From d2d4edbebe07ddb77980656abe7b9bc7a9e0cdf7 Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Wed, 11 Jan 2017 14:59:38 +0900 Subject: perf probe: Fix to show correct locations for events on modules Fix to show correct locations for events on modules by relocating given address instead of retrying after failure. This happens when the module text size is big enough, bigger than sh_addr, because the original code retries with given address + sh_addr if it failed to find CU DIE at the given address. Any address smaller than sh_addr always fails and it retries with the correct address, but addresses bigger than sh_addr will get a CU DIE which is on the given address (not adjusted by sh_addr). In my environment(x86-64), the sh_addr of ".text" section is 0x10030. Since i915 is a huge kernel module, we can see this issue as below. $ grep "[Tt] .*\[i915\]" /proc/kallsyms | sort | head -n1 ffffffffc0270000 t i915_switcheroo_can_switch [i915] ffffffffc0270000 + 0x10030 = ffffffffc0280030, so we'll check symbols cross this boundary. $ grep "[Tt] .*\[i915\]" /proc/kallsyms | grep -B1 ^ffffffffc028\ | head -n 2 ffffffffc027ff80 t haswell_init_clock_gating [i915] ffffffffc0280110 t valleyview_init_clock_gating [i915] So setup probes on both function and see what happen. $ sudo ./perf probe -m i915 -a haswell_init_clock_gating \ -a valleyview_init_clock_gating Added new events: probe:haswell_init_clock_gating (on haswell_init_clock_gating in i915) probe:valleyview_init_clock_gating (on valleyview_init_clock_gating in i915) You can now use it in all perf tools, such as: perf record -e probe:valleyview_init_clock_gating -aR sleep 1 $ sudo ./perf probe -l probe:haswell_init_clock_gating (on haswell_init_clock_gating@gpu/drm/i915/intel_pm.c in i915) probe:valleyview_init_clock_gating (on i915_vga_set_decode:4@gpu/drm/i915/i915_drv.c in i915) As you can see, haswell_init_clock_gating is correctly shown, but valleyview_init_clock_gating is not. With this patch, both events are shown correctly. $ sudo ./perf probe -l probe:haswell_init_clock_gating (on haswell_init_clock_gating@gpu/drm/i915/intel_pm.c in i915) probe:valleyview_init_clock_gating (on valleyview_init_clock_gating@gpu/drm/i915/intel_pm.c in i915) Committer notes: In my case: # perf probe -m i915 -a haswell_init_clock_gating -a valleyview_init_clock_gating Added new events: probe:haswell_init_clock_gating (on haswell_init_clock_gating in i915) probe:valleyview_init_clock_gating (on valleyview_init_clock_gating in i915) You can now use it in all perf tools, such as: perf record -e probe:valleyview_init_clock_gating -aR sleep 1 # perf probe -l probe:haswell_init_clock_gating (on i915_getparam+432@gpu/drm/i915/i915_drv.c in i915) probe:valleyview_init_clock_gating (on __i915_printk+240@gpu/drm/i915/i915_drv.c in i915) # # readelf -SW /lib/modules/4.9.0+/build/vmlinux | egrep -w '.text|Name' [Nr] Name Type Address Off Size ES Flg Lk Inf Al [ 1] .text PROGBITS ffffffff81000000 200000 822fd3 00 AX 0 0 4096 # So both are b0rked, now with the fix: # perf probe -m i915 -a haswell_init_clock_gating -a valleyview_init_clock_gating Added new events: probe:haswell_init_clock_gating (on haswell_init_clock_gating in i915) probe:valleyview_init_clock_gating (on valleyview_init_clock_gating in i915) You can now use it in all perf tools, such as: perf record -e probe:valleyview_init_clock_gating -aR sleep 1 # perf probe -l probe:haswell_init_clock_gating (on haswell_init_clock_gating@gpu/drm/i915/intel_pm.c in i915) probe:valleyview_init_clock_gating (on valleyview_init_clock_gating@gpu/drm/i915/intel_pm.c in i915) # Both looks correct. Signed-off-by: Masami Hiramatsu Tested-by: Arnaldo Carvalho de Melo Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/148411436777.9978.1440275861947194930.stgit@devbox Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/probe-finder.