Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > linux.kernel > #1340903 > unrolled thread
| Started by | Arnaldo Carvalho de Melo <acme@kernel.org> |
|---|---|
| First post | 2016-02-23 18:50 +0100 |
| Last post | 2016-02-24 14:40 +0100 |
| Articles | 6 — 2 participants |
Back to article view | Back to linux.kernel
This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by
below is the oldest one visible, not the original post.
Re: [PATCH 10/48] perf tools: Introduce bpf-output event Arnaldo Carvalho de Melo <acme@kernel.org> - 2016-02-23 18:50 +0100
Re: [PATCH 10/48] perf tools: Introduce bpf-output event "Wangnan (F)" <wangnan0@huawei.com> - 2016-02-24 03:10 +0100
Re: [PATCH 10/48] perf tools: Introduce bpf-output event "Wangnan (F)" <wangnan0@huawei.com> - 2016-02-24 05:10 +0100
Re: [PATCH 10/48] perf tools: Introduce bpf-output event "Wangnan (F)" <wangnan0@huawei.com> - 2016-02-24 06:10 +0100
Re: [PATCH 10/48] perf tools: Introduce bpf-output event "Wangnan (F)" <wangnan0@huawei.com> - 2016-02-24 03:10 +0100
Re: [PATCH 10/48] perf tools: Introduce bpf-output event Arnaldo Carvalho de Melo <acme@kernel.org> - 2016-02-24 14:40 +0100
| From | Arnaldo Carvalho de Melo <acme@kernel.org> |
|---|---|
| Date | 2016-02-23 18:50 +0100 |
| Subject | Re: [PATCH 10/48] perf tools: Introduce bpf-output event |
| Message-ID | <r5pfX-fN-5@gated-at.bofh.it> |
Em Mon, Feb 22, 2016 at 09:10:37AM +0000, Wang Nan escreveu:
> Commit a43eec304259a6c637f4014a6d4767159b6a3aa3 (bpf: introduce
> bpf_perf_event_output() helper) add a helper to enable BPF program
> output data to perf ring buffer through a new type of perf event
> PERF_COUNT_SW_BPF_OUTPUT. This patch enable perf to create perf
> event of that type. Now perf user can use following cmdline to
> receive output data from BPF programs:
>
> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
> # ./perf script
> perf 1560 [004] 347747.086295: evt: ffffffff811fd201 sys_write ...
> perf 1560 [004] 347747.086300: evt: ffffffff811fd201 sys_write ...
> perf 1560 [004] 347747.086315: evt: ffffffff811fd201 sys_write ...
> ...
>
> Test result:
> # cat ./test_bpf_output.c
> /************************ BEGIN **************************/
> #include <uapi/linux/bpf.h>
> struct bpf_map_def {
> unsigned int type;
> unsigned int key_size;
> unsigned int value_size;
> unsigned int max_entries;
> };
>
> #define SEC(NAME) __attribute__((section(NAME), used))
> static u64 (*ktime_get_ns)(void) =
> (void *)BPF_FUNC_ktime_get_ns;
> static int (*trace_printk)(const char *fmt, int fmt_size, ...) =
> (void *)BPF_FUNC_trace_printk;
> static int (*get_smp_processor_id)(void) =
> (void *)BPF_FUNC_get_smp_processor_id;
> static int (*perf_event_output)(void *, struct bpf_map_def *, int, void *, unsigned long) =
> (void *)BPF_FUNC_perf_event_output;
>
> struct bpf_map_def SEC("maps") channel = {
> .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
> .key_size = sizeof(int),
> .value_size = sizeof(u32),
> .max_entries = __NR_CPUS__,
> };
>
> SEC("func_write=sys_write")
> int func_write(void *ctx)
> {
> struct {
> u64 ktime;
> int cpuid;
> } __attribute__((packed)) output_data;
> char error_data[] = "Error: failed to output: %d\n";
>
> output_data.cpuid = get_smp_processor_id();
> output_data.ktime = ktime_get_ns();
> int err = perf_event_output(ctx, &channel, get_smp_processor_id(),
> &output_data, sizeof(output_data));
> if (err)
> trace_printk(error_data, sizeof(error_data), err);
> return 0;
> }
> char _license[] SEC("license") = "GPL";
> int _version SEC("version") = LINUX_VERSION_CODE;
> /************************ END ***************************/
>
> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
> # ./perf script | grep ls
> ls 2242 [003] 347851.557563: evt: ffffffff811fd201 sys_write ...
