/sort.c b/tools/perf/util/sort.c index dd68f115d39..cfd1c0feb32 100644 --- a/tools/perf/util/sort.c +++ b/tools/perf/util/sort.c @@ -263,6 +263,9 @@ static int hist_entry__srcline_snprintf(struct hist_entry *self, char *bf, if (!self->ms.map) goto out_ip; + if (!strncmp(self->ms.map->dso->long_name, "/tmp/perf-", 10)) + goto out_ip; + snprintf(cmd, sizeof(cmd), "addr2line -e %s %016" PRIx64, self->ms.map->dso->long_name, self->ip); fp = popen(cmd, "r"); -- cgit v1.2.3 From 63a1a3d820c619a4dab1781cc16c110a284efded Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Mon, 15 Oct 2012 18:14:35 -0300 Subject: perf hists browser: Fix off-by-two bug on the first column The commit 5395a04841fc ("perf hists: Separate overhead and baseline columns") makes the "Overhead" column no more the first one. So it resulted in the mis-aligned column in the normal (non-diff) output. Signed-off-by: Namhyung Kim Reported-by: Markus Trippelsdorf Acked-by: Jiri Olsa Cc: Ingo Molnar Cc: Jiri Olsa Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/None Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/ui/browsers/hists.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/tools/perf/ui/browsers/hists.c b/tools/perf/ui/browsers/hists.c index 0568536ecf6..fe4430aed63 100644 --- a/tools/perf/ui/browsers/hists.c +++ b/tools/perf/ui/browsers/hists.c @@ -610,6 +610,7 @@ static int hist_browser__show_entry(struct hist_browser *browser, char folded_sign = ' '; bool current_entry = ui_browser__is_current_entry(&browser->b, row); off_t row_offset = entry->row_offset; + bool first = true; if (current_entry) { browser->he_selection = entry; @@ -633,10 +634,11 @@ static int hist_browser__show_entry(struct hist_browser *browser, if (!perf_hpp__format[i].cond) continue; - if (i) { + if (!first) { slsmg_printf(" "); width -= 2; } + first = false; if (perf_hpp__format[i].color) { hpp.ptr = &percent; -- cgit v1.2.3 From 101782ea2c6984cf169631c59df76b8497899caf Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Mon, 1 Oct 2012 20:13:51 -0400 Subject: lib tools traceevent: Add back pevent assignment in __pevent_parse_format() Even though with the change of commit commit 2b29175 "tools lib traceevent: Carve out events format parsing routine", allowed __pevent_parse_format() to parse an event without the need of a pevent handler, the event still needs to assign the pevent handed to it. There's no problem with assigning it if the pevent is NULL, as the event->pevent would be NULL without the assignment. But function parsing handlers may be assigned to the pevent handler to help in parsing the event. If there's no pevent then there would not be any function handlers, but if the pevent isn't assigned first before parsing the event, it wont honor the function handlers that were assigned. Worse yet, the current code crashes if an event has a function that it tries to parse. For example: # perf record -e scsi:scsi_dispatch_cmd_timeout Segmentation fault (core dumped) This happens because the scsi_dispatch_cmd_timeout event format has the following: scsi_trace_parse_cdb(p, __get_dynamic_array(cmnd), REC->cmd_len) which hasn't been defined by the pevent code. Signed-off-by: Steven Rostedt Reviewed-by: Namhyung Kim Cc: Ingo Molnar Cc: Namhyung Kim Link: http://lkml.kernel.org/r/1349136831.22822.133.camel@gandalf.local.home Signed-off-by: Arnaldo Carvalho de Melo --- tools/lib/traceevent/event-parse.