mirror of
https://github.com/sched-ext/scx.git
synced 2024-11-29 06:00:23 +00:00
Merge pull request #508 from multics69/lavd-numa-fix
scx_lavd: fix a potential watchdog timeout error at multi-NUMA/CCX platforms
This commit is contained in:
Changwoo Min
GitHub
commit
1d61dd4c1d
@ -362,6 +362,14 @@ static struct task_ctx *get_task_ctx(struct task_struct *p)
|
||||
return taskc;
|
||||
}
|
||||
|
||||
static s32 get_task_cpu_id(struct task_struct *p)
|
||||
{
|
||||
/*
|
||||
* This code assumes ONFIG_THREAD_INFO_IN_TASK is on in the kernel.
|
||||
*/
|
||||
return READ_ONCE(p->thread_info.cpu);
|
||||
}
|
||||
|
||||
static struct cpu_ctx *get_cpu_ctx(void)
|
||||
{
|
||||
const u32 idx = 0;
|
||||
@ -1277,6 +1285,181 @@ static void update_stat_for_quiescent(struct task_struct *p,
|
||||
cpuc->load_run_time_ns -= clamp_time_slice_ns(taskc->run_time_ns);
|
||||
}
|
||||
|
||||
static bool could_run_on_prev(struct task_struct *p, s32 prev_cpu,
|
||||
struct bpf_cpumask *a_cpumask,
|
||||
struct bpf_cpumask *o_cpumask)
|
||||
{
|
||||
bool ret;
|
||||
|
||||
ret = bpf_cpumask_test_cpu(prev_cpu, p->cpus_ptr) &&
|
||||
(bpf_cpumask_test_cpu(prev_cpu, cast_mask(a_cpumask)) ||
|
||||
bpf_cpumask_test_cpu(prev_cpu, cast_mask(o_cpumask)));
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static s32 pick_cpu(struct task_struct *p, struct task_ctx *taskc,
|
||||
s32 prev_cpu, u64 wake_flags, bool *is_idle)
|
||||
{
|
||||
struct sys_stat *stat_cur = get_sys_stat_cur();
|
||||
struct cpu_ctx *cpuc;
|
||||
struct bpf_cpumask *a_cpumask, *o_cpumask, *t_cpumask;
|
||||
struct bpf_cpumask *active, *ovrflw, *big, *little;
|
||||
s32 cpu_id;
|
||||
|
||||
bpf_rcu_read_lock();
|
||||
|
||||
/*
|
||||
* Prepare cpumaks.
|
||||
*/
|
||||
cpuc = get_cpu_ctx();
|
||||
if (!cpuc || !taskc) {
|
||||
scx_bpf_error("Failed to lookup the current cpu_ctx");
|
||||
cpu_id = prev_cpu;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
a_cpumask = cpuc->tmp_a_mask;
|
||||
o_cpumask = cpuc->tmp_o_mask;
|
||||
t_cpumask = cpuc->tmp_t_mask;
|
||||
active = active_cpumask;
|
||||
ovrflw = ovrflw_cpumask;
|
||||
big = big_cpumask;
|
||||
little = little_cpumask;
|
||||
if (!a_cpumask || !o_cpumask || !t_cpumask ||
|
||||
!active || !ovrflw || !big || !little) {
|
||||
cpu_id = -ENOENT;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
bpf_cpumask_and(a_cpumask, p->cpus_ptr, cast_mask(active));
|
||||
bpf_cpumask_and(o_cpumask, p->cpus_ptr, cast_mask(ovrflw));
|
||||
|
||||
/*
|
||||
* Try to stay on the previous core if it is on active or ovrfw.
|
||||
*/
|
||||
if (could_run_on_prev(p, prev_cpu, a_cpumask, o_cpumask) &&
|
||||
scx_bpf_test_and_clear_cpu_idle(prev_cpu)) {
|
||||
cpu_id = prev_cpu;
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
if (bpf_cpumask_empty(cast_mask(a_cpumask)))
|
||||
goto start_omask;
|
||||
|
||||
/*
|
||||
* Pick a fully idle core among active CPUs with a matching core type.
|
||||
*/
|
||||
if (is_perf_cri(taskc, stat_cur))
|
||||
bpf_cpumask_and(t_cpumask, cast_mask(a_cpumask), cast_mask(big));
|
||||
else
|
||||
bpf_cpumask_and(t_cpumask, cast_mask(a_cpumask), cast_mask(little));
|
||||
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(t_cpumask), SCX_PICK_IDLE_CORE);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick a fully idle core among active CPUs with a matching core type
|
||||
* even if its hypertwin is in use.
|
||||
*/
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(t_cpumask), 0);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick a fully idle core among active CPUs.
