Re: [PATCH bpf-next v2 1/2] bpf: Reject bpf_timer for PREEMPT_RT

[Eduard Zingerman <eddyz87@gmail.com>]

On Mon, 2025-09-08 at 12:20 -0700, Eduard Zingerman wrote:
> On Mon, 2025-09-08 at 12:40 +0800, Leon Hwang wrote:
> > When enable CONFIG_PREEMPT_RT, the kernel will panic when run timer
> > selftests by './test_progs -t timer':

Related discussions:
- https://lore.kernel.org/bpf/b634rejnvxqu6knjqlijosxrcnxbbpagt4de4pl6env6dwldz2@hoofqufparh5/T/
- https://lore.kernel.org/bpf/lhmdi6npaxqeuaumjhmq24ckpul7ufopwzxjbsezhepguqkxag@wolz4r2fazu2/T/

CC'ing Peilin.

> > 
> > [   35.955287] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
> > [   35.955312] in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 120, name: test_progs
> > [   35.955315] preempt_count: 1, expected: 0
> > [   35.955316] RCU nest depth: 0, expected: 0
> > [   35.955317] 2 locks held by test_progs/120:
> > [   35.955319]  #0: ffffffff8f1c3720 (rcu_read_lock_trace){....}-{0:0}, at: bpf_prog_test_run_syscall+0xc9/0x240
> > [   35.955358]  #1: ffff9155fbd331c8 ((&c->lock)){+.+.}-{3:3}, at: ___slab_alloc+0xb0/0xd20
> > [   35.955388] irq event stamp: 100
> > [   35.955389] hardirqs last  enabled at (99): [<ffffffff8dfcd890>] do_syscall_64+0x30/0x2d0
> > [   35.955414] hardirqs last disabled at (100): [<ffffffff8d4a9baa>] __bpf_async_init+0xca/0x310
> > [   35.955428] softirqs last  enabled at (0): [<ffffffff8d296cbb>] copy_process+0x9db/0x2000
> > [   35.955449] softirqs last disabled at (0): [<0000000000000000>] 0x0
> > [   35.955482] CPU: 1 UID: 0 PID: 120 Comm: test_progs Tainted: G           OE       6.17.0-rc1-gc5f5af560d8a #30 PREEMPT_{RT,(full)}
> > [   35.955487] Tainted: [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
> > [   35.955488] Hardware name: QEMU Ubuntu 24.04 PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
> > [   35.955491] Call Trace:
> > [   35.955493]  <TASK>
> > [   35.955499]  dump_stack_lvl+0x73/0xb0
> > [   35.955514]  dump_stack+0x14/0x20
> > [   35.955518]  __might_resched+0x167/0x230
> > [   35.955537]  rt_spin_lock+0x66/0x180
> > [   35.955543]  ? ___slab_alloc+0xb0/0xd20
> > [   35.955549]  ? bpf_map_kmalloc_node+0x7c/0x200
> > [   35.955560]  ___slab_alloc+0xb0/0xd20
> > [   35.955575]  ? __lock_acquire+0x43d/0x2590
> > [   35.955601]  __kmalloc_node_noprof+0x10b/0x410
> > [   35.955605]  ? __kmalloc_node_noprof+0x10b/0x410
> > [   35.955607]  ? bpf_map_kmalloc_node+0x7c/0x200
> > [   35.955616]  bpf_map_kmalloc_node+0x7c/0x200
> > [   35.955624]  __bpf_async_init+0xf8/0x310
> 
> The error is reported because of the kmalloc call in the __bpf_async_init, right?
> Instead of disabling timers for PREEMPT_RT, would it be possible to
> switch implementation to use kernel/bpf/memalloc.c:bpf_mem_alloc() instead?
> 
> > [   35.955633]  bpf_timer_init+0x37/0x40
> > [   35.955637]  bpf_prog_2287350dd5909839_start_cb+0x5d/0x91
> > [   35.955642]  bpf_prog_0d54653d8a74e954_start_timer+0x65/0x8a
> > [   35.955650]  bpf_prog_test_run_syscall+0x111/0x240
> > [   35.955660]  __sys_bpf+0x81c/0x2ab0
> > [   35.955665]  ? __might_fault+0x47/0x90
> > [   35.955700]  __x64_sys_bpf+0x1e/0x30
> > [   35.955703]  x64_sys_call+0x171d/0x20d0
> > [   35.955715]  do_syscall_64+0x6a/0x2d0
> > [   35.955722]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
> > [   35.955728] RIP: 0033:0x7fee4261225d
> > [   35.955734] Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 8b bb 0d 00 f7 d8 64 89 01 48
> > [   35.955736] RSP: 002b:00007fee424e5bd8 EFLAGS: 00000202 ORIG_RAX: 0000000000000141
> > [   35.955742] RAX: ffffffffffffffda RBX: 00007fee424e6cdc RCX: 00007fee4261225d
> > [   35.955744] RDX: 0000000000000050 RSI: 00007fee424e5c20 RDI: 000000000000000a
> > [   35.955745] RBP: 00007fee424e5bf0 R08: 0000000000000003 R09: 00007fee424e5c20
> > [   35.955747] R10: 00007fffc266f910 R11: 0000000000000202 R12: 00007fee424e66c0
> > [   35.955748] R13: ffffffffffffff08 R14: 0000000000000016 R15: 00007fffc266f650
> > [   35.955766]  </TASK>
> 
> [...]