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/tools/lib/traceevent/event-parse.c b/tools/lib/traceevent/event-parse.c index 47264b4652b..f2989c525e4 100644 --- a/tools/lib/traceevent/event-parse.c +++ b/tools/lib/traceevent/event-parse.c @@ -2602,6 +2602,9 @@ find_func_handler(struct pevent *pevent, char *func_name) { struct pevent_function_handler *func; + if (!pevent) + return NULL; + for (func = pevent->func_handlers; func; func = func->next) { if (strcmp(func->name, func_name) == 0) break; @@ -4938,6 +4941,9 @@ enum pevent_errno __pevent_parse_format(struct event_format **eventp, goto event_alloc_failed; } + /* Add pevent to event so that it can be referenced */ + event->pevent = pevent; + ret = event_read_format(event); if (ret < 0) { ret = PEVENT_ERRNO__READ_FORMAT_FAILED; @@ -5041,9 +5047,6 @@ enum pevent_errno pevent_parse_event(struct pevent *pevent, const char *buf, if (event == NULL) return ret; - /* Add pevent to event so that it can be referenced */ - event->pevent = pevent; - if (add_event(pevent, event)) { ret = PEVENT_ERRNO__MEM_ALLOC_FAILED; goto event_add_failed; -- cgit v1.2.3 From 743df75ff10630f1f2a461f0f4b51f601f53ec44 Mon Sep 17 00:00:00 2001 From: Steven Rostedt Date: Mon, 1 Oct 2012 20:23:28 -0400 Subject: tools lib traceevent: Fix missed freeing of subargs in free_arg() in filter Some of args were missed in free_args(), as well as subargs. That is args like FILTER_ARG_NUM have left and right pointers to other args that also need to be freed. Signed-off-by: Steven Rostedt Cc: Ingo Molnar Cc: Namhyung Kim Link: http://lkml.kernel.org/r/1349137408.22822.135.camel@gandalf.local.home Signed-off-by: Arnaldo Carvalho de Melo --- tools/lib/traceevent/parse-filter.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/tools/lib/traceevent/parse-filter.c b/tools/lib/traceevent/parse-filter.c index ad17855528f..5ea4326ad11 100644 --- a/tools/lib/traceevent/parse-filter.c +++ b/tools/lib/traceevent/parse-filter.c @@ -209,7 +209,16 @@ static void free_arg(struct filter_arg *arg) switch (arg->type) { case FILTER_ARG_NONE: case FILTER_ARG_BOOLEAN: + break; + case FILTER_ARG_NUM: + free_arg(arg->num.left); + free_arg(arg->num.right); + break; + + case FILTER_ARG_EXP: + free_arg(arg->exp.left); + free_arg(arg->exp.right); break; case FILTER_ARG_STR: @@ -218,6 +227,12 @@ static void free_arg(struct filter_arg *arg) free(arg->str.buffer); break; + case FILTER_ARG_VALUE: + if (arg->value.type == FILTER_STRING || + arg->value.type == FILTER_CHAR) + free(arg->value.str); + break; + case FILTER_ARG_OP: free_arg(arg->op.left); free_arg(arg->op.right); -- cgit v1.2.3 From 356712f6e296fdae1edae51b96b485ed830bdc0c Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 16 Oct 2012 14:51:04 -0300 Subject: perf python: Initialize 'page_size' variable The commit 0c1fe6b: 'perf tools: Have the page size value available for all tools' Broke the python binding because the global variable 'page_size' is initialized on the main() routine, that is not called when using just the python binding, causing evlist.mmap() to fail because it expects that variable to be initialized to the system's page size. Fix it by initializing it on the binding init routine. Cc: David Ahern Cc: Frederic Weisbecker Cc: Jiri Olsa Cc: Mike Galbraith Cc: Namhyung Kim Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-vrvp3azmbfzexnpmkhmvtzzc@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/python.