|
||||
*/
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(a_cpumask), SCX_PICK_IDLE_CORE);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick an any idle core among active CPUs even if its hypertwin is in
|
||||
* use.
|
||||
*/
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(a_cpumask), 0);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick an any idle core among overflow CPUs.
|
||||
*/
|
||||
start_omask:
|
||||
if (bpf_cpumask_empty(cast_mask(o_cpumask)))
|
||||
goto start_any_mask;
|
||||
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(o_cpumask), 0);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* If there is no idle core under our control, pick random core
|
||||
* either in active of overflow CPUs.
|
||||
*/
|
||||
if (!bpf_cpumask_empty(cast_mask(a_cpumask))) {
|
||||
cpu_id = bpf_cpumask_any_distribute(cast_mask(a_cpumask));
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
if (!bpf_cpumask_empty(cast_mask(o_cpumask))) {
|
||||
cpu_id = bpf_cpumask_any_distribute(cast_mask(o_cpumask));
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* If the task cannot run on either active or overflow cores,
|
||||
* stay on the previous core (if it is okay) or one of its taskset.
|
||||
*/
|
||||
if (bpf_cpumask_test_cpu(prev_cpu, p->cpus_ptr))
|
||||
cpu_id = prev_cpu;
|
||||
else {
|
||||
start_any_mask:
|
||||
cpu_id = bpf_cpumask_any_distribute(p->cpus_ptr);
|
||||
}
|
||||
|
||||
/*
|
||||
* Note that we don't need to kick the picked CPU here since the
|
||||
* ops.select_cpu() path internally triggers kicking cpu if necessary.
|
||||
*/
|
||||
unlock_out:
|
||||
bpf_rcu_read_unlock();
|
||||
return cpu_id;
|
||||
}
|
||||
|
||||
s32 BPF_STRUCT_OPS(lavd_select_cpu, struct task_struct *p, s32 prev_cpu,
|
||||
u64 wake_flags)
|
||||
{
|
||||
bool found_idle = false;
|
||||
struct task_ctx *taskc;
|
||||
s32 cpu_id;
|
||||
|
||||
taskc = get_task_ctx(p);
|
||||
if (!taskc)
|
||||
return prev_cpu;
|
||||
|
||||
taskc->wakeup_ft += !!(wake_flags & SCX_WAKE_SYNC);
|
||||
|
||||
cpu_id = pick_cpu(p, taskc, prev_cpu, wake_flags, &found_idle);
|
||||
if (found_idle) {
|
||||
return cpu_id;
|
||||
}
|
||||
|
||||
return prev_cpu;
|
||||
}
|
||||
|
||||
|
||||
static void calc_when_to_run(struct task_struct *p, struct task_ctx *taskc,
|
||||
u64 enq_flags)
|
||||
{
|
||||
@ -1621,13 +1804,31 @@ static u64 find_proper_dsq(struct task_ctx *taskc, struct cpu_ctx *cpuc)
|
||||
return cpuc->cpdom_alt_id;
|
||||
}
|
||||
|
||||
static void put_cpdom_rq(struct task_struct *p, struct task_ctx *taskc,
|
||||
struct cpu_ctx *cpuc, u64 enq_flags)
|
||||
void BPF_STRUCT_OPS(lavd_enqueue, struct task_struct *p, u64 enq_flags)
|
||||
{
|
||||
struct task_ctx *taskc_run;
|
||||
struct task_struct *p_run;
|
||||
struct cpu_ctx *cpuc_task, *cpuc_cur;
|
||||
struct task_ctx *taskc;
|
||||
s32 cpu_id;
|
||||
u64 dsq_id;
|
||||
|
||||
/*
|
||||
* Place a task to a run queue of current cpu's compute domain.
|
||||
*
|
||||
* If there is an idle CPU at the ops.select_cpu(), the task is already
|
||||
* dispatched at ops.select_cpu(), so ops.enqueue() won't be called.
|
||||
* Hence, the task that is enqueued here are the cases: 1) there is no
|
||||
* idle CPU when ops.select_cpu() or 2) the task is not the case of
|
||||
* being wakened up (i.e., resume after preemption). Therefore, we
|
||||
* always put the task to the global DSQ, so any idle CPU can pick it
|
||||
* up.
|
||||
*/
|
||||
cpu_id = get_task_cpu_id(p);
|
||||
taskc = get_task_ctx(p);
|
||||
cpuc_task = get_cpu_ctx_id(cpu_id);
|
||||
cpuc_cur = get_cpu_ctx();
|
||||
if (!cpuc_cur || !cpuc_task || !taskc)
|
||||
return;
|
||||
|
||||
__sync_fetch_and_add(&nr_queued_task, 1);
|
||||
|
||||
/*
|
||||
@ -1642,6 +1843,8 @@ static void put_cpdom_rq(struct task_struct *p, struct task_ctx *taskc,
|
||||
* If a task is eligible, try to preempt a task.
