Re: [PATCH 10/11] sched_ext: scx_qmap: Add proxy execution support

[John Stultz <jstultz@google.com>]

On Thu, Jul 16, 2026 at 6:23 AM Andrea Righi <arighi@nvidia.com> wrote:
>
> Add a -B option to opt scx_qmap into queueing mutex-blocked tasks for
> proxy execution. Without the option, SCX_OPS_ENQ_BLOCKED remains clear
> and mutex waiters block normally. With -B, blocked donors are passed to
> qmap_enqueue() with SCX_ENQ_BLOCKED.
>
> When scx_qmap receives a blocked donor, dispatch it directly to the
> local DSQ of its current cid with a fresh slice and SCX_ENQ_PREEMPT.
> This places the donor at the head of the DSQ and requests an immediate
> reschedule, allowing the core proxy-exec path to run the mutex owner
> using the donor's scheduling context as soon as the donor is selected.

Hey Andrea!

Thanks again for your great work on all this, I'm working to test your
changes with my full proxy-exec tree, but before that I caught one
issue here.

I was excited to try this out, but when building and running this from
your kernel tree in a VM, (who's kernel is running from the same
kernel tree, built on the host), I get the following error (with or
without -B). The other scx_ schedulers in tools/sched_ext/ seem to
load and run ok.  Maybe is there something I'm forgetting to do here?

