This patchset is the re-spin of the original branch stack sampling
patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
also enables SW based branch filtering support for PPC64 platforms which have
branch stack sampling support. With this new enablement, the branch filter support
for PPC64 platforms have been extended to include all these combinations discussed
below with a sample test application program.
(1) perf record -e branch-misses:u -b ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... ..................... .................... .....................
#
4.42% cprog cprog [k] sw_4_2 cprog [k] lr_addr
4.41% cprog cprog [k] symbol2 cprog [k] hw_1_2
4.41% cprog cprog [k] ctr_addr cprog [k] sw_4_1
4.41% cprog cprog [k] lr_addr cprog [k] sw_4_2
4.41% cprog cprog [k] sw_4_2 cprog [k] callme
4.41% cprog cprog [k] symbol1 cprog [k] hw_1_1
4.41% cprog cprog [k] success_3_1_3 cprog [k] sw_3_1
2.43% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
2.43% cprog cprog [k] hw_1_2 cprog [k] symbol2
2.43% cprog cprog [k] callme cprog [k] hw_1_2
2.43% cprog cprog [k] address1 cprog [k] back1
2.43% cprog cprog [k] back1 cprog [k] callme
2.43% cprog cprog [k] hw_2_1 cprog [k] address1
2.43% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
2.43% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
2.43% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
2.43% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_1
2.43% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_2
2.43% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_3
2.43% cprog cprog [k] callme cprog [k] sw_3_1
2.43% cprog cprog [k] callme cprog [k] sw_4_2
2.43% cprog cprog [k] hw_1_1 cprog [k] symbol1
2.43% cprog cprog [k] callme cprog [k] hw_1_1
2.42% cprog cprog [k] sw_3_1 cprog [k] callme
1.99% cprog cprog [k] success_3_1_1 cprog [k] sw_3_1
1.99% cprog cprog [k] sw_3_1 cprog [k] success_3_1_1
1.99% cprog cprog [k] address2 cprog [k] back2
1.99% cprog cprog [k] hw_2_2 cprog [k] address2
1.99% cprog cprog [k] back2 cprog [k] callme
1.99% cprog cprog [k] callme cprog [k] main
1.99% cprog cprog [k] sw_3_1 cprog [k] success_3_1_3
1.99% cprog cprog [k] hw_1_1 cprog [k] callme
1.99% cprog cprog [k] sw_3_2 cprog [k] callme
1.99% cprog cprog [k] callme cprog [k] sw_3_2
1.99% cprog cprog [k] success_3_1_2 cprog [k] sw_3_1
1.99% cprog cprog [k] sw_3_1 cprog [k] success_3_1_2
1.99% cprog cprog [k] hw_1_2 cprog [k] callme
1.99% cprog cprog [k] sw_4_1 cprog [k] callme
0.02% cprog [unknown] [k] 0xf7ba2328 [unknown] [k] 0xf7ba2320
0.00% cprog libc-2.11.2.so [k] _IO_file_overflow libc-2.11.2.so [k] _IO_file_overflow
0.00% cprog libc-2.11.2.so [k] _IO_file_xsputn libc-2.11.2.so [k] _IO_file_overflow
0.00% cprog cprog [k] callme cprog [k] hw_2_2
PMU filters
-----------
(2) perf record -e branch-misses:u -j any_call ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... ....................... .................... ......................
#
7.82% cprog cprog [k] sw_3_1 cprog [k] success_3_1_2
6.88% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_2
6.88% cprog cprog [k] hw_1_1 cprog [k] symbol1
5.88% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_1
5.88% cprog cprog [k] callme cprog [k] hw_1_1
5.88% cprog cprog [k] sw_3_1 cprog [k] success_3_1_1
5.88% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_3
5.88% cprog cprog [k] callme cprog [k] hw_1_2
5.88% cprog cprog [k] hw_1_2 cprog [k] symbol2
5.88% cprog cprog [k] sw_4_2 cprog [k] lr_addr
5.88% cprog cprog [k] callme cprog [k] sw_4_2
4.88% cprog cprog [k] sw_3_1 cprog [k] success_3_1_3
4.88% cprog cprog [k] callme cprog [k] sw_3_2
4.88% cprog cprog [k] callme cprog [k] hw_2_2
3.94% cprog cprog [k] callme cprog [k] sw_3_1
3.94% cprog cprog [k] callme cprog [k] hw_2_1
2.94% cprog cprog [k] main cprog [k] callme
2.94% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
2.94% cprog cprog [k] callme cprog [k] sw_4_1
0.01% cprog [unknown] [k] 0xf79076c4 [unknown] [k] 0xf78f22c0
0.00% cprog libc-2.11.2.so [k] _IO_file_doallocate libc-2.11.2.so [k] _IO_setb
0.00% cprog libc-2.11.2.so [k] _IO_file_doallocate libc-2.11.2.so [k] mmap
0.00% cprog libc-2.11.2.so [k] _IO_file_xsputn libc-2.11.2.so [k] _IO_default_xsputn
0.00% cprog libc-2.11.2.so [k] _IO_file_overflow libc-2.11.2.so [k] _IO_do_write
0.00% cprog ld-2.11.2.so [k] malloc [unknown] [k] 0xf790b380
(3) perf record -e branch-misses:u -j cond ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... .................. .................... .......................
#
24.85% cprog [unknown] [k] 00000000 cprog [k] callme
15.71% cprog cprog [k] sw_3_1 cprog [k] sw_3_1
7.14% cprog cprog [k] sw_4_2 cprog [k] lr_addr
6.57% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
4.57% cprog cprog [k] hw_2_2 cprog [k] callme
4.57% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
4.57% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
4.57% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
4.57% cprog cprog [k] main cprog [k] hw_1_1
4.57% cprog cprog [k] hw_1_2 cprog [k] hw_1_2
4.57% cprog [unknown] [k] 00000000 cprog [k] main
4.57% cprog cprog [k] hw_2_1 cprog [k] callme
4.57% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
4.57% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
0.01% cprog [unknown] [k] 0xf7aa25dc [unknown] [k] 0xf7aa27e4
0.00% cprog libc-2.11.2.so [k] _IO_doallocbuf libc-2.11.2.so [k] _IO_file_doallocate
0.00% cprog [unknown] [k] 00000000 libc-2.11.2.so [k] _IO_file_doallocate
0.00% cprog [unknown] [k] 00000000 libc-2.11.2.so [k] _IO_file_stat
SW filters
----------
(4) perf record -e branch-misses:u -j any_ret ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... ................. .................... ..............
#
7.91% cprog cprog [k] symbol1 cprog [k] hw_1_1
7.91% cprog cprog [k] success_3_1_3 cprog [k] sw_3_1
7.91% cprog cprog [k] ctr_addr cprog [k] sw_4_1
7.91% cprog cprog [k] lr_addr cprog [k] sw_4_2
7.91% cprog cprog [k] symbol2 cprog [k] hw_1_2
7.90% cprog cprog [k] sw_4_2 cprog [k] callme
4.34% cprog cprog [k] success_3_1_2 cprog [k] sw_3_1
4.33% cprog cprog [k] sw_4_1 cprog [k] callme
4.33% cprog cprog [k] hw_1_2 cprog [k] callme
4.33% cprog cprog [k] success_3_1_1 cprog [k] sw_3_1
4.33% cprog cprog [k] sw_3_2 cprog [k] callme
4.33% cprog cprog [k] back2 cprog [k] callme
4.33% cprog cprog [k] callme cprog [k] main
4.33% cprog cprog [k] hw_1_1 cprog [k] callme
3.58% cprog cprog [k] sw_3_1 cprog [k] callme
3.58% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
3.58% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
3.58% cprog cprog [k] back1 cprog [k] callme
3.57% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
0.00% cprog [unknown] [k] 0xf7abacf4 [unknown] [k] 0xf7abae40
(5) perf record -e branch-misses:u -j ind_call ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... ............. .................... .............
#
63.56% cprog cprog [k] sw_4_2 cprog [k] lr_addr
36.44% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
Mixed filters
-------------
(6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
Error:
The perf.data file has no samples!
NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
branches in that given set. Both the filters are mutually exclussive, so obviously no samples
found in the end profile.
(7) perf record -e branch-misses:u -j any_call,ind_call ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... .............. .................... ..............
#
66.69% cprog cprog [k] sw_4_2 cprog [k] lr_addr
33.31% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
0.00% cprog [unknown] [k] 0x0fe7f264 [unknown] [k] 0x0ff926d0
(8) perf record -e branch-misses:u -j any_call,any_ret,ind_call ./cprog
Error:
The perf.data file has no samples!
(9) perf record -e branch-misses:u -j cond,any_ret ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... .............. .................... .......................
#
46.01% cprog [unknown] [k] 00000000 cprog [k] callme
13.54% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
8.18% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
8.07% cprog [unknown] [k] 00000000 cprog [k] main
8.07% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
8.07% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
8.07% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
0.00% cprog [unknown] [k] 00000000 [unknown] [k] 0xf7c1480c
0.00% cprog libc-2.11.2.so [k] mmap libc-2.11.2.so [k] _IO_file_doallocate
(10) perf record -e branch-misses:u -j cond,ind_call ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... .............. .................... ..............
#
48.11% cprog [unknown] [k] 00000000 cprog [k] callme
13.52% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
12.42% cprog cprog [k] sw_4_2 cprog [k] lr_addr
8.65% cprog [unknown] [k] 00000000 cprog [k] main
8.65% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
8.65% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
0.00% cprog [unknown] [k] 00000000 [unknown] [k] 0xf7a4581c
(11) perf record -e branch-misses:u -j cond,any_ret,ind_call ./cprog
# Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
# ........ ....... .................... .............. .................... .................
#
45.91% cprog [unknown] [k] 00000000 cprog [k] callme
13.26% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
8.17% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
8.17% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
8.17% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
8.17% cprog [unknown] [k] 00000000 cprog [k] main
8.16% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
0.00% cprog [unknown] [k] 00000000 [unknown] [k] 0xf7f87704
0.00% cprog [unknown] [k] 00000000 libc-2.11.2.so [k] _IO_file_sync
Test application program
========================
(1) Makefile:
--------------------------------------------
all: sample.o cprog of.cprog of.sample
sample.o: sample.s
as -o sample.o sample.s
cprog: cprog.c sample.o
gcc -o cprog cprog.c sample.o
of.sample: sample.o
objdump -d sample.o > of.sample
of.cprog: cprog
objdump -d cprog > of.cprog
clean:
rm sample.o cprog of.sample of.cprog
---------------------------------------------
(2) cprog.c
---------------------------------------------
#include <stdio.h>
#define LOOP_COUNT 100000
extern void callme(void);
int main(int argc, char *argv[])
{
int i;
for(i = 0; i < LOOP_COUNT; i++)
callme();
printf("end");
return 0;