> ls 2242 [003] 347851.557571: evt: ffffffff811fd201 sys_write ...
So, there is something strange here:
if (unlikely(event->oncpu != smp_processor_id()))
return -EOPNOTSUPP;
This is where I am hitting, with:
[acme@jouet linux]$ uname -r
4.5.0-rc4
int err = perf_event_output(ctx, &channel, get_smp_processor_id(),
&output_data, sizeof(output_data));
if (err)
trace_printk(error_data, sizeof(error_data), err);
And then:
[root@jouet bpf]# tail /sys/kernel/debug/tracing/trace
perf-13040 [003] d... 12062.807729: : Error: failed to output: -95
perf-13040 [003] d... 12062.807731: : Error: failed to output: -95
perf-13040 [003] d... 12062.807732: : Error: failed to output: -95
perf-13040 [003] d... 12062.807735: : Error: failed to output: -95
perf-13040 [003] d... 12062.807737: : Error: failed to output: -95
perf-13040 [003] d... 12062.807744: : Error: failed to output: -95
gnome-terminal--3091 [001] d... 12062.807773: : Error: failed to output: -95
gnome-terminal--3091 [001] d... 12062.807784: : Error: failed to output: -95
gmain-2830 [002] d... 12062.811791: : Error: failed to output: -95
gmain-2830 [002] d... 12062.811810: : Error: failed to output: -95
[root@jouet bpf]#
Ideas? AFK for a while, will continue investigating.
This already was submitted to Ingo, BTW.
I used, as in the changeset comment tests:
perf record -a -e bpf-output/no-inherit,name=evt/ -e ./test_bpf_output.c/map:channel.event=evt/ ls /
And perf script told me:
[root@jouet bpf]# perf script | tail
perf 13040 [003] 12062.708337: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708339: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708340: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708341: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708343: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708344: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708346: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708347: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708348: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
perf 13040 [003] 12062.708350: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
[root@jouet bpf]#
Wonder where that /lib/modules/4.5.0-rc4/build/vmlinux came from...
[root@jouet bpf]# perf script | cut -d'(' -f2 | sort | uniq -c
1141 /lib/modules/4.5.0-rc4/build/vmlinux)
- Arnaldo
[toc] | [next] | [standalone]
| From | "Wangnan (F)" <wangnan0@huawei.com> |
|---|---|
| Date | 2016-02-24 03:10 +0100 |
| Message-ID | <r5x3Q-656-7@gated-at.bofh.it> |
| In reply to | #1340903 |
On 2016/2/24 9:58, Wangnan (F) wrote:
>
>
> On 2016/2/24 1:45, Arnaldo Carvalho de Melo wrote:
>> Em Mon, Feb 22, 2016 at 09:10:37AM +0000, Wang Nan escreveu:
>>> Commit a43eec304259a6c637f4014a6d4767159b6a3aa3 (bpf: introduce
>>> bpf_perf_event_output() helper) add a helper to enable BPF program
>>> output data to perf ring buffer through a new type of perf event
>>> PERF_COUNT_SW_BPF_OUTPUT. This patch enable perf to create perf
>>> event of that type. Now perf user can use following cmdline to
>>> receive output data from BPF programs:
>>>
>>> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
>>> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
>>> # ./perf script
>>> perf 1560 [004] 347747.086295:
>>> evt: ffffffff811fd201 sys_write ...
>>> perf 1560 [004] 347747.086300:
>>> evt: ffffffff811fd201 sys_write ...
>>> perf 1560 [004] 347747.086315:
>>> evt: ffffffff811fd201 sys_write ...
>>> ...
