JakeHillion/scx-upstream
1
0
Fork 0
mirror of https://github.com/sched-ext/scx.git synced 2024-11-24 11:50:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

scx_bpfland: keep tasks running on full-idle SMT cores

Browse Source 
When a task is the last one running on a CPU and still wants to
continue, allow it to run and replenish its time only if the used CPU is
part a fully idle SMT core.

Signed-off-by: Andrea Righi <andrea.righi@linux.dev>
This commit is contained in:
Andrea Righi 2024-09-29 15:53:25 +02:00
parent c20a19c946
commit 6e24fcc7f0
1 changed files with 34 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
scheds/rust/scx_bpfland/src/bpf/main.bpf.c
View File

@ -411,22 +411,29 @@ static u64 nr_tasks_waiting(void)
}
/*
* Return the task's unused portion of its previously assigned time slice in
* the range a [slice_ns_min .. slice_ns].
* Return a value inversely proportional to the task's weight.
*/
static inline u64 task_slice(struct task_struct *p)
static inline u64 scale_inverse_fair(struct task_struct *p, u64 value)
{
return value * 100 / p->scx.weight;
}
/*
* Evaluate task's time slice in function of the total amount of tasks that are
* waiting to be dispatched and the task's weight.
*/
static inline void task_refill_slice(struct task_struct *p)
{
u64 slice;
/*
* Refresh the amount of waiting tasks to get a more accurate scaling
* factor for the time slice.
*/
nr_waiting = (nr_waiting + nr_tasks_waiting()) / 2;
/*
* Scale the time slice based on the average number of waiting tasks
* (more waiting tasks result in a shorter time slice).
*/
return MAX(slice_ns / (nr_waiting + 1), slice_ns_min);
slice = slice_ns / (nr_waiting + 1);
p->scx.slice = CLAMP(slice, slice_ns_min, slice_ns);
}
/*
@ -817,7 +824,6 @@ s32 BPF_STRUCT_OPS(bpfland_select_cpu, struct task_struct *p, s32 prev_cpu, u64
*/
void BPF_STRUCT_OPS(bpfland_enqueue, struct task_struct *p, u64 enq_flags)
{
struct bpf_cpumask *primary;
struct task_ctx *tctx;
u64 deadline = task_deadline(p);
s32 cpu = scx_bpf_task_cpu(p);
@ -1012,9 +1018,9 @@ void BPF_STRUCT_OPS(bpfland_dispatch, s32 cpu, struct task_struct *prev)
return;
/*
* If the current task expired its time slice, but no other task wants
* to run, simply replenish its time slice and let it run for another
* round on the same CPU.
* If the current task expired its time slice, its CPU is still a
* full-idle SMT core and no other task wants to run, simply replenish
* its time slice and let it run for another round on the same CPU.
*
* Note that bpfland_stopping() won't be called if we replenish the
* time slice here. As a result, the nvcsw statistics won't be updated,
@ -1022,8 +1028,19 @@ void BPF_STRUCT_OPS(bpfland_dispatch, s32 cpu, struct task_struct *prev)
* when the system is overloaded, which isn't the case when there are
* no other tasks to run.
*/
if (prev && (prev->scx.flags & SCX_TASK_QUEUED))
prev->scx.slice = task_slice(prev);
if (prev && (prev->scx.flags & SCX_TASK_QUEUED)) {
const struct cpumask *idle_smtmask;
if (!smt_enabled) {
task_refill_slice(prev);
return;
}
idle_smtmask = scx_bpf_get_idle_smtmask();
if (bpf_cpumask_test_cpu(cpu, idle_smtmask))
task_refill_slice(prev);
scx_bpf_put_idle_cpumask(idle_smtmask);
}
}
/*
@ -1085,7 +1102,7 @@ void BPF_STRUCT_OPS(bpfland_running, struct task_struct *p)
* Refresh task's time slice immediately before it starts to run on its
* assigned CPU.
*/
p->scx.slice = task_slice(p);
task_refill_slice(p);
/*
* Adjust target CPU frequency before the task starts to run.
@ -1146,13 +1163,13 @@ void BPF_STRUCT_OPS(bpfland_stopping, struct task_struct *p, bool runnable)
* Update task vruntime, charging the weighted used time slice.
*/
task_slice = p->se.sum_exec_runtime - tctx->sum_exec_runtime;
p->scx.dsq_vtime += task_slice * 100 / p->scx.weight;
p->scx.dsq_vtime += scale_inverse_fair(p, task_slice);
tctx->sum_exec_runtime = p->se.sum_exec_runtime;
/*
* Update global vruntime.
*/
vtime_now += task_slice * 100 / p->scx.weight;
vtime_now += scale_inverse_fair(p, task_slice);
/*
* Refresh voluntary context switch metrics.