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-) (limited to 'tools') diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c index df4debe564da..0278fe1a4cc6 100644 --- a/tools/perf/util/probe-finder.c +++ b/tools/perf/util/probe-finder.c @@ -1543,16 +1543,12 @@ int debuginfo__find_probe_point(struct debuginfo *dbg, unsigned long addr, Dwarf_Addr _addr = 0, baseaddr = 0; const char *fname = NULL, *func = NULL, *basefunc = NULL, *tmp; int baseline = 0, lineno = 0, ret = 0; - bool reloc = false; -retry: + /* We always need to relocate the address for aranges */ + if (debuginfo__get_text_offset(dbg, &baseaddr) == 0) + addr += baseaddr; /* Find cu die */ if (!dwarf_addrdie(dbg->dbg, (Dwarf_Addr)addr, &cudie)) { - if (!reloc && debuginfo__get_text_offset(dbg, &baseaddr) == 0) { - addr += baseaddr; - reloc = true; - goto retry; - } pr_warning("Failed to find debug information for address %lx\n", addr); ret = -EINVAL; -- cgit v1.2.3 From 3e96dac7c956089d3f23aca98c4dfca57b6aaf8a Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Wed, 11 Jan 2017 15:00:47 +0900 Subject: perf probe: Add error checks to offline probe post-processing Add error check codes on post processing and improve it for offline probe events as: - post processing fails if no matched symbol found in map(-ENOENT) or strdup() failed(-ENOMEM). - Even if the symbol name is the same, it updates symbol address and offset. Signed-off-by: Masami Hiramatsu Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/148411443738.9978.4617979132625405545.stgit@devbox Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/probe-event.c | 50 ++++++++++++++++++++++++++++--------------- 1 file changed, 33 insertions(+), 17 deletions(-) (limited to 'tools') diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c index 4a57c8a60bd9..aa8a9227080a 100644 --- a/tools/perf/util/probe-event.c +++ b/tools/perf/util/probe-event.c @@ -610,6 +610,33 @@ error: return ret ? : -ENOENT; } +/* Adjust symbol name and address */ +static int post_process_probe_trace_point(struct probe_trace_point *tp, + struct map *map, unsigned long offs) +{ + struct symbol *sym; + u64 addr = tp->address + tp->offset - offs; + + sym = map__find_symbol(map, addr); + if (!sym) + return -ENOENT; + + if (strcmp(sym->name, tp->symbol)) { + /* If we have no realname, use symbol for it */ + if (!tp->realname) + tp->realname = tp->symbol; + else + free(tp->symbol); + tp->symbol = strdup(sym->name); + if (!tp->symbol) + return -ENOMEM; + } + tp->offset = addr - sym->start; + tp->address -= offs; + + return 0; +} + /* * Rename DWARF symbols to ELF symbols -- gcc sometimes optimizes functions * and generate new symbols with suffixes such as .constprop.N or .isra.N @@ -622,11 +649,9 @@ static int post_process_offline_probe_trace_events(struct probe_trace_event *tevs, int ntevs, const char *pathname) { - struct symbol *sym; struct map *map; unsigned long stext = 0; - u64 addr; - int i; + int i, ret = 0; /* Prepare a map for offline binary */ map = dso__new_map(pathname); @@ -636,23 +661,14 @@ post_process_offline_probe_trace_events(struct probe_trace_event *tevs, } for (i = 0; i < ntevs; i++) { - addr = tevs[i].point.address + tevs[i].point.offset - stext; - sym = map__find_symbol(map, addr); - if (!sym) - continue; - if (!strcmp(sym->name, tevs[i].point.symbol)) - continue; - /* If