>>>
>>> Test result:
>>> # cat ./test_bpf_output.c
>>> /************************ BEGIN **************************/
>>> #include <uapi/linux/bpf.h>
>>> struct bpf_map_def {
>>> unsigned int type;
>>> unsigned int key_size;
>>> unsigned int value_size;
>>> unsigned int max_entries;
>>> };
>>>
>>> #define SEC(NAME) __attribute__((section(NAME), used))
>>> static u64 (*ktime_get_ns)(void) =
>>> (void *)BPF_FUNC_ktime_get_ns;
>>> static int (*trace_printk)(const char *fmt, int fmt_size, ...) =
>>> (void *)BPF_FUNC_trace_printk;
>>> static int (*get_smp_processor_id)(void) =
>>> (void *)BPF_FUNC_get_smp_processor_id;
>>> static int (*perf_event_output)(void *, struct bpf_map_def *, int,
>>> void *, unsigned long) =
>>> (void *)BPF_FUNC_perf_event_output;
>>>
>>> struct bpf_map_def SEC("maps") channel = {
>>> .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
>>> .key_size = sizeof(int),
>>> .value_size = sizeof(u32),
>>> .max_entries = __NR_CPUS__,
>>> };
>>>
>>> SEC("func_write=sys_write")
>>> int func_write(void *ctx)
>>> {
>>> struct {
>>> u64 ktime;
>>> int cpuid;
>>> } __attribute__((packed)) output_data;
>>> char error_data[] = "Error: failed to output: %d\n";
>>>
>>> output_data.cpuid = get_smp_processor_id();
>>> output_data.ktime = ktime_get_ns();
>>> int err = perf_event_output(ctx, &channel,
>>> get_smp_processor_id(),
>>> &output_data, sizeof(output_data));
>>> if (err)
>>> trace_printk(error_data, sizeof(error_data), err);
>>> return 0;
>>> }
>>> char _license[] SEC("license") = "GPL";
>>> int _version SEC("version") = LINUX_VERSION_CODE;
>>> /************************ END ***************************/
>>>
>>> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
>>> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
>>> # ./perf script | grep ls
>>> ls 2242 [003] 347851.557563: evt: ffffffff811fd201
>>> sys_write ...
>>> ls 2242 [003] 347851.557571: evt: ffffffff811fd201
>>> sys_write ...
>> So, there is something strange here:
>>
>> if (unlikely(event->oncpu != smp_processor_id()))
>> return -EOPNOTSUPP;
>>
>
All failures have 'event->oncpu == -1' here. I guess we should suppress
warning in
this case. But why event->oncpu becomes -1?
Thank you.
[toc] | [prev] | [next] | [standalone]
| From | "Wangnan (F)" <wangnan0@huawei.com> |
|---|---|
| Date | 2016-02-24 05:10 +0100 |
| Message-ID | <r5yVZ-7pJ-39@gated-at.bofh.it> |
| In reply to | #1341237 |
On 2016/2/24 10:04, Wangnan (F) wrote:
>
>
> On 2016/2/24 9:58, Wangnan (F) wrote:
>>
>>
>> On 2016/2/24 1:45, Arnaldo Carvalho de Melo wrote:
>>> Em Mon, Feb 22, 2016 at 09:10:37AM +0000, Wang Nan escreveu:
>>>> Commit a43eec304259a6c637f4014a6d4767159b6a3aa3 (bpf: introduce
>>>> bpf_perf_event_output() helper) add a helper to enable BPF program
>>>> output data to perf ring buffer through a new type of perf event
>>>> PERF_COUNT_SW_BPF_OUTPUT. This patch enable perf to create perf
>>>> event of that type. Now perf user can use following cmdline to
>>>> receive output data from BPF programs:
>>>>
>>>> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
>>>> -e ./test_bpf_output.c/map:channel.event=evt/
>>>> ls /
>>>> # ./perf script
>>>> perf 1560 [004] 347747.086295:
>>>> evt: ffffffff811fd201 sys_write ...
>>>> perf 1560 [004] 347747.086300:
>>>> evt: ffffffff811fd201 sys_write ...
>>>> perf 1560 [004] 347747.086315:
>>>> evt: ffffffff811fd201 sys_write ...
>>>> ...