}
---------------------------------------------
(3) sample.S
---------------------------------------------
# r25, r26, r27 will be used as first level, second level
# and third level stack for LR. Register r20, r21, r22, r23
# r24 will be used for general programming purpose.
.data
msg:
.string "BHRB filter tests\n"
len = . - msg
msg_1_1:
.string "Test: hw_1_1\n"
len_1_1 = 13
msg_1_2:
.string "Test: hw_1_2\n"
len_1_2 = 13
msg_2_1:
.string "Test: hw_2_1\n"
len_2_1 = 13
msg_2_2:
.string "Test: hw_2_2\n"
len_2_2 = 13
msg_3_1:
.string "Test: sw_3_1\n"
len_3_1 = 13
msg_3_1_1:
.string "Test: sw_3_1_1\n"
len_3_1_1 = 15
msg_3_1_2:
.string "Test: sw_3_1_2\n"
len_3_1_2 = 15
msg_3_1_3:
.string "Test: sw_3_1_3\n"
len_3_1_3 = 15
msg_3_2:
.string "Test: sw_3_2\n"
len_3_3 = 13
msg_4_1:
.string "Test: sw_4_1\n"
len_4_1 = 13
msg_4_2:
.string "Test: sw_4_2\n"
len_4_2 = 13
hw_3_1_1_passed:
.string "\thw_3_1_1_passed\n\n"
len_hw_3_1_1_passed = 18
hw_3_1_2_passed:
.string "\thw_3_1_2_passed\n\n"
len_hw_3_1_2_passed = 18
hw_3_1_3_passed:
.string "\thw_3_1_3_passed\n\n"
len_hw_3_1_3_passed = 18
hw_2_1_passed:
.string "\thw_2_1_passed\n\n"
len_hw_2_1_passed = 16
hw_2_2_passed:
.string "\thw_2_2_passed\n\n"
len_hw_2_2_passed = 16
hw_1_1_passed:
.string "\thw_1_1_passed\n\n"
len_hw_1_1_passed = 16
hw_1_2_passed:
.string "\thw_1_2_passed\n\n"
len_hw_1_2_passed = 16
hw_4_1_passed:
.string "\thw_4_1_passed\n\n"
len_hw_4_1_passed = 16
hw_4_2_passed:
.string "\thw_4_2_passed\n\n"
len_hw_4_2_passed = 16
msg_error:
.string "\tError\n"
len_error = 7
.text
.global callme
.global hw_1_1
.global hw_1_2
.global hw_2_1
.global hw_2_2
# HW filter test symbols
symbol1:
# Print "hw_1_1_passed"
li 0, 4
li 3, 1
lis 4, hw_1_1_passed@ha
addi 4, 4, hw_1_1_passed@l
li 5, len_hw_1_1_passed
sc
blr # PERF_SAMPLE_BRANCH_ANY_RET
hw_1_1:
# Save LR - second level
mflr 26
# Print "hw_1_1 called"
li 0, 4
li 3, 1
lis 4, msg_1_1@ha
addi 4, 4, msg_1_1@l
li 5, len_1_1
sc
bl symbol1 # PERF_SAMPLE_BRANCH_ANY_CALL
# Restore LR
mtlr 26
blr # PERF_SAMPLE_BRANCH_ANY_RET
symbol2:
# Print "Symbol2 taken"
li 0, 4
li 3, 1
lis 4, hw_1_2_passed@ha
addi 4, 4, hw_1_2_passed@l
li 5, len_hw_1_2_passed
sc
blr # PERF_SAMPLE_BRANCH_ANY_RET
hw_1_2:
# Save LR - second level
mflr 26
# Print "hw_1_2 called"
li 0, 4
li 3, 1
lis 4, msg_1_2@ha
addi 4, 4, msg_1_2@l
li 5, len_1_2
sc
li 4,20
cmpi 0,4,20
bcl 12, 4*cr0+2, symbol2 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
mtlr 26
blr # PERF_SAMPLE_BRANCH_ANY_RET
# HW filter test
address1:
# Print "hw_2_1_passed"
li 0, 4
li 3, 1
lis 4, hw_2_1_passed@ha
addi 4, 4, hw_2_1_passed@l
li 5, len_hw_2_1_passed
sc
b back1 # PERF_SAMPLE_BRANCH_ANY
hw_2_1:
# Print "hw_2_1 called"
li 0, 4
li 3, 1
lis 4, msg_2_1@ha
addi 4, 4, msg_2_1@l
li 5, len_2_1
sc
# Simple conditional branch (equal)
li 20, 12
cmpi 3, 20, 12
bc 12, 4*cr3+2, address1 # PERF_SAMPLE_BRANCH_COND
back1:
blr # PERF_SAMPLE_BRANCH_ANY_RET
address2:
# Print "hw_2_2_passed"
li 0, 4
li 3, 1
lis 4, hw_2_2_passed@ha
addi 4, 4, hw_2_2_passed@l
li 5, len_hw_2_2_passed
sc
b back2 # PERF_SAMPLE_BRANCH_ANY
hw_2_2:
# Print "hw_2_2 called"
li 0, 4
li 3, 1
lis 4, msg_2_2@ha
addi 4, 4, msg_2_2@l
li 5, len_2_2
sc
# Simple conditional branch (less than)
li 20, 12
cmpi 4, 20, 20
bc 12, 4*cr4+0, address2 # PERF_SAMPLE_BRANCH_COND
back2:
blr # PERF_SAMPLE_BRANCH_ANY_RET
# SW filter test symbols
sw_3_1_1:
# Print "Test: sw_3_1_1"
li 0, 4
li 3, 1
lis 4, msg_3_1_1@ha
addi 4, 4, msg_3_1_1@l
li 5, len_3_1_1
sc
li 22,0
# Test the condition and return
li 21, 10
cmpi 0, 21, 10
bclr 12, 2 # PERF_SAMPLE_BRANCH_ANY_RET | PERF_SAMPLE_BRANCH_COND
# Should not have come here
li 0, 4
li 3, 1
lis 4, msg_error@ha
addi 4, 4, msg_error@l
li 5, len_error
sc
# Mark the error
li 22, 1
# Safe fall back
blr # PERF_SAMPLE_BRANCH_ANY_RET
sw_3_1_2:
# Print "Test: sw_3_1_2"
li 0, 4
li 3, 1
lis 4, msg_3_1_2@ha
addi 4, 4, msg_3_1_2@l
li 5, len_3_1_2
sc
li 23, 0
# Test the condition and return
li 21, 10
cmpi 0, 21, 20
bclr 12, 0 # PERF_SAMPLE_BRANCH_ANY_RET | PERF_SAMPLE_BRANCH_COND
# Should not have come here
li 0, 4
li 3, 1
lis 4, msg_error@ha
addi 4, 4, msg_error@l
li 5, len_error
sc
# Mark the error
li 23, 1
# Safe fall back
blr # PERF_SAMPLE_BRANCH_ANY_RET
sw_3_1_3:
# Print "Test: sw_3_1_3"
li 0, 4
li 3, 1
lis 4, msg_3_1_3@ha
addi 4, 4, msg_3_1_3@l
li 5, len_3_1_3
sc
li 24, 0
# Test the condition and return
li 21, 10
cmpi 0, 21, 5
bclr 12, 1 # PERF_SAMPLE_BRANCH_ANY_RET | PERF_SAMPLE_BRANCH_COND
# Mark the error
li 24, 1
# Should not have come here
li 0, 4
li 3, 1
lis 4, msg_error@ha
addi 4, 4, msg_error@l
li 5, len_error
sc
# Safe fall back
blr # PERF_SAMPLE_BRANCH_ANY_RET
success_3_1_1:
li 0, 4
li 3, 1
lis 4, hw_3_1_1_passed@ha
addi 4, 4, hw_3_1_1_passed@l
li 5, len_hw_3_1_1_passed
sc
blr
success_3_1_2:
li 0, 4
li 3, 1
lis 4, hw_3_1_2_passed@ha
addi 4, 4, hw_3_1_2_passed@l
li 5, len_hw_3_1_2_passed
sc
blr
success_3_1_3:
li 0, 4
li 3, 1
lis 4, hw_3_1_3_passed@ha
addi 4, 4, hw_3_1_3_passed@l
li 5, len_hw_3_1_3_passed
sc
blr
sw_3_1:
# Save LR
mflr 26
# Print "Test: sw_3_1"
li 0, 4
li 3, 1
lis 4, msg_3_1@ha
addi 4, 4, msg_3_1@l
li 5, len_3_1
sc
# Equal comparison condition
bl sw_3_1_1 # PERF_SAMPLE_BRANCH_ANY_CALL
cmpi 0, 22, 0
bcl 12, 2, success_3_1_1 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
# LT comparison condition
bl sw_3_1_2 # PERF_SAMPLE_BRANCH_ANY_CALL
cmpi 0, 23, 0
bcl 12, 2, success_3_1_2 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
# GT comparison condition
bl sw_3_1_3 # PERF_SAMPLE_BRANCH_ANY_CALL
cmpi 0, 24, 0
bcl 12, 2, success_3_1_3 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
mtlr 26
blr # PERF_SAMPLE_BRANCH_ANY_RET
sw_3_2:
# Print "Test: sw_3_2"
li 0, 4
li 3, 1
lis 4, msg_3_2@ha
addi 4, 4, msg_3_2@l
li 5, len_3_1
sc
# FIXME: Anything more here ?
blr # PERF_SAMPLE_BRANCH_ANY_RET
# Indirect call tests
# CTR
ctr_addr:
# Print "bcctr taken"
li 0, 4
li 3, 1
lis 4, hw_4_1_passed@ha
addi 4, 4, hw_4_1_passed@l
li 5, len_hw_4_1_passed
sc
blr # PERF_SAMPLE_BRANCH_ANY_RET
sw_4_1:
# Save LR
mflr 26
# Print "sw_4_1 called"
li 0, 4
li 3, 1
lis 4, msg_4_1@ha
addi 4, 4, msg_4_1@l
li 5, len_4_1
sc
# Save address in CTR
lis 20, ctr_addr@ha
addi 20, 20, ctr_addr@l
mtctr 20
# Compare and jump to CTR
li 21, 10
cmpi 0, 21, 10
bcctrl 12, 4*cr0+2 # PERF_SAMPLE_BRANCH_IND_CALL
mtlr 26
blr # PERF_SAMPLE_BRANCH_ANY_RET
# LR
lr_addr:
# Print "bclrl taken"
li 0, 4
li 3, 1
lis 4, hw_4_2_passed@ha
addi 4, 4, hw_4_2_passed@l
li 5, len_hw_4_2_passed
sc
blr # PERF_SAMPLE_BRANCH_ANY_RET
sw_4_2:
# Save LR
mflr 26
# Print "Test: sw_4_2"
li 0, 4
li 3, 1
lis 4, msg_4_2@ha
addi 4, 4, msg_4_2@l
li 5, len_4_2
sc
# Save address in LR
lis 20, lr_addr@ha
addi 20, 20, lr_addr@l
mtlr 20
# Compare and jump to CTR
li 21, 10
cmpi 0, 21, 10
bclrl 12, 4*cr0+2 # PERF_SAMPLE_BRANCH_IND_CALL
# Restore LR
mtlr 26
blr # PERF_SAMPLE_BRANCH_ANY_RET
callme:
# Save LR
mflr 25
# Print "Branch filter Test"
li 0, 4
li 3, 1
lis 4, msg@ha
addi 4, 4, msg@l
li 5, len
sc
# PERF_SAMPLE_BRANCH_ANY_CALL
bl hw_1_1 # PERF_SAMPLE_BRANCH_ANY_CALL
bl hw_1_2 # PERF_SAMPLE_BRANCH_ANY_CALL
# PERF_SAMPLE_BRANCH_COND
bl hw_2_1 # PERF_SAMPLE_BRANCH_ANY_CALL
bl hw_2_2 # PERF_SAMPLE_BRANCH_ANY_CALL
# PERF_SAMPLE_BRANCH_ANY_RET
bl sw_3_1 # PERF_SAMPLE_BRANCH_ANY_CALL
bl sw_3_2 # PERF_SAMPLE_BRANCH_ANY_CALL
# PERF_SAMPLE_BRANCH_IND_CALL
bl sw_4_1 # PERF_SAMPLE_BRANCH_ANY_CALL
bl sw_4_2 # PERF_SAMPLE_BRANCH_ANY_CALL
# Restore LR
mtlr 25
blr # PERF_SAMPLE_BRANCH_ANY_RET
--------------------------------------------------------------------
Changes in V2
--------------
(1) Enabled PPC64 SW branch filtering support
(2) Incorporated changes required for all previous comments
Anshuman Khandual (6):
perf: New conditional branch filter criteria in branch stack sampling
powerpc, perf: Enable conditional branch filter for POWER8
perf, tool: Conditional branch filter 'cond' added to perf record
x86, perf: Add conditional branch filtering support
perf, documentation: Description for conditional branch filter
powerpc, perf: Enable SW filtering in branch stack sampling framework
arch/powerpc/include/asm/perf_event_server.h | 2 +-
arch/powerpc/perf/core-book3s.c | 200 +++++++++++++++++++++++++--
arch/powerpc/perf/power8-pmu.c | 25 ++--
arch/x86/kernel/cpu/perf_event_intel_lbr.c | 5 +
include/uapi/linux/perf_event.h | 3 +-
tools/perf/Documentation/perf-record.txt | 3 +-
tools/perf/builtin-record.c | 1 +
7 files changed, 216 insertions(+), 23 deletions(-)
--
1.7.11.7
POWER8 PMU based BHRB supports filtering for conditional branches.
This patch introduces new branch filter PERF_SAMPLE_BRANCH_COND which
will extend the existing perf ABI. Other architectures can provide
this functionality with either HW filtering support (if present) or
with SW filtering of instructions.
Signed-off-by: Anshuman Khandual <[email protected]>
Reviewed-by: Stephane Eranian <[email protected]>
---
include/uapi/linux/perf_event.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 0b1df41..5da52b6 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -160,8 +160,9 @@ enum perf_branch_sample_type {
PERF_SAMPLE_BRANCH_ABORT_TX = 1U << 7, /* transaction aborts */
PERF_SAMPLE_BRANCH_IN_TX = 1U << 8, /* in transaction */
PERF_SAMPLE_BRANCH_NO_TX = 1U << 9, /* not in transaction */
+ PERF_SAMPLE_BRANCH_COND = 1U << 10, /* conditional branches */
- PERF_SAMPLE_BRANCH_MAX = 1U << 10, /* non-ABI */
+ PERF_SAMPLE_BRANCH_MAX = 1U << 11, /* non-ABI */
};
#define PERF_SAMPLE_BRANCH_PLM_ALL \
--
1.7.11.7
This patch adds conditional branch filtering support,
enabling it for PERF_SAMPLE_BRANCH_COND in perf branch
stack sampling framework by utilizing an available
software filter X86_BR_JCC.
Signed-off-by: Anshuman Khandual <[email protected]>
Reviewed-by: Stephane Eranian <[email protected]>
---
arch/x86/kernel/cpu/perf_event_intel_lbr.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/arch/x86/kernel/cpu/perf_event_intel_lbr.c b/arch/x86/kernel/cpu/perf_event_intel_lbr.c
index d5be06a..9723773 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_lbr.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_lbr.c
@@ -371,6 +371,9 @@ static void intel_pmu_setup_sw_lbr_filter(struct perf_event *event)
if (br_type & PERF_SAMPLE_BRANCH_NO_TX)
mask |= X86_BR_NO_TX;
+ if (br_type & PERF_SAMPLE_BRANCH_COND)
+ mask |= X86_BR_JCC;
+
/*
* stash actual user request into reg, it may
* be used by fixup code for some CPU
@@ -665,6 +668,7 @@ static const int nhm_lbr_sel_map[PERF_SAMPLE_BRANCH_MAX] = {
* NHM/WSM erratum: must include IND_JMP to capture IND_CALL
*/
[PERF_SAMPLE_BRANCH_IND_CALL] = LBR_IND_CALL | LBR_IND_JMP,
+ [PERF_SAMPLE_BRANCH_COND] = LBR_JCC,
};
static const int snb_lbr_sel_map[PERF_SAMPLE_BRANCH_MAX] = {
@@ -676,6 +680,7 @@ static const int snb_lbr_sel_map[PERF_SAMPLE_BRANCH_MAX] = {
[PERF_SAMPLE_BRANCH_ANY_CALL] = LBR_REL_CALL | LBR_IND_CALL
| LBR_FAR,
[PERF_SAMPLE_BRANCH_IND_CALL] = LBR_IND_CALL,
+ [PERF_SAMPLE_BRANCH_COND] = LBR_JCC,
};
/* core */
--
1.7.11.7
This patch enables SW based post processing of BHRB captured branches
to be able to meet more user defined branch filtration criteria in perf
branch stack sampling framework. This changes increase the number of
filters and their valid combinations on powerpc64 platform with BHRB
support. Summary of code changes described below.
(1) struct cpu_hw_events
Introduced two new variables and modified one to track various filters.
a) bhrb_hw_filter Tracks PMU based HW branch filter flags.
Computed from PMU dependent call back.
b) bhrb_sw_filter Tracks SW based instruction filter flags
Computed from PPC64 generic SW filter.
c) filter_mask Tracks overall filter flags for PPC64
(2) Creating HW event with BHRB request
Kernel would try to figure out supported HW filters through a PMU call
back ppmu->bhrb_filter_map(). Here it would only invalidate unsupported
HW filter combinations. In future we could process one element from the
combination in HW and one in SW. Meanwhile cpuhw->filter_mask would be
tracking the overall supported branch filter requests on the PMU.
Kernel would also process the user request against available SW filters
for PPC64. Then we would process filter_mask to verify whether all the
user requested branch filters have been taken care of either in HW or in
SW.
(3) BHRB SW filter processing
During the BHRB data capture inside the PMU interrupt context, each
of the captured "perf_branch_entry.from" would be checked for compliance
with applicable SW branch filters. If the entry does not confirm to the
filter requirements, it would be discarded from the final perf branch
stack buffer.
(4) Instruction classification for proposed SW filters
Here are the list of category of instructions which have been classified
under the proposed SW filters.
(a) PERF_SAMPLE_BRANCH_ANY_RETURN
(i) [Un]conditional branch to LR without setting the LR
(1) blr
(2) bclr
(3) btlr
(4) bflr
(5) bdnzlr
(6) bdnztlr
(7) bdnzflr
(8) bdzlr
(9) bdztlr
(10) bdzflr
(11) bltlr
(12) blelr
(13) beqlr
(14) bgelr
(15) bgtlr
(16) bnllr
(17) bnelr
(18) bnglr
(19) bsolr
(20) bnslr
(21) biclr
(22) bnilr
(23) bunlr
(24) bnulr
(b) PERF_SAMPLE_BRANCH_IND_CALL
(i) [Un]conditional branch to CTR with setting the link
(1) bctrl
(2) bcctrl
(3) btctrl
(4) bfctrl
(5) bltctrl
(6) blectrl
(7) beqctrl
(8) bgectrl
(9) bgtctrl
(10) bnlctrl
(11) bnectrl
(12) bngctrl
(13) bsoctrl
(14) bnsctrl
(15) bicctrl
(16) bnictrl
(17) bunctrl
(18) bnuctrl
(ii) [Un]conditional branch to LR setting the link
(0) bclrl
(1) blrl
(2) btlrl
(3) bflrl
(4) bdnzlrl
(5) bdnztlrl
(6) bdnzflrl
(7) bdzlrl
(8) bdztlrl
(9) bdzflrl
(10) bltlrl
(11) blelrl
(12) beqlrl
(13) bgelrl
(14) bgtlrl
(15) bnllrl
(16) bnelrl
(17) bnglrl
(18) bsolrl
(19) bnslrl
(20) biclrl
(21) bnilrl
(22) bunlrl
(23) bnulrl
(iii) [Un]conditional branch to TAR setting the link
(1) btarl
(2) bctarl
Signed-off-by: Anshuman Khandual <[email protected]>
---
arch/powerpc/include/asm/perf_event_server.h | 2 +-
arch/powerpc/perf/core-book3s.c | 200 +++++++++++++++++++++++++--
arch/powerpc/perf/power8-pmu.c | 19 ++-
3 files changed, 198 insertions(+), 23 deletions(-)
diff --git a/arch/powerpc/include/asm/perf_event_server.h b/arch/powerpc/include/asm/perf_event_server.h
index 8b24926..5fc798b 100644
--- a/arch/powerpc/include/asm/perf_event_server.h
+++ b/arch/powerpc/include/asm/perf_event_server.h
@@ -34,7 +34,7 @@ struct power_pmu {
unsigned long *valp);
int (*get_alternatives)(u64 event_id, unsigned int flags,
u64 alt[]);
- u64 (*bhrb_filter_map)(u64 branch_sample_type);
+ u64 (*bhrb_filter_map)(u64 branch_sample_type, u64 *filter_mask);
void (*config_bhrb)(u64 pmu_bhrb_filter);
void (*disable_pmc)(unsigned int pmc, unsigned long mmcr[]);
int (*limited_pmc_event)(u64 event_id);
diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index eeae308..81c4a1d 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -26,6 +26,10 @@
#define BHRB_PREDICTION 0x0000000000000001
#define BHRB_EA 0xFFFFFFFFFFFFFFFC
+#define for_each_branch_sample_type(x) \
+ for ((x) = PERF_SAMPLE_BRANCH_USER; \
+ (x) < PERF_SAMPLE_BRANCH_MAX; (x) <<= 1)
+
struct cpu_hw_events {
int n_events;
int n_percpu;
@@ -47,7 +51,9 @@ struct cpu_hw_events {
int n_txn_start;
/* BHRB bits */
- u64 bhrb_filter; /* BHRB HW branch filter */
+ u64 bhrb_hw_filter; /* BHRB HW branch filter */
+ u64 bhrb_sw_filter; /* BHRB SW branch filter */
+ u64 filter_mask; /* Branch filter mask */
int bhrb_users;
void *bhrb_context;
struct perf_branch_stack bhrb_stack;
@@ -400,6 +406,101 @@ static __u64 power_pmu_bhrb_to(u64 addr)
return target - (unsigned long)&instr + addr;
}
+#define BRANCH_LINK 0x00000001
+#define BRANCH_LR 0x4C000020
+#define BRANCH_CTR 0x4C000420
+#define BRANCH_TAR 0x4C000460
+
+/* Check the instruction opcodes */
+static bool validate_instruction(unsigned int *addr, u64 bhrb_sw_filter)
+{
+ if (bhrb_sw_filter & PERF_SAMPLE_BRANCH_ANY_RETURN) {
+ /* Link is not set */
+ if (!(*addr & BRANCH_LINK)) {
+ /*
+ * Conditional and unconditional
+ * branch to LR.
+ */
+ if ((*addr & BRANCH_LR) == BRANCH_LR)
+ return true;
+
+ /* Everything else */
+ return false;
+ }
+
+ /* Link is set */
+ return false;
+ }
+
+ if (bhrb_sw_filter & PERF_SAMPLE_BRANCH_IND_CALL) {
+ /* Link is set */
+ if (*addr & BRANCH_LINK) {
+ /*
+ * Conditional and unconditional
+ * branch to CTR.
+ */
+ if ((*addr & BRANCH_CTR) == BRANCH_CTR)
+ return true;
+ /*
+ * Conditional and unconditional
+ * branch to LR.
+ */
+ if ((*addr & BRANCH_LR) == BRANCH_LR)
+ return true;
+ /*
+ * Conditional and unconditional
+ * branch to TAR.
+ */
+ if ((*addr & BRANCH_TAR) == BRANCH_TAR)
+ return true;
+
+ /* Everything else */
+ return false;
+ }
+
+ /* Link is not set */
+ return false;
+ }
+
+ /* No software branch filter, control
+ * should not have come here.
+ */
+ return true;
+}
+
+/* Extract the instruction from the address */
+static bool check_instruction(u64 addr, u64 bhrb_sw_filter)
+{
+ unsigned int instr;
+ bool ret;
+
+ if (bhrb_sw_filter == 0)
+ return true;
+
+ if (is_kernel_addr(addr)) {
+ ret = validate_instruction((unsigned int *) addr, bhrb_sw_filter);
+ } else {
+ /*
+ * Userspace address need to copied first
+ * before analysis.
+ */
+ pagefault_disable();
+ ret = __get_user_inatomic(instr, (unsigned int __user *)addr);
+
+ /*
+ * If the instruction could not be accessible
+ * from user space, we still OKAY the entry.