|
||||
*/
|
||||
if (is_eligible(taskc)) {
|
||||
struct task_ctx *taskc_run;
|
||||
struct task_struct *p_run;
|
||||
/*
|
||||
* Try to find and kick a victim CPU, which runs a less urgent
|
||||
* task. The kick will be done asynchronously.
|
||||
@ -1654,216 +1857,17 @@ static void put_cpdom_rq(struct task_struct *p, struct task_ctx *taskc,
|
||||
p_run = bpf_get_current_task_btf();
|
||||
taskc_run = try_get_task_ctx(p_run);
|
||||
if (taskc_run && p_run->scx.slice != 0)
|
||||
try_yield_current_cpu(p_run, cpuc, taskc_run);
|
||||
try_yield_current_cpu(p_run, cpuc_cur, taskc_run);
|
||||
}
|
||||
|
||||
/*
|
||||
* Enqueue the task to one of the DSQs based on its virtual deadline.
|
||||
* Enqueue the task to one of task's DSQs based on its virtual deadline.
|
||||
*/
|
||||
dsq_id = find_proper_dsq(taskc, cpuc);
|
||||
dsq_id = find_proper_dsq(taskc, cpuc_task);
|
||||
scx_bpf_dispatch_vtime(p, dsq_id, LAVD_SLICE_UNDECIDED,
|
||||
taskc->vdeadline_log_clk, enq_flags);
|
||||
}
|
||||
|
||||
static bool could_run_on_prev(struct task_struct *p, s32 prev_cpu,
|
||||
struct bpf_cpumask *a_cpumask,
|
||||
struct bpf_cpumask *o_cpumask)
|
||||
{
|
||||
bool ret;
|
||||
|
||||
ret = bpf_cpumask_test_cpu(prev_cpu, p->cpus_ptr) &&
|
||||
(bpf_cpumask_test_cpu(prev_cpu, cast_mask(a_cpumask)) ||
|
||||
bpf_cpumask_test_cpu(prev_cpu, cast_mask(o_cpumask)));
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static s32 pick_cpu(struct task_struct *p, struct task_ctx *taskc,
|
||||
s32 prev_cpu, u64 wake_flags, bool *is_idle)
|
||||
{
|
||||
struct sys_stat *stat_cur = get_sys_stat_cur();
|
||||
struct cpu_ctx *cpuc;
|
||||
struct bpf_cpumask *a_cpumask, *o_cpumask, *t_cpumask;
|
||||
struct bpf_cpumask *active, *ovrflw, *big, *little;
|
||||
s32 cpu_id;
|
||||
|
||||
bpf_rcu_read_lock();
|
||||
|
||||
/*
|
||||
* Prepare cpumaks.
|
||||
*/
|
||||
cpuc = get_cpu_ctx();
|
||||
if (!cpuc || !taskc) {
|
||||
scx_bpf_error("Failed to lookup the current cpu_ctx");
|
||||
cpu_id = prev_cpu;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
a_cpumask = cpuc->tmp_a_mask;
|
||||
o_cpumask = cpuc->tmp_o_mask;
|
||||
t_cpumask = cpuc->tmp_t_mask;
|
||||
active = active_cpumask;
|
||||
ovrflw = ovrflw_cpumask;
|
||||
big = big_cpumask;
|
||||
little = little_cpumask;
|
||||
if (!a_cpumask || !o_cpumask || !t_cpumask ||
|
||||
!active || !ovrflw || !big || !little) {
|
||||
cpu_id = -ENOENT;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
bpf_cpumask_and(a_cpumask, p->cpus_ptr, cast_mask(active));
|
||||
bpf_cpumask_and(o_cpumask, p->cpus_ptr, cast_mask(ovrflw));
|
||||
|
||||
/*
|
||||
* Try to stay on the previous core if it is on active or ovrfw.
|
||||
*/
|
||||
if (could_run_on_prev(p, prev_cpu, a_cpumask, o_cpumask) &&
|
||||
scx_bpf_test_and_clear_cpu_idle(prev_cpu)) {
|
||||
cpu_id = prev_cpu;
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
if (bpf_cpumask_empty(cast_mask(a_cpumask)))
|
||||
goto start_omask;
|
||||
|
||||
/*
|
||||
* Pick a fully idle core among active CPUs with a matching core type.