>>>>
>>>> Test result:
>>>> # cat ./test_bpf_output.c
>>>> /************************ BEGIN **************************/
>>>> #include <uapi/linux/bpf.h>
>>>> struct bpf_map_def {
>>>> unsigned int type;
>>>> unsigned int key_size;
>>>> unsigned int value_size;
>>>> unsigned int max_entries;
>>>> };
>>>>
>>>> #define SEC(NAME) __attribute__((section(NAME), used))
>>>> static u64 (*ktime_get_ns)(void) =
>>>> (void *)BPF_FUNC_ktime_get_ns;
>>>> static int (*trace_printk)(const char *fmt, int fmt_size, ...) =
>>>> (void *)BPF_FUNC_trace_printk;
>>>> static int (*get_smp_processor_id)(void) =
>>>> (void *)BPF_FUNC_get_smp_processor_id;
>>>> static int (*perf_event_output)(void *, struct bpf_map_def *,
>>>> int, void *, unsigned long) =
>>>> (void *)BPF_FUNC_perf_event_output;
>>>>
>>>> struct bpf_map_def SEC("maps") channel = {
>>>> .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
>>>> .key_size = sizeof(int),
>>>> .value_size = sizeof(u32),
>>>> .max_entries = __NR_CPUS__,
>>>> };
>>>>
>>>> SEC("func_write=sys_write")
>>>> int func_write(void *ctx)
>>>> {
>>>> struct {
>>>> u64 ktime;
>>>> int cpuid;
>>>> } __attribute__((packed)) output_data;
>>>> char error_data[] = "Error: failed to output: %d\n";
>>>>
>>>> output_data.cpuid = get_smp_processor_id();
>>>> output_data.ktime = ktime_get_ns(); supr
>>>> int err = perf_event_output(ctx, &channel,
>>>> get_smp_processor_id(),
>>>> &output_data, sizeof(output_data));
>>>> if (err)
>>>> trace_printk(error_data, sizeof(error_data), err);
>>>> return 0;
>>>> }
>>>> char _license[] SEC("license") = "GPL";
>>>> int _version SEC("version") = LINUX_VERSION_CODE;
>>>> /************************ END ***************************/
>>>>
>>>> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
>>>> -e ./test_bpf_output.c/map:channel.event=evt/
>>>> ls /
>>>> # ./perf script | grep ls
>>>> ls 2242 [003] 347851.557563: evt: ffffffff811fd201
>>>> sys_write ...
>>>> ls 2242 [003] 347851.557571: evt: ffffffff811fd201
>>>> sys_write ...
>>> So, there is something strange here:
>>>
>>> if (unlikely(event->oncpu != smp_processor_id()))
>>> return -EOPNOTSUPP;
>>>
>>
>
> All failures have 'event->oncpu == -1' here. I guess we should
> suppress warning in
> this case. But why event->oncpu becomes -1?
>
For this specific test it is not surprising to see these error messages.
In this test
we create bpf-output channel on 'ls' process only, but the BPF script is
triggered
on all procs (BPF triggering is not related to perf event scheduling).
Trying to
output data through 'ls' specific bpf-output channel should fail if this
'sys_write'
is not issued by 'ls' or its children. So it is a correct behavior.
However, I also see them in system wide channel:
# echo "" > /sys/kernel/debug/tracing/trace
# ./perf record -a -e bpf-output/no-inherit,name=evt/ \
-e ./test_bpf_output.c/map:channel.event=evt/
-a
^C[ perf record: Woken up 0 times to write data ]
[ perf record: Captured and wrote 17.534 MB perf.data (264326 samples) ]
# cat /sys/kernel/debug/tracing/trace | tail
rs:main Q:Reg-582 [000] d..2 4858.711225: : Error: failed to
output: -95
rs:main Q:Reg-582 [000] d..2 4858.711241: : Error: failed to
output: -95
gmain-1858 [003] d..2 4858.711436: : Error: failed to
output: -95
gmain-1858 [003] d..2 4858.711441: : Error: failed to
output: -95
gmain-1858 [003] d..2 4858.711473: : Error: failed to
output: -95
rs:main Q:Reg-582 [002] d..2 4858.712215: : Error: failed to
output: -95
rs:main Q:Reg-582 [002] d..2 4858.712224: : Error: failed to
output: -95
gmain-1858 [003] d..2 4858.712230: : Error: failed to
output: -95
rs:main Q:Reg-582 [002] d..2 4858.712235: : Error: failed to
output: -95
rs:main Q:Reg-582 [002] d..2 4858.712239: : Error: failed to
output: -95
System wide events can also be scheduled in and out. If the bpf-output
events
are scheduled out, trying to output data through it causes the above
failure.