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c index 9181bf212fb..a2657fd9683 100644 --- a/tools/perf/util/python.c +++ b/tools/perf/util/python.c @@ -1015,6 +1015,8 @@ PyMODINIT_FUNC initperf(void) pyrf_cpu_map__setup_types() < 0) return; + page_size = sysconf(_SC_PAGE_SIZE); + Py_INCREF(&pyrf_evlist__type); PyModule_AddObject(module, "evlist", (PyObject*)&pyrf_evlist__type); -- cgit v1.2.3 From f8de2885eba35d20a062325bcdfe5d71e9dc2222 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Tue, 16 Oct 2012 14:53:24 -0300 Subject: perf python: Link with libtraceevent The evsel methods to read tracepoint fields uses libtraceevent functions, becoming needed by the python binding as well. Cc: David Ahern Cc: Frederic Weisbecker Cc: Jiri Olsa Cc: Mike Galbraith Cc: Namhyung Kim Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-j3o4v7jyvp9ke9n230l96a1m@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/setup.py | 1 + 1 file changed, 1 insertion(+) diff --git a/tools/perf/util/setup.py b/tools/perf/util/setup.py index d0f9f29cf18..09c3cea95d3 100644 --- a/tools/perf/util/setup.py +++ b/tools/perf/util/setup.py @@ -31,6 +31,7 @@ perf = Extension('perf', sources = ext_sources, include_dirs = ['util/include'], extra_compile_args = cflags, + extra_objects = [build_tmp + '/../../libtraceevent.a'], ) setup(name='perf', -- cgit v1.2.3 From 77626081849c9050b20670e5d832aca54c966936 Mon Sep 17 00:00:00 2001 From: David Miller Date: Wed, 17 Oct 2012 01:06:56 -0400 Subject: perf tools: Fix build on sparc. More UAPI stuff. Signed-off-by: David S. Miller Cc: Ingo Molnar Cc: Paul Mackerras Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/20121017.010656.383828471689899431.davem@davemloft.net Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/perf.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/perf.h b/tools/perf/perf.h index 87f4ec6d1f3..8421c20a25f 100644 --- a/tools/perf/perf.h +++ b/tools/perf/perf.h @@ -57,7 +57,7 @@ void get_term_dimensions(struct winsize *ws); #endif #ifdef __sparc__ -#include "../../arch/sparc/include/asm/unistd.h" +#include "../../arch/sparc/include/uapi/asm/unistd.h" #define rmb() asm volatile("":::"memory") #define cpu_relax() asm volatile("":::"memory") #define CPUINFO_PROC "cpu" -- cgit v1.2.3 From 88a21d2f07d2a4bec2e3e03dd50a39683b938b10 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Wed, 17 Oct 2012 13:54:08 -0300 Subject: perf hists browser: Add back callchain folding symbol The commit 5395a04841fc ("perf hists: Separate overhead and baseline columns") makes the "Overhead" column no more the first one, this caused the test that checks if it is time to show if a histogram entry has callchains never hits. Fix it by checking if the 'i' variable is equal to PERF_HPP__OVERHEAD instead of 0. Cc: David Ahern Cc: Frederic Weisbecker Cc: Jiri Olsa Cc: Mike Galbraith Cc: Namhyung Kim Cc: Paul Mackerras Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lkml.kernel.org/n/tip-w3lcbx0fx1fnh3l2cbq40q2e@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/ui/browsers/hists.