we have no realname, use symbol for it */ - if (!tevs[i].point.realname) - tevs[i].point.realname = tevs[i].point.symbol; - else - free(tevs[i].point.symbol); - tevs[i].point.symbol = strdup(sym->name); - tevs[i].point.offset = addr - sym->start; + ret = post_process_probe_trace_point(&tevs[i].point, + map, stext); + if (ret < 0) + break; } map__put(map); - return 0; + return ret; } static int add_exec_to_probe_trace_events(struct probe_trace_event *tevs, -- cgit v1.2.3 From 613f050d68a8ed3c0b18b9568698908ef7bbc1f7 Mon Sep 17 00:00:00 2001 From: Masami Hiramatsu Date: Wed, 11 Jan 2017 15:01:57 +0900 Subject: perf probe: Fix to probe on gcc generated functions in modules Fix to probe on gcc generated functions on modules. Since probing on a module is based on its symbol name, it should be adjusted on actual symbols. E.g. without this fix, perf probe shows probe definition on non-exist symbol as below. $ perf probe -m build-x86_64/net/netfilter/nf_nat.ko -F in_range* in_range.isra.12 $ perf probe -m build-x86_64/net/netfilter/nf_nat.ko -D in_range p:probe/in_range nf_nat:in_range+0 With this fix, perf probe correctly shows a probe on gcc-generated symbol. $ perf probe -m build-x86_64/net/netfilter/nf_nat.ko -D in_range p:probe/in_range nf_nat:in_range.isra.12+0 This also fixes same problem on online module as below. $ perf probe -m i915 -D assert_plane p:probe/assert_plane i915:assert_plane.constprop.134+0 Signed-off-by: Masami Hiramatsu Tested-by: Arnaldo Carvalho de Melo Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/148411450673.9978.14905987549651656075.stgit@devbox Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/probe-event.c | 45 +++++++++++++++++++++++++++--------------- tools/perf/util/probe-finder.c | 7 +++++-- tools/perf/util/probe-finder.h | 3 +++ 3 files changed, 37 insertions(+), 18 deletions(-) (limited to 'tools') diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c index aa8a9227080a..6a6f44dd594b 100644 --- a/tools/perf/util/probe-event.c +++ b/tools/perf/util/probe-event.c @@ -698,18 +698,31 @@ static int add_exec_to_probe_trace_events(struct probe_trace_event *tevs, return ret; } -static int add_module_to_probe_trace_events(struct probe_trace_event *tevs, - int ntevs, const char *module) +static int +post_process_module_probe_trace_events(struct probe_trace_event *tevs, + int ntevs, const char *module, + struct debuginfo *dinfo) { + Dwarf_Addr text_offs = 0; int i, ret = 0; char *mod_name = NULL; + struct map *map; if (!module) return 0; - mod_name = find_module_name(module); + map = get_target_map(module, false); + if (!map || debuginfo__get_text_offset(dinfo, &text_offs, true) < 0) { + pr_warning("Failed to get ELF symbols for %s\n", module); + return -EINVAL; + } + mod_name = find_module_name(module); for (i = 0; i < ntevs; i++) { + ret = post_process_probe_trace_point(&tevs[i].point, + map, (unsigned long)text_offs); + if (ret < 0) + break; tevs[i].point.module = strdup(mod_name ? mod_name : module); if (!tevs[i].point.module) { @@ -719,6 +732,8 @@ static int add_module_to_probe_trace_events(struct probe_trace_event *tevs, } free(mod_name); + map__put(map); + return ret; } @@ -776,7 +791,7 @@ arch__post_process_probe_trace_events(struct perf_probe_event *pev __maybe_unuse static int post_process_probe_trace_events(struct perf_probe_event *pev, struct probe_trace_event *tevs, int ntevs, const char *module, - bool uprobe) + bool uprobe, struct debuginfo *dinfo) { int ret; @@ -784,7 +799,8 @@ static int post_process_probe_trace_events(struct perf_probe_event *pev, ret = add_exec_to_probe_trace_events(tevs, ntevs, module); else if (module) /* Currently ref_reloc_sym based probe is not for drivers */ - ret = add_module_to_probe_trace_events(tevs, ntevs, module); + ret = post_process_module_probe_trace_events(tevs, ntevs, + module, dinfo); else ret = post_process_kernel_probe_trace_events(tevs, ntevs); @@ -828,30 +844,27 @@ static int try_to_find_probe_trace_events(struct perf_probe_event *pev, } } - debuginfo__delete(dinfo); - if (ntevs > 0) { /* Succeeded to find trace events */ pr_debug("Found %d probe_trace_events.