Implement sched::sched_getaffinity() by thib-ack · Pull Request #1148 · nix-rust/nix

thib-ack · 2019-10-30T21:07:19Z

Hello,

I found the function sched::sched_setaffinity() but I also needed to get the process affinity and I did not find the function sched::sched_getaffinity()
So I added the function which maps sched_getaffinity(2) which "get a process's CPU affinity mask" to the sched.rs file.
I hope the code match your guidelines (returned Result<CpuSet> instead of pointer parameter)
If that's not the case, tell me, I will fix asap.

I hope this will help !

Thanks,
Thibaut

asomers

Could you also please add a CHANGELOG entry and a test?

asomers · 2019-11-01T03:05:46Z

+            libc::sched_getaffinity(
+                pid.into(),
+                mem::size_of::<CpuSet>() as libc::size_t,
+                & mut cpuset.cpu_set,


Style nit: there should be no space between & and mut.

asomers · 2019-11-01T03:07:15Z

+            )
+        };
+
+        match res {


I think you can do this more simply by Errno::result(res).and(Ok(cpuset))

thib-ack · 2019-11-01T09:55:28Z

Issues fixed
test added (in new dedicated file test_sched.rs)
changelog added

Thanks

asomers

Could you please also add docs for sched_getaffinity? And since you're here, it would be great if you could do the same for sched_setaffinity and for the CpuSet methods.

asomers · 2019-11-02T04:07:36Z

+    let initial_affinity = sched_getaffinity(Pid::from_raw(0)).unwrap();
+    let mut at_least_one_cpu = false;
+    let mut last_valid_cpu = 0;
+    for field in 0..(8 * mem::size_of::<libc::cpu_set_t>()) {


What does the magic number 8 mean? You should remove it if possible, or define it as a const if not.

This value comes from the CpuSet::set() and CpuSet::is_set() methods.
https://github.com/nix-rust/nix/blob/master/src/sched.rs#L63

pub fn is_set(&self, field: usize) -> Result<bool> { if field >= 8 * mem::size_of::<libc::cpu_set_t>() { Err(Error::Sys(Errno::EINVAL)) } else { Ok(unsafe { libc::CPU_ISSET(field, &self.cpu_set) }) } }

That's because there is no easy way to know the size of a CpuSet and in the test I want to check each bits of the CpuSet.

I can do differently:

add a method CpuSet::size() which will return this value.

call CpuSet::is_set() until I get a Errno::EINVAL
What do you prefer ?

Better to add a method to CpuSet. But I would call it count, because that's what the C macros call it.

asomers · 2019-11-02T04:08:12Z

@@ -0,0 +1,36 @@
+
+#[cfg(any(target_os = "android",


You should move the #[cfg()] into test/test.rs, for consistency with other files.

thib-ack · 2019-11-02T17:00:16Z

Hi,

I'll be AFK for the next 2 weeks unfortunately. I'll continue to work on this when I come back.
For example, I'm thinking the sched::sched_getaffinity() should take a &CpuSet as parameter instead of building one internally. That way, users could allocate the CpuSet once and use it many times in sched_getaffinity().
You agree ?

Thanks,

asomers · 2019-11-02T17:18:28Z

Hi,

I'll be AFK for the next 2 weeks unfortunately. I'll continue to work on this when I come back.
For example, I'm thinking the sched::sched_getaffinity() should take a &CpuSet as parameter instead of building one internally. That way, users could allocate the CpuSet once and use it many times in sched_getaffinity().
You agree ?

Thanks,

No, I disgree. Return arguments are very C-ish. It's much Rustier to use return values. I doubt there will be any performance difference, because rustc is able to do what C++ people call "return by move".

asomers · 2019-11-02T17:57:41Z

You'll have to rebase to fix the build error on nightly.

asomers

I like the docs! I think we're getting close now.

asomers · 2019-11-02T21:55:43Z

        }
+
+        /// Return the maximum number of CPU in CpuSet
+        #[inline]


You can remove the #[inline]. The compiler is smart enough to figure it out for itself.

asomers · 2019-11-02T21:56:40Z

+        /// `field` is the CPU id to remove
        pub fn unset(&mut self, field: usize) -> Result<()> {
-            if field >= 8 * mem::size_of::<libc::cpu_set_t>() {
+            CpuSet::count();


Why this line?

oops, copy-paste mistake

asomers · 2019-11-03T13:36:55Z

+    ///
+    /// Binding the current thread to CPU 0 can be done as follows:
+    ///
+    /// ```rust


I would probably make this rust,no_run. Otherwise, you'll be changing the affinity of the doc test process.

sched_getaffinity(2) get a process's CPU affinity mask

asomers

bors r+

1148: Implement sched::sched_getaffinity() r=asomers a=thib-ack Hello, I found the function `sched::sched_setaffinity()` but I also needed to get the process affinity and I did not find the function `sched::sched_getaffinity()` So I added the function which maps [sched_getaffinity(2)](https://linux.die.net/man/2/sched_getaffinity) which "get a process's CPU affinity mask" to the sched.rs file. I hope the code match your guidelines (returned `Result<CpuSet>` instead of pointer parameter) If that's not the case, tell me, I will fix asap. I hope this will help ! Thanks, Thibaut Co-authored-by: Thibaut Ackermann <thibaut@ackermann.dev>

bors · 2019-11-24T17:58:13Z

Build succeeded

FreeBSD 11.2
continuous-integration/travis-ci/push

thib-ack force-pushed the add_sched_getaffinity branch from 32feaef to 8f9956b Compare October 31, 2019 16:45

asomers requested changes Nov 1, 2019

View reviewed changes

thib-ack force-pushed the add_sched_getaffinity branch from 30aad7c to dde8156 Compare November 1, 2019 09:54

thib-ack force-pushed the add_sched_getaffinity branch from dde8156 to d6fbf81 Compare November 1, 2019 09:57

thib-ack requested a review from asomers November 1, 2019 11:39

asomers requested changes Nov 2, 2019

View reviewed changes

thib-ack force-pushed the add_sched_getaffinity branch 2 times, most recently from 5895dc1 to 47e9413 Compare November 2, 2019 09:42

thib-ack force-pushed the add_sched_getaffinity branch from 47e9413 to 3d6ba05 Compare November 2, 2019 18:23

thib-ack requested a review from asomers November 2, 2019 19:17

asomers requested changes Nov 2, 2019

View reviewed changes

thib-ack force-pushed the add_sched_getaffinity branch from 3d6ba05 to 9934abc Compare November 3, 2019 07:06

thib-ack requested a review from asomers November 3, 2019 07:47

asomers reviewed Nov 3, 2019

View reviewed changes

Implement sched::sched_getaffinity()

18c2038

sched_getaffinity(2) get a process's CPU affinity mask

thib-ack force-pushed the add_sched_getaffinity branch from 9934abc to 18c2038 Compare November 21, 2019 07:37

thib-ack requested a review from asomers November 21, 2019 08:14

asomers approved these changes Nov 24, 2019

View reviewed changes

bors bot merged commit 18c2038 into nix-rust:master Nov 24, 2019

Conversation

thib-ack commented Oct 30, 2019

Uh oh!

asomers left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

thib-ack commented Nov 1, 2019

Uh oh!

asomers left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

thib-ack commented Nov 2, 2019

Uh oh!

asomers commented Nov 2, 2019

Uh oh!

asomers commented Nov 2, 2019

Uh oh!

asomers left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

asomers left a comment

Choose a reason for hiding this comment

Uh oh!

bors bot commented Nov 24, 2019

Build succeeded

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants