Re: [PATCH V5 1/1] perf: event preserve and create across cpu hotplug

From: Mukesh Ojha
Date: Tue Sep 24 2019 - 10:31:00 EST



On 8/12/2019 4:12 PM, Jiri Olsa wrote:
On Fri, Aug 02, 2019 at 12:16:53AM +0530, Mukesh Ojha wrote:
Perf framework doesn't allow preserving CPU events across
CPU hotplugs. The events are scheduled out as and when the
CPU walks offline. Moreover, the framework also doesn't
allow the clients to create events on an offline CPU. As
a result, the clients have to keep on monitoring the CPU
state until it comes back online.

Therefore, introducing the perf framework to support creation
and preserving of (CPU) events for offline CPUs. Through
this, the CPU's online state would be transparent to the
client and it not have to worry about monitoring the CPU's
state. Success would be returned to the client even while
creating the event on an offline CPU. If during the lifetime
of the event the CPU walks offline, the event would be
preserved and would continue to count as soon as (and if) the
CPU comes back online.

Co-authored-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Signed-off-by: Raghavendra Rao Ananta <rananta@xxxxxxxxxxxxxx>
Signed-off-by: Mukesh Ojha <mojha@xxxxxxxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
Cc: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
Cc: Alexei Starovoitov <ast@xxxxxxxxxx>
---
Change in V5:
=============
- Rebased it.
note that we might need to change how we store cpu topology,
now that it can change during the sampling.. like below it's
the comparison of header data with and without cpu 1

I think some of the report code checks on topology or caches
and it might get confused

perhaps we could watch cpu topology in record and update the
data as we see it changing.. future TODO list ;-)

Hi Jiri,

Can we do something like below to address issue related to header update while perf record with offline cpus.

--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1432,7 +1432,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ opts->no_bpf_event = true;
ÂÂÂÂÂÂÂ }

-ÂÂÂÂÂÂ err = record__synthesize(rec, false);
+ÂÂÂÂÂÂ err = record__synthesize(rec, true);
ÂÂÂÂÂÂÂ if (err < 0)
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ goto out_child;

@@ -1652,7 +1652,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
ÂÂÂÂÂÂÂ } else
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ status = err;

-ÂÂÂÂÂÂ record__synthesize(rec, true);
+ÂÂÂÂÂÂ record__synthesize(rec, false);
ÂÂÂÂÂÂÂ /* this will be recalculated during process_buildids() */
ÂÂÂÂÂÂÂ rec->samples = 0;

Thanks.
Mukesh



perf stat is probably fine

jirka


---
-# nrcpus online : 39
+# nrcpus online : 40
# nrcpus avail : 40
# cpudesc : Intel(R) Xeon(R) Silver 4114 CPU @ 2.20GHz
# cpuid : GenuineIntel,6,85,4
...
# sibling sockets : 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38
-# sibling sockets : 3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
+# sibling sockets : 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
# sibling dies : 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38
-# sibling dies : 3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
+# sibling dies : 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
# sibling threads : 0,20
+# sibling threads : 1,21
# sibling threads : 2,22
# sibling threads : 3,23
# sibling threads : 4,24
@@ -38,9 +39,8 @@
# sibling threads : 17,37
# sibling threads : 18,38
# sibling threads : 19,39
-# sibling threads : 21
# CPU 0: Core ID 0, Die ID 0, Socket ID 0
-# CPU 1: Core ID -1, Die ID -1, Socket ID -1
+# CPU 1: Core ID 0, Die ID 0, Socket ID 1
# CPU 2: Core ID 4, Die ID 0, Socket ID 0
# CPU 3: Core ID 4, Die ID 0, Socket ID 1
# CPU 4: Core ID 1, Die ID 0, Socket ID 0
@@ -79,14 +79,16 @@
# CPU 37: Core ID 9, Die ID 0, Socket ID 1
# CPU 38: Core ID 10, Die ID 0, Socket ID 0
# CPU 39: Core ID 10, Die ID 0, Socket ID 1
-# node0 meminfo : total = 47391616 kB, free = 46536844 kB
+# node0 meminfo : total = 47391616 kB, free = 46548348 kB
# node0 cpu list : 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38
-# node1 meminfo : total = 49539612 kB, free = 48908820 kB
-# node1 cpu list : 3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
+# node1 meminfo : total = 49539612 kB, free = 48897176 kB
+# node1 cpu list : 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39
# pmu mappings: intel_pt = 8, uncore_cha_1 = 25, uncore_irp_3 = 49, software = 1, uncore_imc_5 = 18, uncore_m3upi_0 = 21, uncore_iio_free_running_5 = 45, uncore_irp_1 = 47, uncore_m2m_1 = 12, uncore_imc_3 = 16, uncore_cha_8 = 32, uncore_iio_free_running_3 = 43, uncore_imc_1 = 14, uncore_upi_1 = 20, power = 10, uncore_cha_6 = 30, uncore_iio_free_running_1 = 41, uncore_iio_4 = 38, uprobe = 7, cpu = 4, uncore_cha_4 = 28, uncore_iio_2 = 36, cstate_core = 53, breakpoint = 5, uncore_cha_2 = 26, uncore_irp_4 = 50, uncore_m3upi_1 = 22, uncore_iio_0 = 34, tracepoint = 2, uncore_cha_0 = 24, uncore_irp_2 = 48, cstate_pkg = 54, uncore_imc_4 = 17, uncore_cha_9 = 33, uncore_iio_free_running_4 = 44, uncore_ubox = 23, uncore_irp_0 = 46, uncore_m2m_0 = 11, uncore_imc_2 = 15, kprobe = 6, uncore_cha_7 = 31, uncore_iio_free_running_2 = 42, uncore_iio_5 = 39, uncore_imc_0 = 13, uncore_upi_0 = 19, uncore_cha_5 = 29, uncore_iio_free_running_0 = 40, uncore_pcu = 52, msr = 9, uncore_iio_3 = 37, uncore_cha_3 = 27, uncore_irp_5 = 51, uncore_iio_1 = 35
# CPU cache info:
# L1 Data 32K [0,20]
# L1 Instruction 32K [0,20]
+# L1 Data 32K [1,21]
+# L1 Instruction 32K [1,21]
# L1 Data 32K [2,22]
# L1 Instruction 32K [2,22]
# L1 Data 32K [3,23]
@@ -123,9 +125,8 @@
# L1 Instruction 32K [18,38]
# L1 Data 32K [19,39]
# L1 Instruction 32K [19,39]
-# L1 Data 32K [21]
-# L1 Instruction 32K [21]
# L2 Unified 1024K [0,20]
+# L2 Unified 1024K [1,21]
# L2 Unified 1024K [2,22]
# L2 Unified 1024K [3,23]
# L2 Unified 1024K [4,24]
@@ -144,12 +145,11 @@
# L2 Unified 1024K [17,37]
# L2 Unified 1024K [18,38]
# L2 Unified 1024K [19,39]
-# L2 Unified 1024K [21]
# L3 Unified 14080K [0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38]
-# L3 Unified 14080K [3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39]
...