From 8d14d54fa98b6141c6520616f515844f00830503 Mon Sep 17 00:00:00 2001
From: 4lDO2 <4lDO2@protonmail.com>
Date: Fri, 26 Aug 2022 11:54:47 +0200
Subject: [PATCH] Support scheduler CPU affinity.
---
src/context/context.rs | 9 ++++++++-
src/context/switch.rs | 3 ++-
src/scheme/proc.rs | 24 +++++++++++++++++++++---
src/scheme/sys/context.rs | 11 +++++++++--
4 files changed, 40 insertions(+), 7 deletions(-)
diff --git a/src/context/context.rs b/src/context/context.rs
index 566448c4..374edc23 100644
--- a/src/context/context.rs
+++ b/src/context/context.rs
@@ -117,6 +117,7 @@ pub struct ContextSnapshot {
pub running: bool,
pub cpu_id: Option<usize>,
pub cpu_time: u128,
+ pub sched_affinity: Option<usize>,
pub syscall: Option<(usize, usize, usize, usize, usize, usize)>,
// Clone fields
//TODO: is there a faster way than allocation?
@@ -161,6 +162,7 @@ impl ContextSnapshot {
running: context.running,
cpu_id: context.cpu_id,
cpu_time: context.cpu_time,
+ sched_affinity: context.sched_affinity,
syscall: context.syscall,
name,
files,
@@ -198,12 +200,16 @@ pub struct Context {
pub status_reason: &'static str,
/// Context running or not
pub running: bool,
- /// CPU ID, if locked
+ /// Current CPU ID
pub cpu_id: Option<usize>,
/// Time this context was switched to
pub switch_time: u128,
/// Amount of CPU time used
pub cpu_time: u128,
+ /// Scheduler CPU affinity. If set, [`cpu_id`] can except [`None`] never be anything else than
+ /// this value.
+ // TODO: bitmask (selection of multiple allowed CPUs)?
+ pub sched_affinity: Option<usize>,
/// Current system call
pub syscall: Option<(usize, usize, usize, usize, usize, usize)>,
/// Head buffer to use when system call buffers are not page aligned
@@ -352,6 +358,7 @@ impl Context {
cpu_id: None,
switch_time: 0,
cpu_time: 0,
+ sched_affinity: None,
syscall: None,
syscall_head,
syscall_tail,
diff --git a/src/context/switch.rs b/src/context/switch.rs
index 6ba12b69..22c2fd3e 100644
--- a/src/context/switch.rs
+++ b/src/context/switch.rs
@@ -17,7 +17,8 @@ use crate::time;
unsafe fn update(context: &mut Context, cpu_id: usize) {
// Take ownership if not already owned
- if context.cpu_id == None {
+ // TODO: Support unclaiming context, while still respecting the CPU affinity.
+ if context.cpu_id == None && context.sched_affinity.map_or(true, |id| id == crate::cpu_id()) {
context.cpu_id = Some(cpu_id);
// println!("{}: take {} {}", cpu_id, context.id, *context.name.read());
}
diff --git a/src/scheme/proc.rs b/src/scheme/proc.rs
index d6f80a0a..b90b5d9e 100644
--- a/src/scheme/proc.rs
+++ b/src/scheme/proc.rs
@@ -141,6 +141,7 @@ enum Operation {
// FD to access the file descriptor behind grants.
GrantHandle { description: Arc<RwLock<FileDescription>> },
+ SchedAffinity,
Sigactions(Arc<RwLock<Vec<(SigAction, usize)>>>),
CurrentSigactions,
AwaitingSigactionsChange(Arc<RwLock<Vec<(SigAction, usize)>>>),
@@ -307,6 +308,7 @@ impl ProcScheme {
Some("sigactions") => Operation::Sigactions(Arc::clone(&get_context(pid)?.read().actions)),
Some("current-sigactions") => Operation::CurrentSigactions,
Some("mmap-min-addr") => Operation::MmapMinAddr(Arc::clone(get_context(pid)?.read().addr_space().map_err(|_| Error::new(ENOENT))?)),
+ Some("sched-affinity") => Operation::SchedAffinity,
_ => return Err(Error::new(EINVAL))
};
@@ -834,6 +836,12 @@ impl Scheme for ProcScheme {
*buf.array_chunks_mut::<{mem::size_of::<usize>()}>().next().unwrap() = usize::to_ne_bytes(val);
Ok(mem::size_of::<usize>())
}
+ Operation::SchedAffinity => {
+ // TODO: Deduplicate code
+ let val = context::contexts().get(info.pid).ok_or(Error::new(EBADFD))?.read().sched_affinity.map_or(usize::MAX, |a| a % crate::cpu_count());
+ *buf.array_chunks_mut::<{mem::size_of::<usize>()}>().next().unwrap() = usize::to_ne_bytes(val);
+ Ok(mem::size_of::<usize>())
+ }
// TODO: Replace write() with SYS_DUP_FORWARD.
