perf tool: Refactoring IO data files code
RFC PATCH 1/1 perf tool: Refactoring IO data files code
Hi,
based on following discussion:
http://marc.info/?l=linux-kernel&m=131731212425421&w=2
http://marc.info/?l=linux-kernel&m=131736842010972&w=2
I made some changes to have event memory mapping and input/output data code in one place. The reason was to make the change for multiple event data files at least possible ;)
This patchset is marked as RFC since it's quite big change and I expect some better design might come up.
However, patches 1 and 2 are not big deal, since they are just small (not so related) changes.
Patches 3-4 are the main change of the patchset.
1/5 perf tool: Fix session host_nachine retrieval 2/5 perf tool: Initialize events IDs in a single function 3/5 perf tool: Introducing perf_mmap object
4/5 perf tool: Introducing perf_data object
5/5 perf tool: Putting mmap support to perf_data object
I did some overall testing of all the changed command, and so far so good. Just hit issue with 'diff' command, but this one seems not to work even without my changes.
Also the current perf python binsing code is broken wit missing dependencies so I could not test my changes there.
thanks for comments,
jirka
--- tools/perf/Makefile | 4 + tools/perf/builtin-annotate.c | 3 +- tools/perf/builtin-buildid-list.c | 4 +- tools/perf/builtin-diff.c | 6 +- tools/perf/builtin-evlist.c | 3 +- tools/perf/builtin-inject.c | 9 +- tools/perf/builtin-kmem.c | 18 +- tools/perf/builtin-lock.c | 3 +- tools/perf/builtin-record.c | 206 ++++++++----------------- tools/perf/builtin-report.c | 7 +- tools/perf/builtin-sched.c | 6 +- tools/perf/builtin-script.c | 3 +- tools/perf/builtin-test.c | 17 ++- tools/perf/builtin-timechart.c | 7 +- tools/perf/builtin-top.c | 52 ++++--- tools/perf/perf.h | 26 --- tools/perf/util/data.c | 303 +++++++++++++++++++++++++++++++++++++ tools/perf/util/data.h | 78 ++++++++++ tools/perf/util/event.c | 14 +- tools/perf/util/evlist.c | 268 +++------------------------------ tools/perf/util/evlist.h | 8 +- tools/perf/util/evsel.h | 4 +- tools/perf/util/header.c | 35 +++-- tools/perf/util/header.h | 6 +- tools/perf/util/mmap.c | 138 +++++++++++++++++ tools/perf/util/mmap.h | 48 ++++++ tools/perf/util/python.c | 23 ++- tools/perf/util/session.c | 115 +++++--------- tools/perf/util/session.h | 38 ++++-- 29 files changed, 855 insertions(+), 597 deletions(-)
PATCH 1/5 perf tool: Fix session host_nachine retrieval
The host_machine is statically included inside perf_session struct so it is always present.
Changing function:
'perf_session__find_host_machine' to 'perf_session__host_machine'
Removing NULL checking code after this function, since it is not needed.
--- tools/perf/builtin-inject.c | 6 +----- tools/perf/builtin-kmem.c | 12 +++++------- tools/perf/builtin-record.c | 6 +----- tools/perf/builtin-top.c | 4 ++-- tools/perf/util/event.c | 14 +++++--------- tools/perf/util/session.h | 6 +++--- 6 files changed, 17 insertions(+), 31 deletions(-)
tools/perf/builtin-inject.c +1 -5
@@ -111,11 +111,7 @@ static int dso__inject_build_id(struct dso *self, struct perf_session *session)return -1 ; } - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) { -pr_err ("Can't find machine for session\n" ); -return -1 ; - } + machine =perf_session__host_machine (session);if (self->kernel) misc =PERF_RECORD_MISC_KERNEL ;
tools/perf/builtin-kmem.c +5 -7
@@ -342,7 +342,6 @@ static void __print_result(struct rb_root *root, struct perf_session *session, int n_lines, int is_caller) {struct rb_node *next; -struct machine *machine;printf ("%.102s\n" , graph_dotted_line);printf (" %-34s |" , is_caller ?"Callsite" :"Alloc Ptr" );
@@ -351,11 +350,6 @@ static void __print_result(struct rb_root *root, struct perf_session *session, next =rb_first (root); - machine =perf_session__find_host_machine (session); -if (!machine) { -pr_err ("__print_result: couldn't find kernel information\n" ); -return ; - }while (next && n_lines--) { struct alloc_stat *data = rb_entry(next, struct alloc_stat, node);
@@ -366,8 +360,12 @@ static void __print_result(struct rb_root *root, struct perf_session *session,if (is_caller) { addr = data->call_site; -if (!raw_ip) +if (!raw_ip) { +struct machine *machine; + + machine =perf_session__host_machine (session); sym =machine__find_kernel_function (machine, addr, &map,NULL ); + } }else addr = data->ptr;
tools/perf/builtin-record.c +1 -5
@@ -662,11 +662,7 @@ static int __cmd_record(int argc, const char **argv) } } - machine =perf_session__find_host_machine (session); -if (!machine) { -pr_err ("Couldn't find native kernel information.\n" ); -return -1 ; - } + machine =perf_session__host_machine (session); err =perf_event__synthesize_kernel_mmap (process_synthesized_event, session, machine,"_text" );
tools/perf/builtin-top.c +2 -2
@@ -692,13 +692,13 @@ static void perf_event__process_sample(const union perf_event *event, ++top.us_samples;if (top.hide_user_symbols)return ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session);break ;case PERF_RECORD_MISC_KERNEL : ++top.kernel_samples;if (top.hide_kernel_symbols)return ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session);break ;case PERF_RECORD_MISC_GUEST_KERNEL : ++top.guest_kernel_samples;
tools/perf/util/event.c +5 -9
@@ -589,12 +589,12 @@ int perf_event__process_mmap(union perf_event *event,return 0 ; } - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) - goto out_problem; thread =perf_session__findnew (session, event->mmap.pid);if (thread ==NULL ) goto out_problem; + + machine =perf_session__host_machine (session); + map = map__new(&machine->user_dsos, event->mmap.start, event->mmap.len, event->mmap.pgoff, event->mmap.pid, event->mmap.filename,
@@ -672,15 +672,11 @@ void thread__find_addr_map(struct thread *self,if (cpumode ==PERF_RECORD_MISC_KERNEL && perf_host) { al->level ='k' ; - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) { - al->map =NULL ; -return ; - } + machine =perf_session__host_machine (session); mg = &machine->kmaps; }else if (cpumode ==PERF_RECORD_MISC_USER && perf_host) { al->level ='.' ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session); }else if (cpumode ==PERF_RECORD_MISC_GUEST_KERNEL && perf_guest) { al->level ='g' ; machine =perf_session__find_machine (session, pid);
tools/perf/util/session.h +3 -3
@@ -121,7 +121,7 @@ void perf_session__update_sample_type(struct perf_session *self);void perf_session__remove_thread (struct perf_session *self,struct thread *th);static inline -struct machine *perf_session__find_host_machine (struct perf_session *self) +struct machine *perf_session__host_machine (struct perf_session *self) {return &self->host_machine; }
@@ -130,7 +130,7 @@ static inlinestruct machine *perf_session__find_machine (struct perf_session *self,pid_t pid) {if (pid ==HOST_KERNEL_ID ) -return &self->host_machine; +return perf_session__host_machine (self);return machines__find (&self->machines, pid); }
@@ -138,7 +138,7 @@ static inlinestruct machine *perf_session__findnew_machine (struct perf_session *self,pid_t pid) {if (pid ==HOST_KERNEL_ID ) -return &self->host_machine; +return perf_session__host_machine (self);return machines__findnew (&self->machines, pid); }
Signature · 1 line
1.7.4
Re: [PATCH 1/5] perf tool: Fix session host_nachine retrieval
Em Fri, Nov 18, 2011 at 02:46:41PM +0100, Jiri Olsa escreveu:
Quoted 7 lines
The
host_machineis statically included insideperf_sessionstruct so it is always present.Changing function:
'perf_session__find_host_machine' to 'perf_session__host_machine'Removing NULL checking code after this function, since it is not needed.
I'm working on finding the machine in perf_session__process_events and then passing it to the perf_event_ops methods, will fold this in there.
Quoted 173 lines
Signed-off-by Jiri Olsa <[email protected]>--- tools/perf/builtin-inject.c | 6 +----- tools/perf/builtin-kmem.c | 12 +++++------- tools/perf/builtin-record.c | 6 +----- tools/perf/builtin-top.c | 4 ++-- tools/perf/util/event.c | 14 +++++--------- tools/perf/util/session.h | 6 +++--- 6 files changed, 17 insertions(+), 31 deletions(-)tools/perf/builtin-inject.c +1
-5@@ -111,11 +111,7 @@ static int dso__inject_build_id(struct dso *self, struct perf_session *session)return -1 ; } - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) { -pr_err ("Can't find machine for session\n" ); -return -1 ; - } + machine =perf_session__host_machine (session);if (self->kernel) misc =PERF_RECORD_MISC_KERNEL ;tools/perf/builtin-kmem.c +5
-7@@ -342,7 +342,6 @@ static void __print_result(struct rb_root *root, struct perf_session *session, int n_lines, int is_caller) {struct rb_node *next; -struct machine *machine;printf ("%.102s\n" , graph_dotted_line);printf (" %-34s |" , is_caller ?"Callsite" :"Alloc Ptr" );@@ -351,11 +350,6 @@ static void __print_result(struct rb_root *root, struct perf_session *session, next =rb_first (root); - machine =perf_session__find_host_machine (session); -if (!machine) { -pr_err ("__print_result: couldn't find kernel information\n" ); -return ; - }while (next && n_lines--) { struct alloc_stat *data = rb_entry(next, struct alloc_stat, node);@@ -366,8 +360,12 @@ static void __print_result(struct rb_root *root, struct perf_session *session,if (is_caller) { addr = data->call_site; -if (!raw_ip) +if (!raw_ip) { +struct machine *machine; + + machine =perf_session__host_machine (session); sym =machine__find_kernel_function (machine, addr, &map,NULL ); + } }else addr = data->ptr;tools/perf/builtin-record.c +1
-5@@ -662,11 +662,7 @@ static int __cmd_record(int argc, const char **argv) } } - machine =perf_session__find_host_machine (session); -if (!machine) { -pr_err ("Couldn't find native kernel information.\n" ); -return -1 ; - } + machine =perf_session__host_machine (session); err =perf_event__synthesize_kernel_mmap (process_synthesized_event, session, machine,"_text" );tools/perf/builtin-top.c +2
-2@@ -692,13 +692,13 @@ static void perf_event__process_sample(const union perf_event *event, ++top.us_samples;if (top.hide_user_symbols)return ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session);break ;case PERF_RECORD_MISC_KERNEL : ++top.kernel_samples;if (top.hide_kernel_symbols)return ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session);break ;case PERF_RECORD_MISC_GUEST_KERNEL : ++top.guest_kernel_samples;tools/perf/util/event.c +5
-9@@ -589,12 +589,12 @@ int perf_event__process_mmap(union perf_event *event,return 0 ; } - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) - goto out_problem; thread =perf_session__findnew (session, event->mmap.pid);if (thread ==NULL ) goto out_problem; + + machine =perf_session__host_machine (session); + map = map__new(&machine->user_dsos, event->mmap.start, event->mmap.len, event->mmap.pgoff, event->mmap.pid, event->mmap.filename,@@ -672,15 +672,11 @@ void thread__find_addr_map(struct thread *self,if (cpumode ==PERF_RECORD_MISC_KERNEL && perf_host) { al->level ='k' ; - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) { - al->map =NULL ; -return ; - } + machine =perf_session__host_machine (session); mg = &machine->kmaps; }else if (cpumode ==PERF_RECORD_MISC_USER && perf_host) { al->level ='.' ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session); }else if (cpumode ==PERF_RECORD_MISC_GUEST_KERNEL && perf_guest) { al->level ='g' ; machine =perf_session__find_machine (session, pid);tools/perf/util/session.h +3
-3@@ -121,7 +121,7 @@ void perf_session__update_sample_type(struct perf_session *self);void perf_session__remove_thread (struct perf_session *self,struct thread *th);static inline -struct machine *perf_session__find_host_machine (struct perf_session *self) +struct machine *perf_session__host_machine (struct perf_session *self) {return &self->host_machine; }@@ -130,7 +130,7 @@ static inlinestruct machine *perf_session__find_machine (struct perf_session *self,pid_t pid) {if (pid ==HOST_KERNEL_ID ) -return &self->host_machine; +return perf_session__host_machine (self);return machines__find (&self->machines, pid); }@@ -138,7 +138,7 @@ static inlinestruct machine *perf_session__findnew_machine (struct perf_session *self,pid_t pid) {if (pid ==HOST_KERNEL_ID ) -return &self->host_machine; +return perf_session__host_machine (self);return machines__findnew (&self->machines, pid); }Signature · 1 line
1.7.4
PATCH 2/5 perf tool: Initialize events IDs in a single function
In reply to Jiri OlsaCurrently events' IDs initialization is scattered among the code. Adding function 'perf_evlist__init_ids' to allocate and retrive events' IDs in one place.