+ */
+ if (ret) {
+ pagefault_enable();
+ return true;
+ }
+ pagefault_enable();
+ ret = validate_instruction(&instr, bhrb_sw_filter);
+ }
+ return ret;
+}
+
/* Processing BHRB entries */
void power_pmu_bhrb_read(struct cpu_hw_events *cpuhw)
{
@@ -459,14 +560,28 @@ void power_pmu_bhrb_read(struct cpu_hw_events *cpuhw)
addr = 0;
}
cpuhw->bhrb_entries[u_index].from = addr;
+
+ /* Apply SW filter */
+ if (!check_instruction(cpuhw->
+ bhrb_entries[u_index].from,
+ cpuhw->bhrb_sw_filter))
+ u_index--;
} else {
/* Branches to immediate field
(ie I or B form) */
cpuhw->bhrb_entries[u_index].from = addr;
- cpuhw->bhrb_entries[u_index].to =
- power_pmu_bhrb_to(addr);
- cpuhw->bhrb_entries[u_index].mispred = pred;
- cpuhw->bhrb_entries[u_index].predicted = ~pred;
+ if (check_instruction(cpuhw->
+ bhrb_entries[u_index].from,
+ cpuhw->bhrb_sw_filter)) {
+ cpuhw->bhrb_entries[u_index].
+ to = power_pmu_bhrb_to(addr);
+ cpuhw->bhrb_entries[u_index].
+ mispred = pred;
+ cpuhw->bhrb_entries[u_index].
+ predicted = ~pred;
+ } else {
+ u_index--;
+ }
}
u_index++;
@@ -1159,7 +1274,7 @@ static void power_pmu_enable(struct pmu *pmu)
out:
if (cpuhw->bhrb_users)
- ppmu->config_bhrb(cpuhw->bhrb_filter);
+ ppmu->config_bhrb(cpuhw->bhrb_hw_filter);
local_irq_restore(flags);
}
@@ -1191,6 +1306,26 @@ static int collect_events(struct perf_event *group, int max_count,
return n;
}
+/* SW based branch filters */
+static u64 branch_filter_map(u64 branch_sample_type, u64 *filter_mask)
+{
+ u64 branch_sw_filter = 0;
+
+ if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY) {
+ WARN_ON(*filter_mask != PERF_SAMPLE_BRANCH_ANY);
+ return branch_sw_filter;
+ }
+ if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_RETURN) {
+ branch_sw_filter |= PERF_SAMPLE_BRANCH_ANY_RETURN;
+ *filter_mask |= PERF_SAMPLE_BRANCH_ANY_RETURN;
+ }
+ if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL) {
+ branch_sw_filter |= PERF_SAMPLE_BRANCH_IND_CALL;
+ *filter_mask |= PERF_SAMPLE_BRANCH_IND_CALL;
+ }
+ return branch_sw_filter;
+}
+
/*
* Add a event to the PMU.
* If all events are not already frozen, then we disable and
@@ -1254,8 +1389,11 @@ nocheck:
out:
if (has_branch_stack(event)) {
power_pmu_bhrb_enable(event);
- cpuhw->bhrb_filter = ppmu->bhrb_filter_map(
- event->attr.branch_sample_type);
+
+ cpuhw->bhrb_hw_filter = ppmu->bhrb_filter_map
+ (event->attr.branch_sample_type, &cpuhw->filter_mask);
+ cpuhw->bhrb_sw_filter = branch_filter_map
+ (event->attr.branch_sample_type, &cpuhw->filter_mask);
}
perf_pmu_enable(event->pmu);
@@ -1531,6 +1669,35 @@ static int hw_perf_cache_event(u64 config, u64 *eventp)
return 0;
}
+/* Validate requested filters either in PMU or in SW */
+static int match_filters(u64 branch_sample_type, u64 filter_mask)
+{
+ u64 x;
+
+ if (filter_mask == PERF_SAMPLE_BRANCH_ANY)
+ return true;
+
+ for_each_branch_sample_type(x) {
+ if (!(branch_sample_type & x))
+ continue;
+ /*
+ * Privilege filter requests have been already
+ * taken care during base PMU configuration.
+ */
+ if (x == PERF_SAMPLE_BRANCH_USER)
+ continue;
+ if (x == PERF_SAMPLE_BRANCH_KERNEL)
+ continue;
+ if (x == PERF_SAMPLE_BRANCH_HV)
+ continue;
+
+ /* Requested filter not available */
+ if (!(filter_mask & x))
+ return false;
+ }
+ return true;
+}
+
static int power_pmu_event_init(struct perf_event *event)
{
u64 ev;
@@ -1637,10 +1804,21 @@ static int power_pmu_event_init(struct perf_event *event)
err = power_check_constraints(cpuhw, events, cflags, n + 1);
if (has_branch_stack(event)) {
- cpuhw->bhrb_filter = ppmu->bhrb_filter_map(
- event->attr.branch_sample_type);
+ /* PMU supported branch filters */
+ cpuhw->bhrb_hw_filter = ppmu->bhrb_filter_map
+ (event->attr.branch_sample_type, &cpuhw->filter_mask);
+
+ /* ABI - PMU does not support filter combination */
+ if (cpuhw->bhrb_hw_filter == -1)
+ return -EOPNOTSUPP;
+
+ /* SW supported branch filters */
+ cpuhw->bhrb_sw_filter = branch_filter_map
+ (event->attr.branch_sample_type, &cpuhw->filter_mask);
- if(cpuhw->bhrb_filter == -1)
+ /* ABI - Requested filters are not present */
+ if(!match_filters(event->attr.branch_sample_type,
+ cpuhw->filter_mask))
return -EOPNOTSUPP;
}
diff --git a/arch/powerpc/perf/power8-pmu.c b/arch/powerpc/perf/power8-pmu.c
index 6e28587..e02027b 100644
--- a/arch/powerpc/perf/power8-pmu.c
+++ b/arch/powerpc/perf/power8-pmu.c
@@ -558,9 +558,10 @@ static int power8_generic_events[] = {
[PERF_COUNT_HW_BRANCH_MISSES] = PM_BR_MPRED_CMPL,
};
-static u64 power8_bhrb_filter_map(u64 branch_sample_type)
+static u64 power8_bhrb_filter_map(u64 branch_sample_type, u64 *filter_mask)
{
u64 pmu_bhrb_filter = 0;
+ *filter_mask = 0;
/* BHRB and regular PMU events share the same privilege state
* filter configuration. BHRB is always recorded along with a
@@ -570,15 +571,10 @@ static u64 power8_bhrb_filter_map(u64 branch_sample_type)
*/
/* No branch filter requested */
- if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY)
+ if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY) {
+ *filter_mask = PERF_SAMPLE_BRANCH_ANY;
return pmu_bhrb_filter;
-
- /* Invalid branch filter options - HW does not support */
- if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_RETURN)
- return -1;
-
- if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL)
- return -1;
+ }
/* Invalid branch filter combination - HW does not support */
if ((branch_sample_type & PERF_SAMPLE_BRANCH_ANY_CALL) &&
@@ -587,16 +583,17 @@ static u64 power8_bhrb_filter_map(u64 branch_sample_type)
if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_CALL) {
pmu_bhrb_filter |= POWER8_MMCRA_IFM1;
+ *filter_mask |= PERF_SAMPLE_BRANCH_ANY_CALL;
return pmu_bhrb_filter;
}
if (branch_sample_type & PERF_SAMPLE_BRANCH_COND) {
pmu_bhrb_filter |= POWER8_MMCRA_IFM3;
+ *filter_mask |= PERF_SAMPLE_BRANCH_COND;
return pmu_bhrb_filter;
}
- /* Every thing else is unsupported */
- return -1;
+ return pmu_bhrb_filter;
}
static void power8_config_bhrb(u64 pmu_bhrb_filter)
--
1.7.11.7
Adding documentation support for conditional branch filter.
Signed-off-by: Anshuman Khandual <[email protected]>
Reviewed-by: Stephane Eranian <[email protected]>
---
tools/perf/Documentation/perf-record.txt | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index e297b74..59ca8d0 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -163,12 +163,13 @@ following filters are defined:
- any_call: any function call or system call
- any_ret: any function return or system call return
- ind_call: any indirect branch
+ - cond: conditional branches
- u: only when the branch target is at the user level
- k: only when the branch target is in the kernel
- hv: only when the target is at the hypervisor level
+
-The option requires at least one branch type among any, any_call, any_ret, ind_call.
+The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
The privilege levels may be omitted, in which case, the privilege levels of the associated
event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
levels are subject to permissions. When sampling on multiple events, branch stack sampling
--
1.7.11.7
Adding perf record support for new branch stack filter criteria
PERF_SAMPLE_BRANCH_COND.
Signed-off-by: Anshuman Khandual <[email protected]>
---
tools/perf/builtin-record.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index ecca62e..802d11d 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -625,6 +625,7 @@ static const struct branch_mode branch_modes[] = {
BRANCH_OPT("any_call", PERF_SAMPLE_BRANCH_ANY_CALL),
BRANCH_OPT("any_ret", PERF_SAMPLE_BRANCH_ANY_RETURN),
BRANCH_OPT("ind_call", PERF_SAMPLE_BRANCH_IND_CALL),
+ BRANCH_OPT("cond", PERF_SAMPLE_BRANCH_COND),
BRANCH_END
};
--
1.7.11.7
Enables conditional branch filter support for POWER8
utilizing MMCRA register based filter and also invalidates
a BHRB branch filter combination involving conditional
branches.
Signed-off-by: Anshuman Khandual <[email protected]>
---
arch/powerpc/perf/power8-pmu.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/arch/powerpc/perf/power8-pmu.c b/arch/powerpc/perf/power8-pmu.c
index 2ee4a70..6e28587 100644
--- a/arch/powerpc/perf/power8-pmu.c
+++ b/arch/powerpc/perf/power8-pmu.c
@@ -580,11 +580,21 @@ static u64 power8_bhrb_filter_map(u64 branch_sample_type)
if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL)
return -1;
+ /* Invalid branch filter combination - HW does not support */
+ if ((branch_sample_type & PERF_SAMPLE_BRANCH_ANY_CALL) &&
+ (branch_sample_type & PERF_SAMPLE_BRANCH_COND))
+ return -1;
+
if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_CALL) {
pmu_bhrb_filter |= POWER8_MMCRA_IFM1;
return pmu_bhrb_filter;
}
+ if (branch_sample_type & PERF_SAMPLE_BRANCH_COND) {
+ pmu_bhrb_filter |= POWER8_MMCRA_IFM3;
+ return pmu_bhrb_filter;
+ }
+
/* Every thing else is unsupported */
return -1;
}
--
1.7.11.7
2013/8/30 Anshuman Khandual <[email protected]>
>
> This patchset is the re-spin of the original branch stack sampling
> patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
> also enables SW based branch filtering support for PPC64 platforms which have
> branch stack sampling support. With this new enablement, the branch filter support
> for PPC64 platforms have been extended to include all these combinations discussed
> below with a sample test application program.
>
>
I am trying to understand which HW has support for capturing the
branches: PPC7 or PPC8.
Then it seems you're saying that only PPC8 has the filtering support.
On PPC7 you use the
SW filter. Did I get this right?
I will look at the patch set.
>
> (1) perf record -e branch-misses:u -b ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... ..................... .................... .....................
