On 1/21/26 2:43 PM, Andrew Morton wrote:

On Wed, 21 Jan 2026 14:10:36 -0500 Waiman Long <longman@redhat.com> wrote:

Commit 3acb913c9d5b ("mm/mm_init: use deferred_init_memmap_chunk()
in deferred_grow_zone()") made deferred_grow_zone() call
deferred_init_memmap_chunk() within a pgdat_resize_lock() critical
section with irqs disabled. It did check for irqs_disabled() in
deferred_init_memmap_chunk() to avoid calling cond_resched(). For a
PREEMPT_RT kernel build, however, spin_lock_irqsave() does not disable
interrupt but rcu_read_lock() is called. This leads to the following
bug report.

  BUG: sleeping function called from invalid context at mm/mm_init.c:2091
  in_atomic(): 0, irqs_disabled(): 0, non_block: 0, pid: 1, name: swapper/0
  preempt_count: 0, expected: 0
  RCU nest depth: 1, expected: 0
  3 locks held by swapper/0/1:
   #0: ffff80008471b7a0 (sched_domains_mutex){+.+.}-{4:4}, at: sched_domains_mutex_lock+0x28/0x40
   #1: ffff003bdfffef48 (&pgdat->node_size_lock){+.+.}-{3:3}, at: deferred_grow_zone+0x140/0x278
   #2: ffff800084acf600 (rcu_read_lock){....}-{1:3}, at: rt_spin_lock+0x1b4/0x408
  CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Tainted: G        W           6.19.0-rc6-test #1 PREEMPT_{RT,(full)
}
  Tainted: [W]=WARN
  Call trace:
   show_stack+0x20/0x38 (C)
   dump_stack_lvl+0xdc/0xf8
   dump_stack+0x1c/0x28
   __might_resched+0x384/0x530
   deferred_init_memmap_chunk+0x560/0x688
   deferred_grow_zone+0x190/0x278
   _deferred_grow_zone+0x18/0x30
   get_page_from_freelist+0x780/0xf78
   __alloc_frozen_pages_noprof+0x1dc/0x348
   alloc_slab_page+0x30/0x110
   allocate_slab+0x98/0x2a0
   new_slab+0x4c/0x80
   ___slab_alloc+0x5a4/0x770
   __slab_alloc.constprop.0+0x88/0x1e0
   __kmalloc_node_noprof+0x2c0/0x598
   __sdt_alloc+0x3b8/0x728
   build_sched_domains+0xe0/0x1260
   sched_init_domains+0x14c/0x1c8
   sched_init_smp+0x9c/0x1d0
   kernel_init_freeable+0x218/0x358
   kernel_init+0x28/0x208
   ret_from_fork+0x10/0x20

Fix it by checking rcu_preempt_depth() as well to prevent calling
cond_resched(). Note that CONFIG_PREEMPT_RCU should always be enabled
in a PREEMPT_RT kernel.

...

--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -2085,7 +2085,12 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
 
 			spfn = chunk_end;
 
-			if (irqs_disabled())
+			/*
+			 * pgdat_resize_lock() only disables irqs in non-RT
+			 * kernels but calls rcu_read_lock() in a PREEMPT_RT
+			 * kernel.
+			 */
+			if (irqs_disabled() || rcu_preempt_depth())
 				touch_nmi_watchdog();

rcu_preempt_depth() seems a fairly internal low-level thing - it's
rarely used.

That is true. Beside the scheduler, workqueue also use rcu_preempt_depth(). This API is included in "include/linux/rcupdate.h" which is included directly or indirectly by many kernel files. So even though it is rarely used, but it is still a public API.


Is there a more official way of detecting this condition?  Maybe even
#ifdef CONFIG_PREEMPT_RCU?

I am not aware of a more official way of detecting this. Maybe Sebastian has some ideas. rcu_preempt_count() is defined whether CONFIG_PREEMPT_RCU is defined or not. So we don't need a "#ifdef CONFIG_PREEMPT_RCU". Maybe I should explicitly include "include/linux/rcupdate.h" in mm/mm_init.c just to be sure.

CONFIG_PREEMPT_RCU defaults to on if PREMPT_RT is set. With !CONFIG_PREEMPT_RCU, rcu_preempt_depth() is hard-coded to 0 and will be optimized out.

Cheers,
Longman