--- tools/perf/util/evlist.c | 50 +++++++++++++++++++++++++++++---------------- 1 files changed, 32 insertions(+), 18 deletions(-)
tools/perf/util/evlist.c +32 -18
@@ -326,10 +326,6 @@ static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mif (ioctl (fd,PERF_EVENT_IOC_SET_OUTPUT , output) !=0 ) goto out_unmap; } - -if ((evsel->attr.read_format &PERF_FORMAT_ID ) && -perf_evlist__id_add_fd (evlist, evsel, cpu, thread, fd) <0 ) - goto out_unmap; } } }
@@ -366,10 +362,6 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, inif (ioctl (fd,PERF_EVENT_IOC_SET_OUTPUT , output) !=0 ) goto out_unmap; } - -if ((evsel->attr.read_format &PERF_FORMAT_ID ) && -perf_evlist__id_add_fd (evlist, evsel,0 , thread, fd) <0 ) - goto out_unmap; } }
@@ -385,6 +377,33 @@ out_unmap:return -1 ; } +static int perf_evlist__init_ids (struct perf_evlist *evlist) +{ +struct perf_evsel *evsel; +struct thread_map *threads = evlist->threads; +struct cpu_map *cpus = evlist->cpus; +int cpu, thread; + +list_for_each_entry (evsel, &evlist->entries, node) { +if (!(evsel->attr.read_format &PERF_FORMAT_ID )) +continue ; + +if ((!evsel->sample_id) && +perf_evsel__alloc_id (evsel, cpus->nr, threads->nr) <0 ) +return -ENOMEM ; + +for (cpu =0 ; cpu < cpus->nr; cpu++) +for (thread =0 ; thread < threads->nr; thread++) { +int fd =FD (evsel, cpu, thread); +if (perf_evlist__id_add_fd (evlist, evsel, + cpu, thread, fd)) +return -EINVAL ; + } + } + +return 0 ; +} +/** perf_evlist__mmap - Create per cpu maps to receive events * * @evlist - list of events
@@ -403,10 +422,8 @@ out_unmap:int perf_evlist__mmap (struct perf_evlist *evlist,int pages,bool overwrite) {unsigned int page_size =sysconf (_SC_PAGE_SIZE); -int mask = pages * page_size -1 ; -struct perf_evsel *evsel; +int mask = pages * page_size -1 , ret;const struct cpu_map *cpus = evlist->cpus; -const struct thread_map *threads = evlist->threads;int prot =PROT_READ | (overwrite ?0 :PROT_WRITE );if (evlist->mmap ==NULL &&perf_evlist__alloc_mmap (evlist) <0 )
@@ -418,14 +435,11 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) evlist->overwrite = overwrite; evlist->mmap_len = (pages +1 ) * page_size; -list_for_each_entry (evsel, &evlist->entries, node) { -if ((evsel->attr.read_format &PERF_FORMAT_ID ) && - evsel->sample_id ==NULL && -perf_evsel__alloc_id (evsel, cpus->nr, threads->nr) <0 ) -return -ENOMEM ; - } + ret =perf_evlist__init_ids (evlist); +if (ret) +return ret; -if (evlist->cpus->map[0 ] ==-1 ) +if (cpus->map[0 ] ==-1 )return perf_evlist__mmap_per_thread (evlist, prot, mask);return perf_evlist__mmap_per_cpu (evlist, prot, mask);
Signature · 1 line
1.7.4
Re: [PATCH 2/5] perf tool: Initialize events IDs in a single function
Em Fri, Nov 18, 2011 at 02:46:42PM +0100, Jiri Olsa escreveu:
Quoted 3 lines
Currently events' IDs initialization is scattered among the code. Adding function '
perf_evlist__init_ids' to allocate and retrive events' IDs in one place.
I'm ok with this one, will merge it in my perf/core branch together with a fix for a bug I noticed in the original code and that remains with your changes: consider what happens if we do a second perf_evlist__mmap, we will call perf_evlist__id_add_fd again and thus bump evsel->ids++ over nr_threads * nr_cpus, overflowing it.
- Arnaldo
Re: [PATCH 2/5] perf tool: Initialize events IDs in a single function
Quoted 10 lines · Arnaldo Carvalho de Melo
Em Fri, Nov 18, 2011 at 02:46:42PM +0100, Jiri Olsa escreveu:
Quoted 3 lines
Currently events' IDs initialization is scattered among the code. Adding function '
perf_evlist__init_ids' to allocate and retrive events' IDs in one place.I'm ok with this one, will merge it in my perf/core branch together with a fix for a bug I noticed in the original code and that remains with your changes: consider what happens if we do a second
perf_evlist__mmap, we will callperf_evlist__id_add_fdagain and thus bump evsel->ids++ overnr_threads*nr_cpus, overflowing it.
is there a valid scenario for this function to be called more than once? I'd think that just ensuring this is called once is enough
jirka
Re: [PATCH 2/5] perf tool: Initialize events IDs in a single function
Em Sat, Nov 19, 2011 at 08:36:21PM -0500, Jiri Olsa escreveu:
Quoted 6 lines · Arnaldo Carvalho de Melo
Quoted 4 lines
Em Fri, Nov 18, 2011 at 02:46:42PM +0100, Jiri Olsa escreveu:
Quoted 3 lines
Currently events' IDs initialization is scattered among the code. Adding function '
perf_evlist__init_ids' to allocate and retrive events' IDs in one place.
Quoted 8 lines
Quoted 5 lines
I'm ok with this one, will merge it in my perf/core branch together with a fix for a bug I noticed in the original code and that remains with your changes: consider what happens if we do a second
perf_evlist__mmap, we will callperf_evlist__id_add_fdagain and thus bump evsel->ids++ overnr_threads*nr_cpus, overflowing it.is there a valid scenario for this function to be called more than once? I'd think that just ensuring this is called once is enough
I don't know, I just don't want to preclude that, crystal balls are not that reliable :-)
- Arnaldo
PATCH 3/5 perf tool: Introducing perf_mmap object
In reply to Jiri OlsaAdding perf_mmap object to handle event memory maps.
All the memory map related functions originally scatered through the whole code arenow place in perf_mmap object.
To map and unmap perf_mmap:
perf_mmap__open
perf_mmap__close
For reading events via callback:
perf_mmap__process
For reading events directly:
perf_mmap__read
following helpers were moved in from perf.h:
perf_mmap__read_head
perf_mmap__write_tail
--- tools/perf/Makefile | 2 + tools/perf/builtin-record.c | 57 ++++------------- tools/perf/builtin-test.c | 6 ++- tools/perf/builtin-top.c | 19 ++++-- tools/perf/perf.h | 26 -------- tools/perf/util/evlist.c | 117 +++++++----------------------------- tools/perf/util/evlist.h | 2 +- tools/perf/util/mmap.c | 138 +++++++++++++++++++++++++++++++++++++++++++ tools/perf/util/mmap.h | 45 ++++++++++++++ 9 files changed, 239 insertions(+), 173 deletions(-) create mode 100644 tools/perf/util/mmap.c create mode 100644 tools/perf/util/mmap.h
tools/perf/Makefile +2 -0
@@ -274,6 +274,7 @@ LIB_H += util/xyarray.h LIB_H += util/header.h LIB_H += util/help.h LIB_H += util/session.h +LIB_H += util/mmap.h LIB_H += util/strbuf.h LIB_H += util/strlist.h LIB_H += util/strfilter.h
@@ -337,6 +338,7 @@ LIB_OBJS += $(OUTPUT)util/debug.o LIB_OBJS += $(OUTPUT)util/map.o LIB_OBJS += $(OUTPUT)util/pstack.o LIB_OBJS += $(OUTPUT)util/session.o +LIB_OBJS += $(OUTPUT)util/mmap.o LIB_OBJS += $(OUTPUT)util/thread.o LIB_OBJS += $(OUTPUT)util/thread_map.o LIB_OBJS += $(OUTPUT)util/trace-event-parse.o
tools/perf/builtin-record.c +14 -43
94 lines not on this page · raw message
tools/perf/builtin-test.c +5 -1
@@ -12,6 +12,7 @@#include "util/parse-events.h" #include "util/symbol.h" #include "util/thread_map.h" +#include "util/mmap.h" #include "../../include/linux/hw_breakpoint.h" static long page_size;
@@ -476,6 +477,7 @@ static int test__basic_mmap(void) expected_nr_events[nsyscalls], i, j;struct perf_evsel *evsels[nsyscalls], *evsel;int sample_size =__perf_evsel__sample_size (attr.sample_type); +struct perf_mmap *md;for (i =0 ; i < nsyscalls; ++i) {char name[64 ];
@@ -551,7 +553,9 @@ static int test__basic_mmap(void) ++foo; } -while ((event =perf_evlist__mmap_read (evlist,0 )) !=NULL ) { + md = &evlist->mmap[0 ]; + +while ((event =perf_mmap__read (md)) !=NULL ) {struct perf_sample sample;if (event->header.type !=PERF_RECORD_SAMPLE ) {
tools/perf/builtin-top.c +12 -7
@@ -38,6 +38,7 @@#include "util/cpumap.h" #include "util/xyarray.h" #include "util/sort.h" +#include "util/mmap.h" #include "util/debug.h"
@@ -804,14 +805,16 @@ static void perf_event__process_sample(const union perf_event *event,return ; } -static void perf_session__mmap_read_idx (struct perf_session *self,int idx) +static void session_mmap_read (struct perf_session *self, +struct perf_mmap *md) { -struct perf_sample sample; -struct perf_evsel *evsel;union perf_event *event; -int ret; -while ((event =perf_evlist__mmap_read (top.evlist, idx)) !=NULL ) { +while ((event =perf_mmap__read (md)) !=NULL ) { +struct perf_sample sample; +struct perf_evsel *evsel; +int ret; + ret =perf_session__parse_sample (self, event, &sample);if (ret) {pr_err ("Can't parse sample, err = %d\n" , ret);
@@ -835,8 +838,10 @@ static void perf_session__mmap_read(struct perf_session *self) {int i; -for (i =0 ; i < top.evlist->nr_mmaps; i++) -perf_session__mmap_read_idx (self, i); +for (i =0 ; i < top.evlist->nr_mmaps; i++) { +struct perf_mmap *md = &top.evlist->mmap[i]; +session_mmap_read (self, md); + } }static void start_counters (struct perf_evlist *evlist)
tools/perf/perf.h +0 -26
@@ -104,32 +104,6 @@ void get_term_dimensions(struct winsize *ws);#include "util/types.h" #include <stdbool.h> -struct perf_mmap { -void *base; -int mask; -unsigned int prev; -}; - -static inline unsigned int perf_mmap__read_head (struct perf_mmap *mm) -{ -struct perf_event_mmap_page *pc = mm->base; -int head = pc->data_head; -rmb (); -return head; -} - -static inline void perf_mmap__write_tail (struct perf_mmap *md, -unsigned long tail) -{ -struct perf_event_mmap_page *pc = md->base; - -/* -* ensure all reads are done before we write the tail out. -*/ -/* mb(); */ - pc->data_tail = tail; -} -/* * prctl(PR_TASK_PERF_EVENTS_DISABLE) will (cheaply) disable all * counters in the current task.
tools/perf/util/evlist.c +22 -95
195 lines not on this page · raw message
tools/perf/util/evlist.h +1 -1
@@ -18,7 +18,7 @@ struct perf_evlist {int nr_entries;int nr_fds;int nr_mmaps; -int mmap_len; +int pages;bool overwrite;union perf_event event_copy;struct perf_mmap *mmap;
tools/perf/util/mmap.c new file +138 -0
138 lines not on this page · raw message
tools/perf/util/mmap.h new file +45 -0
45 lines not on this page · raw message
Signature · 1 line
1.7.4
Re: [PATCH 3/5] perf tool: Introducing perf_mmap object
Em Fri, Nov 18, 2011 at 02:46:43PM +0100, Jiri Olsa escreveu:
Quoted 10 lines
+++
b/tools/perf/builtin-test.ctools/perf/builtin-test.c +1
-0@@ -476,6 +477,7 @@ static int test__basic_mmap(void) expected_nr_events[nsyscalls], i, j;struct perf_evsel *evsels[nsyscalls], *evsel;int sample_size =__perf_evsel__sample_size (attr.sample_type); +struct perf_mmap *md;for (i =0 ; i < nsyscalls; ++i) {char name[64 ];
tools/perf/builtin-test.c static int test__basic_mmap(void) · line 561
Quoted 10 lines
tools/perf/builtin-test.c +3
-1@@ -551,7 +553,9 @@ static int test__basic_mmap(void) ++foo; } -while ((event =perf_evlist__mmap_read (evlist,0 )) !=NULL ) { + md = &evlist->mmap[0 ]; + +while ((event =perf_mmap__read (md)) !=NULL ) {struct perf_sample sample;if (event->header.type !=PERF_RECORD_SAMPLE ) {
Please keep perf_evlist__mmap_read() as just a wrapper for the above operation, that way you reduce the patch size by not touching the functions that use perf_evlist__mmap_read().
Later, if we think this is the right thing to do, i.e. to open code it like you're doing above, we can elliminate it, but I think its better to keep it as perf_evlist__mmap_read(evlist, 0) as it uses just one line instead of the 4 above :-)
Quoted 21 lines
tools/perf/builtin-top.c +3
-2@@ -38,6 +38,7 @@#include "util/cpumap.h" #include "util/xyarray.h" #include "util/sort.h" +#include "util/mmap.h" #include "util/debug.h" @@ -804,14 +805,16 @@ static void perf_event__process_sample(const union perf_event *event,return ; } -static void perf_session__mmap_read_idx (struct perf_session *self,int idx) +static void session_mmap_read (struct perf_session *self, +struct perf_mmap *md) { -struct perf_sample sample;
tools/perf/builtin-top.c static void perf_event__process_sample(const union perf_event *event, · line 817
Quoted 10 lines
tools/perf/builtin-top.c +5
-3-struct perf_evsel *evsel;union perf_event *event; -int ret; -while ((event =perf_evlist__mmap_read (top.evlist, idx)) !=NULL ) { +while ((event =perf_mmap__read (md)) !=NULL ) { +struct perf_sample sample; +struct perf_evsel *evsel; +int ret; +
Ditto, all the above is not needed, just keep perf_evlist__mmap_read()
tools/perf/builtin-top.c static void perf_session__mmap_read(struct perf_session *self) · line 844
Quoted 13 lines
tools/perf/builtin-top.c +4
-2ret =perf_session__parse_sample (self, event, &sample);if (ret) {pr_err ("Can't parse sample, err = %d\n" , ret);@@ -835,8 +838,10 @@ static void perf_session__mmap_read(struct perf_session *self) {int i; -for (i =0 ; i < top.evlist->nr_mmaps; i++) -perf_session__mmap_read_idx (self, i); +for (i =0 ; i < top.evlist->nr_mmaps; i++) { +struct perf_mmap *md = &top.evlist->mmap[i]; +session_mmap_read (self, md); + }
ditto
Quoted 27 lines
tools/perf/builtin-top.c +0
-0}static void start_counters (struct perf_evlist *evlist)tools/perf/perf.h +0
-16@@ -104,32 +104,6 @@ void get_term_dimensions(struct winsize *ws);#include "util/types.h" #include <stdbool.h> -struct perf_mmap { -void *base; -int mask; -unsigned int prev; -}; - -static inline unsigned int perf_mmap__read_head (struct perf_mmap *mm) -{ -struct perf_event_mmap_page *pc = mm->base; -int head = pc->data_head; -rmb (); -return head; -} - -static inline void perf_mmap__write_tail (struct perf_mmap *md, -unsigned long tail)
tools/perf/perf.h void get_term_dimensions(struct winsize *ws); · line 132
Quoted 10 lines
tools/perf/perf.h +0
-10-{ -struct perf_event_mmap_page *pc = md->base; - -/* -* ensure all reads are done before we write the tail out. -*/ -/* mb(); */ - pc->data_tail = tail; -} -
Ok, the above just moved to tools/perf/util/mmap.c, I guess, reading on...