> #
> 4.42% cprog cprog [k] sw_4_2 cprog [k] lr_addr
> 4.41% cprog cprog [k] symbol2 cprog [k] hw_1_2
> 4.41% cprog cprog [k] ctr_addr cprog [k] sw_4_1
> 4.41% cprog cprog [k] lr_addr cprog [k] sw_4_2
> 4.41% cprog cprog [k] sw_4_2 cprog [k] callme
> 4.41% cprog cprog [k] symbol1 cprog [k] hw_1_1
> 4.41% cprog cprog [k] success_3_1_3 cprog [k] sw_3_1
> 2.43% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
> 2.43% cprog cprog [k] hw_1_2 cprog [k] symbol2
> 2.43% cprog cprog [k] callme cprog [k] hw_1_2
> 2.43% cprog cprog [k] address1 cprog [k] back1
> 2.43% cprog cprog [k] back1 cprog [k] callme
> 2.43% cprog cprog [k] hw_2_1 cprog [k] address1
> 2.43% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
> 2.43% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
> 2.43% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
> 2.43% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_1
> 2.43% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_2
> 2.43% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_3
> 2.43% cprog cprog [k] callme cprog [k] sw_3_1
> 2.43% cprog cprog [k] callme cprog [k] sw_4_2
> 2.43% cprog cprog [k] hw_1_1 cprog [k] symbol1
> 2.43% cprog cprog [k] callme cprog [k] hw_1_1
> 2.42% cprog cprog [k] sw_3_1 cprog [k] callme
> 1.99% cprog cprog [k] success_3_1_1 cprog [k] sw_3_1
> 1.99% cprog cprog [k] sw_3_1 cprog [k] success_3_1_1
> 1.99% cprog cprog [k] address2 cprog [k] back2
> 1.99% cprog cprog [k] hw_2_2 cprog [k] address2
> 1.99% cprog cprog [k] back2 cprog [k] callme
> 1.99% cprog cprog [k] callme cprog [k] main
> 1.99% cprog cprog [k] sw_3_1 cprog [k] success_3_1_3
> 1.99% cprog cprog [k] hw_1_1 cprog [k] callme
> 1.99% cprog cprog [k] sw_3_2 cprog [k] callme
> 1.99% cprog cprog [k] callme cprog [k] sw_3_2
> 1.99% cprog cprog [k] success_3_1_2 cprog [k] sw_3_1
> 1.99% cprog cprog [k] sw_3_1 cprog [k] success_3_1_2
> 1.99% cprog cprog [k] hw_1_2 cprog [k] callme
> 1.99% cprog cprog [k] sw_4_1 cprog [k] callme
> 0.02% cprog [unknown] [k] 0xf7ba2328 [unknown] [k] 0xf7ba2320
> 0.00% cprog libc-2.11.2.so [k] _IO_file_overflow libc-2.11.2.so [k] _IO_file_overflow
> 0.00% cprog libc-2.11.2.so [k] _IO_file_xsputn libc-2.11.2.so [k] _IO_file_overflow
> 0.00% cprog cprog [k] callme cprog [k] hw_2_2
>
> PMU filters
> -----------
> (2) perf record -e branch-misses:u -j any_call ./cprog
>
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... ....................... .................... ......................
> #
> 7.82% cprog cprog [k] sw_3_1 cprog [k] success_3_1_2
> 6.88% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_2
> 6.88% cprog cprog [k] hw_1_1 cprog [k] symbol1
> 5.88% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_1
> 5.88% cprog cprog [k] callme cprog [k] hw_1_1
> 5.88% cprog cprog [k] sw_3_1 cprog [k] success_3_1_1
> 5.88% cprog cprog [k] sw_3_1 cprog [k] sw_3_1_3
> 5.88% cprog cprog [k] callme cprog [k] hw_1_2
> 5.88% cprog cprog [k] hw_1_2 cprog [k] symbol2
> 5.88% cprog cprog [k] sw_4_2 cprog [k] lr_addr
> 5.88% cprog cprog [k] callme cprog [k] sw_4_2
> 4.88% cprog cprog [k] sw_3_1 cprog [k] success_3_1_3
> 4.88% cprog cprog [k] callme cprog [k] sw_3_2
> 4.88% cprog cprog [k] callme cprog [k] hw_2_2
> 3.94% cprog cprog [k] callme cprog [k] sw_3_1
> 3.94% cprog cprog [k] callme cprog [k] hw_2_1
> 2.94% cprog cprog [k] main cprog [k] callme
> 2.94% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
> 2.94% cprog cprog [k] callme cprog [k] sw_4_1
> 0.01% cprog [unknown] [k] 0xf79076c4 [unknown] [k] 0xf78f22c0
> 0.00% cprog libc-2.11.2.so [k] _IO_file_doallocate libc-2.11.2.so [k] _IO_setb
> 0.00% cprog libc-2.11.2.so [k] _IO_file_doallocate libc-2.11.2.so [k] mmap
> 0.00% cprog libc-2.11.2.so [k] _IO_file_xsputn libc-2.11.2.so [k] _IO_default_xsputn
> 0.00% cprog libc-2.11.2.so [k] _IO_file_overflow libc-2.11.2.so [k] _IO_do_write
> 0.00% cprog ld-2.11.2.so [k] malloc [unknown] [k] 0xf790b380
>
>
> (3) perf record -e branch-misses:u -j cond ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... .................. .................... .......................
> #
> 24.85% cprog [unknown] [k] 00000000 cprog [k] callme
> 15.71% cprog cprog [k] sw_3_1 cprog [k] sw_3_1
> 7.14% cprog cprog [k] sw_4_2 cprog [k] lr_addr
> 6.57% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
> 4.57% cprog cprog [k] hw_2_2 cprog [k] callme
> 4.57% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
> 4.57% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
> 4.57% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
> 4.57% cprog cprog [k] main cprog [k] hw_1_1
> 4.57% cprog cprog [k] hw_1_2 cprog [k] hw_1_2
> 4.57% cprog [unknown] [k] 00000000 cprog [k] main
> 4.57% cprog cprog [k] hw_2_1 cprog [k] callme
> 4.57% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
> 4.57% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
> 0.01% cprog [unknown] [k] 0xf7aa25dc [unknown] [k] 0xf7aa27e4
> 0.00% cprog libc-2.11.2.so [k] _IO_doallocbuf libc-2.11.2.so [k] _IO_file_doallocate
> 0.00% cprog [unknown] [k] 00000000 libc-2.11.2.so [k] _IO_file_doallocate
> 0.00% cprog [unknown] [k] 00000000 libc-2.11.2.so [k] _IO_file_stat
>
> SW filters
> ----------
> (4) perf record -e branch-misses:u -j any_ret ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... ................. .................... ..............
> #
> 7.91% cprog cprog [k] symbol1 cprog [k] hw_1_1
> 7.91% cprog cprog [k] success_3_1_3 cprog [k] sw_3_1
> 7.91% cprog cprog [k] ctr_addr cprog [k] sw_4_1
> 7.91% cprog cprog [k] lr_addr cprog [k] sw_4_2
> 7.91% cprog cprog [k] symbol2 cprog [k] hw_1_2
> 7.90% cprog cprog [k] sw_4_2 cprog [k] callme
> 4.34% cprog cprog [k] success_3_1_2 cprog [k] sw_3_1
> 4.33% cprog cprog [k] sw_4_1 cprog [k] callme
> 4.33% cprog cprog [k] hw_1_2 cprog [k] callme
> 4.33% cprog cprog [k] success_3_1_1 cprog [k] sw_3_1
> 4.33% cprog cprog [k] sw_3_2 cprog [k] callme
> 4.33% cprog cprog [k] back2 cprog [k] callme
> 4.33% cprog cprog [k] callme cprog [k] main
> 4.33% cprog cprog [k] hw_1_1 cprog [k] callme
> 3.58% cprog cprog [k] sw_3_1 cprog [k] callme
> 3.58% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
> 3.58% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
> 3.58% cprog cprog [k] back1 cprog [k] callme
> 3.57% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
> 0.00% cprog [unknown] [k] 0xf7abacf4 [unknown] [k] 0xf7abae40
>
>
> (5) perf record -e branch-misses:u -j ind_call ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... ............. .................... .............
> #
> 63.56% cprog cprog [k] sw_4_2 cprog [k] lr_addr
> 36.44% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
>
>
> Mixed filters
> -------------
> (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
> Error:
> The perf.data file has no samples!
>
> NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
> branches in that given set. Both the filters are mutually exclussive, so obviously no samples
> found in the end profile.
>
> (7) perf record -e branch-misses:u -j any_call,ind_call ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... .............. .................... ..............
> #
> 66.69% cprog cprog [k] sw_4_2 cprog [k] lr_addr
> 33.31% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
> 0.00% cprog [unknown] [k] 0x0fe7f264 [unknown] [k] 0x0ff926d0
>
>
> (8) perf record -e branch-misses:u -j any_call,any_ret,ind_call ./cprog
> Error:
> The perf.data file has no samples!
>
> (9) perf record -e branch-misses:u -j cond,any_ret ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... .............. .................... .......................
> #
> 46.01% cprog [unknown] [k] 00000000 cprog [k] callme
> 13.54% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
> 8.18% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
> 8.07% cprog [unknown] [k] 00000000 cprog [k] main
> 8.07% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
> 8.07% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
> 8.07% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
> 0.00% cprog [unknown] [k] 00000000 [unknown] [k] 0xf7c1480c
> 0.00% cprog libc-2.11.2.so [k] mmap libc-2.11.2.so [k] _IO_file_doallocate
>
> (10) perf record -e branch-misses:u -j cond,ind_call ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... .............. .................... ..............
> #
> 48.11% cprog [unknown] [k] 00000000 cprog [k] callme
> 13.52% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
> 12.42% cprog cprog [k] sw_4_2 cprog [k] lr_addr
> 8.65% cprog [unknown] [k] 00000000 cprog [k] main
> 8.65% cprog cprog [k] sw_4_1 cprog [k] ctr_addr
> 8.65% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
> 0.00% cprog [unknown] [k] 00000000 [unknown] [k] 0xf7a4581c
>
>
> (11) perf record -e branch-misses:u -j cond,any_ret,ind_call ./cprog
> # Overhead Command Source Shared Object Source Symbol Target Shared Object Target Symbol
> # ........ ....... .................... .............. .................... .................
> #
> 45.91% cprog [unknown] [k] 00000000 cprog [k] callme
> 13.26% cprog [unknown] [k] 00000000 cprog [k] sw_4_2
> 8.17% cprog cprog [k] sw_3_1_3 cprog [k] sw_3_1
> 8.17% cprog [unknown] [k] 00000000 cprog [k] sw_4_1
> 8.17% cprog cprog [k] sw_3_1_2 cprog [k] sw_3_1
> 8.17% cprog [unknown] [k] 00000000 cprog [k] main
> 8.16% cprog cprog [k] sw_3_1_1 cprog [k] sw_3_1
> 0.00% cprog [unknown] [k] 00000000 [unknown] [k] 0xf7f87704
> 0.00% cprog [unknown] [k] 00000000 libc-2.11.2.so [k] _IO_file_sync
>
> Test application program
> ========================
> (1) Makefile:
> --------------------------------------------
> all: sample.o cprog of.cprog of.sample
>
> sample.o: sample.s
> as -o sample.o sample.s
> cprog: cprog.c sample.o
> gcc -o cprog cprog.c sample.o
> of.sample: sample.o
> objdump -d sample.o > of.sample
> of.cprog: cprog
> objdump -d cprog > of.cprog
> clean:
> rm sample.o cprog of.sample of.cprog
> ---------------------------------------------
> (2) cprog.c
> ---------------------------------------------
> #include <stdio.h>
> #define LOOP_COUNT 100000
>
> extern void callme(void);
>
> int main(int argc, char *argv[])
> {
> int i;
> for(i = 0; i < LOOP_COUNT; i++)
> callme();
>
> printf("end");
> return 0;
> }
> ---------------------------------------------
> (3) sample.S
> ---------------------------------------------
> # r25, r26, r27 will be used as first level, second level
> # and third level stack for LR. Register r20, r21, r22, r23
> # r24 will be used for general programming purpose.