I don't think it is a problem.
Peter, Could you please give some infomation? In which case a system wide
bpf output channel would be scheduled out?
Thank you.
[toc] | [prev] | [next] | [standalone]
| From | "Wangnan (F)" <wangnan0@huawei.com> |
|---|---|
| Date | 2016-02-24 06:10 +0100 |
| Message-ID | <r5zS2-84d-15@gated-at.bofh.it> |
| In reply to | #1341315 |
On 2016/2/24 12:03, Wangnan (F) wrote: > > > On 2016/2/24 10:04, Wangnan (F) wrote: >> >> >> On 2016/2/24 9:58, Wangnan (F) wrote: >>> [SNIP] >>> So, there is something strange here: >>>> >>>> if (unlikely(event->oncpu != smp_processor_id())) >>>> return -EOPNOTSUPP; >>>> >>> >> >> All failures have 'event->oncpu == -1' here. I guess we should >> suppress warning in >> this case. But why event->oncpu becomes -1? >> > > For this specific test it is not surprising to see these error > messages. In this test > we create bpf-output channel on 'ls' process only, but the BPF script > is triggered > on all procs (BPF triggering is not related to perf event scheduling). > Trying to > output data through 'ls' specific bpf-output channel should fail if > this 'sys_write' > is not issued by 'ls' or its children. So it is a correct behavior. > > However, I also see them in system wide channel: > > # echo "" > /sys/kernel/debug/tracing/trace > # ./perf record -a -e bpf-output/no-inherit,name=evt/ \ > -e ./test_bpf_output.c/map:channel.event=evt/ > -a > ^C[ perf record: Woken up 0 times to write data ] > [ perf record: Captured and wrote 17.534 MB perf.data (264326 samples) ] > # cat /sys/kernel/debug/tracing/trace | tail > rs:main Q:Reg-582 [000] d..2 4858.711225: : Error: failed to > output: -95 > rs:main Q:Reg-582 [000] d..2 4858.711241: : Error: failed to > output: -95 > gmain-1858 [003] d..2 4858.711436: : Error: failed to > output: -95 > gmain-1858 [003] d..2 4858.711441: : Error: failed to > output: -95 > gmain-1858 [003] d..2 4858.711473: : Error: failed to > output: -95 > rs:main Q:Reg-582 [002] d..2 4858.712215: : Error: failed to > output: -95 > rs:main Q:Reg-582 [002] d..2 4858.712224: : Error: failed to > output: -95 > gmain-1858 [003] d..2 4858.712230: : Error: failed to > output: -95 > rs:main Q:Reg-582 [002] d..2 4858.712235: : Error: failed to > output: -95 > rs:main Q:Reg-582 [002] d..2 4858.712239: : Error: failed to > output: -95 > > System wide events can also be scheduled in and out. If the bpf-output > events > are scheduled out, trying to output data through it causes the above > failure. > I don't think it is a problem. > > Peter, Could you please give some infomation? In which case a system wide > bpf output channel would be scheduled out? > Sorry, I think my brain is not quite well. Actually this is a easy question: all ENOTSUPP results are generated before PERF_EVENT_IOC_ENABLE or after PERF_EVENT_IOC_DISABLE. You saw so many failure messages because it is sys_write, and perf itself needs it. So you can simply ignore these messages. Thank you.
[toc] | [prev] | [next] | [standalone]
| From | "Wangnan (F)" <wangnan0@huawei.com> |
|---|---|
| Date | 2016-02-24 03:10 +0100 |
| Message-ID | <r5x3Q-656-9@gated-at.bofh.it> |
| In reply to | #1340903 |
On 2016/2/24 1:45, Arnaldo Carvalho de Melo wrote:
> Em Mon, Feb 22, 2016 at 09:10:37AM +0000, Wang Nan escreveu:
>> Commit a43eec304259a6c637f4014a6d4767159b6a3aa3 (bpf: introduce
>> bpf_perf_event_output() helper) add a helper to enable BPF program
>> output data to perf ring buffer through a new type of perf event
>> PERF_COUNT_SW_BPF_OUTPUT. This patch enable perf to create perf
>> event of that type. Now perf user can use following cmdline to
>> receive output data from BPF programs:
>>
>> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
>> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
>> # ./perf script
>> perf 1560 [004] 347747.086295: evt: ffffffff811fd201 sys_write ...