Quoted 18 lines
tools/perf/perf.h +0
-0/* * prctl(PR_TASK_PERF_EVENTS_DISABLE) will (cheaply) disable all * counters in the current task. tools/perf/util/evlist.c +1
-0@@ -12,6 +12,7 @@#include "evlist.h" #include "evsel.h" #include "util.h" +#include "mmap.h" #include <sys/mman.h> @@ -200,82 +201,14 @@ struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id)return NULL ; }
tools/perf/util/evlist.c struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) · line 211
Quoted 10 lines
tools/perf/util/evlist.c +0
-9-union perf_event *perf_evlist__mmap_read (struct perf_evlist *evlist,int idx) -{ -/* XXX Move this to perf.c, making it generally available */ -unsigned int page_size =sysconf (_SC_PAGE_SIZE); -struct perf_mmap *md = &evlist->mmap[idx]; -unsigned int head =perf_mmap__read_head (md); -unsigned int old = md->prev; -unsigned char *data = md->base + page_size; -union perf_event *event =NULL ;
Just keep this as a simple wrapper to the functions moved to the perf_mmap__ class
Quoted 60 lines
tools/perf/util/evlist.c +0
-58- -if (evlist->overwrite) { -/* -* If we're further behind than half the buffer, there's a chance -* the writer will bite our tail and mess up the samples under us. -* -* If we somehow ended up ahead of the head, we got messed up. -* -* In either case, truncate and restart at head. -*/ -int diff = head - old; -if (diff > md->mask /2 || diff <0 ) { -fprintf (stderr,"WARNING: failed to keep up with mmap data.\n" ); - -/* -* head points to a known good entry, start there. -*/ - old = head; - } - } - -if (old != head) { -size_t size; - - event = (union perf_event *)&data[old & md->mask]; - size = event->header.size; - -/* -* Event straddles the mmap boundary -- header should always -* be inside due to u64 alignment of output. -*/ -if ((old & md->mask) + size != ((old + size) & md->mask)) { -unsigned int offset = old; -unsigned int len =min (sizeof (*event), size), cpy; -void *dst = &evlist->event_copy; - -do { - cpy =min (md->mask +1 - (offset & md->mask), len); -memcpy (dst, &data[offset & md->mask], cpy); - offset += cpy; - dst += cpy; - len -= cpy; - }while (len); - - event = &evlist->event_copy; - } - - old += size; - } - - md->prev = old; - -if (!evlist->overwrite) -perf_mmap__write_tail (md, old); - -return event; -} -void perf_evlist__munmap (struct perf_evlist *evlist) {
tools/perf/util/evlist.c struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) · line 211
Quoted 10 lines
tools/perf/util/evlist.c +3
-4int i;for (i =0 ; i < evlist->nr_mmaps; i++) { -if (evlist->mmap[i].base !=NULL ) { -munmap (evlist->mmap[i].base, evlist->mmap_len); - evlist->mmap[i].base =NULL ; - } +struct perf_mmap *m = &evlist->mmap[i]; +if (m->base !=NULL ) +perf_mmap__close (m);
Wouldn't it be perf_mmap__munmap() ?
Quoted 8 lines
tools/perf/util/evlist.c +0
-1}free (evlist->mmap);@@ -292,20 +225,18 @@ int perf_evlist__alloc_mmap(struct perf_evlist *evlist) } static int __perf_evlist__mmap(struct perf_evlist *evlist, -int idx,int prot,int mask,int fd)
tools/perf/util/evlist.c int perf_evlist__alloc_mmap(struct perf_evlist *evlist) · line 232
Quoted 10 lines
tools/perf/util/evlist.c +4
-5+ int idx, int fd) { - evlist->mmap[idx].prev =0 ; - evlist->mmap[idx].mask = mask; - evlist->mmap[idx].base =mmap (NULL , evlist->mmap_len, prot, -MAP_SHARED , fd,0 ); -if (evlist->mmap[idx].base ==MAP_FAILED ) +struct perf_mmap *m = &evlist->mmap[idx]; + +if (perf_mmap__open (m, fd, evlist->overwrite, evlist->pages))
And here perf_mmap__mmap or at least perf_mmap__map and the other perf_mmap__unmap?
Quoted 21 lines
tools/perf/util/evlist.c +2
-2return -1 ;perf_evlist__add_pollfd (evlist, fd);return 0 ; } -static int perf_evlist__mmap_per_cpu (struct perf_evlist *evlist,int prot,int mask) +static int perf_evlist__mmap_per_cpu (struct perf_evlist *evlist) {struct perf_evsel *evsel;int cpu, thread;@@ -320,7 +251,7 @@ static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mif (output ==-1 ) { output = fd;if (__perf_evlist__mmap (evlist, cpu, - prot, mask, output) <0 ) + output) <0 ) goto out_unmap; }else {if (ioctl (fd,PERF_EVENT_IOC_SET_OUTPUT , output) !=0 )
tools/perf/util/evlist.c static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int m · line 270
Quoted 10 lines
tools/perf/util/evlist.c +3
-4@@ -334,15 +265,14 @@ static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int m out_unmap:for (cpu =0 ; cpu < evlist->cpus->nr; cpu++) { -if (evlist->mmap[cpu].base !=NULL ) { -munmap (evlist->mmap[cpu].base, evlist->mmap_len); - evlist->mmap[cpu].base =NULL ; - } +struct perf_mmap *m = &evlist->mmap[cpu]; +if (m->base !=NULL ) +perf_mmap__close (m);
ditto
Quoted 274 lines
tools/perf/util/evlist.c +9
-12}return -1 ; } -static int perf_evlist__mmap_per_thread (struct perf_evlist *evlist,int prot,int mask) +static int perf_evlist__mmap_per_thread (struct perf_evlist *evlist) {struct perf_evsel *evsel;int thread;@@ -356,7 +286,7 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, inif (output ==-1 ) { output = fd;if (__perf_evlist__mmap (evlist, thread, - prot, mask, output) <0 ) + output) <0 ) goto out_unmap; }else {if (ioctl (fd,PERF_EVENT_IOC_SET_OUTPUT , output) !=0 )@@ -369,10 +299,9 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, in out_unmap:for (thread =0 ; thread < evlist->threads->nr; thread++) { -if (evlist->mmap[thread].base !=NULL ) { -munmap (evlist->mmap[thread].base, evlist->mmap_len); - evlist->mmap[thread].base =NULL ; - } +struct perf_mmap *m = &evlist->mmap[thread]; +if (m->base !=NULL ) +perf_mmap__close (m); }return -1 ; }@@ -421,10 +350,8 @@ static int perf_evlist__init_ids(struct perf_evlist *evlist)*/ int perf_evlist__mmap (struct perf_evlist *evlist,int pages,bool overwrite) { -unsigned int page_size =sysconf (_SC_PAGE_SIZE); -int mask = pages * page_size -1 , ret;const struct cpu_map *cpus = evlist->cpus; -int prot =PROT_READ | (overwrite ?0 :PROT_WRITE ); +int ret;if (evlist->mmap ==NULL &&perf_evlist__alloc_mmap (evlist) <0 )return -ENOMEM ;@@ -433,16 +360,16 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite)return -ENOMEM ; evlist->overwrite = overwrite; - evlist->mmap_len = (pages +1 ) * page_size; + evlist->pages = pages; ret =perf_evlist__init_ids (evlist);if (ret)return ret;if (cpus->map[0 ] ==-1 ) -return perf_evlist__mmap_per_thread (evlist, prot, mask); +return perf_evlist__mmap_per_thread (evlist); -return perf_evlist__mmap_per_cpu (evlist, prot, mask); +return perf_evlist__mmap_per_cpu (evlist); } int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid,tools/perf/util/evlist.h +1
-1@@ -18,7 +18,7 @@ struct perf_evlist {int nr_entries;int nr_fds;int nr_mmaps; -int mmap_len; +int pages;bool overwrite;union perf_event event_copy;struct perf_mmap *mmap;tools/perf/util/mmap.c new file +138
-0@@ -0,0 +1,138 @@ +#include <string.h> +#include <stdio.h> +#include "mmap.h" + +int perf_mmap__open (struct perf_mmap *m,int fd,bool overwrite,int pages) +{ +unsigned int page_size =sysconf (_SC_PAGE_SIZE); +int mask, len, prot; +void *base; + + mask = pages * page_size -1 ; + len = (pages +1 ) * page_size; + prot =PROT_READ | (overwrite ?0 :PROT_WRITE ); + + base =mmap (NULL , len, prot,MAP_SHARED , fd,0 ); +if (base ==MAP_FAILED ) +return -1 ; + +memset (m,0 ,sizeof (*m)); + m->mask = mask; + m->len = len; + m->base = base; + m->fd = fd; + m->owrt = overwrite; +return 0 ; +} + +int perf_mmap__close (struct perf_mmap *m) +{ +int ret =munmap (m->base, m->len); + +memset (m,0x0 ,sizeof (*m)); +return ret; +} + +int perf_mmap__process (struct perf_mmap *md,perf_mmap_process_t process) +{ +unsigned int head, old, page_size =sysconf (_SC_PAGE_SIZE); +unsigned char *data = md->base + page_size; +unsigned long size; +void *buf; + + head =perf_mmap__read_head (md); + old = md->prev; + +if (old == head) +return 0 ; + + size = head - old; + +if ((old & md->mask) + size != (head & md->mask)) { + buf = &data[old & md->mask]; + size = md->mask +1 - (old & md->mask); + old += size; + +process (md, buf, size); + } + + buf = &data[old & md->mask]; + size = head - old; + old += size; + +process (md, buf, size); + + md->prev = old; +perf_mmap__write_tail (md, old); +return 1 ; +} + +union perf_event *perf_mmap__read (struct perf_mmap *md) +{ +unsigned int head, old, page_size =sysconf (_SC_PAGE_SIZE); +unsigned char *data = md->base + page_size; +union perf_event *event =NULL ; + + head =perf_mmap__read_head (md); + old = md->prev; + +if (md->owrt) { +/* +* If we're further behind than half the buffer, there's +* a chance the writer will bite our tail and mess up the +* samples under us. +* +* If we somehow ended up ahead of the head, we got messed up. +* +* In either case, truncate and restart at head. +*/ +int diff = head - old; +if (diff > md->mask /2 || diff <0 ) { +fprintf (stderr,"WARNING: failed to keep up " +"with mmap data.\n" ); + +/* +* head points to a known good entry, start there. +*/ + old = head; + } + } + +if (old != head) { +size_t size; + + event = (union perf_event *)&data[old & md->mask]; + size = event->header.size; + +/* +* Event straddles the mmap boundary -- header should always +* be inside due to u64 alignment of output. +*/ +if ((old & md->mask) + size != ((old + size) & md->mask)) { +unsigned int offset = old; +unsigned int len =min (sizeof (*event), size), cpy; +static union perf_event event_copy; +void *dst = &event_copy; + +do { + cpy =min (md->mask +1 - (offset & md->mask), + len); +memcpy (dst, &data[offset & md->mask], cpy); + offset += cpy; + dst += cpy; + len -= cpy; + }while (len); + + event = &event_copy; + } + + old += size; + } + + md->prev = old; + +if (!md->owrt) +perf_mmap__write_tail (md, old); + +return event; +}tools/perf/util/mmap.h new file +45
-0@@ -0,0 +1,45 @@ +#ifndef __PERF_MMAP_H +#define __PERF_MMAP_H + +#include <sys/mman.h> +#include "event.h" +#include "../perf.h" + +struct perf_mmap { +void *base; +int mask; +u_int prev; +int len; +int fd; +bool owrt; +}; + +typedef void (*perf_mmap_process_t )(struct perf_mmap *m, +void *buf,unsigned long size); + +int perf_mmap__open (struct perf_mmap *m,int fd,bool overwrite,int pages); +int perf_mmap__close (struct perf_mmap *m); +int perf_mmap__process (struct perf_mmap *m,perf_mmap_process_t process); +union perf_event *perf_mmap__read (struct perf_mmap *md); + +static inline unsigned int perf_mmap__read_head (struct perf_mmap *mm) +{ +struct perf_event_mmap_page *pc = mm->base; +int head = pc->data_head; +rmb (); +return head; +} + +static inline void perf_mmap__write_tail (struct perf_mmap *md, +unsigned long tail) +{ +struct perf_event_mmap_page *pc = md->base; + +/* +* ensure all reads are done before we write the tail out. +*/ +/* mb(); */ + pc->data_tail = tail; +} + +#endif /* __PERF_MMAP_H */ Signature · 1 line
1.7.4
Re: [PATCH 3/5] perf tool: Introducing perf_mmap object
Quoted 26 lines · Arnaldo Carvalho de Melo
Em Fri, Nov 18, 2011 at 02:46:43PM +0100, Jiri Olsa escreveu:
Quoted 10 lines
+++
b/tools/perf/builtin-test.ctools/perf/builtin-test.c +1
-0@@ -476,6 +477,7 @@ static int test__basic_mmap(void) expected_nr_events[nsyscalls], i, j;struct perf_evsel *evsels[nsyscalls], *evsel;int sample_size =__perf_evsel__sample_size (attr.sample_type); +struct perf_mmap *md;for (i =0 ; i < nsyscalls; ++i) {char name[64 ];tools/perf/builtin-test.c static int test__basic_mmap(void) · line 561
Quoted 10 lines
tools/perf/builtin-test.c +3
-1@@ -551,7 +553,9 @@ static int test__basic_mmap(void) ++foo; } -while ((event =perf_evlist__mmap_read (evlist,0 )) !=NULL ) { + md = &evlist->mmap[0 ]; + +while ((event =perf_mmap__read (md)) !=NULL ) {struct perf_sample sample;if (event->header.type !=PERF_RECORD_SAMPLE ) {Please keep
perf_evlist__mmap_read()as just a wrapper for the above operation, that way you reduce the patch size by not touching the functions that useperf_evlist__mmap_read().
Quoted 4 lines
Later, if we think this is the right thing to do, i.e. to open code it like you're doing above, we can elliminate it, but I think its better to keep it as
perf_evlist__mmap_read(evlist, 0)as it uses just one line instead of the 4 above :-)
right, I'll sent v2 shortly
Quoted 6 lines
Quoted 6 lines
tools/perf/builtin-top.c +0
-0@@ -38,6 +38,7 @@
SNIP
Quoted 21 lines
tools/perf/builtin-top.c
Quoted 12 lines
void
perf_evlist__munmap(struct perf_evlist *evlist){tools/perf/builtin-top.c +3
-4int i;for (i =0 ; i < evlist->nr_mmaps; i++) { -if (evlist->mmap[i].base !=NULL ) { -munmap (evlist->mmap[i].base, evlist->mmap_len); - evlist->mmap[i].base =NULL ; - } +struct perf_mmap *m = &evlist->mmap[i]; +if (m->base !=NULL ) +perf_mmap__close (m);Wouldn't it be
perf_mmap__munmap()?Quoted 6 lines
tools/perf/builtin-top.c +0
-0}free (evlist->mmap);@@ -292,20 +225,18 @@ int perf_evlist__alloc_mmap(struct perf_evlist *evlist) }
Quoted 15 lines
tools/perf/builtin-top.c int perf_evlist__alloc_mmap(struct perf_evlist *evlist) · line 232
Quoted 12 lines
tools/perf/builtin-top.c +4
-6static int __perf_evlist__mmap (struct perf_evlist *evlist, -int idx,int prot,int mask,int fd) +int idx,int fd) { - evlist->mmap[idx].prev =0 ; - evlist->mmap[idx].mask = mask; - evlist->mmap[idx].base =mmap (NULL , evlist->mmap_len, prot, -MAP_SHARED , fd,0 ); -if (evlist->mmap[idx].base ==MAP_FAILED ) +struct perf_mmap *m = &evlist->mmap[idx]; + +if (perf_mmap__open (m, fd, evlist->overwrite, evlist->pages))And here
perf_mmap__mmapor at leastperf_mmap__mapand the otherperf_mmap__unmap?
not sure, perf_mmap__open and perf_mmap__close actually does map/umap.. maybe we can rename them ;)
PATCH 4/5 perf tool: Introducing perf_data object
In reply to Jiri OlsaAdding perf_data object to handle input/output data files. The objective is to have this functionality centralized and ready for supporting multiple event data streams.
