perfmon doesn't send SIGPROF in kernel 2.5.64
Eric,
Quoted 7 lines · Eric Piel
Quoted 5 lines · Stephane Eranian
The attached patch against 2.5.67 corrects the problem. Note that this patch also includes other updates.
Let me know how it goes.
Very good, it worked!:
I glad to see that it worked for you as well this time.
Quoted 15 lines
Also, after a quick glance at the patch I noticed this update:
-/* pmc8 */ {PFM_REG_CONFIG, 0,0xffffffff3fffffffUL,0xffffffff9fffffffUL, NULL,pfm_mck_pmc_check, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
-/* pmc9 */ {PFM_REG_CONFIG, 0,0xffffffff3ffffffcUL,0xffffffff9fffffffUL, NULL,pfm_mck_pmc_check, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
+/* pmc8 */ {PFM_REG_CONFIG, 0,0xffffffff3fffffffUL,0xffffffff3fffffffUL, NULL,pfm_mck_pmc_check, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
+/* pmc9 */ {PFM_REG_CONFIG, 0,0xffffffff3ffffffcUL,0xffffffff3fffffffUL, NULL,pfm_mck_pmc_check, {0UL,0UL, 0UL, 0UL}, {0UL,0UL, 0UL, 0UL}},
Quoted 4 lines
Could you describe shortly what this is suppose to fix? Recently I had some troubles with some PMU counters on Itanium2 (hard to trigger) and I wonder if this could be solved by that patch...
This only affects measurements made with the opcode matchers. The UIUC folks reported a problem with certain opcode pattern not being counted properly. I was off by one bit in the mask. This has no affect on measurements NOT using the opcode matchers.
Signature · 1 line
-Stephane