>> perf 1560 [004] 347747.086300: evt: ffffffff811fd201 sys_write ...
>> perf 1560 [004] 347747.086315: evt: ffffffff811fd201 sys_write ...
>> ...
>>
>> Test result:
>> # cat ./test_bpf_output.c
>> /************************ BEGIN **************************/
>> #include <uapi/linux/bpf.h>
>> struct bpf_map_def {
>> unsigned int type;
>> unsigned int key_size;
>> unsigned int value_size;
>> unsigned int max_entries;
>> };
>>
>> #define SEC(NAME) __attribute__((section(NAME), used))
>> static u64 (*ktime_get_ns)(void) =
>> (void *)BPF_FUNC_ktime_get_ns;
>> static int (*trace_printk)(const char *fmt, int fmt_size, ...) =
>> (void *)BPF_FUNC_trace_printk;
>> static int (*get_smp_processor_id)(void) =
>> (void *)BPF_FUNC_get_smp_processor_id;
>> static int (*perf_event_output)(void *, struct bpf_map_def *, int, void *, unsigned long) =
>> (void *)BPF_FUNC_perf_event_output;
>>
>> struct bpf_map_def SEC("maps") channel = {
>> .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
>> .key_size = sizeof(int),
>> .value_size = sizeof(u32),
>> .max_entries = __NR_CPUS__,
>> };
>>
>> SEC("func_write=sys_write")
>> int func_write(void *ctx)
>> {
>> struct {
>> u64 ktime;
>> int cpuid;
>> } __attribute__((packed)) output_data;
>> char error_data[] = "Error: failed to output: %d\n";
>>
>> output_data.cpuid = get_smp_processor_id();
>> output_data.ktime = ktime_get_ns();
>> int err = perf_event_output(ctx, &channel, get_smp_processor_id(),
>> &output_data, sizeof(output_data));
>> if (err)
>> trace_printk(error_data, sizeof(error_data), err);
>> return 0;
>> }
>> char _license[] SEC("license") = "GPL";
>> int _version SEC("version") = LINUX_VERSION_CODE;
>> /************************ END ***************************/
>>
>> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
>> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
>> # ./perf script | grep ls
>> ls 2242 [003] 347851.557563: evt: ffffffff811fd201 sys_write ...
>> ls 2242 [003] 347851.557571: evt: ffffffff811fd201 sys_write ...
> So, there is something strange here:
>
> if (unlikely(event->oncpu != smp_processor_id()))
> return -EOPNOTSUPP;
>
> This is where I am hitting, with:
>
> [acme@jouet linux]$ uname -r
> 4.5.0-rc4
>
> int err = perf_event_output(ctx, &channel, get_smp_processor_id(),
> &output_data, sizeof(output_data));
> if (err)
> trace_printk(error_data, sizeof(error_data), err);
>
> And then:
>
> [root@jouet bpf]# tail /sys/kernel/debug/tracing/trace
> perf-13040 [003] d... 12062.807729: : Error: failed to output: -95
> perf-13040 [003] d... 12062.807731: : Error: failed to output: -95
> perf-13040 [003] d... 12062.807732: : Error: failed to output: -95
> perf-13040 [003] d... 12062.807735: : Error: failed to output: -95
> perf-13040 [003] d... 12062.807737: : Error: failed to output: -95
> perf-13040 [003] d... 12062.807744: : Error: failed to output: -95
> gnome-terminal--3091 [001] d... 12062.807773: : Error: failed to output: -95
> gnome-terminal--3091 [001] d... 12062.807784: : Error: failed to output: -95
> gmain-2830 [002] d... 12062.811791: : Error: failed to output: -95
> gmain-2830 [002] d... 12062.811810: : Error: failed to output: -95
> [root@jouet bpf]#
>
> Ideas? AFK for a while, will continue investigating.
I also noticed this output, but didn't digg into it because all events
I concerned is okay. I'll look into this today.
> This already was submitted to Ingo, BTW.