All the input/output file related functions originally scatered through the whole code are now placed in perf_mmap object.
To open/close data file:
perf_data__open
perf_data__close
Plus several helper functions:
perf_data__is_pipe
perf_data__is_ro
perf_data__is_new
perf_data__size
The perf_data object handles input/output file open/create processing, plus check for pipe intput/output.
The perf_session object was modified to contains perf_data. Thus for perf_session users the change is almost invisible apart from newly defined open mode enums:
PERF_DATA_NONE
PERF_DATA_READ
PERF_DATA_WRITE_TRUNC
PERF_DATA_WRITE_APPEND
The PERF_DATA_NONE serves for 'top' session, where no storage is needed. The rest is selfexplanatory.
--- tools/perf/Makefile | 2 + tools/perf/builtin-annotate.c | 3 +- tools/perf/builtin-buildid-list.c | 4 +- tools/perf/builtin-diff.c | 6 +- tools/perf/builtin-evlist.c | 3 +- tools/perf/builtin-inject.c | 3 +- tools/perf/builtin-kmem.c | 6 +- tools/perf/builtin-lock.c | 3 +- tools/perf/builtin-record.c | 134 ++++++++----------------- tools/perf/builtin-report.c | 7 +- tools/perf/builtin-sched.c | 6 +- tools/perf/builtin-script.c | 3 +- tools/perf/builtin-timechart.c | 7 +- tools/perf/builtin-top.c | 12 +-- tools/perf/util/data.c | 203 +++++++++++++++++++++++++++++++++++++ tools/perf/util/data.h | 65 ++++++++++++ tools/perf/util/header.c | 35 ++++--- tools/perf/util/header.h | 6 +- tools/perf/util/session.c | 115 ++++++++-------------- tools/perf/util/session.h | 26 ++++-- 20 files changed, 429 insertions(+), 220 deletions(-) create mode 100644 tools/perf/util/data.c create mode 100644 tools/perf/util/data.h
tools/perf/Makefile +2 -0
@@ -275,6 +275,7 @@ LIB_H += util/header.h LIB_H += util/help.h LIB_H += util/session.h LIB_H += util/mmap.h +LIB_H += util/data.h LIB_H += util/strbuf.h LIB_H += util/strlist.h LIB_H += util/strfilter.h
@@ -339,6 +340,7 @@ LIB_OBJS += $(OUTPUT)util/map.o LIB_OBJS += $(OUTPUT)util/pstack.o LIB_OBJS += $(OUTPUT)util/session.o LIB_OBJS += $(OUTPUT)util/mmap.o +LIB_OBJS += $(OUTPUT)util/data.o LIB_OBJS += $(OUTPUT)util/thread.o LIB_OBJS += $(OUTPUT)util/thread_map.o LIB_OBJS += $(OUTPUT)util/trace-event-parse.o
tools/perf/builtin-annotate.c +2 -1
@@ -183,7 +183,8 @@ static int __cmd_annotate(void)struct perf_evsel *pos;u64 total_nr_samples; - session =perf_session__new (input_name,O_RDONLY , force, false, &event_ops); + session =perf_session__new (input_name,PERF_DATA_READ , force, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-buildid-list.c +2 -2
@@ -43,8 +43,8 @@ static int perf_session__list_build_ids(void) {struct perf_session *session; - session =perf_session__new (input_name,O_RDONLY , force, false, - &build_id__mark_dso_hit_ops); + session =perf_session__new (input_name,PERF_DATA_READ , force, + false, &build_id__mark_dso_hit_ops);if (session ==NULL )return -1 ;
tools/perf/builtin-diff.c +4 -2
@@ -145,8 +145,10 @@ static int __cmd_diff(void)int ret, i;struct perf_session *session[2 ]; - session[0 ] =perf_session__new (input_old,O_RDONLY , force, false, &event_ops); - session[1 ] =perf_session__new (input_new,O_RDONLY , force, false, &event_ops); + session[0 ] =perf_session__new (input_old,PERF_DATA_READ , force, + false, &event_ops); + session[1 ] =perf_session__new (input_new,PERF_DATA_READ , force, + false, &event_ops);if (session[0 ] ==NULL || session[1 ] ==NULL )return -ENOMEM ;
tools/perf/builtin-evlist.c +2 -1
@@ -22,7 +22,8 @@ static int __cmd_evlist(void)struct perf_session *session;struct perf_evsel *pos; - session =perf_session__new (input_name,O_RDONLY ,0 , false,NULL ); + session =perf_session__new (input_name,PERF_DATA_READ , + false, false,NULL );if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-inject.c +2 -1
@@ -205,7 +205,8 @@ static int __cmd_inject(void) inject_ops.tracing_data = perf_event__repipe_tracing_data; } - session =perf_session__new (input_name,O_RDONLY , false, true, &inject_ops); + session =perf_session__new (input_name,PERF_DATA_READ , false, + true, &inject_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-kmem.c +4 -2
@@ -480,8 +480,10 @@ static void sort_result(void)static int __cmd_kmem (void ) {int err = -EINVAL ; -struct perf_session *session =perf_session__new (input_name,O_RDONLY , -0 , false, &event_ops); +struct perf_session *session; + + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-lock.c +2 -1
@@ -871,7 +871,8 @@ static struct perf_event_ops eops = {static int read_events (void ) { - session =perf_session__new (input_name,O_RDONLY ,0 , false, &eops); + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &eops);if (!session)die ("Initializing perf session failed\n" );
tools/perf/builtin-record.c +43 -91
298 lines not on this page · raw message
tools/perf/builtin-report.c +4 -3
@@ -35,7 +35,7 @@#include <linux/bitmap.h> -static char const *input_name ="perf.data" ; +static char const *input_name;static bool force, use_tui, use_stdio;static bool hide_unresolved;
@@ -264,7 +264,8 @@ static int __cmd_report(void)signal (SIGINT , sig_handler); - session =perf_session__new (input_name,O_RDONLY , force, false, &event_ops); + session =perf_session__new (input_name,PERF_DATA_READ , force, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
@@ -514,7 +515,7 @@ int cmd_report(int argc, const char **argv, const char *prefix __used)if (inverted_callchain) callchain_param.order =ORDER_CALLER ; -if (strcmp (input_name,"-" ) !=0 ) +if (input_name &&strcmp (input_name,"-" ) !=0 )setup_browser (true);else use_browser =0 ;
tools/perf/builtin-sched.c +4 -2
@@ -1640,8 +1640,10 @@ static struct perf_event_ops event_ops = {static void read_events (bool destroy,struct perf_session **psession) {int err = -EINVAL ; -struct perf_session *session =perf_session__new (input_name,O_RDONLY , -0 , false, &event_ops); +struct perf_session *session; + + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )die ("No Memory" );
tools/perf/builtin-script.c +2 -1
@@ -1261,7 +1261,8 @@ int cmd_script(int argc, const char **argv, const char *prefix __used)if (!script_name)setup_pager (); - session =perf_session__new (input_name,O_RDONLY ,0 , false, &event_ops); + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-timechart.c +4 -3
@@ -984,10 +984,11 @@ static struct perf_event_ops event_ops = {static int __cmd_timechart (void ) { -struct perf_session *session =perf_session__new (input_name,O_RDONLY , -0 , false, &event_ops); -int ret = -EINVAL ; +struct perf_session *session; +int ret =EINVAL ; + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-top.c +3 -9
@@ -959,11 +959,9 @@ static int __cmd_top(void) {pthread_t thread;int ret; -/* -* FIXME: perf_session__new should allow passing a O_MMAP, so that all this -* mmap reading, etc is encapsulated in it. Use O_WRONLY for now. -*/ - top.session =perf_session__new (NULL ,O_WRONLY , false, false,NULL ); + + top.session =perf_session__new (NULL ,PERF_DATA_NONE , false, + false,NULL );if (top.session ==NULL )return -ENOMEM ;
@@ -1248,10 +1246,6 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) pos->attr.sample_period = default_interval; } -if (perf_evlist__alloc_pollfd (top.evlist) <0 || -perf_evlist__alloc_mmap (top.evlist) <0 ) - goto out_free_fd; - top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node); symbol_conf.priv_size =sizeof (struct annotation );
tools/perf/util/data.c new file +203 -0
203 lines not on this page · raw message
tools/perf/util/data.h new file +65 -0
65 lines not on this page · raw message
tools/perf/util/header.c +18 -17
@@ -1123,7 +1123,7 @@ int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full) {struct header_print_data hd;struct perf_header *header = &session->header; -int fd = session->fd; +int fd =perf_session__fd (session); hd.fp = fp; hd.full = full;
@@ -1525,13 +1525,13 @@ int perf_header__write_pipe(int fd)int perf_session__write_header (struct perf_session *session,struct perf_evlist *evlist, -int fd,bool at_exit) +bool at_exit) {struct perf_file_header f_header;struct perf_file_attr f_attr;struct perf_header *header = &session->header;struct perf_evsel *attr, *pair =NULL ; -int err; +int err, fd =perf_session__fd (session);lseek (fd,sizeof (f_header),SEEK_SET );
@@ -1960,10 +1960,11 @@ static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,return 0 ; } -static int perf_header__read_pipe (struct perf_session *session,int fd) +static int perf_header__read_pipe (struct perf_session *session) {struct perf_header *header = &session->header;struct perf_pipe_file_header f_header; +int fd =perf_session__fd (session);if (perf_file_header__read_pipe (&f_header, header, fd, session->repipe) <0 ) {
@@ -1971,25 +1972,24 @@ static int perf_header__read_pipe(struct perf_session *session, int fd)return -EINVAL ; } - session->fd = fd; -return 0 ; } -int perf_session__read_header (struct perf_session *session,int fd) +int perf_session__read_header (struct perf_session *session) {struct perf_header *header = &session->header;struct perf_file_header f_header;struct perf_file_attr f_attr;u64 f_id;int nr_attrs, nr_ids, i, j; +int fd =perf_session__fd (session); session->evlist =perf_evlist__new (NULL ,NULL );if (session->evlist ==NULL )return -ENOMEM ; -if (session->fd_pipe) -return perf_header__read_pipe (session, fd); +if (perf_session__is_pipe (session)) +return perf_header__read_pipe (session);if (perf_file_header__read (&f_header, header, fd) <0 ) {pr_debug ("incompatible file format\n" );
@@ -2213,14 +2213,14 @@ int perf_event__process_event_type(union perf_event *event,return 0 ; } -int perf_event__synthesize_tracing_data (int fd,struct perf_evlist *evlist, -perf_event__handler_t process, - struct perf_session *session __unused) +int perf_event__synthesize_tracing_data (struct perf_evlist *evlist, +perf_event__handler_t process, +struct perf_session *session) {union perf_event ev;struct tracing_data *tdata;ssize_t size =0 , aligned_size =0 , padding; -int err__used =0 ; +int fd =perf_session__fd (session);/* * We are going to store the size of the data followed
@@ -2263,18 +2263,19 @@ int perf_event__process_tracing_data(union perf_event *event, struct perf_session *session) {ssize_t size_read, padding, size = event->tracing_data.size; -off_t offset =lseek (session->fd,0 ,SEEK_CUR ); +int fd =perf_session__fd (session); +off_t offset =lseek (fd,0 ,SEEK_CUR );char buf[BUFSIZ ];/* setup for reading amidst mmap */ -lseek (session->fd, offset +sizeof (struct tracing_data_event ), +lseek (fd, offset +sizeof (struct tracing_data_event ),SEEK_SET ); - size_read =trace_report (session->fd, session->repipe); + size_read =trace_report (fd, session->repipe); padding =ALIGN (size_read,sizeof (u64)) - size_read; -if (read (session->fd, buf, padding) <0 ) +if (read (fd, buf, padding) <0 )die ("reading input file" );if (session->repipe) {int retw =write (STDOUT_FILENO , buf, padding);
tools/perf/util/header.h +3 -3
@@ -69,10 +69,10 @@ struct perf_header {struct perf_evlist ; -int perf_session__read_header (struct perf_session *session,int fd); +int perf_session__read_header (struct perf_session *session);int perf_session__write_header (struct perf_session *session,struct perf_evlist *evlist, -int fd,bool at_exit); +bool at_exit);int perf_header__write_pipe (int fd);int perf_header__push_event (u64 id,const char *name);
@@ -111,7 +111,7 @@ int perf_event__synthesize_event_types(perf_event__handler_t process,int perf_event__process_event_type (union perf_event *event,struct perf_session *session); -int perf_event__synthesize_tracing_data (int fd,struct perf_evlist *evlist, +int perf_event__synthesize_tracing_data (struct perf_evlist *evlist,perf_event__handler_t process,struct perf_session *session); int perf_event__process_tracing_data(union perf_event *event,
tools/perf/util/session.c +41 -74
227 lines not on this page · raw message
tools/perf/util/session.h +19 -7
@@ -6,6 +6,7 @@#include "header.h" #include "symbol.h" #include "thread.h" +#include "data.h" #include <linux/rbtree.h> #include "../../../include/linux/perf_event.h"
@@ -28,7 +29,7 @@ struct ordered_samples {struct perf_session {struct perf_header header; -unsigned long size; +struct perf_data data;unsigned long mmap_window;struct rb_root threads;struct list_head dead_threads;
@@ -45,8 +46,6 @@ struct perf_session {struct hists hists;u64 sample_type;int sample_size; -int fd; -bool fd_pipe;bool repipe;bool sample_id_all;u16 id_hdr_size;
@@ -54,7 +53,6 @@ struct perf_session {char *cwd;struct ordered_samples ordered_samples;struct callchain_cursor callchain_cursor; -char filename[0 ]; };struct perf_evsel ;
@@ -88,9 +86,8 @@ struct perf_event_ops {bool ordering_requires_timestamps; }; -struct perf_session *perf_session__new (const char *filename,int mode, -bool force,bool repipe, -struct perf_event_ops *ops); +struct perf_session *perf_session__new (const char *name,int mode,bool force, +bool repipe,struct perf_event_ops *ops);void perf_session__delete (struct perf_session *self);void perf_event_header__bswap (struct perf_event_header *self);
@@ -179,4 +176,19 @@ int perf_session__cpu_bitmap(struct perf_session *session, const char *cpu_list, unsigned long *cpu_bitmap);void perf_session__fprintf_info (struct perf_session *s,FILE *fp,bool full); + +static inline bool perf_session__is_new (struct perf_session *self) +{ +return perf_data__is_new (&self->data); +} + +static inline int perf_session__fd (struct perf_session *self) +{ +return perf_data__fd (&self->data); +} + +static inline int perf_session__is_pipe (struct perf_session *self) +{ +return perf_data__is_pipe (&self->data); +}#endif /* __PERF_SESSION_H */
Signature · 1 line
1.7.4
PATCH 5/5 perf tool: Putting mmap support to perf_data object
In reply to Jiri OlsaAdding event memory maps management under the perf_data object. The objective is to have this functionality centralized and dont mix this functionality into event list.