thanks
-john

# ./build/bin/scx_qmap -B
libbpf: prog 'qmap_dequeue': BPF program load failed: -EACCES
libbpf: prog 'qmap_dequeue': -- BEGIN PROG LOAD LOG --
0: R1=ctx() R10=fp0
; void BPF_STRUCT_OPS(qmap_dequeue, struct task_struct *p, u64
deq_flags) @ scx_qmap.bpf.c:584
0: (79) r2 = *(u64 *)(r1 +0)
func 'dequeue' arg0 has btf_id 142670 type STRUCT 'task_struct'
1: R1=ctx() R2=trusted_ptr_task_struct()
1: (79) r4 = *(u64 *)(r1 +8)          ; R1=ctx() R4=scalar()
; __sync_fetch_and_add(&qa.nr_dequeued, 1); @ scx_qmap.bpf.c:588
2: (18) r1 = 0x10000fff7000           ; R1=scalar()
4: (18) r3 = 0x10000fff7000           ; R3=scalar()
6: (07) r3 += 32                      ; R3=scalar()
7: (bf) r5 = addr_space_cast(r3, 0, 1)        ; R3=scalar() R5=arena
8: (b7) r3 = 1                        ; R3=1
9: (b7) r0 = 1                        ; R0=1
10: (db) lock *(u64 *)(r5 +0) += r0   ; R0=1 R5=arena
11: (18) r5 = 0x100000000             ; R5=0x100000000
; if (deq_flags & SCX_DEQ_CORE_SCHED_EXEC) @ scx_qmap.bpf.c:589
13: (5f) r4 &= r5                     ;
R4=scalar(smin=smin32=0,smax=umax=0x100000000,smax32=umax32=0,var_off=(0x0;
0x100000000))
14: (15) if r4 == 0x0 goto pc+3       ; R4=0x100000000
; __sync_fetch_and_add(&qa.nr_core_sched_execed, 1); @ scx_qmap.bpf.c:590
15: (07) r1 += 48                     ; R1=scalar()
16: (bf) r1 = addr_space_cast(r1, 0, 1)       ; R1=arena
17: (db) lock *(u64 *)(r1 +0) += r3   ;
; QMAP_TOUCH_ARENA(); @ scx_qmap.bpf.c:268
18: (18) r1 = 0xffff888104842c00      ; R1=map_ptr(map=arena,ks=0,vs=0)
; v = bpf_task_storage_get(&task_ctx_stor, p, 0, 0); @ scx_qmap.bpf.c:270
20: (18) r1 = 0xffff88810c333600      ; R1=map_ptr(map=task_ctx_stor,ks=4,vs=8)
22: (b7) r3 = 0                       ; R3=0
23: (b7) r4 = 0                       ; R4=0
24: (85) call bpf_task_storage_get#156        ;
R0=map_value_or_null(id=1,map=task_ctx_stor,ks=4,vs=8)
; if (!v || !v->taskc) @ scx_qmap.bpf.c:271
25: (15) if r0 == 0x0 goto pc+76      ;
R0=map_value(id=1,map=task_ctx_stor,ks=4,vs=8)
26: (79) r7 = *(u64 *)(r0 +0)         ;
R0=map_value(id=1,map=task_ctx_stor,ks=4,vs=8) R7=scalar()
27: (15) if r7 == 0x0 goto pc+74      ; R7=scalar(umin=1)
28: (bf) r9 = addr_space_cast(r7, 0, 1)       ; R7=scalar(umin=1) R9=arena
; if (taskc && taskc->fifo) { @ scx_qmap.bpf.c:593
29: (79) r8 = *(u64 *)(r9 +24)        ; R8=scalar() R9=arena
30: (15) if r8 == 0x0 goto pc+71      ; R8=scalar(umin=1)
; if (taskc->highpri) @ scx_qmap.bpf.c:594
31: (71) r1 = *(u8 *)(r9 +45)         ;
R1=scalar(smin=smin32=0,smax=umax=smax32=umax32=255,var_off=(0x0;
0xff)) R9=arena
32: (56) if w1 != 0x1 goto pc+9       ; R1=1
33: (b7) r1 = 1                       ; R1=1
; __sync_fetch_and_sub(&qa.nr_highpri_queued, 1); @ scx_qmap.bpf.c:595
34: (87) r1 = -r1                     ; R1=-1
35: (18) r2 = 0x10000fff7000          ; R2=scalar()
37: (07) r2 += 88                     ; R2=scalar()
38: (bf) r2 = addr_space_cast(r2, 0, 1)       ; R2=arena
39: (db) lock *(u64 *)(r2 +0) += r1   ; R1=-1 R2=arena
; struct qmap_fifo __arena *fifo = taskc->fifo; @ scx_qmap.bpf.c:320
40: (79) r8 = *(u64 *)(r9 +24)        ; R8=scalar() R9=arena
; if (!fifo) @ scx_qmap.bpf.c:323
41: (15) if r8 == 0x0 goto pc+60      ; R8=scalar(umin=1)
; lock = qa_q_lock(fifo->idx); @ scx_qmap.bpf.c:326
42: (bf) r2 = addr_space_cast(r8, 0, 1)       ; R2=arena R8=scalar(umin=1)
43: (61) r1 = *(u32 *)(r2 +16)        ;
R1=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff))
R2=arena
; switch (qid) { @ scx_qmap.bpf.c:129
44: (66) if w1 s> 0x1 goto pc+8       ;
R1=scalar(smin=0,smax=umax=0xffffffff,smax32=1,var_off=(0x0;
0xffffffff))
45: (18) r6 = 0xffff88810c3339f8      ;
R6=map_value(map=.data.qa_q_lock,ks=4,vs=4)
47: (16) if w1 == 0x0 goto pc+17      ;
R1=scalar(smin=0,smax=umax=0xffffffff,smax32=1,var_off=(0x0;
0xffffffff))
48: (16) if w1 == 0x1 goto pc+1 50: R2=arena R7=scalar(umin=1)
R8=scalar(umin=1) R9=arena R10=fp0
; switch (qid) { @ scx_qmap.bpf.c:129
50: (18) r6 = 0xffff88810c333bf8      ;
R6=map_value(map=.data.qa_q_lock,ks=4,vs=4)
52: (05) goto pc+12
65: (7b) *(u64 *)(r10 -16) = r2       ; R2=arena R10=fp0 fp-16=arena
; if (bpf_res_spin_lock(lock)) { @ scx_qmap.bpf.c:212
66: (bf) r1 = r6                      ;
R1=map_value(map=.data.qa_q_lock,ks=4,vs=4)
R6=map_value(map=.data.qa_q_lock,ks=4,vs=4)
67: (85) call bpf_res_spin_lock#85478         ; R0=0
68: (16) if w0 == 0x0 goto pc+9       ; R0=0
; if (taskc->fifo != fifo) { @ scx_qmap.bpf.c:331
78: (79) r1 = *(u64 *)(r9 +24)        ; R1=scalar() R9=arena
79: (1d) if r1 == r8 goto pc+1 81:
R6=map_value(map=.data.qa_q_lock,ks=4,vs=4) R7=scalar(umin=1)
R8=scalar(umin=1) R9=arena R10=fp0 fp-16=arena
; bpf_res_spin_unlock(lock); @ scx_qmap.bpf.c:332
81: (79) r2 = *(u64 *)(r9 +16)        ; R2=scalar() R9=arena
; if (taskc->q_next) @ scx_qmap.bpf.c:336
82: (79) r1 = *(u64 *)(r9 +8)         ; R1=scalar() R9=arena
83: (15) if r1 == 0x0 goto pc+4       ; R1=scalar(umin=1)
84: (bf) r3 = addr_space_cast(r1, 0, 1)       ; R1=scalar(umin=1) R3=arena
; taskc->q_next->q_prev = taskc->q_prev; @ scx_qmap.bpf.c:337
85: (7b) *(u64 *)(r3 +16) = r2        ; R2=scalar() R3=arena
; if (taskc->q_prev) @ scx_qmap.bpf.c:340
86: (79) r2 = *(u64 *)(r9 +16)        ; R2=scalar() R9=arena
; taskc->q_next->q_prev = taskc->q_prev; @ scx_qmap.bpf.c:337
87: (05) goto pc+2
; fifo->tail = taskc->q_prev; @ scx_qmap.bpf.c:339
90: (07) r7 += 8                      ; R7=scalar()
; if (taskc->q_prev) @ scx_qmap.bpf.c:340
91: (15) if r2 == 0x0 goto pc+2       ; R2=scalar(umin=1)
92: (07) r2 += 8                      ; R2=scalar()
93: (bf) r8 = r2                      ; R8=scalar(id=2)
94: (bf) r2 = addr_space_cast(r8, 0, 1)       ; R2=arena R8=scalar(id=2)
95: (7b) *(u64 *)(r2 +0) = r1         ; R1=scalar(umin=1) R2=arena
96: (b7) r1 = 0                       ; R1=0
; taskc->q_prev = NULL; @ scx_qmap.bpf.c:345
97: (7b) *(u64 *)(r7 +16) = r1
R7 invalid mem access 'scalar'
processed 71 insns (limit 1000000) max_states_per_insn 1 total_states
5 peak_states 5 mark_read 0
-- END PROG LOAD LOG --
libbpf: prog 'qmap_dequeue': failed to load: -EACCES
libbpf: failed to load object 'scx_qmap'
libbpf: failed to load BPF skeleton 'scx_qmap': -EACCES
[SCX_BUG] scx_qmap.c:497 (Permission denied)
Failed to load skel