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/ui/browsers/hists.c b/tools/perf/ui/browsers/hists.c index fe4430aed63..ef2f93ca749 100644 --- a/tools/perf/ui/browsers/hists.c +++ b/tools/perf/ui/browsers/hists.c @@ -647,7 +647,7 @@ static int hist_browser__show_entry(struct hist_browser *browser, ui_browser__set_percent_color(&browser->b, percent, current_entry); - if (i == 0 && symbol_conf.use_callchain) { + if (i == PERF_HPP__OVERHEAD && symbol_conf.use_callchain) { slsmg_printf("%c ", folded_sign); width -= 2; } -- cgit v1.2.3 From 45bff41a9a6f22af28e4ba22f83c87f619e573a8 Mon Sep 17 00:00:00 2001 From: Arnaldo Carvalho de Melo Date: Thu, 18 Oct 2012 11:38:35 -0300 Subject: perf python: Properly link with libtraceevent Namhyung Kim reported that the build fails with: GEN python/perf.so gcc: error: python_ext_build/tmp//../../libtraceevent.a: No such file or directory error: command 'gcc' failed with exit status 1 cp: cannot stat `python_ext_build/lib/perf.so': No such file or directory make: *** [python/perf.so] Error 1 We need to propagate the TE_PATH variable to the setup.py file. Reported-by: Namhyung Kim Signed-off-by: Arnaldo Carvalho de Melo Link: http://lkml.kernel.org/n/tip-8umiPbm4sxpknKivbjgykhut@git.kernel.org [ Fixed superfluous variable build error. ] Signed-off-by: Ingo Molnar --- tools/perf/Makefile | 24 +++++++++++++----------- tools/perf/util/python.c | 2 -- tools/perf/util/setup.py | 3 ++- 3 files changed, 15 insertions(+), 14 deletions(-) diff --git a/tools/perf/Makefile b/tools/perf/Makefile index f7c968ad517..00deed4d615 100644 --- a/tools/perf/Makefile +++ b/tools/perf/Makefile @@ -184,9 +184,22 @@ SCRIPT_SH += perf-archive.sh grep-libs = $(filter -l%,$(1)) strip-libs = $(filter-out -l%,$(1)) +TRACE_EVENT_DIR = ../lib/traceevent/ + +ifneq ($(OUTPUT),) + TE_PATH=$(OUTPUT) +else + TE_PATH=$(TRACE_EVENT_DIR) +endif + +LIBTRACEEVENT = $(TE_PATH)libtraceevent.a +TE_LIB := -L$(TE_PATH) -ltraceevent + PYTHON_EXT_SRCS := $(shell grep -v ^\# util/python-ext-sources) PYTHON_EXT_DEPS := util/python-ext-sources util/setup.py +export LIBTRACEEVENT + $(OUTPUT)python/perf.so: $(PYTHON_EXT_SRCS) $(PYTHON_EXT_DEPS) $(QUIET_GEN)CFLAGS='$(BASIC_CFLAGS)' $(PYTHON_WORD) util/setup.py \ --quiet build_ext; \ @@ -198,17 +211,6 @@ $(OUTPUT)python/perf.so: $(PYTHON_EXT_SRCS) $(PYTHON_EXT_DEPS) SCRIPTS = $(patsubst %.sh,%,$(SCRIPT_SH)) -TRACE_EVENT_DIR = ../lib/traceevent/ - -ifneq ($(OUTPUT),) - TE_PATH=$(OUTPUT) -else - TE_PATH=$(TRACE_EVENT_DIR) -endif - -LIBTRACEEVENT = $(TE_PATH)libtraceevent.a -TE_LIB := -L$(TE_PATH) -ltraceevent - # # Single 'perf' binary right now: # diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c index a2657fd9683..9181bf212fb 100644 --- a/tools/perf/util/python.c +++ b/tools/perf/util/python.c @@ -1015,8 +1015,6 @@ PyMODINIT_FUNC initperf(void) pyrf_cpu_map__setup_types() < 0) return; - page_size = sysconf(_SC_PAGE_SIZE); - Py_INCREF(&pyrf_evlist__type); PyModule_AddObject(module, "evlist", (PyObject*)&pyrf_evlist__type); diff --git a/tools/perf/util/setup.py b/tools/perf/util/setup.py index 09c3cea95d3..73d51026978 100644 --- a/tools/perf/util/setup.py +++ b/tools/perf/util/setup.py @@ -23,6 +23,7 @@ cflags += getenv('CFLAGS', '').split() build_lib = getenv('PYTHON_EXTBUILD_LIB') build_tmp = getenv('PYTHON_EXTBUILD_TMP') +libtraceevent = getenv('LIBTRACEEVENT') ext_sources = [f.strip() for f in file('util/python-ext-sources') if len(f.strip()) > 0 and f[0] != '#'] @@ -31,7 +32,7 @@ perf = Extension('perf', sources = ext_sources, include_dirs = ['util/include'], extra_compile_args = cflags, - extra_objects = [build_tmp + '/../../libtraceevent.a'], + extra_objects = [libtraceevent], ) setup(name='perf', -- cgit v1.2.3