New function 'perf_data__mmap' is added into perf_data. It prepares allocates/prepares both perf_mmap and pollfd structures, and does the actuall mmap-ing.
Each perf_mmap now contains perf_data_file structure, to have memory map connected directly with the storage.
--- tools/perf/builtin-record.c | 29 +++++-- tools/perf/builtin-test.c | 13 +++- tools/perf/builtin-top.c | 21 ++++-- tools/perf/util/data.c | 100 ++++++++++++++++++++++++ tools/perf/util/data.h | 13 +++ tools/perf/util/evlist.c | 181 ++----------------------------------------- tools/perf/util/evlist.h | 8 +-- tools/perf/util/evsel.h | 4 +- tools/perf/util/mmap.h | 3 + tools/perf/util/python.c | 23 ++++-- tools/perf/util/session.h | 6 ++ 11 files changed, 192 insertions(+), 209 deletions(-)
tools/perf/builtin-record.c +20 -9
@@ -318,8 +318,9 @@ try_again:exit (-1 ); } -if (perf_evlist__mmap (evlist, mmap_pages, false) <0 ) -die ("failed to mmap with %d (%s)\n" , errno,strerror (errno)); +if (perf_evlist__init_ids (evlist)) +die ("failed to init sample IDs %d (%s)\n" , + errno,strerror (errno));if (perf_session__is_new (session)) session->evlist = evlist;
@@ -402,18 +403,18 @@ static struct perf_event_header finished_round_event = { .type = PERF_RECORD_FINISHED_ROUND, }; -static void mmap_read (struct perf_mmap *m__used ,void *buf,unsigned long size) +static void mmap_read (struct perf_mmap *m,void *buf,unsigned long size) { -int fd =perf_session__fd (session); +int fd =perf_data__fd_file (m->file);write_output (fd, buf, size); } -static int mmap_read_all (void ) +static int mmap_read_all (struct perf_data *data) {int i, ret =0 ; -for (i =0 ; i < evsel_list->nr_mmaps; i++) { -struct perf_mmap *m = &evsel_list->mmap[i]; +for (i =0 ; i < data->nr_mmaps; i++) { +struct perf_mmap *m = &data->mmaps[i]; ret +=perf_mmap__process (m, mmap_read); }
@@ -540,6 +541,13 @@ static int __cmd_record(int argc, const char **argv)open_counters (evsel_list); +if (perf_data__mmap (perf_session__data (session), + evsel_list, mmap_pages, false) <0 ) { +pr_err ("Failed to mmap with %d (%s)\n" , + errno,strerror (errno)); + goto out_delete_session; + } +/* * perf_session__delete(session) will be called at atexit_header() */
@@ -642,10 +650,13 @@ static int __cmd_record(int argc, const char **argv)close (go_pipe[1 ]);for (;;) { -if (!mmap_read_all ()) { +struct perf_data *data =perf_session__data (session); + +if (!mmap_read_all (data)) { +if (done)break ; - err =poll (evsel_list->pollfd, evsel_list->nr_fds,-1 ); + err =poll (data->fds, data->nr_fds,-1 ); waking++; }
tools/perf/builtin-test.c +10 -3
@@ -478,6 +478,9 @@ static int test__basic_mmap(void)struct perf_evsel *evsels[nsyscalls], *evsel;int sample_size =__perf_evsel__sample_size (attr.sample_type);struct perf_mmap *md; +struct perf_data data; + +PERF_DATA__NONE (data);for (i =0 ; i < nsyscalls; ++i) {char name[64 ];
@@ -541,7 +544,11 @@ static int test__basic_mmap(void) } } -if (perf_evlist__mmap (evlist,128 , true) <0 ) { + err =perf_evlist__init_ids (evlist); +if (err) + goto out_close_fd; + +if (perf_data__mmap (&data, evlist,128 , true)) {pr_debug ("failed to mmap events: %d (%s)\n" , errno,strerror (errno)); goto out_close_fd;
@@ -553,7 +560,7 @@ static int test__basic_mmap(void) ++foo; } - md = &evlist->mmap[0 ]; + md = &data.mmaps[0 ];while ((event =perf_mmap__read (md)) !=NULL ) {struct perf_sample sample;
@@ -591,7 +598,7 @@ static int test__basic_mmap(void) err =0 ; out_munmap: -perf_evlist__munmap (evlist); +perf_data__close (&data); out_close_fd:for (i =0 ; i < nsyscalls; ++i)perf_evsel__close_fd (evsels[i],1 , threads->nr);
tools/perf/builtin-top.c +14 -7
@@ -836,15 +836,16 @@ static void session_mmap_read(struct perf_session *self,static void perf_session__mmap_read (struct perf_session *self) { +struct perf_data *data =perf_session__data (self);int i; -for (i =0 ; i < top.evlist->nr_mmaps; i++) { -struct perf_mmap *md = &top.evlist->mmap[i]; +for (i =0 ; i < data->nr_mmaps; i++) { +struct perf_mmap *md = &data->mmaps[i];session_mmap_read (self, md); } } -static void start_counters (struct perf_evlist *evlist) +static void start_counters (struct perf_data *data,struct perf_evlist *evlist) {struct perf_evsel *counter, *first;
@@ -925,7 +926,10 @@ try_again: } } -if (perf_evlist__mmap (evlist, mmap_pages, false) <0 ) { +if (perf_evlist__init_ids (evlist)) + goto out_err; + +if (perf_data__mmap (data, evlist, mmap_pages, false) <0 ) {ui__warning ("Failed to mmap with %d (%s)\n" , errno,strerror (errno)); goto out_err;
@@ -957,6 +961,7 @@ static int setup_sample_type(void)static int __cmd_top (void ) { +struct perf_data *data;pthread_t thread;int ret;
@@ -975,12 +980,14 @@ static int __cmd_top(void)else perf_event__synthesize_threads (perf_event__process, top.session); -start_counters (top.evlist); + data =perf_session__data (top.session); + +start_counters (data, top.evlist); top.session->evlist = top.evlist;perf_session__update_sample_type (top.session);/* Wait for a minimal set of events before starting the snapshot */ -poll (top.evlist->pollfd, top.evlist->nr_fds,100 ); +poll (data->fds, data->nr_fds,100 );perf_session__mmap_read (top.session);
@@ -1006,7 +1013,7 @@ static int __cmd_top(void)perf_session__mmap_read (top.session);if (hits == top.samples) - ret =poll (top.evlist->pollfd, top.evlist->nr_fds,100 ); + ret =poll (data->fds, data->nr_fds,100 ); } out_delete:
tools/perf/util/data.c +100 -0
103 lines not on this page · raw message
tools/perf/util/data.h +13 -0
@@ -28,12 +28,20 @@ struct perf_data {int mode;bool force; +struct perf_mmap *mmaps; +struct pollfd *fds; +int nr_mmaps; +int nr_fds; +struct list_head files; };int perf_data__open (struct perf_data *data,const char *name,int mode,bool force);void perf_data__close (struct perf_data *data); +int perf_data__mmap (struct perf_data *data, +struct perf_evlist *evlist, +int pages,bool overwrite);#define PERF_DATA__NONE (__data) \ perf_data__open(&__data, NULL, PERF_DATA_NONE, false);
@@ -62,4 +70,9 @@ static inline int perf_data__fd(struct perf_data *data) {return data->header.fd; } + +static inline int perf_data__fd_file (struct perf_data_file *file) +{ +return file->fd; +}#endif /* __PERF_DATA_H */
tools/perf/util/evlist.c +7 -174
236 lines not on this page · raw message
tools/perf/util/evlist.h +1 -7
@@ -16,13 +16,6 @@ struct perf_evlist {struct list_head entries;struct hlist_head heads[PERF_EVLIST__HLIST_SIZE ];int nr_entries; -int nr_fds; -int nr_mmaps; -int pages; -bool overwrite; -union perf_event event_copy; -struct perf_mmap *mmap; -struct pollfd *pollfd;struct thread_map *threads;struct cpu_map *cpus;struct perf_evsel *selected;
@@ -74,6 +67,7 @@ int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, pid_t target_tid, const char *cpu_list);void perf_evlist__delete_maps (struct perf_evlist *evlist);int perf_evlist__set_filters (struct perf_evlist *evlist); +int perf_evlist__init_ids (struct perf_evlist *evlist);u64 perf_evlist__sample_type (const struct perf_evlist *evlist);bool perf_evlist__sample_id_all (const const struct perf_evlist *evlist);
tools/perf/util/evsel.h +3 -1
@@ -8,7 +8,9 @@#include "xyarray.h" #include "cgroup.h" #include "hist.h" - + +#define PERF_EVSEL_FD (e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) + struct perf_counts_values { union { struct {
tools/perf/util/mmap.h +3 -0
@@ -4,6 +4,7 @@#include <sys/mman.h> #include "event.h" #include "../perf.h" +#include "data.h" struct perf_mmap {void *base;
@@ -12,6 +13,8 @@ struct perf_mmap {int len;int fd;bool owrt; + +struct perf_data_file *file; }; typedef void (*perf_mmap_process_t)(struct perf_mmap *m,
tools/perf/util/python.c +15 -8
@@ -7,6 +7,7 @@#include "event.h" #include "cpumap.h" #include "thread_map.h" +#include "data.h" /* Define PyVarObject_HEAD_INIT for python 2.5 */ #ifndef PyVarObject_HEAD_INIT
@@ -598,7 +599,6 @@ static int pyrf_evsel__init(struct pyrf_evsel *pevsel,static void pyrf_evsel__delete (struct pyrf_evsel *pevsel) { -perf_evsel__exit (&pevsel->evsel); pevsel->ob_type->tp_free ((PyObject *)pevsel); }
@@ -668,6 +668,7 @@ static int pyrf_evsel__setup_types(void)struct pyrf_evlist {PyObject_HEAD + struct perf_data data;struct perf_evlist evlist; };
@@ -684,12 +685,12 @@ static int pyrf_evlist__init(struct pyrf_evlist *pevlist, threads = ((struct pyrf_thread_map *)pthreads)->threads; cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;perf_evlist__init (&pevlist->evlist, cpus, threads); +PERF_DATA__NONE (pevlist->data);return 0 ; }static void pyrf_evlist__delete (struct pyrf_evlist *pevlist) { -perf_evlist__exit (&pevlist->evlist); pevlist->ob_type->tp_free ((PyObject *)pevlist); }
@@ -697,6 +698,7 @@ static PyObject *pyrf_evlist__mmap(struct pyrf_evlist *pevlist, PyObject *args, PyObject *kwargs) {struct perf_evlist *evlist = &pevlist->evlist; +struct perf_data *data = &pevlist->data;static char *kwlist[] = {"pages" ,"overwrite" ,NULL };int pages =128 , overwrite = false;
@@ -704,7 +706,12 @@ static PyObject *pyrf_evlist__mmap(struct pyrf_evlist *pevlist, &pages, &overwrite))return NULL ; -if (perf_evlist__mmap (evlist, pages, overwrite) <0 ) { +if (perf_evlist__init_ids (evlist)) { +PyErr_SetFromErrno (PyExc_OSError); +return NULL ; + } + +if (perf_data__mmap (data, evlist, pages, overwrite) <0 ) {PyErr_SetFromErrno (PyExc_OSError);return NULL ; }
@@ -716,14 +723,14 @@ static PyObject *pyrf_evlist__mmap(struct pyrf_evlist *pevlist,static PyObject *pyrf_evlist__poll (struct pyrf_evlist *pevlist,PyObject *args,PyObject *kwargs) { -struct perf_evlist *evlist = &pevlist->evlist; +struct perf_data *data = &pevlist->data;static char *kwlist[] = {"timeout" ,NULL };int timeout =-1 , n;if (!PyArg_ParseTupleAndKeywords (args, kwargs,"|i" , kwlist, &timeout))return NULL ; - n =poll (evlist->pollfd, evlist->nr_fds, timeout); + n =poll (data->fds, data->nr_fds, timeout);if (n <0 ) {PyErr_SetFromErrno (PyExc_OSError);return NULL ;
@@ -735,13 +742,13 @@ static PyObject *pyrf_evlist__poll(struct pyrf_evlist *pevlist,static PyObject *pyrf_evlist__get_pollfd (struct pyrf_evlist *pevlist,PyObject *args__used ,PyObject *kwargs__used ) { -struct perf_evlist *evlist = &pevlist->evlist; +struct perf_data *data = &pevlist->data;PyObject *list =PyList_New (0 );int i; -for (i =0 ; i < evlist->nr_fds; ++i) { +for (i =0 ; i < data->nr_fds; ++i) {PyObject *file; -FILE *fp =fdopen (evlist->pollfd[i].fd,"r" ); +FILE *fp =fdopen (data->fds[i].fd,"r" );if (fp ==NULL ) goto free_list;
tools/perf/util/session.h +6 -0
@@ -191,4 +191,10 @@ static inline int perf_session__is_pipe(struct perf_session *self) {return perf_data__is_pipe (&self->data); } + +static inline struct perf_data *perf_session__data (struct perf_session *self) +{ +return &self->data; +} +#endif /* __PERF_SESSION_H */
Signature · 1 line
1.7.4
Em Fri, Nov 18, 2011 at 02:46:40PM +0100, Jiri Olsa escreveu:
Quoted 16 lines
Hi,
based on following discussion:
http://marc.info/?l=linux-kernel&m=131731212425421&w=2
http://marc.info/?l=linux-kernel&m=131736842010972&w=2I made some changes to have event memory mapping and input/output data code in one place. The reason was to make the change for multiple event data files at least possible ;)
This patchset is marked as RFC since it's quite big change and I expect some better design might come up.