>
> I used, as in the changeset comment tests:
>
> perf record -a -e bpf-output/no-inherit,name=evt/ -e ./test_bpf_output.c/map:channel.event=evt/ ls /
>
> And perf script told me:
>
> [root@jouet bpf]# perf script | tail
> perf 13040 [003] 12062.708337: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708339: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708340: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708341: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708343: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708344: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708346: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708347: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708348: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> perf 13040 [003] 12062.708350: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> [root@jouet bpf]#
>
> Wonder where that /lib/modules/4.5.0-rc4/build/vmlinux came from...
>
> [root@jouet bpf]# perf script | cut -d'(' -f2 | sort | uniq -c
> 1141 /lib/modules/4.5.0-rc4/build/vmlinux)
It's a standard directory for perf searching vmlinux. Isn't it?
tools/perf/util/symbol.c:
static const char * const vmlinux_paths_upd[] = {
"/boot/vmlinux-%s",
"/usr/lib/debug/boot/vmlinux-%s",
"/lib/modules/%s/build/vmlinux",
"/usr/lib/debug/lib/modules/%s/vmlinux",
"/usr/lib/debug/boot/vmlinux-%s.debug"
};
So what's your problem?
Thank you.
[toc] | [prev] | [next] | [standalone]
| From | Arnaldo Carvalho de Melo <acme@kernel.org> |
|---|---|
| Date | 2016-02-24 14:40 +0100 |
| Message-ID | <r5HPC-5eb-57@gated-at.bofh.it> |
| In reply to | #1341239 |
Em Wed, Feb 24, 2016 at 09:58:34AM +0800, Wangnan (F) escreveu:
>
>
> On 2016/2/24 1:45, Arnaldo Carvalho de Melo wrote:
> >Em Mon, Feb 22, 2016 at 09:10:37AM +0000, Wang Nan escreveu:
> >>Commit a43eec304259a6c637f4014a6d4767159b6a3aa3 (bpf: introduce
> >>bpf_perf_event_output() helper) add a helper to enable BPF program
> >>output data to perf ring buffer through a new type of perf event
> >>PERF_COUNT_SW_BPF_OUTPUT. This patch enable perf to create perf
> >>event of that type. Now perf user can use following cmdline to
> >>receive output data from BPF programs:
> >>
> >> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
> >> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
> >> # ./perf script
> >> perf 1560 [004] 347747.086295: evt: ffffffff811fd201 sys_write ...
> >> perf 1560 [004] 347747.086300: evt: ffffffff811fd201 sys_write ...
> >> perf 1560 [004] 347747.086315: evt: ffffffff811fd201 sys_write ...
> >> ...
> >>
> >>Test result:
> >> # cat ./test_bpf_output.c
> >> /************************ BEGIN **************************/
> >> #include <uapi/linux/bpf.h>
> >> struct bpf_map_def {
> >> unsigned int type;
> >> unsigned int key_size;
> >> unsigned int value_size;
> >> unsigned int max_entries;
> >> };
> >>
> >> #define SEC(NAME) __attribute__((section(NAME), used))
> >> static u64 (*ktime_get_ns)(void) =
> >> (void *)BPF_FUNC_ktime_get_ns;
> >> static int (*trace_printk)(const char *fmt, int fmt_size, ...) =
> >> (void *)BPF_FUNC_trace_printk;
> >> static int (*get_smp_processor_id)(void) =
> >> (void *)BPF_FUNC_get_smp_processor_id;
> >> static int (*perf_event_output)(void *, struct bpf_map_def *, int, void *, unsigned long) =
> >> (void *)BPF_FUNC_perf_event_output;
> >>
> >> struct bpf_map_def SEC("maps") channel = {
> >> .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
> >> .key_size = sizeof(int),
> >> .value_size = sizeof(u32),
> >> .max_entries = __NR_CPUS__,
> >> };
> >>
> >> SEC("func_write=sys_write")
> >> int func_write(void *ctx)
> >> {
> >> struct {
> >> u64 ktime;
> >> int cpuid;
> >> } __attribute__((packed)) output_data;
> >> char error_data[] = "Error: failed to output: %d\n";
> >>
> >> output_data.cpuid = get_smp_processor_id();
> >> output_data.ktime = ktime_get_ns();
> >> int err = perf_event_output(ctx, &channel, get_smp_processor_id(),
> >> &output_data, sizeof(output_data));
> >> if (err)
> >> trace_printk(error_data, sizeof(error_data), err);
> >> return 0;
> >> }
> >> char _license[] SEC("license") = "GPL";
> >> int _version SEC("version") = LINUX_VERSION_CODE;
> >> /************************ END ***************************/
> >>
> >> # ./perf record -a -e bpf-output/no-inherit,name=evt/ \
> >> -e ./test_bpf_output.c/map:channel.event=evt/ ls /
> >> # ./perf script | grep ls
> >> ls 2242 [003] 347851.557563: evt: ffffffff811fd201 sys_write ...
