Skip to contentSkip to conversation

Performance Monitor Internals:

Linux-ia64 Jonathan Case Nicklin 5 participants 5 messages Nov 7, 2000 – Nov 13, 2000

lore
Share & tools
lore
#0 lore raw

Greetings,

My colleague and I are working with the IA64 performance monitors
and would like to get involved
or at least learn more about the development effort of the syscall
interface and its internals. Has
any person or group been assigned to this task? If so, please email
me or the list so we can
possibly collaborate on the design/implementation effort.

Sincerely,
Jonathan Case Nicklin / John Baboval

Mission Critical Linux
www.missioncriticallinux.com

#1 lore raw

Hi,

Quoted 9 lines

My colleague and I are working with the IA64 performance monitors
and would like to get involved
or at least learn more about the development effort of the syscall
interface and its internals. Has
any person or group been assigned to this task? If so, please email
me or the list so we can
possibly collaborate on the design/implementation effort.

I am currently working on the perfmon support for Linux/ia64 as you may have seen for the recent patches. The goal is to support the entire set of features of the Itanium PMU including EARS and BTB and not just counting. Also support is planned for per process AND system-wide monitoring on both UP and SMP systems.

So I you have any ideas, concerns or want to contribute something, please feel free to contact me.

Regards,

+--------------------------------------------------------------------+
| Ste'phane ERANIAN                       | Email [email protected] |
| Hewlett-Packard Laboratories            |                          |
| 1501, Page Mill Road MS 1U-15           |                          |
| Palo  Alto, CA 94303-096                |                          |
| USA                                     |                          |
| Tel : (650) 857-7174                    |                          |
| Fax : (650) 857-5548                    |                          |
+--------------------------------------------------------------------+
#2 lore raw
In reply to Jonathan Case Nicklin
Quoted 1 line

I am currently working on the perfmon support for Linux/ia64 as you

may

Quoted 1 line

have seen for the recent patches. The goal is to support the entire

set

Quoted 1 line

of features of the Itanium PMU including EARS and BTB and not just

counting.

Quoted 1 line

Also support is planned for per process AND system-wide monitoring

on

Quoted 3 lines

both UP and SMP systems.

So I you have any ideas, concerns or want to contribute something,

please

Quoted 1 line

feel free to contact me.

I'm also interested in using the performance monitor, but my only concern is if user processes will able to access the performance counters directly with a "mov 1,PMD[r2]" instruction, instead of having to use some sort of kernel service to access the counter. Essentially, all I'm looking for is a replacement for the IA-32's "RDTSC" instruction in the code I'm porting.

Ross Ridge

#3 lore raw
In reply to Jonathan Case Nicklin
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Ross Ridge

Ross> I'm also interested in using the performance monitor, but my Ross> only concern is if user processes will able to access the Ross> performance counters directly with a "mov 1,PMD[r2]" Ross> instruction, instead of having to use some sort of kernel Ross> service to access the counter. Essentially, all I'm looking Ross> for is a replacement for the IA-32's "RDTSC" instruction in Ross> the code I'm porting.

You can use the attached code.

	--david

static __inline__ unsigned long
ia64_get_itc (void)

{
	unsigned long result;

	__asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
	while (__builtin_expect ((int) result = -1, 0))
		__asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
	return result;
}
#4 lore raw
In reply to Jonathan Case Nicklin
Quoted 10 lines · [email protected] said:

static __inline__ unsigned long
ia64_get_itc (void)

{
	unsigned long result;

	__asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
	while (__builtin_expect ((int) result = -1, 0))
		__asm__ __volatile__("mov %0=ar.itc" : "=r"(result) :: "memory");
	return result;
}

Why is it no good to return result when it satisfies:

    (result & 0xffffffff) = 0xffffffff

? Curiosity.

		-- Pete