However, patches 1 and 2 are not big deal, since they are just small (not so related) changes.
Quoted 7 lines
Patches 3-4 are the main change of the patchset.
1/5 perf tool: Fix session
host_nachineretrieval 2/5 perf tool: Initialize events IDs in a single function 3/5 perf tool: Introducingperf_mmapobject
4/5 perf tool: Introducingperf_dataobject
5/5 perf tool: Putting mmap support toperf_dataobject
Reading those patches now.
Quoted 3 lines
I did some overall testing of all the changed command, and so far so good. Just hit issue with 'diff' command, but this one seems not to work even without my changes.
Yeah, it needs more work to properly suport multi-event perf.data files.
Quoted 3 lines
Also the current perf python binsing code is broken wit missing dependencies so I could not test my changes there.
Which ones? I fixed some and posted to my perf/urgent branch, waiting for Ingo to merge.
- Arnaldo
RFC PATCH v2 1/1 perf tool: Refactoring IO data files code
In reply to Jiri OlsaHi,
based on following discussion:
http://marc.info/?l=linux-kernel&m=131731212425421&w=2
http://marc.info/?l=linux-kernel&m=131736842010972&w=2
I made some changes to have event memory mapping and input/output data code in one place. The reason was to make the change for multiple event data files at least possible ;)
This patchset is marked as RFC since it's quite big change and I expect some better design might come up.
However, patches 1 and 2 are not big deal, since they are just small (not so related) changes.
Patches 3-4 are the main change of the patchset.
1/5 perf tool: Fix session host_nachine retrieval 2/5 perf tool: Initialize events IDs in a single function 3/5 perf tool: Introducing perf_mmap object
4/5 perf tool: Introducing perf_data object
5/5 perf tool: Putting mmap support to perf_data object
I did some overall testing of all the changed command, and so far so good. Just hit issue with 'diff' command, but this one seems not to work even without my changes.
Also the current perf python binsing code is broken wit missing dependencies so I could not test my changes there.
v2 changes:
- 3/5 kept the perf_evlist__mmap_read function as a wrapper around perf_mmap__read
- 4/5 removed unlink call from file_backup, because it's not needed when followed with rename
thanks for comments,
jirka
PATCH v2 1/5 perf tool: Fix session host_nachine retrieval
The host_machine is statically included inside perf_session struct so it is always present.
Changing function:
'perf_session__find_host_machine' to 'perf_session__host_machine'
Removing NULL checking code after this function, since it is not needed.
--- tools/perf/builtin-inject.c | 6 +----- tools/perf/builtin-kmem.c | 12 +++++------- tools/perf/builtin-record.c | 6 +----- tools/perf/builtin-top.c | 4 ++-- tools/perf/util/event.c | 14 +++++--------- tools/perf/util/session.h | 6 +++--- 6 files changed, 17 insertions(+), 31 deletions(-)
tools/perf/builtin-inject.c +1 -5
@@ -111,11 +111,7 @@ static int dso__inject_build_id(struct dso *self, struct perf_session *session)return -1 ; } - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) { -pr_err ("Can't find machine for session\n" ); -return -1 ; - } + machine =perf_session__host_machine (session);if (self->kernel) misc =PERF_RECORD_MISC_KERNEL ;
tools/perf/builtin-kmem.c +5 -7
@@ -342,7 +342,6 @@ static void __print_result(struct rb_root *root, struct perf_session *session, int n_lines, int is_caller) {struct rb_node *next; -struct machine *machine;printf ("%.102s\n" , graph_dotted_line);printf (" %-34s |" , is_caller ?"Callsite" :"Alloc Ptr" );
@@ -351,11 +350,6 @@ static void __print_result(struct rb_root *root, struct perf_session *session, next =rb_first (root); - machine =perf_session__find_host_machine (session); -if (!machine) { -pr_err ("__print_result: couldn't find kernel information\n" ); -return ; - }while (next && n_lines--) { struct alloc_stat *data = rb_entry(next, struct alloc_stat, node);
@@ -366,8 +360,12 @@ static void __print_result(struct rb_root *root, struct perf_session *session,if (is_caller) { addr = data->call_site; -if (!raw_ip) +if (!raw_ip) { +struct machine *machine; + + machine =perf_session__host_machine (session); sym =machine__find_kernel_function (machine, addr, &map,NULL ); + } }else addr = data->ptr;
tools/perf/builtin-record.c +1 -5
@@ -662,11 +662,7 @@ static int __cmd_record(int argc, const char **argv) } } - machine =perf_session__find_host_machine (session); -if (!machine) { -pr_err ("Couldn't find native kernel information.\n" ); -return -1 ; - } + machine =perf_session__host_machine (session); err =perf_event__synthesize_kernel_mmap (process_synthesized_event, session, machine,"_text" );
tools/perf/builtin-top.c +2 -2
@@ -692,13 +692,13 @@ static void perf_event__process_sample(const union perf_event *event, ++top.us_samples;if (top.hide_user_symbols)return ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session);break ;case PERF_RECORD_MISC_KERNEL : ++top.kernel_samples;if (top.hide_kernel_symbols)return ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session);break ;case PERF_RECORD_MISC_GUEST_KERNEL : ++top.guest_kernel_samples;
tools/perf/util/event.c +5 -9
@@ -589,12 +589,12 @@ int perf_event__process_mmap(union perf_event *event,return 0 ; } - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) - goto out_problem; thread =perf_session__findnew (session, event->mmap.pid);if (thread ==NULL ) goto out_problem; + + machine =perf_session__host_machine (session); + map = map__new(&machine->user_dsos, event->mmap.start, event->mmap.len, event->mmap.pgoff, event->mmap.pid, event->mmap.filename,
@@ -672,15 +672,11 @@ void thread__find_addr_map(struct thread *self,if (cpumode ==PERF_RECORD_MISC_KERNEL && perf_host) { al->level ='k' ; - machine =perf_session__find_host_machine (session); -if (machine ==NULL ) { - al->map =NULL ; -return ; - } + machine =perf_session__host_machine (session); mg = &machine->kmaps; }else if (cpumode ==PERF_RECORD_MISC_USER && perf_host) { al->level ='.' ; - machine =perf_session__find_host_machine (session); + machine =perf_session__host_machine (session); }else if (cpumode ==PERF_RECORD_MISC_GUEST_KERNEL && perf_guest) { al->level ='g' ; machine =perf_session__find_machine (session, pid);
tools/perf/util/session.h +3 -3
@@ -121,7 +121,7 @@ void perf_session__update_sample_type(struct perf_session *self);void perf_session__remove_thread (struct perf_session *self,struct thread *th);static inline -struct machine *perf_session__find_host_machine (struct perf_session *self) +struct machine *perf_session__host_machine (struct perf_session *self) {return &self->host_machine; }
@@ -130,7 +130,7 @@ static inlinestruct machine *perf_session__find_machine (struct perf_session *self,pid_t pid) {if (pid ==HOST_KERNEL_ID ) -return &self->host_machine; +return perf_session__host_machine (self);return machines__find (&self->machines, pid); }
@@ -138,7 +138,7 @@ static inlinestruct machine *perf_session__findnew_machine (struct perf_session *self,pid_t pid) {if (pid ==HOST_KERNEL_ID ) -return &self->host_machine; +return perf_session__host_machine (self);return machines__findnew (&self->machines, pid); }
Signature · 1 line
1.7.6.4
PATCH v2 2/5 perf tool: Initialize events IDs in a single function
In reply to Jiri OlsaCurrently events' IDs initialization is scattered among the code. Adding function 'perf_evlist__init_ids' to allocate and retrive events' IDs in one place.
--- tools/perf/util/evlist.c | 50 +++++++++++++++++++++++++++++---------------- 1 files changed, 32 insertions(+), 18 deletions(-)
tools/perf/util/evlist.c +32 -18
@@ -326,10 +326,6 @@ static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mif (ioctl (fd,PERF_EVENT_IOC_SET_OUTPUT , output) !=0 ) goto out_unmap; } - -if ((evsel->attr.read_format &PERF_FORMAT_ID ) && -perf_evlist__id_add_fd (evlist, evsel, cpu, thread, fd) <0 ) - goto out_unmap; } } }
@@ -366,10 +362,6 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, inif (ioctl (fd,PERF_EVENT_IOC_SET_OUTPUT , output) !=0 ) goto out_unmap; } - -if ((evsel->attr.read_format &PERF_FORMAT_ID ) && -perf_evlist__id_add_fd (evlist, evsel,0 , thread, fd) <0 ) - goto out_unmap; } }
@@ -385,6 +377,33 @@ out_unmap:return -1 ; } +static int perf_evlist__init_ids (struct perf_evlist *evlist) +{ +struct perf_evsel *evsel; +struct thread_map *threads = evlist->threads; +struct cpu_map *cpus = evlist->cpus; +int cpu, thread; + +list_for_each_entry (evsel, &evlist->entries, node) { +if (!(evsel->attr.read_format &PERF_FORMAT_ID )) +continue ; + +if ((!evsel->sample_id) && +perf_evsel__alloc_id (evsel, cpus->nr, threads->nr) <0 ) +return -ENOMEM ; + +for (cpu =0 ; cpu < cpus->nr; cpu++) +for (thread =0 ; thread < threads->nr; thread++) { +int fd =FD (evsel, cpu, thread); +if (perf_evlist__id_add_fd (evlist, evsel, + cpu, thread, fd)) +return -EINVAL ; + } + } + +return 0 ; +} +/** perf_evlist__mmap - Create per cpu maps to receive events * * @evlist - list of events
@@ -403,10 +422,8 @@ out_unmap:int perf_evlist__mmap (struct perf_evlist *evlist,int pages,bool overwrite) {unsigned int page_size =sysconf (_SC_PAGE_SIZE); -int mask = pages * page_size -1 ; -struct perf_evsel *evsel; +int mask = pages * page_size -1 , ret;const struct cpu_map *cpus = evlist->cpus; -const struct thread_map *threads = evlist->threads;int prot =PROT_READ | (overwrite ?0 :PROT_WRITE );if (evlist->mmap ==NULL &&perf_evlist__alloc_mmap (evlist) <0 )
@@ -418,14 +435,11 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) evlist->overwrite = overwrite; evlist->mmap_len = (pages +1 ) * page_size; -list_for_each_entry (evsel, &evlist->entries, node) { -if ((evsel->attr.read_format &PERF_FORMAT_ID ) && - evsel->sample_id ==NULL && -perf_evsel__alloc_id (evsel, cpus->nr, threads->nr) <0 ) -return -ENOMEM ; - } + ret =perf_evlist__init_ids (evlist); +if (ret) +return ret; -if (evlist->cpus->map[0 ] ==-1 ) +if (cpus->map[0 ] ==-1 )return perf_evlist__mmap_per_thread (evlist, prot, mask);return perf_evlist__mmap_per_cpu (evlist, prot, mask);
Signature · 1 line
1.7.6.4
PATCH v2 3/5 perf tool: Introducing perf_mmap object
In reply to Jiri OlsaAdding perf_mmap object to handle event memory maps.
All the memory map related functions originally scatered through the whole code arenow place in perf_mmap object.
To map and unmap perf_mmap:
perf_mmap__open
perf_mmap__close
For reading events via callback:
perf_mmap__process
For reading events directly:
perf_mmap__read
following helpers were moved in from perf.h:
perf_mmap__read_head
perf_mmap__write_tail
--- tools/perf/Makefile | 2 + tools/perf/builtin-record.c | 57 ++++------------- tools/perf/perf.h | 26 -------- tools/perf/util/evlist.c | 116 ++++++++---------------------------- tools/perf/util/evlist.h | 2 +- tools/perf/util/mmap.c | 138 +++++++++++++++++++++++++++++++++++++++++++ tools/perf/util/mmap.h | 45 ++++++++++++++ 7 files changed, 225 insertions(+), 161 deletions(-) create mode 100644 tools/perf/util/mmap.c create mode 100644 tools/perf/util/mmap.h
tools/perf/Makefile +2 -0
@@ -274,6 +274,7 @@ LIB_H += util/xyarray.h LIB_H += util/header.h LIB_H += util/help.h LIB_H += util/session.h +LIB_H += util/mmap.h LIB_H += util/strbuf.h LIB_H += util/strlist.h LIB_H += util/strfilter.h
@@ -337,6 +338,7 @@ LIB_OBJS += $(OUTPUT)util/debug.o LIB_OBJS += $(OUTPUT)util/map.o LIB_OBJS += $(OUTPUT)util/pstack.o LIB_OBJS += $(OUTPUT)util/session.o +LIB_OBJS += $(OUTPUT)util/mmap.o LIB_OBJS += $(OUTPUT)util/thread.o LIB_OBJS += $(OUTPUT)util/thread_map.o LIB_OBJS += $(OUTPUT)util/trace-event-parse.o
tools/perf/builtin-record.c +14 -43
94 lines not on this page · raw message
tools/perf/perf.h +0 -26
@@ -104,32 +104,6 @@ void get_term_dimensions(struct winsize *ws);#include "util/types.h" #include <stdbool.h> -struct perf_mmap { -void *base; -int mask; -unsigned int prev; -}; - -static inline unsigned int perf_mmap__read_head (struct perf_mmap *mm) -{ -struct perf_event_mmap_page *pc = mm->base; -int head = pc->data_head; -rmb (); -return head; -} - -static inline void perf_mmap__write_tail (struct perf_mmap *md, -unsigned long tail) -{ -struct perf_event_mmap_page *pc = md->base; - -/* -* ensure all reads are done before we write the tail out. -*/ -/* mb(); */ - pc->data_tail = tail; -} -/* * prctl(PR_TASK_PERF_EVENTS_DISABLE) will (cheaply) disable all * counters in the current task.
tools/perf/util/evlist.c +25 -91
195 lines not on this page · raw message
tools/perf/util/evlist.h +1 -1
@@ -18,7 +18,7 @@ struct perf_evlist {int nr_entries;int nr_fds;int nr_mmaps; -int mmap_len; +int pages;bool overwrite;union perf_event event_copy;struct perf_mmap *mmap;
tools/perf/util/mmap.c new file +138 -0
138 lines not on this page · raw message
tools/perf/util/mmap.h new file +45 -0
45 lines not on this page · raw message
Signature · 1 line
1.7.6.4
PATCH v2 4/5 perf tool: Introducing perf_data object
In reply to Jiri OlsaAdding perf_data object to handle input/output data files. The objective is to have this functionality centralized and ready for supporting multiple event data streams.
All the input/output file related functions originally scatered through the whole code are now placed in perf_mmap object.