\n", ntevs); ret = post_process_probe_trace_events(pev, *tevs, ntevs, - pev->target, pev->uprobes); + pev->target, pev->uprobes, dinfo); if (ret < 0 || ret == ntevs) { + pr_debug("Post processing failed or all events are skipped. (%d)\n", ret); clear_probe_trace_events(*tevs, ntevs); zfree(tevs); + ntevs = 0; } - if (ret != ntevs) - return ret < 0 ? ret : ntevs; - ntevs = 0; - /* Fall through */ } + debuginfo__delete(dinfo); + if (ntevs == 0) { /* No error but failed to find probe point. */ pr_warning("Probe point '%s' not found.\n", synthesize_perf_probe_point(&pev->point)); return -ENOENT; - } - /* Error path : ntevs < 0 */ - pr_debug("An error occurred in debuginfo analysis (%d).\n", ntevs); - if (ntevs < 0) { + } else if (ntevs < 0) { + /* Error path : ntevs < 0 */ + pr_debug("An error occurred in debuginfo analysis (%d).\n", ntevs); if (ntevs == -EBADF) pr_warning("Warning: No dwarf info found in the vmlinux - " "please rebuild kernel with CONFIG_DEBUG_INFO=y.\n"); diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c index 0278fe1a4cc6..0d9d6e0803b8 100644 --- a/tools/perf/util/probe-finder.c +++ b/tools/perf/util/probe-finder.c @@ -1501,7 +1501,8 @@ int debuginfo__find_available_vars_at(struct debuginfo *dbg, } /* For the kernel module, we need a special code to get a DIE */ -static int debuginfo__get_text_offset(struct debuginfo *dbg, Dwarf_Addr *offs) +int debuginfo__get_text_offset(struct debuginfo *dbg, Dwarf_Addr *offs, + bool adjust_offset) { int n, i; Elf32_Word shndx; @@ -1530,6 +1531,8 @@ static int debuginfo__get_text_offset(struct debuginfo *dbg, Dwarf_Addr *offs) if (!shdr) return -ENOENT; *offs = shdr->sh_addr; + if (adjust_offset) + *offs -= shdr->sh_offset; } } return 0; @@ -1545,7 +1548,7 @@ int debuginfo__find_probe_point(struct debuginfo *dbg, unsigned long addr, int baseline = 0, lineno = 0, ret = 0; /* We always need to relocate the address for aranges */ - if (debuginfo__get_text_offset(dbg, &baseaddr) == 0) + if (debuginfo__get_text_offset(dbg, &baseaddr, false) == 0) addr += baseaddr; /* Find cu die */ if (!dwarf_addrdie(dbg->dbg, (Dwarf_Addr)addr, &cudie)) { diff --git a/tools/perf/util/probe-finder.h b/tools/perf/util/probe-finder.h index f1d8558f498e..2956c5198652 100644 --- a/tools/perf/util/probe-finder.h +++ b/tools/perf/util/probe-finder.h @@ -46,6 +46,9 @@ int debuginfo__find_trace_events(struct debuginfo *dbg, int debuginfo__find_probe_point(struct debuginfo *dbg, unsigned long addr, struct perf_probe_point *ppt); +int debuginfo__get_text_offset(struct debuginfo *dbg, Dwarf_Addr *offs, + bool adjust_offset); + /* Find a line range */ int debuginfo__find_line_range(struct debuginfo *dbg, struct line_range *lr); -- cgit v1.2.3 From df21d2fa733035e4d414379960f94b2516b41296 Mon Sep 17 00:00:00 2001 From: Madhavan Srinivasan Date: Mon, 19 Dec 2016 17:46:53 +0530 Subject: selftest/powerpc: Wrong PMC initialized in pmc56_overflow test Test uses PMC2 to count the event. But PMC1 is being initialized. Patch to fix it. Fixes: 3752e453f6ba ('selftests/powerpc: Add tests of PMU EBBs') Signed-off-by: Madhavan Srinivasan Signed-off-by: Michael Ellerman --- tools/testing/selftests/powerpc/pmu/ebb/pmc56_overflow_test.