>
> .data
>
> msg:
> .string "BHRB filter tests\n"
> len = . - msg
> msg_1_1:
> .string "Test: hw_1_1\n"
> len_1_1 = 13
> msg_1_2:
> .string "Test: hw_1_2\n"
> len_1_2 = 13
> msg_2_1:
> .string "Test: hw_2_1\n"
> len_2_1 = 13
> msg_2_2:
> .string "Test: hw_2_2\n"
> len_2_2 = 13
> msg_3_1:
> .string "Test: sw_3_1\n"
> len_3_1 = 13
> msg_3_1_1:
> .string "Test: sw_3_1_1\n"
> len_3_1_1 = 15
> msg_3_1_2:
> .string "Test: sw_3_1_2\n"
> len_3_1_2 = 15
> msg_3_1_3:
> .string "Test: sw_3_1_3\n"
> len_3_1_3 = 15
> msg_3_2:
> .string "Test: sw_3_2\n"
> len_3_3 = 13
> msg_4_1:
> .string "Test: sw_4_1\n"
> len_4_1 = 13
> msg_4_2:
> .string "Test: sw_4_2\n"
> len_4_2 = 13
>
> hw_3_1_1_passed:
> .string "\thw_3_1_1_passed\n\n"
> len_hw_3_1_1_passed = 18
> hw_3_1_2_passed:
> .string "\thw_3_1_2_passed\n\n"
> len_hw_3_1_2_passed = 18
> hw_3_1_3_passed:
> .string "\thw_3_1_3_passed\n\n"
> len_hw_3_1_3_passed = 18
>
> hw_2_1_passed:
> .string "\thw_2_1_passed\n\n"
> len_hw_2_1_passed = 16
>
> hw_2_2_passed:
> .string "\thw_2_2_passed\n\n"
> len_hw_2_2_passed = 16
>
> hw_1_1_passed:
> .string "\thw_1_1_passed\n\n"
> len_hw_1_1_passed = 16
>
> hw_1_2_passed:
> .string "\thw_1_2_passed\n\n"
> len_hw_1_2_passed = 16
>
> hw_4_1_passed:
> .string "\thw_4_1_passed\n\n"
> len_hw_4_1_passed = 16
>
> hw_4_2_passed:
> .string "\thw_4_2_passed\n\n"
> len_hw_4_2_passed = 16
>
> msg_error:
> .string "\tError\n"
> len_error = 7
> .text
> .global callme
> .global hw_1_1
> .global hw_1_2
> .global hw_2_1
> .global hw_2_2
>
> # HW filter test symbols
> symbol1:
> # Print "hw_1_1_passed"
> li 0, 4
> li 3, 1
> lis 4, hw_1_1_passed@ha
> addi 4, 4, hw_1_1_passed@l
> li 5, len_hw_1_1_passed
> sc
>
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> hw_1_1:
> # Save LR - second level
> mflr 26
>
> # Print "hw_1_1 called"
> li 0, 4
> li 3, 1
> lis 4, msg_1_1@ha
> addi 4, 4, msg_1_1@l
> li 5, len_1_1
> sc
>
> bl symbol1 # PERF_SAMPLE_BRANCH_ANY_CALL
>
> # Restore LR
> mtlr 26
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> symbol2:
> # Print "Symbol2 taken"
> li 0, 4
> li 3, 1
> lis 4, hw_1_2_passed@ha
> addi 4, 4, hw_1_2_passed@l
> li 5, len_hw_1_2_passed
> sc
>
> blr # PERF_SAMPLE_BRANCH_ANY_RET
> hw_1_2:
> # Save LR - second level
> mflr 26
>
> # Print "hw_1_2 called"
> li 0, 4
> li 3, 1
> lis 4, msg_1_2@ha
> addi 4, 4, msg_1_2@l
> li 5, len_1_2
> sc
>
> li 4,20
> cmpi 0,4,20
> bcl 12, 4*cr0+2, symbol2 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
>
> mtlr 26
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> # HW filter test
>
> address1:
> # Print "hw_2_1_passed"
> li 0, 4
> li 3, 1
> lis 4, hw_2_1_passed@ha
> addi 4, 4, hw_2_1_passed@l
> li 5, len_hw_2_1_passed
> sc
> b back1 # PERF_SAMPLE_BRANCH_ANY
>
> hw_2_1:
> # Print "hw_2_1 called"
> li 0, 4
> li 3, 1
> lis 4, msg_2_1@ha
> addi 4, 4, msg_2_1@l
> li 5, len_2_1
> sc
>
> # Simple conditional branch (equal)
> li 20, 12
> cmpi 3, 20, 12
> bc 12, 4*cr3+2, address1 # PERF_SAMPLE_BRANCH_COND
>
> back1:
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> address2:
> # Print "hw_2_2_passed"
> li 0, 4
> li 3, 1
> lis 4, hw_2_2_passed@ha
> addi 4, 4, hw_2_2_passed@l
> li 5, len_hw_2_2_passed
> sc
> b back2 # PERF_SAMPLE_BRANCH_ANY
>
> hw_2_2:
> # Print "hw_2_2 called"
> li 0, 4
> li 3, 1
> lis 4, msg_2_2@ha
> addi 4, 4, msg_2_2@l
> li 5, len_2_2
> sc
>
> # Simple conditional branch (less than)
> li 20, 12
> cmpi 4, 20, 20
> bc 12, 4*cr4+0, address2 # PERF_SAMPLE_BRANCH_COND
> back2:
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> # SW filter test symbols
> sw_3_1_1:
> # Print "Test: sw_3_1_1"
> li 0, 4
> li 3, 1
> lis 4, msg_3_1_1@ha
> addi 4, 4, msg_3_1_1@l
> li 5, len_3_1_1
> sc
>
> li 22,0
> # Test the condition and return
> li 21, 10
> cmpi 0, 21, 10
> bclr 12, 2 # PERF_SAMPLE_BRANCH_ANY_RET | PERF_SAMPLE_BRANCH_COND
>
> # Should not have come here
> li 0, 4
> li 3, 1
> lis 4, msg_error@ha
> addi 4, 4, msg_error@l
> li 5, len_error
> sc
>
> # Mark the error
> li 22, 1
>
> # Safe fall back
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> sw_3_1_2:
> # Print "Test: sw_3_1_2"
> li 0, 4
> li 3, 1
> lis 4, msg_3_1_2@ha
> addi 4, 4, msg_3_1_2@l
> li 5, len_3_1_2
> sc
>
> li 23, 0
> # Test the condition and return
> li 21, 10
> cmpi 0, 21, 20
> bclr 12, 0 # PERF_SAMPLE_BRANCH_ANY_RET | PERF_SAMPLE_BRANCH_COND
>
> # Should not have come here
> li 0, 4
> li 3, 1
> lis 4, msg_error@ha
> addi 4, 4, msg_error@l
> li 5, len_error
> sc
>
> # Mark the error
> li 23, 1
>
> # Safe fall back
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> sw_3_1_3:
> # Print "Test: sw_3_1_3"
> li 0, 4
> li 3, 1
> lis 4, msg_3_1_3@ha
> addi 4, 4, msg_3_1_3@l
> li 5, len_3_1_3
> sc
>
> li 24, 0
> # Test the condition and return
> li 21, 10
> cmpi 0, 21, 5
> bclr 12, 1 # PERF_SAMPLE_BRANCH_ANY_RET | PERF_SAMPLE_BRANCH_COND
>
> # Mark the error
> li 24, 1
>
> # Should not have come here
> li 0, 4
> li 3, 1
> lis 4, msg_error@ha
> addi 4, 4, msg_error@l
> li 5, len_error
> sc
>
> # Safe fall back
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> success_3_1_1:
> li 0, 4
> li 3, 1
> lis 4, hw_3_1_1_passed@ha
> addi 4, 4, hw_3_1_1_passed@l
> li 5, len_hw_3_1_1_passed
> sc
> blr
>
> success_3_1_2:
> li 0, 4
> li 3, 1
> lis 4, hw_3_1_2_passed@ha
> addi 4, 4, hw_3_1_2_passed@l
> li 5, len_hw_3_1_2_passed
> sc
> blr
>
> success_3_1_3:
> li 0, 4
> li 3, 1
> lis 4, hw_3_1_3_passed@ha
> addi 4, 4, hw_3_1_3_passed@l
> li 5, len_hw_3_1_3_passed
> sc
> blr
>
> sw_3_1:
> # Save LR
> mflr 26
>
> # Print "Test: sw_3_1"
> li 0, 4
> li 3, 1
> lis 4, msg_3_1@ha
> addi 4, 4, msg_3_1@l
> li 5, len_3_1
> sc
>
> # Equal comparison condition
> bl sw_3_1_1 # PERF_SAMPLE_BRANCH_ANY_CALL
> cmpi 0, 22, 0
> bcl 12, 2, success_3_1_1 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
>
> # LT comparison condition
> bl sw_3_1_2 # PERF_SAMPLE_BRANCH_ANY_CALL
> cmpi 0, 23, 0
> bcl 12, 2, success_3_1_2 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
>
> # GT comparison condition
> bl sw_3_1_3 # PERF_SAMPLE_BRANCH_ANY_CALL
> cmpi 0, 24, 0
> bcl 12, 2, success_3_1_3 # PERF_SAMPLE_BRANCH_ANY_CALL | PERF_SAMPLE_BRANCH_COND
>
> mtlr 26
> blr # PERF_SAMPLE_BRANCH_ANY_RET
> sw_3_2:
> # Print "Test: sw_3_2"
> li 0, 4
> li 3, 1
> lis 4, msg_3_2@ha
> addi 4, 4, msg_3_2@l
> li 5, len_3_1
> sc
>
> # FIXME: Anything more here ?
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> # Indirect call tests
>
> # CTR
> ctr_addr:
> # Print "bcctr taken"
> li 0, 4
> li 3, 1
> lis 4, hw_4_1_passed@ha
> addi 4, 4, hw_4_1_passed@l
> li 5, len_hw_4_1_passed
> sc
>
> blr # PERF_SAMPLE_BRANCH_ANY_RET
> sw_4_1:
> # Save LR
> mflr 26
>
> # Print "sw_4_1 called"
> li 0, 4
> li 3, 1
> lis 4, msg_4_1@ha
> addi 4, 4, msg_4_1@l
> li 5, len_4_1
> sc
>
> # Save address in CTR
> lis 20, ctr_addr@ha
> addi 20, 20, ctr_addr@l
> mtctr 20
>
>
> # Compare and jump to CTR
> li 21, 10
> cmpi 0, 21, 10
> bcctrl 12, 4*cr0+2 # PERF_SAMPLE_BRANCH_IND_CALL
>
> mtlr 26
> blr # PERF_SAMPLE_BRANCH_ANY_RET
> # LR
> lr_addr:
> # Print "bclrl taken"
> li 0, 4
> li 3, 1
> lis 4, hw_4_2_passed@ha
> addi 4, 4, hw_4_2_passed@l
> li 5, len_hw_4_2_passed
> sc
>
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> sw_4_2:
> # Save LR
> mflr 26
>
> # Print "Test: sw_4_2"
> li 0, 4
> li 3, 1
> lis 4, msg_4_2@ha
> addi 4, 4, msg_4_2@l
> li 5, len_4_2
> sc
>
> # Save address in LR
> lis 20, lr_addr@ha
> addi 20, 20, lr_addr@l
> mtlr 20
>
>
> # Compare and jump to CTR
> li 21, 10
> cmpi 0, 21, 10
> bclrl 12, 4*cr0+2 # PERF_SAMPLE_BRANCH_IND_CALL
>
> # Restore LR
> mtlr 26
> blr # PERF_SAMPLE_BRANCH_ANY_RET
>
> callme:
> # Save LR
> mflr 25
>
> # Print "Branch filter Test"
> li 0, 4
> li 3, 1
> lis 4, msg@ha
> addi 4, 4, msg@l
> li 5, len
> sc
>
> # PERF_SAMPLE_BRANCH_ANY_CALL
> bl hw_1_1 # PERF_SAMPLE_BRANCH_ANY_CALL
> bl hw_1_2 # PERF_SAMPLE_BRANCH_ANY_CALL
> # PERF_SAMPLE_BRANCH_COND
> bl hw_2_1 # PERF_SAMPLE_BRANCH_ANY_CALL
> bl hw_2_2 # PERF_SAMPLE_BRANCH_ANY_CALL
>
> # PERF_SAMPLE_BRANCH_ANY_RET
> bl sw_3_1 # PERF_SAMPLE_BRANCH_ANY_CALL
> bl sw_3_2 # PERF_SAMPLE_BRANCH_ANY_CALL
> # PERF_SAMPLE_BRANCH_IND_CALL
> bl sw_4_1 # PERF_SAMPLE_BRANCH_ANY_CALL
> bl sw_4_2 # PERF_SAMPLE_BRANCH_ANY_CALL
>
> # Restore LR
> mtlr 25
> blr # PERF_SAMPLE_BRANCH_ANY_RET
> --------------------------------------------------------------------
>
> Changes in V2
> --------------
> (1) Enabled PPC64 SW branch filtering support
> (2) Incorporated changes required for all previous comments
>
> Anshuman Khandual (6):
> perf: New conditional branch filter criteria in branch stack sampling
> powerpc, perf: Enable conditional branch filter for POWER8
> perf, tool: Conditional branch filter 'cond' added to perf record
> x86, perf: Add conditional branch filtering support
> perf, documentation: Description for conditional branch filter
> powerpc, perf: Enable SW filtering in branch stack sampling framework
>
> arch/powerpc/include/asm/perf_event_server.h | 2 +-
> arch/powerpc/perf/core-book3s.c | 200 +++++++++++++++++++++++++--
> arch/powerpc/perf/power8-pmu.c | 25 ++--
> arch/x86/kernel/cpu/perf_event_intel_lbr.c | 5 +
> include/uapi/linux/perf_event.h | 3 +-
> tools/perf/Documentation/perf-record.txt | 3 +-
> tools/perf/builtin-record.c | 1 +
> 7 files changed, 216 insertions(+), 23 deletions(-)
>
> --
> 1.7.11.7
>
On 08/30/2013 05:18 PM, Stephane Eranian wrote:
> 2013/8/30 Anshuman Khandual <[email protected]>
>> >
>> > This patchset is the re-spin of the original branch stack sampling
>> > patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>> > also enables SW based branch filtering support for PPC64 platforms which have
>> > branch stack sampling support. With this new enablement, the branch filter support
>> > for PPC64 platforms have been extended to include all these combinations discussed
>> > below with a sample test application program.