To open/close data file:
perf_data__open
perf_data__close
Plus several helper functions:
perf_data__is_pipe
perf_data__is_ro
perf_data__is_new
perf_data__size
The perf_data object handles input/output file open/create processing, plus check for pipe intput/output.
The perf_session object was modified to contains perf_data. Thus for perf_session users the change is almost invisible apart from newly defined open mode enums:
PERF_DATA_NONE
PERF_DATA_READ
PERF_DATA_WRITE_TRUNC
PERF_DATA_WRITE_APPEND
The PERF_DATA_NONE serves for 'top' session, where no storage is needed. The rest is selfexplanatory.
--- tools/perf/Makefile | 2 + tools/perf/builtin-annotate.c | 3 +- tools/perf/builtin-buildid-list.c | 4 +- tools/perf/builtin-diff.c | 6 +- tools/perf/builtin-evlist.c | 3 +- tools/perf/builtin-inject.c | 3 +- tools/perf/builtin-kmem.c | 6 +- tools/perf/builtin-lock.c | 3 +- tools/perf/builtin-record.c | 134 ++++++++----------------- tools/perf/builtin-report.c | 7 +- tools/perf/builtin-sched.c | 6 +- tools/perf/builtin-script.c | 3 +- tools/perf/builtin-timechart.c | 7 +- tools/perf/builtin-top.c | 12 +-- tools/perf/util/data.c | 199 +++++++++++++++++++++++++++++++++++++ tools/perf/util/data.h | 65 ++++++++++++ tools/perf/util/header.c | 35 ++++--- tools/perf/util/header.h | 6 +- tools/perf/util/session.c | 115 ++++++++-------------- tools/perf/util/session.h | 26 ++++-- 20 files changed, 425 insertions(+), 220 deletions(-) create mode 100644 tools/perf/util/data.c create mode 100644 tools/perf/util/data.h
tools/perf/Makefile +2 -0
@@ -275,6 +275,7 @@ LIB_H += util/header.h LIB_H += util/help.h LIB_H += util/session.h LIB_H += util/mmap.h +LIB_H += util/data.h LIB_H += util/strbuf.h LIB_H += util/strlist.h LIB_H += util/strfilter.h
@@ -339,6 +340,7 @@ LIB_OBJS += $(OUTPUT)util/map.o LIB_OBJS += $(OUTPUT)util/pstack.o LIB_OBJS += $(OUTPUT)util/session.o LIB_OBJS += $(OUTPUT)util/mmap.o +LIB_OBJS += $(OUTPUT)util/data.o LIB_OBJS += $(OUTPUT)util/thread.o LIB_OBJS += $(OUTPUT)util/thread_map.o LIB_OBJS += $(OUTPUT)util/trace-event-parse.o
tools/perf/builtin-annotate.c +2 -1
@@ -183,7 +183,8 @@ static int __cmd_annotate(void)struct perf_evsel *pos;u64 total_nr_samples; - session =perf_session__new (input_name,O_RDONLY , force, false, &event_ops); + session =perf_session__new (input_name,PERF_DATA_READ , force, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-buildid-list.c +2 -2
@@ -43,8 +43,8 @@ static int perf_session__list_build_ids(void) {struct perf_session *session; - session =perf_session__new (input_name,O_RDONLY , force, false, - &build_id__mark_dso_hit_ops); + session =perf_session__new (input_name,PERF_DATA_READ , force, + false, &build_id__mark_dso_hit_ops);if (session ==NULL )return -1 ;
tools/perf/builtin-diff.c +4 -2
@@ -145,8 +145,10 @@ static int __cmd_diff(void)int ret, i;struct perf_session *session[2 ]; - session[0 ] =perf_session__new (input_old,O_RDONLY , force, false, &event_ops); - session[1 ] =perf_session__new (input_new,O_RDONLY , force, false, &event_ops); + session[0 ] =perf_session__new (input_old,PERF_DATA_READ , force, + false, &event_ops); + session[1 ] =perf_session__new (input_new,PERF_DATA_READ , force, + false, &event_ops);if (session[0 ] ==NULL || session[1 ] ==NULL )return -ENOMEM ;
tools/perf/builtin-evlist.c +2 -1
@@ -22,7 +22,8 @@ static int __cmd_evlist(void)struct perf_session *session;struct perf_evsel *pos; - session =perf_session__new (input_name,O_RDONLY ,0 , false,NULL ); + session =perf_session__new (input_name,PERF_DATA_READ , + false, false,NULL );if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-inject.c +2 -1
@@ -205,7 +205,8 @@ static int __cmd_inject(void) inject_ops.tracing_data = perf_event__repipe_tracing_data; } - session =perf_session__new (input_name,O_RDONLY , false, true, &inject_ops); + session =perf_session__new (input_name,PERF_DATA_READ , false, + true, &inject_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-kmem.c +4 -2
@@ -480,8 +480,10 @@ static void sort_result(void)static int __cmd_kmem (void ) {int err = -EINVAL ; -struct perf_session *session =perf_session__new (input_name,O_RDONLY , -0 , false, &event_ops); +struct perf_session *session; + + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-lock.c +2 -1
@@ -871,7 +871,8 @@ static struct perf_event_ops eops = {static int read_events (void ) { - session =perf_session__new (input_name,O_RDONLY ,0 , false, &eops); + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &eops);if (!session)die ("Initializing perf session failed\n" );
tools/perf/builtin-record.c +43 -91
298 lines not on this page · raw message
tools/perf/builtin-report.c +4 -3
@@ -35,7 +35,7 @@#include <linux/bitmap.h> -static char const *input_name ="perf.data" ; +static char const *input_name;static bool force, use_tui, use_stdio;static bool hide_unresolved;
@@ -264,7 +264,8 @@ static int __cmd_report(void)signal (SIGINT , sig_handler); - session =perf_session__new (input_name,O_RDONLY , force, false, &event_ops); + session =perf_session__new (input_name,PERF_DATA_READ , force, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
@@ -514,7 +515,7 @@ int cmd_report(int argc, const char **argv, const char *prefix __used)if (inverted_callchain) callchain_param.order =ORDER_CALLER ; -if (strcmp (input_name,"-" ) !=0 ) +if (input_name &&strcmp (input_name,"-" ) !=0 )setup_browser (true);else use_browser =0 ;
tools/perf/builtin-sched.c +4 -2
@@ -1640,8 +1640,10 @@ static struct perf_event_ops event_ops = {static void read_events (bool destroy,struct perf_session **psession) {int err = -EINVAL ; -struct perf_session *session =perf_session__new (input_name,O_RDONLY , -0 , false, &event_ops); +struct perf_session *session; + + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )die ("No Memory" );
tools/perf/builtin-script.c +2 -1
@@ -1261,7 +1261,8 @@ int cmd_script(int argc, const char **argv, const char *prefix __used)if (!script_name)setup_pager (); - session =perf_session__new (input_name,O_RDONLY ,0 , false, &event_ops); + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-timechart.c +4 -3
@@ -984,10 +984,11 @@ static struct perf_event_ops event_ops = {static int __cmd_timechart (void ) { -struct perf_session *session =perf_session__new (input_name,O_RDONLY , -0 , false, &event_ops); -int ret = -EINVAL ; +struct perf_session *session; +int ret =EINVAL ; + session =perf_session__new (input_name,PERF_DATA_READ , false, + false, &event_ops);if (session ==NULL )return -ENOMEM ;
tools/perf/builtin-top.c +3 -9
@@ -954,11 +954,9 @@ static int __cmd_top(void) {pthread_t thread;int ret; -/* -* FIXME: perf_session__new should allow passing a O_MMAP, so that all this -* mmap reading, etc is encapsulated in it. Use O_WRONLY for now. -*/ - top.session =perf_session__new (NULL ,O_WRONLY , false, false,NULL ); + + top.session =perf_session__new (NULL ,PERF_DATA_NONE , false, + false,NULL );if (top.session ==NULL )return -ENOMEM ;
@@ -1243,10 +1241,6 @@ int cmd_top(int argc, const char **argv, const char *prefix __used) pos->attr.sample_period = default_interval; } -if (perf_evlist__alloc_pollfd (top.evlist) <0 || -perf_evlist__alloc_mmap (top.evlist) <0 ) - goto out_free_fd; - top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node); symbol_conf.priv_size =sizeof (struct annotation );
tools/perf/util/data.c new file +199 -0
199 lines not on this page · raw message
tools/perf/util/data.h new file +65 -0
65 lines not on this page · raw message
tools/perf/util/header.c +18 -17
@@ -1123,7 +1123,7 @@ int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full) {struct header_print_data hd;struct perf_header *header = &session->header; -int fd = session->fd; +int fd =perf_session__fd (session); hd.fp = fp; hd.full = full;
@@ -1525,13 +1525,13 @@ int perf_header__write_pipe(int fd)int perf_session__write_header (struct perf_session *session,struct perf_evlist *evlist, -int fd,bool at_exit) +bool at_exit) {struct perf_file_header f_header;struct perf_file_attr f_attr;struct perf_header *header = &session->header;struct perf_evsel *attr, *pair =NULL ; -int err; +int err, fd =perf_session__fd (session);lseek (fd,sizeof (f_header),SEEK_SET );
@@ -1960,10 +1960,11 @@ static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,return 0 ; } -static int perf_header__read_pipe (struct perf_session *session,int fd) +static int perf_header__read_pipe (struct perf_session *session) {struct perf_header *header = &session->header;struct perf_pipe_file_header f_header; +int fd =perf_session__fd (session);if (perf_file_header__read_pipe (&f_header, header, fd, session->repipe) <0 ) {
@@ -1971,25 +1972,24 @@ static int perf_header__read_pipe(struct perf_session *session, int fd)return -EINVAL ; } - session->fd = fd; -return 0 ; } -int perf_session__read_header (struct perf_session *session,int fd) +int perf_session__read_header (struct perf_session *session) {struct perf_header *header = &session->header;struct perf_file_header f_header;struct perf_file_attr f_attr;u64 f_id;int nr_attrs, nr_ids, i, j; +int fd =perf_session__fd (session); session->evlist =perf_evlist__new (NULL ,NULL );if (session->evlist ==NULL )return -ENOMEM ; -if (session->fd_pipe) -return perf_header__read_pipe (session, fd); +if (perf_session__is_pipe (session)) +return perf_header__read_pipe (session);if (perf_file_header__read (&f_header, header, fd) <0 ) {pr_debug ("incompatible file format\n" );
@@ -2213,14 +2213,14 @@ int perf_event__process_event_type(union perf_event *event,return 0 ; } -int perf_event__synthesize_tracing_data (int fd,struct perf_evlist *evlist, -perf_event__handler_t process, - struct perf_session *session __unused) +int perf_event__synthesize_tracing_data (struct perf_evlist *evlist, +perf_event__handler_t process, +struct perf_session *session) {union perf_event ev;struct tracing_data *tdata;ssize_t size =0 , aligned_size =0 , padding; -int err__used =0 ; +int fd =perf_session__fd (session);/* * We are going to store the size of the data followed
@@ -2263,18 +2263,19 @@ int perf_event__process_tracing_data(union perf_event *event, struct perf_session *session) {ssize_t size_read, padding, size = event->tracing_data.size; -off_t offset =lseek (session->fd,0 ,SEEK_CUR ); +int fd =perf_session__fd (session); +off_t offset =lseek (fd,0 ,SEEK_CUR );char buf[BUFSIZ ];/* setup for reading amidst mmap */ -lseek (session->fd, offset +sizeof (struct tracing_data_event ), +lseek (fd, offset +sizeof (struct tracing_data_event ),SEEK_SET ); - size_read =trace_report (session->fd, session->repipe); + size_read =trace_report (fd, session->repipe); padding =ALIGN (size_read,sizeof (u64)) - size_read; -if (read (session->fd, buf, padding) <0 ) +if (read (fd, buf, padding) <0 )die ("reading input file" );if (session->repipe) {int retw =write (STDOUT_FILENO , buf, padding);
tools/perf/util/header.h +3 -3
@@ -69,10 +69,10 @@ struct perf_header {struct perf_evlist ; -int perf_session__read_header (struct perf_session *session,int fd); +int perf_session__read_header (struct perf_session *session);int perf_session__write_header (struct perf_session *session,struct perf_evlist *evlist, -int fd,bool at_exit); +bool at_exit);int perf_header__write_pipe (int fd);int perf_header__push_event (u64 id,const char *name);
@@ -111,7 +111,7 @@ int perf_event__synthesize_event_types(perf_event__handler_t process,int perf_event__process_event_type (union perf_event *event,struct perf_session *session); -int perf_event__synthesize_tracing_data (int fd,struct perf_evlist *evlist, +int perf_event__synthesize_tracing_data (struct perf_evlist *evlist,perf_event__handler_t process,struct perf_session *session); int perf_event__process_tracing_data(union perf_event *event,
tools/perf/util/session.c +41 -74
227 lines not on this page · raw message
tools/perf/util/session.h +19 -7
@@ -6,6 +6,7 @@#include "header.h" #include "symbol.h" #include "thread.h" +#include "data.h" #include <linux/rbtree.h> #include "../../../include/linux/perf_event.h"
@@ -28,7 +29,7 @@ struct ordered_samples {struct perf_session {struct perf_header header; -unsigned long size; +struct perf_data data;unsigned long mmap_window;struct rb_root threads;struct list_head dead_threads;
@@ -45,8 +46,6 @@ struct perf_session {struct hists hists;u64 sample_type;int sample_size; -int fd; -bool fd_pipe;bool repipe;bool sample_id_all;u16 id_hdr_size;
@@ -54,7 +53,6 @@ struct perf_session {char *cwd;struct ordered_samples ordered_samples;struct callchain_cursor callchain_cursor; -char filename[0 ]; };struct perf_evsel ;
@@ -88,9 +86,8 @@ struct perf_event_ops {bool ordering_requires_timestamps; }; -struct perf_session *perf_session__new (const char *filename,int mode, -bool force,bool repipe, -struct perf_event_ops *ops); +struct perf_session *perf_session__new (const char *name,int mode,bool force, +bool repipe,struct perf_event_ops *ops);void perf_session__delete (struct perf_session *self);void perf_event_header__bswap (struct perf_event_header *self);
@@ -179,4 +176,19 @@ int perf_session__cpu_bitmap(struct perf_session *session, const char *cpu_list, unsigned long *cpu_bitmap);void perf_session__fprintf_info (struct perf_session *s,FILE *fp,bool full); + +static inline bool perf_session__is_new (struct perf_session *self) +{ +return perf_data__is_new (&self->data); +} + +static inline int perf_session__fd (struct perf_session *self) +{ +return perf_data__fd (&self->data); +} + +static inline int perf_session__is_pipe (struct perf_session *self) +{ +return perf_data__is_pipe (&self->data); +}#endif /* __PERF_SESSION_H */
Signature · 1 line
1.7.6.4
PATCH v2 5/5 perf tool: Putting mmap support to perf_data object
In reply to Jiri OlsaAdding event memory maps management under the perf_data object. The objective is to have this functionality centralized and dont mix this functionality into event list.