// TODO: Find a better way to switch address spaces, since they also require switching
// the instruction and stack pointer. Maybe remove `<pid>/regs` altogether and replace it
@@ -1081,6 +1089,13 @@ impl Scheme for ProcScheme {
addrspace.write().mmap_min = val;
Ok(mem::size_of::<usize>())
}
+ // TODO: Deduplicate code.
+ Operation::SchedAffinity => {
+ let val = usize::from_ne_bytes(<[u8; mem::size_of::<usize>()]>::try_from(buf).map_err(|_| Error::new(EINVAL))?);
+ context::contexts().get(info.pid).ok_or(Error::new(EBADFD))?.write().sched_affinity = if val == usize::MAX { None } else { Some(val % crate::cpu_count()) };
+ Ok(mem::size_of::<usize>())
+ }
+
_ => Err(Error::new(EBADF)),
}
}
@@ -1131,6 +1146,7 @@ impl Scheme for ProcScheme {
Operation::CurrentSigactions => "current-sigactions",
Operation::OpenViaDup => "open-via-dup",
Operation::MmapMinAddr(_) => "mmap-min-addr",
+ Operation::SchedAffinity => "sched-affinity",
_ => return Err(Error::new(EOPNOTSUPP)),
});
@@ -1359,6 +1375,9 @@ fn inherit_context() -> Result<ContextId> {
let mut new_context = new_context_lock.write();
new_context.status = Status::Stopped(SIGSTOP);
+
+ // TODO: Move all of these IDs into somewhere in userspace. Processes as an abstraction
+ // needs not be in the kernel; contexts are sufficient.
new_context.euid = current_context.euid;
new_context.egid = current_context.egid;
new_context.ruid = current_context.ruid;
@@ -1368,10 +1387,9 @@ fn inherit_context() -> Result<ContextId> {
new_context.ppid = current_context.id;
new_context.pgid = current_context.pgid;
new_context.umask = current_context.umask;
- new_context.sigmask = current_context.sigmask;
- new_context.cpu_id = current_context.cpu_id;
- // TODO: More to copy?
+ // TODO: Force userspace to copy sigmask. Start with "all signals blocked".
+ new_context.sigmask = current_context.sigmask;
new_context.id
};
diff --git a/src/scheme/sys/context.rs b/src/scheme/sys/context.rs
index 0f27aad5..dacc831b 100644
--- a/src/scheme/sys/context.rs
+++ b/src/scheme/sys/context.rs
@@ -5,7 +5,7 @@ use crate::context;
use crate::syscall::error::Result;
pub fn resource() -> Result<Vec<u8>> {
- let mut string = format!("{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<12}{:<8}{}\n",
+ let mut string = format!("{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<12}{:<8}{}\n",
"PID",
"PGID",
"PPID",
@@ -17,6 +17,7 @@ pub fn resource() -> Result<Vec<u8>> {
"ENS",
"STAT",
"CPU",
+ "AFF",
"TIME",
"MEM",
"NAME");
@@ -62,6 +63,11 @@ pub fn resource() -> Result<Vec<u8>> {
} else {
format!("?")
};
+ let affinity = if let Some(aff) = context.sched_affinity {
+ format!("{}", aff)
+ } else {
+ format!("?")
+ };
let cpu_time_s = context.cpu_time / crate::time::NANOS_PER_SEC;
let cpu_time_ns = context.cpu_time % crate::time::NANOS_PER_SEC;
@@ -95,7 +101,7 @@ pub fn resource() -> Result<Vec<u8>> {
format!("{} B", memory)
};
- string.push_str(&format!("{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<12}{:<8}{}\n",
+ string.push_str(&format!("{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<6}{:<12}{:<8}{}\n",
context.id.into(),
context.pgid.into(),
context.ppid.into(),
@@ -107,6 +113,7 @@ pub fn resource() -> Result<Vec<u8>> {
context.ens.into(),
stat_string,
cpu_string,
+ affinity,
cpu_time_string,
memory_string,
*context.name.read()));
--
GitLab