Fix a few fallouts

PATCH In review Daniel Wagner 3 participants 7 messages Jul 19, 2021 – Aug 30, 2021

Share & tools

Daniel Wagner

author

2021-07-19 09:34 UTC #0 lore raw

v3 cover Fix a few fallouts

v1:
- initial version
v2:
- update error message for sched_getaffinity()
v3:
- updated commit message for sched_getaffinity()

Daniel Wagner (1):
rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

 src/lib/rt-numa.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

Signature · 1 line

2.32.0

Daniel Wagner

author

2021-07-19 09:30 UTC #1 lore raw

PATCH v3 1/1 rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

pthread_getaffinity_np() prevents static builds as glibc does not expose it for this configuration. Instead use sched_getaffinity() directly as pthread_getaffinity_np() is just a fancy glibc wrapper around sched_getaffinity().

Static builds helps in in situation where the rootfs is minimal and does for example not ship libnuma.

The pthread_getaffinity_np call was introduced by commit f240656b056b ("rt-tests: cyclictest: Fix -t without a user specified [NUM]")

Signed-off-by Daniel Wagner <[email protected]>

---
 src/lib/rt-numa.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

src/lib/rt-numa.c +2 -4

index babcc634d57e..bb0121a65eca 100644
@@ -68,15 +68,13 @@ int cpu_for_thread_sp(int thread_num, int max_cpus, struct bitmask *cpumask)
 int cpu_for_thread_ua(int thread_num, int max_cpus)
 {
 	int res, num_cpus, i, m, cpu;
-	pthread_t thread;
 	cpu_set_t cpuset;
 
-	thread = pthread_self();
 	CPU_ZERO(&cpuset);
 
-	res = pthread_getaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
+	res = sched_getaffinity(0, sizeof(cpu_set_t), &cpuset);
 	if (res != 0)
-		fatal("pthread_getaffinity_np failed: %s\n", strerror(res));
+		fatal("sched_getaffinity failed: %s\n", strerror(res));
 
 	num_cpus = CPU_COUNT(&cpuset);
 	m = thread_num % num_cpus;

Signature · 1 line

2.32.0

Daniel Wagner

2021-08-27 07:31 UTC #2 lore raw

Re: [PATCH rt-tests v3 1/1] rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

Quoted 11 lines · Daniel Wagner

pthread_getaffinity_np() prevents static builds as glibc does not expose it for this configuration. Instead use sched_getaffinity() directly as pthread_getaffinity_np() is just a fancy glibc wrapper around sched_getaffinity().
Static builds helps in in situation where the rootfs is minimal and does for example not ship libnuma.
The pthread_getaffinity_np call was introduced by commit f240656b056b ("rt-tests: cyclictest: Fix -t without a user specified [NUM]")

ping

John Kacur

2021-08-27 17:09 UTC #3 lore raw

Re: [PATCH rt-tests v3 1/1] rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

In reply to Daniel Wagner

Quoted 43 lines · Daniel Wagner

pthread_getaffinity_np() prevents static builds as glibc does not expose it for this configuration. Instead use sched_getaffinity() directly as pthread_getaffinity_np() is just a fancy glibc wrapper around sched_getaffinity().
Static builds helps in in situation where the rootfs is minimal and does for example not ship libnuma.
The pthread_getaffinity_np call was introduced by commit f240656b056b ("rt-tests: cyclictest: Fix -t without a user specified [NUM]")
Signed-off-by Daniel Wagner <[email protected]>
---
 src/lib/rt-numa.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
src/lib/rt-numa.c +2 -4
index babcc634d57e..bb0121a65eca 100644
@@ -68,15 +68,13 @@ int cpu_for_thread_sp(int thread_num, int max_cpus, struct bitmask *cpumask)
 int cpu_for_thread_ua(int thread_num, int max_cpus)
 {
 	int res, num_cpus, i, m, cpu;
-	pthread_t thread;
 	cpu_set_t cpuset;
 
-	thread = pthread_self();
 	CPU_ZERO(&cpuset);
 
-	res = pthread_getaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
+	res = sched_getaffinity(0, sizeof(cpu_set_t), &cpuset);
 	if (res != 0)
-		fatal("pthread_getaffinity_np failed: %s\n", strerror(res));
+		fatal("sched_getaffinity failed: %s\n", strerror(res));
 
 	num_cpus = CPU_COUNT(&cpuset);
 	m = thread_num % num_cpus;
Signature · 1 line
2.32.0

-Some minor fixes to the description

Signed-off-by John Kacur <[email protected]>

Daniel Wagner

2021-08-29 19:58 UTC #4 lore raw

Re: [PATCH rt-tests v3 1/1] rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

Hi John,

Quoted 1 line · John Kacur

-Some minor fixes to the description

Thanks a lot. Do you plan to do a release soon? I'd like to update the versions in LAVA's test-definition test suite. And it's way simpler to base it on a released version.

Thanks,
Daniel

John Kacur

2021-08-30 12:54 UTC #5 lore raw

Re: [PATCH rt-tests v3 1/1] rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

Quoted 12 lines · Daniel Wagner

Hi John,
Quoted 1 line · John Kacur
-Some minor fixes to the description
Thanks a lot. Do you plan to do a release soon? I'd like to update the versions in LAVA's test-definition test suite. And it's way simpler to base it on a released version.
Thanks,
Daniel

I hadn't planned on doing so, but I can do so if it will help you out. I can probably get to this, this afternoon my time, or if not, then tomorrow.

Could you provide a link to LAVA again so I don't have to go looking for it.?

Thanks

John

Daniel Wagner

author

2021-08-30 13:10 UTC #6 lore raw

Re: [PATCH rt-tests v3 1/1] rt-numa: Use sched_getaffinity() instead of pthread_getaffinity_np()

Hi John,

Quoted 3 lines · John Kacur

I hadn't planned on doing so, but I can do so if it will help you out. I can probably get to this, this afternoon my time, or if not, then tomorrow.

Great. Thanks!

Quoted 2 lines

Could you provide a link to LAVA again so I don't have to go looking for it.?

https://github.com/Linaro/test-definitions

and the main script for cyclictest is here:

https://github.com/Linaro/test-definitions/tree/master/automated/linux/cyclictest

There is the 'bin' folder which contains a static build binary. I know, this is a bit strange. I'll try to argue with upstream and see if we get this dropped eventually. But for this update I'd like to get the JSON output feature enabled first and not get lost in the discussion why the static build binary should be dropped.

BTW, I've queued the JSON changes in my tree:

https://github.com/igaw/test-definitions/tree/rt-tests-json

I think I could get this ready before Plumbers, so that we can discuss the next steps during a MC/BoF.

Thanks,
Daniel