New function 'perf_data__mmap' is added into perf_data. It prepares allocates/prepares both perf_mmap and pollfd structures, and does the actuall mmap-ing.
Each perf_mmap now contains perf_data_file structure, to have memory map connected directly with the storage.
--- tools/perf/builtin-record.c | 29 +++++-- tools/perf/builtin-test.c | 14 +++- tools/perf/builtin-top.c | 26 ++++-- tools/perf/util/data.c | 107 ++++++++++++++++++++++++ tools/perf/util/data.h | 15 ++++ tools/perf/util/evlist.c | 188 ++----------------------------------------- tools/perf/util/evlist.h | 10 +-- tools/perf/util/evsel.h | 4 +- tools/perf/util/mmap.h | 3 + tools/perf/util/python.c | 23 ++++-- tools/perf/util/session.h | 6 ++ 11 files changed, 205 insertions(+), 220 deletions(-)
tools/perf/builtin-record.c +20 -9
@@ -318,8 +318,9 @@ try_again:exit (-1 ); } -if (perf_evlist__mmap (evlist, mmap_pages, false) <0 ) -die ("failed to mmap with %d (%s)\n" , errno,strerror (errno)); +if (perf_evlist__init_ids (evlist)) +die ("failed to init sample IDs %d (%s)\n" , + errno,strerror (errno));if (perf_session__is_new (session)) session->evlist = evlist;
@@ -402,18 +403,18 @@ static struct perf_event_header finished_round_event = { .type = PERF_RECORD_FINISHED_ROUND, }; -static void mmap_read (struct perf_mmap *m__used ,void *buf,unsigned long size) +static void mmap_read (struct perf_mmap *m,void *buf,unsigned long size) { -int fd =perf_session__fd (session); +int fd =perf_data__fd_file (m->file);write_output (fd, buf, size); } -static int mmap_read_all (void ) +static int mmap_read_all (struct perf_data *data) {int i, ret =0 ; -for (i =0 ; i < evsel_list->nr_mmaps; i++) { -struct perf_mmap *m = &evsel_list->mmap[i]; +for (i =0 ; i < data->nr_mmaps; i++) { +struct perf_mmap *m = &data->mmaps[i]; ret +=perf_mmap__process (m, mmap_read); }
@@ -540,6 +541,13 @@ static int __cmd_record(int argc, const char **argv)open_counters (evsel_list); +if (perf_data__mmap (perf_session__data (session), + evsel_list, mmap_pages, false) <0 ) { +pr_err ("Failed to mmap with %d (%s)\n" , + errno,strerror (errno)); + goto out_delete_session; + } +/* * perf_session__delete(session) will be called at atexit_header() */
@@ -642,10 +650,13 @@ static int __cmd_record(int argc, const char **argv)close (go_pipe[1 ]);for (;;) { -if (!mmap_read_all ()) { +struct perf_data *data =perf_session__data (session); + +if (!mmap_read_all (data)) { +if (done)break ; - err =poll (evsel_list->pollfd, evsel_list->nr_fds,-1 ); + err =poll (data->fds, data->nr_fds,-1 ); waking++; }
tools/perf/builtin-test.c +11 -3
@@ -12,6 +12,7 @@#include "util/parse-events.h" #include "util/symbol.h" #include "util/thread_map.h" +#include "util/data.h" #include "../../include/linux/hw_breakpoint.h" static long page_size;
@@ -476,6 +477,9 @@ static int test__basic_mmap(void) expected_nr_events[nsyscalls], i, j;struct perf_evsel *evsels[nsyscalls], *evsel;int sample_size =__perf_evsel__sample_size (attr.sample_type); +struct perf_data data; + +PERF_DATA__NONE (data);for (i =0 ; i < nsyscalls; ++i) {char name[64 ];
@@ -539,7 +543,11 @@ static int test__basic_mmap(void) } } -if (perf_evlist__mmap (evlist,128 , true) <0 ) { + err =perf_evlist__init_ids (evlist); +if (err) + goto out_close_fd; + +if (perf_data__mmap (&data, evlist,128 , true)) {pr_debug ("failed to mmap events: %d (%s)\n" , errno,strerror (errno)); goto out_close_fd;
@@ -551,7 +559,7 @@ static int test__basic_mmap(void) ++foo; } -while ((event =perf_evlist__mmap_read (evlist,0 )) !=NULL ) { +while ((event =perf_data__mmap_read (&data,0 )) !=NULL ) {struct perf_sample sample;if (event->header.type !=PERF_RECORD_SAMPLE ) {
@@ -587,7 +595,7 @@ static int test__basic_mmap(void) err =0 ; out_munmap: -perf_evlist__munmap (evlist); +perf_data__close (&data); out_close_fd:for (i =0 ; i < nsyscalls; ++i)perf_evsel__close_fd (evsels[i],1 , threads->nr);
tools/perf/builtin-top.c +17 -9
@@ -804,14 +804,15 @@ static void perf_event__process_sample(const union perf_event *event,return ; } -static void perf_session__mmap_read_idx (struct perf_session *self,int idx) +static void perf_session__mmap_read_idx (struct perf_session *self, +struct perf_data *data,int idx) {struct perf_sample sample;struct perf_evsel *evsel;union perf_event *event;int ret; -while ((event =perf_evlist__mmap_read (top.evlist, idx)) !=NULL ) { +while ((event =perf_data__mmap_read (data, idx)) !=NULL ) { ret =perf_session__parse_sample (self, event, &sample);if (ret) {pr_err ("Can't parse sample, err = %d\n" , ret);
@@ -833,13 +834,14 @@ static void perf_session__mmap_read_idx(struct perf_session *self, int idx)static void perf_session__mmap_read (struct perf_session *self) { +struct perf_data *data =perf_session__data (self);int i; -for (i =0 ; i < top.evlist->nr_mmaps; i++) -perf_session__mmap_read_idx (self, i); +for (i =0 ; i < data->nr_mmaps; i++) +perf_session__mmap_read_idx (self, data, i); } -static void start_counters (struct perf_evlist *evlist) +static void start_counters (struct perf_data *data,struct perf_evlist *evlist) {struct perf_evsel *counter, *first;
@@ -920,7 +922,10 @@ try_again: } } -if (perf_evlist__mmap (evlist, mmap_pages, false) <0 ) { +if (perf_evlist__init_ids (evlist)) + goto out_err; + +if (perf_data__mmap (data, evlist, mmap_pages, false) <0 ) {ui__warning ("Failed to mmap with %d (%s)\n" , errno,strerror (errno)); goto out_err;
@@ -952,6 +957,7 @@ static int setup_sample_type(void)static int __cmd_top (void ) { +struct perf_data *data;pthread_t thread;int ret;
@@ -970,12 +976,14 @@ static int __cmd_top(void)else perf_event__synthesize_threads (perf_event__process, top.session); -start_counters (top.evlist); + data =perf_session__data (top.session); + +start_counters (data, top.evlist); top.session->evlist = top.evlist;perf_session__update_sample_type (top.session);/* Wait for a minimal set of events before starting the snapshot */ -poll (top.evlist->pollfd, top.evlist->nr_fds,100 ); +poll (data->fds, data->nr_fds,100 );perf_session__mmap_read (top.session);
@@ -1001,7 +1009,7 @@ static int __cmd_top(void)perf_session__mmap_read (top.session);if (hits == top.samples) - ret =poll (top.evlist->pollfd, top.evlist->nr_fds,100 ); + ret =poll (data->fds, data->nr_fds,100 ); } out_delete:
tools/perf/util/data.c +107 -0
110 lines not on this page · raw message
tools/perf/util/data.h +15 -0
@@ -28,12 +28,22 @@ struct perf_data {int mode;bool force; +struct perf_mmap *mmaps; +struct pollfd *fds; +int nr_mmaps; +int nr_fds; +struct list_head files; };int perf_data__open (struct perf_data *data,const char *name,int mode,bool force);void perf_data__close (struct perf_data *data); +int perf_data__mmap (struct perf_data *data, +struct perf_evlist *evlist, +int pages,bool overwrite); +union perf_event *perf_data__mmap_read (struct perf_data *data, +int idx);#define PERF_DATA__NONE (__data) \ perf_data__open(&__data, NULL, PERF_DATA_NONE, false);
@@ -62,4 +72,9 @@ static inline int perf_data__fd(struct perf_data *data) {return data->header.fd; } + +static inline int perf_data__fd_file (struct perf_data_file *file) +{ +return file->fd; +}#endif /* __PERF_DATA_H */
tools/perf/util/evlist.c +7 -181
243 lines not on this page · raw message
tools/perf/util/evlist.h +1 -9
@@ -16,13 +16,6 @@ struct perf_evlist {struct list_head entries;struct hlist_head heads[PERF_EVLIST__HLIST_SIZE ];int nr_entries; -int nr_fds; -int nr_mmaps; -int pages; -bool overwrite; -union perf_event event_copy; -struct perf_mmap *mmap; -struct pollfd *pollfd;struct thread_map *threads;struct cpu_map *cpus;struct perf_evsel *selected;
@@ -48,8 +41,6 @@ void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd);struct perf_evsel *perf_evlist__id2evsel (struct perf_evlist *evlist,u64 id); -union perf_event *perf_evlist__mmap_read (struct perf_evlist *self,int idx); -int perf_evlist__open (struct perf_evlist *evlist,bool group);int perf_evlist__alloc_mmap (struct perf_evlist *evlist);
@@ -74,6 +65,7 @@ int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, pid_t target_tid, const char *cpu_list);void perf_evlist__delete_maps (struct perf_evlist *evlist);int perf_evlist__set_filters (struct perf_evlist *evlist); +int perf_evlist__init_ids (struct perf_evlist *evlist);u64 perf_evlist__sample_type (const struct perf_evlist *evlist);bool perf_evlist__sample_id_all (const const struct perf_evlist *evlist);
tools/perf/util/evsel.h +3 -1
@@ -8,7 +8,9 @@#include "xyarray.h" #include "cgroup.h" #include "hist.h" - + +#define PERF_EVSEL_FD (e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) + struct perf_counts_values { union { struct {
tools/perf/util/mmap.h +3 -0
@@ -4,6 +4,7 @@#include <sys/mman.h> #include "event.h" #include "../perf.h" +#include "data.h" struct perf_mmap {void *base;
@@ -12,6 +13,8 @@ struct perf_mmap {int len;int fd;bool owrt; + +struct perf_data_file *file; }; typedef void (*perf_mmap_process_t)(struct perf_mmap *m,
tools/perf/util/python.c +15 -8
@@ -7,6 +7,7 @@#include "event.h" #include "cpumap.h" #include "thread_map.h" +#include "data.h" /* Define PyVarObject_HEAD_INIT for python 2.5 */ #ifndef PyVarObject_HEAD_INIT
@@ -598,7 +599,6 @@ static int pyrf_evsel__init(struct pyrf_evsel *pevsel,static void pyrf_evsel__delete (struct pyrf_evsel *pevsel) { -perf_evsel__exit (&pevsel->evsel); pevsel->ob_type->tp_free ((PyObject *)pevsel); }
@@ -668,6 +668,7 @@ static int pyrf_evsel__setup_types(void)struct pyrf_evlist {PyObject_HEAD + struct perf_data data;struct perf_evlist evlist; };
@@ -684,12 +685,12 @@ static int pyrf_evlist__init(struct pyrf_evlist *pevlist, threads = ((struct pyrf_thread_map *)pthreads)->threads; cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;perf_evlist__init (&pevlist->evlist, cpus, threads); +PERF_DATA__NONE (pevlist->data);return 0 ; }static void pyrf_evlist__delete (struct pyrf_evlist *pevlist) { -perf_evlist__exit (&pevlist->evlist); pevlist->ob_type->tp_free ((PyObject *)pevlist); }
@@ -697,6 +698,7 @@ static PyObject *pyrf_evlist__mmap(struct pyrf_evlist *pevlist, PyObject *args, PyObject *kwargs) {struct perf_evlist *evlist = &pevlist->evlist; +struct perf_data *data = &pevlist->data;static char *kwlist[] = {"pages" ,"overwrite" ,NULL };int pages =128 , overwrite = false;
@@ -704,7 +706,12 @@ static PyObject *pyrf_evlist__mmap(struct pyrf_evlist *pevlist, &pages, &overwrite))return NULL ; -if (perf_evlist__mmap (evlist, pages, overwrite) <0 ) { +if (perf_evlist__init_ids (evlist)) { +PyErr_SetFromErrno (PyExc_OSError); +return NULL ; + } + +if (perf_data__mmap (data, evlist, pages, overwrite) <0 ) {PyErr_SetFromErrno (PyExc_OSError);return NULL ; }
@@ -716,14 +723,14 @@ static PyObject *pyrf_evlist__mmap(struct pyrf_evlist *pevlist,static PyObject *pyrf_evlist__poll (struct pyrf_evlist *pevlist,PyObject *args,PyObject *kwargs) { -struct perf_evlist *evlist = &pevlist->evlist; +struct perf_data *data = &pevlist->data;static char *kwlist[] = {"timeout" ,NULL };int timeout =-1 , n;if (!PyArg_ParseTupleAndKeywords (args, kwargs,"|i" , kwlist, &timeout))return NULL ; - n =poll (evlist->pollfd, evlist->nr_fds, timeout); + n =poll (data->fds, data->nr_fds, timeout);if (n <0 ) {PyErr_SetFromErrno (PyExc_OSError);return NULL ;
@@ -735,13 +742,13 @@ static PyObject *pyrf_evlist__poll(struct pyrf_evlist *pevlist,static PyObject *pyrf_evlist__get_pollfd (struct pyrf_evlist *pevlist,PyObject *args__used ,PyObject *kwargs__used ) { -struct perf_evlist *evlist = &pevlist->evlist; +struct perf_data *data = &pevlist->data;PyObject *list =PyList_New (0 );int i; -for (i =0 ; i < evlist->nr_fds; ++i) { +for (i =0 ; i < data->nr_fds; ++i) {PyObject *file; -FILE *fp =fdopen (evlist->pollfd[i].fd,"r" ); +FILE *fp =fdopen (data->fds[i].fd,"r" );if (fp ==NULL ) goto free_list;
tools/perf/util/session.h +6 -0
@@ -191,4 +191,10 @@ static inline int perf_session__is_pipe(struct perf_session *self) {return perf_data__is_pipe (&self->data); } + +static inline struct perf_data *perf_session__data (struct perf_session *self) +{ +return &self->data; +} +#endif /* __PERF_SESSION_H */
Signature · 1 line
1.7.6.4