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'tools') diff --git a/tools/testing/selftests/powerpc/pmu/ebb/pmc56_overflow_test.c b/tools/testing/selftests/powerpc/pmu/ebb/pmc56_overflow_test.c index c22860ab9733..30e1ac62e8cb 100644 --- a/tools/testing/selftests/powerpc/pmu/ebb/pmc56_overflow_test.c +++ b/tools/testing/selftests/powerpc/pmu/ebb/pmc56_overflow_test.c @@ -66,7 +66,7 @@ int pmc56_overflow(void) FAIL_IF(ebb_event_enable(&event)); - mtspr(SPRN_PMC1, pmc_sample_period(sample_period)); + mtspr(SPRN_PMC2, pmc_sample_period(sample_period)); mtspr(SPRN_PMC5, 0); mtspr(SPRN_PMC6, 0); -- cgit v1.2.3 From 21f5eda9b8671744539c8295b9df62991fffb2ce Mon Sep 17 00:00:00 2001 From: Halil Pasic Date: Mon, 29 Aug 2016 18:25:22 +0200 Subject: tools/virtio/ringtest: fix run-on-all.sh for offline cpus Since ef1b144d ("tools/virtio/ringtest: fix run-on-all.sh to work without /dev/cpu") run-on-all.sh uses seq 0 $HOST_AFFINITY as the list of ids of the CPUs to run the command on (assuming ids of online CPUs are consecutive and start from 0), where $HOST_AFFINITY is the highest CPU id in the system previously determined using lscpu. This can fail on systems with offline CPUs. Instead let's use lscpu to determine the list of online CPUs. Signed-off-by: Halil Pasic Fixes: ef1b144d ("tools/virtio/ringtest: fix run-on-all.sh to work without /dev/cpu") Reviewed-by: Sascha Silbe Signed-off-by: Cornelia Huck --- tools/virtio/ringtest/run-on-all.sh | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'tools') diff --git a/tools/virtio/ringtest/run-on-all.sh b/tools/virtio/ringtest/run-on-all.sh index 2e69ca812b4c..29b0d3920bfc 100755 --- a/tools/virtio/ringtest/run-on-all.sh +++ b/tools/virtio/ringtest/run-on-all.sh @@ -1,12 +1,13 @@ #!/bin/sh +CPUS_ONLINE=$(lscpu --online -p=cpu|grep -v -e '#') #use last CPU for host. Why not the first? #many devices tend to use cpu0 by default so #it tends to be busier -HOST_AFFINITY=$(lscpu -p=cpu | tail -1) +HOST_AFFINITY=$(echo "${CPUS_ONLINE}"|tail -n 1) #run command on all cpus -for cpu in $(seq 0 $HOST_AFFINITY) +for cpu in $CPUS_ONLINE do #Don't run guest and host on same CPU #It actually works ok if using signalling -- cgit v1.2.3 From 47a4c49af6cc1982ce613c8ee79aab459d04c44c Mon Sep 17 00:00:00 2001 From: Halil Pasic Date: Fri, 2 Sep 2016 17:59:36 +0200 Subject: tools/virtio/ringtest: tweaks for s390 Make ringtest work on s390 too. Signed-off-by: Halil Pasic Acked-by: Sascha Silbe Signed-off-by: Cornelia Huck --- tools/virtio/ringtest/main.h | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'tools') diff --git a/tools/virtio/ringtest/main.h b/tools/virtio/ringtest/main.h index 34e63cc4c572..14142faf040b 100644 --- a/tools/virtio/ringtest/main.h +++ b/tools/virtio/ringtest/main.h @@ -26,6 +26,16 @@ static inline void wait_cycles(unsigned long long cycles) #define VMEXIT_CYCLES 500 #define VMENTRY_CYCLES 500 +#elif defined(__s390x__) +static inline void wait_cycles(unsigned long long cycles) +{ + asm volatile("0: brctg %0,0b" : : "d" (cycles)); +} + +/* tweak me */ +#define VMEXIT_CYCLES 200 +#define VMENTRY_CYCLES 200 + #else static inline void wait_cycles(unsigned long long cycles) { @@ -81,6 +91,8 @@ extern unsigned ring_size; /* Is there a portable way to do this? */ #if defined(__x86_64__) || defined(__i386__) #define cpu_relax() asm ("rep; nop" ::: "memory") +#elif defined(__s390x__) +#define cpu_relax() barrier() #else #define cpu_relax() assert(0) #endif -- cgit v1.2.3