>> >
>> >
> I am trying to understand which HW has support for capturing the
> branches: PPC7 or PPC8.
> Then it seems you're saying that only PPC8 has the filtering support.
> On PPC7 you use the
> SW filter. Did I get this right?
>
> I will look at the patch set.
>
Hey Stephane,
POWER7 does not have BHRB support required to capture the branches. Right
now its only POWER8 (which has BHRB) can capture branches in HW. It has some
PMU level branch filters and rest we have implemented in SW. But these SW
filters cannot be applied in POWER7 as it does not support branch stack
sampling because of lack of BHRB. I have mentioned PPC64 support in the
sense that this SW filtering code could be used in existing or future generation
powerpc processors which would have PMU support for branch stack sampling. My
apologies if the description for the patchset was ambiguous.
Regards
Anshuman
On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
> This patchset is the re-spin of the original branch stack sampling
> patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
> also enables SW based branch filtering support for PPC64 platforms which have
> branch stack sampling support. With this new enablement, the branch filter support
> for PPC64 platforms have been extended to include all these combinations discussed
> below with a sample test application program.
...
> Mixed filters
> -------------
> (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
> Error:
> The perf.data file has no samples!
>
> NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
> branches in that given set. Both the filters are mutually exclussive, so obviously no samples
> found in the end profile.
The semantics of multiple filters is not clear to me. It could be an OR,
or an AND. You have implemented AND, does that match existing behaviour
on x86 for example?
cheers
On 09/10/2013 07:36 AM, Michael Ellerman wrote:
> On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
>> This patchset is the re-spin of the original branch stack sampling
>> patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>> also enables SW based branch filtering support for PPC64 platforms which have
>> branch stack sampling support. With this new enablement, the branch filter support
>> for PPC64 platforms have been extended to include all these combinations discussed
>> below with a sample test application program.
>
> ...
>
>> Mixed filters
>> -------------
>> (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
>> Error:
>> The perf.data file has no samples!
>>
>> NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
>> branches in that given set. Both the filters are mutually exclussive, so obviously no samples
>> found in the end profile.
>
> The semantics of multiple filters is not clear to me. It could be an OR,
> or an AND. You have implemented AND, does that match existing behaviour
> on x86 for example?
I believe it does match. X86 code drops the branch records (originally captured
in the LBR) while applying the SW filters.
Regards
Anshuman
On 08/30/2013 05:18 PM, Stephane Eranian wrote:
> 2013/8/30 Anshuman Khandual <[email protected]>
>> >
>> > This patchset is the re-spin of the original branch stack sampling
>> > patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>> > also enables SW based branch filtering support for PPC64 platforms which have
>> > branch stack sampling support. With this new enablement, the branch filter support
>> > for PPC64 platforms have been extended to include all these combinations discussed
>> > below with a sample test application program.
>> >
>> >
> I am trying to understand which HW has support for capturing the
> branches: PPC7 or PPC8.
> Then it seems you're saying that only PPC8 has the filtering support.
> On PPC7 you use the
> SW filter. Did I get this right?
>
> I will look at the patch set.
>
Hey Stephane,
Just wondering if you got a chance to go though the patchset ?
Regards
Anshuman
On 09/21/2013 12:11 PM, Anshuman Khandual wrote:
> On 08/30/2013 05:18 PM, Stephane Eranian wrote:
>> 2013/8/30 Anshuman Khandual <[email protected]>
>>>>
>>>> This patchset is the re-spin of the original branch stack sampling
>>>> patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>>>> also enables SW based branch filtering support for PPC64 platforms which have
>>>> branch stack sampling support. With this new enablement, the branch filter support
>>>> for PPC64 platforms have been extended to include all these combinations discussed
>>>> below with a sample test application program.
>>>>
>>>>
>> I am trying to understand which HW has support for capturing the
>> branches: PPC7 or PPC8.
>> Then it seems you're saying that only PPC8 has the filtering support.
>> On PPC7 you use the
>> SW filter. Did I get this right?
>>
>> I will look at the patch set.
>>
>
> Hey Stephane,
>
> Just wondering if you got a chance to go though the patchset ?
s/though/through/
On Tue, Sep 10, 2013 at 4:06 AM, Michael Ellerman
<[email protected]> wrote:
>
> On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
> > This patchset is the re-spin of the original branch stack sampling
> > patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
> > also enables SW based branch filtering support for PPC64 platforms which have
> > branch stack sampling support. With this new enablement, the branch filter support
> > for PPC64 platforms have been extended to include all these combinations discussed
> > below with a sample test application program.
>
> ...
>
> > Mixed filters
> > -------------
> > (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
> > Error:
> > The perf.data file has no samples!
> >
> > NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
> > branches in that given set. Both the filters are mutually exclussive, so obviously no samples
> > found in the end profile.
>
> The semantics of multiple filters is not clear to me. It could be an OR,
> or an AND. You have implemented AND, does that match existing behaviour
> on x86 for example?
>
The semantic on the API is OR. AND does not make sense: CALL & RETURN?
On x86, the HW filter is an OR (default: ALL, set bit to disable a
type). I suspect
it is similar on PPC.
>
> cheers
>
>
On 09/21/2013 12:25 PM, Stephane Eranian wrote:
> On Tue, Sep 10, 2013 at 4:06 AM, Michael Ellerman
> <[email protected]> wrote:
>> >
>> > On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
>>> > > This patchset is the re-spin of the original branch stack sampling
>>> > > patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>>> > > also enables SW based branch filtering support for PPC64 platforms which have
>>> > > branch stack sampling support. With this new enablement, the branch filter support
>>> > > for PPC64 platforms have been extended to include all these combinations discussed
>>> > > below with a sample test application program.
>> >
>> > ...
>> >
>>> > > Mixed filters
>>> > > -------------
>>> > > (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
>>> > > Error:
>>> > > The perf.data file has no samples!
>>> > >
>>> > > NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
>>> > > branches in that given set. Both the filters are mutually exclussive, so obviously no samples
>>> > > found in the end profile.
>> >
>> > The semantics of multiple filters is not clear to me. It could be an OR,
>> > or an AND. You have implemented AND, does that match existing behaviour
>> > on x86 for example?
>> >
> The semantic on the API is OR. AND does not make sense: CALL & RETURN?
> On x86, the HW filter is an OR (default: ALL, set bit to disable a
> type). I suspect
> it is similar on PPC.
Hey Stephane,
In POWER8 BHRB, we have got three HW PMU filters out of which we are trying
to use two of them PERF_SAMPLE_BRANCH_ANY_CALL and PERF_SAMPLE_BRANCH_COND
respectively.
(1) These filters are exclusive of each other and cannot be OR-ed with each other
(2) The SW filters are applied on the branch record set captured with BHRB
which have the HW filters applied. So the working set is already reduced
with the HW PMU filters. SW filter goes through the working set and figures
out which one of them satisfy the SW filter criteria and gets picked up. The
SW filter cannot find out branches records which matches the criteria outside
of BHRB captured set. So we cannot OR the filters.
This makes the combination of HW and SW filter inherently an "AND" not OR.
(3) But once we have captured the BHRB filtered data with HW PMU filter, multiple SW
filters (if requested) can be applied either in OR or AND manner.
It should be either like
(1) (HW_FILTER_1) && (SW_FILTER_1) && (SW_FILTER_2)
or like
(2) (HW_FILTER_1) && (SW_FILTER_1 || SW_FILTER_2)
NOTE: I admit that the current validate_instruction() function does not do
either of them correctly. Will fix it in the next iteration.
(4) These combination of filters are not supported right now because
(a) We are unable to process two HW PMU filters simultaneously
(b) We have not worked on replacement SW filter for either of the HW filters
(1) (HW_FILTER_1), (HW_FILTER_2)
(2) (HW_FILTER_1), (HW_FILTER_2), (SW_FILTER_1)
(3) (HW_FILTER_1), (HW_FILTER_2), (SW_FILTER_1), (SW_FILTER_2)
How ever these combination of filters can be supported right now.
(1) (HW_FILTER_1)
(2) (HW_FILTER_2)
(3) (SW_FILTER_1)
(4) (SW_FILTER_2)
(5) (SW_FILTER_1), (SW_FILTER_2)
(6) (HW_FILTER_1), (SW_FILTER_1)
(7) (HW_FILTER_1), (SW_FILTER_2)
(8) (HW_FILTER_1), (SW_FILTER_1), (SW_FILTER_2)
(9) (HW_FILTER_2), (SW_FILTER_1)
(10) (HW_FILTER_2), (SW_FILTER_2)
(11) (HW_FILTER_2), (SW_FILTER_1), (SW_FILTER_2)
Given the situation as explained here, which semantic would be better for single
HW and multiple SW filters. Accordingly validate_instruction() function will have
to be re-implemented. But I believe OR-ing the SW filters will be preferable.
(1) (HW_FILTER_1) && (SW_FILTER_1) && (SW_FILTER_2)
or
(2) (HW_FILTER_1) && (SW_FILTER_1 || SW_FILTER_2)
Please let me know your inputs and suggestions on this. Thank you.
Regards
Anshuman
On Mon, 2013-09-23 at 14:45 +0530, Anshuman Khandual wrote:
> On 09/21/2013 12:25 PM, Stephane Eranian wrote:
> > On Tue, Sep 10, 2013 at 4:06 AM, Michael Ellerman
> > <[email protected]> wrote:
> >> >
> >> > On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
> >>> > > This patchset is the re-spin of the original branch stack sampling
> >>> > > patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
> >>> > > also enables SW based branch filtering support for PPC64 platforms which have
> >>> > > branch stack sampling support. With this new enablement, the branch filter support
> >>> > > for PPC64 platforms have been extended to include all these combinations discussed
> >>> > > below with a sample test application program.
> >> >
> >> > ...
> >> >
> >>> > > Mixed filters
> >>> > > -------------
> >>> > > (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
> >>> > > Error:
> >>> > > The perf.data file has no samples!
> >>> > >
> >>> > > NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
> >>> > > branches in that given set. Both the filters are mutually exclussive, so obviously no samples
> >>> > > found in the end profile.
> >> >
> >> > The semantics of multiple filters is not clear to me. It could be an OR,
> >> > or an AND. You have implemented AND, does that match existing behaviour
> >> > on x86 for example?
> >
> > The semantic on the API is OR. AND does not make sense: CALL & RETURN?
> > On x86, the HW filter is an OR (default: ALL, set bit to disable a
> > type). I suspect
> > it is similar on PPC.