|
||||
*/
|
||||
if (is_perf_cri(taskc, stat_cur))
|
||||
bpf_cpumask_and(t_cpumask, cast_mask(a_cpumask), cast_mask(big));
|
||||
else
|
||||
bpf_cpumask_and(t_cpumask, cast_mask(a_cpumask), cast_mask(little));
|
||||
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(t_cpumask), SCX_PICK_IDLE_CORE);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick a fully idle core among active CPUs with a matching core type
|
||||
* even if its hypertwin is in use.
|
||||
*/
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(t_cpumask), 0);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick a fully idle core among active CPUs.
|
||||
*/
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(a_cpumask), SCX_PICK_IDLE_CORE);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick an any idle core among active CPUs even if its hypertwin is in
|
||||
* use.
|
||||
*/
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(a_cpumask), 0);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* Pick an any idle core among overflow CPUs.
|
||||
*/
|
||||
start_omask:
|
||||
if (bpf_cpumask_empty(cast_mask(o_cpumask)))
|
||||
goto start_any_mask;
|
||||
|
||||
cpu_id = scx_bpf_pick_idle_cpu(cast_mask(o_cpumask), 0);
|
||||
if (cpu_id >= 0) {
|
||||
*is_idle = true;
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* If there is no idle core under our control, pick random core
|
||||
* either in active of overflow CPUs.
|
||||
*/
|
||||
if (!bpf_cpumask_empty(cast_mask(a_cpumask))) {
|
||||
cpu_id = bpf_cpumask_any_distribute(cast_mask(a_cpumask));
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
if (!bpf_cpumask_empty(cast_mask(o_cpumask))) {
|
||||
cpu_id = bpf_cpumask_any_distribute(cast_mask(o_cpumask));
|
||||
goto unlock_out;
|
||||
}
|
||||
|
||||
/*
|
||||
* If the task cannot run on either active or overflow cores,
|
||||
* stay on the previous core (if it is okay) or one of its taskset.
|
||||
*/
|
||||
if (bpf_cpumask_test_cpu(prev_cpu, p->cpus_ptr))
|
||||
cpu_id = prev_cpu;
|
||||
else {
|
||||
start_any_mask:
|
||||
cpu_id = bpf_cpumask_any_distribute(p->cpus_ptr);
|
||||
}
|
||||
|
||||
/*
|
||||
* Note that we don't need to kick the picked CPU here since the
|
||||
* ops.select_cpu() path internally triggers kicking cpu if necessary.
|
||||
*/
|
||||
unlock_out:
|
||||
bpf_rcu_read_unlock();
|
||||
return cpu_id;
|
||||
}
|
||||
|
||||
s32 BPF_STRUCT_OPS(lavd_select_cpu, struct task_struct *p, s32 prev_cpu,
|
||||
u64 wake_flags)
|
||||
{
|
||||
bool found_idle = false;
|
||||
struct task_ctx *taskc;
|
||||
s32 cpu_id;
|
||||
|
||||
taskc = get_task_ctx(p);
|
||||
if (!taskc)
|
||||
return prev_cpu;
|
||||
|
||||
taskc->wakeup_ft += !!(wake_flags & SCX_WAKE_SYNC);
|
||||
|
||||
cpu_id = pick_cpu(p, taskc, prev_cpu, wake_flags, &found_idle);
|
||||
if (found_idle) {
|
||||
return cpu_id;
|
||||
}
|
||||
|
||||
return prev_cpu;
|
||||
}
|
||||
|
||||
void BPF_STRUCT_OPS(lavd_enqueue, struct task_struct *p, u64 enq_flags)
|
||||
{
|
||||
struct cpu_ctx *cpuc;
|
||||
struct task_ctx *taskc;
|
||||
|
||||
/*
|
||||
* If there is an idle CPU at the ops.select_cpu(), the task is already
|
||||
* dispatched at ops.select_cpu(), so ops.enqueue() won't be called.
|
||||
* Hence, the task that is enqueued here are the cases: 1) there is no
|
||||
* idle CPU when ops.select_cpu() or 2) the task is not the case of
|
||||
* being wakened up (i.e., resume after preemption). Therefore, we
|
||||
* always put the task to the global DSQ, so any idle CPU can pick it
|
||||
* up.
|
||||
*/
|
||||
cpuc = get_cpu_ctx();
|
||||
taskc = get_task_ctx(p);
|
||||
if (!cpuc || !taskc)
|
||||
return;
|
||||
|
||||
/*
|
||||
* Place a task to a run queue of current cpu's compute domain.
|
||||
*/
|
||||
put_cpdom_rq(p, taskc, cpuc, enq_flags);
|
||||
}
|
||||
|
||||
static bool is_kernel_task(struct task_struct *p)
|
||||
{
|
||||
return !!(p->flags & PF_KTHREAD);
|
||||
|
||||
Loading…
Reference in New Issue
Block a user