> >> ls 2242 [003] 347851.557571: evt: ffffffff811fd201 sys_write ...
> >So, there is something strange here:
> >
> > if (unlikely(event->oncpu != smp_processor_id()))
> > return -EOPNOTSUPP;
> >
> >This is where I am hitting, with:
> >
> >[acme@jouet linux]$ uname -r
> >4.5.0-rc4
> >
> > int err = perf_event_output(ctx, &channel, get_smp_processor_id(),
> > &output_data, sizeof(output_data));
> > if (err)
> > trace_printk(error_data, sizeof(error_data), err);
> >
> >And then:
> >
> >[root@jouet bpf]# tail /sys/kernel/debug/tracing/trace
> > perf-13040 [003] d... 12062.807729: : Error: failed to output: -95
> > perf-13040 [003] d... 12062.807731: : Error: failed to output: -95
> > perf-13040 [003] d... 12062.807732: : Error: failed to output: -95
> > perf-13040 [003] d... 12062.807735: : Error: failed to output: -95
> > perf-13040 [003] d... 12062.807737: : Error: failed to output: -95
> > perf-13040 [003] d... 12062.807744: : Error: failed to output: -95
> > gnome-terminal--3091 [001] d... 12062.807773: : Error: failed to output: -95
> > gnome-terminal--3091 [001] d... 12062.807784: : Error: failed to output: -95
> > gmain-2830 [002] d... 12062.811791: : Error: failed to output: -95
> > gmain-2830 [002] d... 12062.811810: : Error: failed to output: -95
> >[root@jouet bpf]#
> >
> >Ideas? AFK for a while, will continue investigating.
>
> I also noticed this output, but didn't digg into it because all events
> I concerned is okay. I'll look into this today.
>
> >This already was submitted to Ingo, BTW.
> >
> >I used, as in the changeset comment tests:
> >
> >perf record -a -e bpf-output/no-inherit,name=evt/ -e ./test_bpf_output.c/map:channel.event=evt/ ls /
> >
> >And perf script told me:
> >
> >[root@jouet bpf]# perf script | tail
> > perf 13040 [003] 12062.708337: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708339: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708340: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708341: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708343: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708344: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708346: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708347: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708348: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> > perf 13040 [003] 12062.708350: evt: ffffffff81234eb1 sys_write (/lib/modules/4.5.0-rc4/build/vmlinux)
> >[root@jouet bpf]#
> >
> >Wonder where that /lib/modules/4.5.0-rc4/build/vmlinux came from...
> >
> >[root@jouet bpf]# perf script | cut -d'(' -f2 | sort | uniq -c
> > 1141 /lib/modules/4.5.0-rc4/build/vmlinux)
>
> It's a standard directory for perf searching vmlinux. Isn't it?
Nah, that was me being confused by 'perf script's output, it looked like
what was enclosed in () right after the sys_write was a parameter for
that function (sys_write), when in fact it is the DSO where sys_write is
in, duh.
- Arnaldo
> tools/perf/util/symbol.c:
>
> static const char * const vmlinux_paths_upd[] = {
> "/boot/vmlinux-%s",
> "/usr/lib/debug/boot/vmlinux-%s",
> "/lib/modules/%s/build/vmlinux",
> "/usr/lib/debug/lib/modules/%s/vmlinux",
> "/usr/lib/debug/boot/vmlinux-%s.debug"
> };
>
> So what's your problem?
>
> Thank you.
[toc] | [prev] | [standalone]
Back to top | Article view | linux.kernel
csiph-web