>
> Given the situation as explained here, which semantic would be better for single
> HW and multiple SW filters. Accordingly validate_instruction() function will have
> to be re-implemented. But I believe OR-ing the SW filters will be preferable.
>
> (1) (HW_FILTER_1) && (SW_FILTER_1) && (SW_FILTER_2)
> or
> (2) (HW_FILTER_1) && (SW_FILTER_1 || SW_FILTER_2)
>
> Please let me know your inputs and suggestions on this. Thank you.
You need to implement the correct semantics, regardless of how the
hardware happens to work.
That means if multiple filters are specified you need to do all the
filtering in software.
cheers
On 09/25/2013 07:49 AM, Michael Ellerman wrote:
> On Mon, 2013-09-23 at 14:45 +0530, Anshuman Khandual wrote:
>> On 09/21/2013 12:25 PM, Stephane Eranian wrote:
>>> On Tue, Sep 10, 2013 at 4:06 AM, Michael Ellerman
>>> <[email protected]> wrote:
>>>>>
>>>>> On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
>>>>>>> This patchset is the re-spin of the original branch stack sampling
>>>>>>> patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>>>>>>> also enables SW based branch filtering support for PPC64 platforms which have
>>>>>>> branch stack sampling support. With this new enablement, the branch filter support
>>>>>>> for PPC64 platforms have been extended to include all these combinations discussed
>>>>>>> below with a sample test application program.
>>>>>
>>>>> ...
>>>>>
>>>>>>> Mixed filters
>>>>>>> -------------
>>>>>>> (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
>>>>>>> Error:
>>>>>>> The perf.data file has no samples!
>>>>>>>
>>>>>>> NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
>>>>>>> branches in that given set. Both the filters are mutually exclussive, so obviously no samples
>>>>>>> found in the end profile.
>>>>>
>>>>> The semantics of multiple filters is not clear to me. It could be an OR,
>>>>> or an AND. You have implemented AND, does that match existing behaviour
>>>>> on x86 for example?
>>>
>>> The semantic on the API is OR. AND does not make sense: CALL & RETURN?
>>> On x86, the HW filter is an OR (default: ALL, set bit to disable a
>>> type). I suspect
>>> it is similar on PPC.
>>
>> Given the situation as explained here, which semantic would be better for single
>> HW and multiple SW filters. Accordingly validate_instruction() function will have
>> to be re-implemented. But I believe OR-ing the SW filters will be preferable.
>>
>> (1) (HW_FILTER_1) && (SW_FILTER_1) && (SW_FILTER_2)
>> or
>> (2) (HW_FILTER_1) && (SW_FILTER_1 || SW_FILTER_2)
>>
>> Please let me know your inputs and suggestions on this. Thank you.
>
> You need to implement the correct semantics, regardless of how the
> hardware happens to work.
>
> That means if multiple filters are specified you need to do all the
> filtering in software.
Hello Stephane,
I looked at the X86 code on branch filtering implementation.
(1) During event creation intel_pmu_hw_config calls intel_pmu_setup_lbr_filter when LBR sampling
is required, intel_pmu_setup_lbr_filter calls these two functions
(a) intel_pmu_setup_sw_lbr_filter
"event->hw.branch_reg.reg" contains all the SW filter masks which can be
supported for the user requested filters event->attr.branch_sample_type (even
if some of them could implemented in PMU HW)
(b) intel_pmu_setup_hw_lbr_filter (when HW filtering is present)
"event->hw.branch_reg.config" contains all the PMU HW filter masks corresponding
to the requested filters in event->attr.branch_sample_type. One point to note
here is that if the user has requested for some branch filter which is not supported
in the HW LBR filter, the event creation request is rejected with EOPNOTSUPP. This
not true for the filters which can be ignored in the PMU.
(2) When the event is enabled in the PMU
(a) cpuc->lbr_sel->config gets into the HW register to enable the filtering of branches
which was determined in the function intel_pmu_setup_hw_lbr_filter.
(3) After the IRQ happened, intel_pmu_lbr_read reads all the entries from the LBR HW and then
applies the filter in the function intel_pmu_lbr_filter.
(a) intel_pmu_lbr_filter functions take into account cpuc->br_sel (which is nothing but
event->hw.branch_reg.reg as determined in the function intel_pmu_setup_sw_lbr_filter)
which contains the entire branch filter request set in terms applicable SW filter. Here
the semantic is OR when we look at from SW filter implementation point of view.
BUT what branch record set we are working on right now ? A set which was captured with LBR HW
with cpuc->lbr_sel->config filters enabled on it. So to me the X86 implementation of the semantics
look something like this.
A - Branch filter set requested by the user
B - Subset of A which can be supported in HW
C - Subset of A which can be supported in SW
(B) && (C)
NOTE: Individual filters are OR-ed inside both B and C sets.
So here the semantics is not a true OR. This is my understanding till now which may be wrong. Please
help me understand if the semantics is something otherwise than what is explained above.
In POWER8 because we cannot OR individual HW PMU supported filters, till now the semantics looked a bit odd.
But as Michael has pointed out here that if there are multiple branch filter requests implement all of them
in SW. Only in case where the user requests for an individual filter and if it happen to be supported in HW
PMU, we will use the PMU filters.
Regards
Anshuman
On Mon, Sep 23, 2013 at 11:15 AM, Anshuman Khandual
<[email protected]> wrote:
> On 09/21/2013 12:25 PM, Stephane Eranian wrote:
>> On Tue, Sep 10, 2013 at 4:06 AM, Michael Ellerman
>> <[email protected]> wrote:
>>> >
>>> > On Fri, 2013-08-30 at 09:54 +0530, Anshuman Khandual wrote:
>>>> > > This patchset is the re-spin of the original branch stack sampling
>>>> > > patchset which introduced new PERF_SAMPLE_BRANCH_COND filter. This patchset
>>>> > > also enables SW based branch filtering support for PPC64 platforms which have
>>>> > > branch stack sampling support. With this new enablement, the branch filter support
>>>> > > for PPC64 platforms have been extended to include all these combinations discussed
>>>> > > below with a sample test application program.
>>> >
>>> > ...
>>> >
>>>> > > Mixed filters
>>>> > > -------------
>>>> > > (6) perf record -e branch-misses:u -j any_call,any_ret ./cprog
>>>> > > Error:
>>>> > > The perf.data file has no samples!
>>>> > >
>>>> > > NOTE: As expected. The HW filters all the branches which are calls and SW tries to find return
>>>> > > branches in that given set. Both the filters are mutually exclussive, so obviously no samples
>>>> > > found in the end profile.
>>> >
>>> > The semantics of multiple filters is not clear to me. It could be an OR,
>>> > or an AND. You have implemented AND, does that match existing behaviour
>>> > on x86 for example?
>>> >
>> The semantic on the API is OR. AND does not make sense: CALL & RETURN?
>> On x86, the HW filter is an OR (default: ALL, set bit to disable a
>> type). I suspect
>> it is similar on PPC.
>
> Hey Stephane,
>
> In POWER8 BHRB, we have got three HW PMU filters out of which we are trying
> to use two of them PERF_SAMPLE_BRANCH_ANY_CALL and PERF_SAMPLE_BRANCH_COND
> respectively.
>
> (1) These filters are exclusive of each other and cannot be OR-ed with each other
>
So you are saying that the HW filter is exclusive. That seems odd. But
I think it is
because of the choices is ANY. ANY covers all the types of branches. Therefore
it does not make a difference whether you add COND or not. And
vice-versa, if you
set COND, you need to disable ANY. I bet if you add other filters such
as CALL, RETURN,
then you could OR them and say: I want RETURN or CALLS.
But that's okay. The API operates in OR mode but if the HW does not
support it, you
can check the mask and reject if more than one type is set. That is
arch-specific code.
The alternative, if to only capture ANY and emulate the filter in SW.
This will work, of
course. But the downside, is that you lose the way to appreciate how
many, for instance,
COND branches you sampled out of the total number of COND branches
retired. Unless
you can count COND branches separately.
> (2) The SW filters are applied on the branch record set captured with BHRB
> which have the HW filters applied. So the working set is already reduced
> with the HW PMU filters. SW filter goes through the working set and figures
> out which one of them satisfy the SW filter criteria and gets picked up. The
> SW filter cannot find out branches records which matches the criteria outside
> of BHRB captured set. So we cannot OR the filters.
>
Yes, you can if you set the HW filter to ANY. And then filter the
branches by type
based on the SW mask. You need to decode each sampled branch for that. This
is done in X86 to work around HW bugs in the HW filter, for instance.
> This makes the combination of HW and SW filter inherently an "AND" not OR.
>
> (3) But once we have captured the BHRB filtered data with HW PMU filter, multiple SW
> filters (if requested) can be applied either in OR or AND manner.
>
> It should be either like
> (1) (HW_FILTER_1) && (SW_FILTER_1) && (SW_FILTER_2)
> or like
> (2) (HW_FILTER_1) && (SW_FILTER_1 || SW_FILTER_2)
>
> NOTE: I admit that the current validate_instruction() function does not do
> either of them correctly. Will fix it in the next iteration.
>
Just set the HW filter to ANY and filter in SW.
Isn't that possible?
> (4) These combination of filters are not supported right now because
>
> (a) We are unable to process two HW PMU filters simultaneously
> (b) We have not worked on replacement SW filter for either of the HW filters
>
> (1) (HW_FILTER_1), (HW_FILTER_2)
> (2) (HW_FILTER_1), (HW_FILTER_2), (SW_FILTER_1)
> (3) (HW_FILTER_1), (HW_FILTER_2), (SW_FILTER_1), (SW_FILTER_2)
>
> How ever these combination of filters can be supported right now.
>
> (1) (HW_FILTER_1)
> (2) (HW_FILTER_2)
>
> (3) (SW_FILTER_1)
> (4) (SW_FILTER_2)
> (5) (SW_FILTER_1), (SW_FILTER_2)
>
> (6) (HW_FILTER_1), (SW_FILTER_1)
> (7) (HW_FILTER_1), (SW_FILTER_2)
> (8) (HW_FILTER_1), (SW_FILTER_1), (SW_FILTER_2)
> (9) (HW_FILTER_2), (SW_FILTER_1)
> (10) (HW_FILTER_2), (SW_FILTER_2)
> (11) (HW_FILTER_2), (SW_FILTER_1), (SW_FILTER_2)
>
>
> Given the situation as explained here, which semantic would be better for single
> HW and multiple SW filters. Accordingly validate_instruction() function will have
> to be re-implemented. But I believe OR-ing the SW filters will be preferable.
>
> (1) (HW_FILTER_1) && (SW_FILTER_1) && (SW_FILTER_2)
> or
> (2) (HW_FILTER_1) && (SW_FILTER_1 || SW_FILTER_2)
>
> Please let me know your inputs and suggestions on this. Thank you.
>
> Regards
> Anshuman
>
On 09/26/2013 04:44 PM, Stephane Eranian wrote:
> So you are saying that the HW filter is exclusive. That seems odd. But
> I think it is
> because of the choices is ANY. ANY covers all the types of branches. Therefore
> it does not make a difference whether you add COND or not. And
> vice-versa, if you
> set COND, you need to disable ANY. I bet if you add other filters such
> as CALL, RETURN,
> then you could OR them and say: I want RETURN or CALLS.
>
> But that's okay. The API operates in OR mode but if the HW does not
> support it, you
> can check the mask and reject if more than one type is set. That is
> arch-specific code.
> The alternative, if to only capture ANY and emulate the filter in SW.
> This will work, of
> course. But the downside, is that you lose the way to appreciate how
> many, for instance,
> COND branches you sampled out of the total number of COND branches
> retired. Unless
> you can count COND branches separately.
Hey Stephane,
Thanks for your reply. I am working on a solution where PMU will process
all the requested branch filters in HW only if it can filter all of them in an
OR manner else it will just leave the entire thing upto the SW to process and
do no filtering itself. This implies that branch filtering will either happen
completely in HW or completely in SW and never in a mixed manner. This way
it will conform to the OR mode defined in the API. I will post the revised
patch set soon.
Regards
Anshuman