Re: [PATCH] arm64/mm: Disable barrier batching in interrupt contexts

[Catalin Marinas <catalin.marinas@arm.com>]

On Mon, May 12, 2025 at 11:22:40AM +0100, Ryan Roberts wrote:
> @@ -79,7 +83,9 @@ static inline void queue_pte_barriers(void)
>  #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
>  static inline void arch_enter_lazy_mmu_mode(void)
>  {
> -	VM_WARN_ON(in_interrupt());
> +	if (in_interrupt())
> +		return;
> +
>  	VM_WARN_ON(test_thread_flag(TIF_LAZY_MMU));

I still get this warning trigger with some debugging enabled (more
specifically, CONFIG_DEBUG_PAGEALLOC). Patch applied on top of the arm64
for-kernelci.

Is it because the unmap code uses arch_enter_lazy_mmu_mode() already and
__apply_to_page_range() via __kernel_map_pages() is attempting another
nested call? I think it's still safe, we just drop the optimisation in
the outer code and issue the barriers immediately. So maybe drop this
warning as well but add a comment on how nesting works.

------------[ cut here ]------------
WARNING: CPU: 6 PID: 1 at arch/arm64/include/asm/pgtable.h:89 __apply_to_page_range+0x85c/0x9f8
Modules linked in: ip_tables x_tables ipv6
CPU: 6 UID: 0 PID: 1 Comm: systemd Not tainted 6.15.0-rc5-00075-g676795fe9cf6 #1 PREEMPT 
Hardware name: QEMU KVM Virtual Machine, BIOS 2024.08-4 10/25/2024
pstate: 40400005 (nZcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : __apply_to_page_range+0x85c/0x9f8
lr : __apply_to_page_range+0x2b4/0x9f8
sp : ffff80008009b3c0
x29: ffff80008009b460 x28: ffff0000c43a3000 x27: ffff0001ff62b108
x26: ffff0000c43a4000 x25: 0000000000000001 x24: 0010000000000001
x23: ffffbf24c9c209c0 x22: ffff80008009b4d0 x21: ffffbf24c74a3b20
x20: ffff0000c43a3000 x19: ffff0001ff609d18 x18: 0000000000000001
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000003
x14: 0000000000000028 x13: ffffbf24c97c1000 x12: ffff0000c43a3fff
x11: ffffbf24cacc9a70 x10: ffff0000c43a3fff x9 : ffff0001fffff018
x8 : 0000000000000012 x7 : ffff0000c43a4000 x6 : ffff0000c43a4000
x5 : ffffbf24c9c209c0 x4 : ffff0000c43a3fff x3 : ffff0001ff609000
x2 : 0000000000000d18 x1 : ffff0000c03e8000 x0 : 0000000080000000
Call trace:
 __apply_to_page_range+0x85c/0x9f8 (P)
 apply_to_page_range+0x14/0x20
 set_memory_valid+0x5c/0xd8
 __kernel_map_pages+0x84/0xc0
 get_page_from_freelist+0x1110/0x1340
 __alloc_frozen_pages_noprof+0x114/0x1178
 alloc_pages_mpol+0xb8/0x1d0
 alloc_frozen_pages_noprof+0x48/0xc0
 alloc_pages_noprof+0x10/0x60
 get_free_pages_noprof+0x14/0x90
 __tlb_remove_folio_pages_size.isra.0+0xe4/0x140
 __tlb_remove_folio_pages+0x10/0x20
 unmap_page_range+0xa1c/0x14c0
 unmap_single_vma.isra.0+0x48/0x90
 unmap_vmas+0xe0/0x200
 vms_clear_ptes+0xf4/0x140
 vms_complete_munmap_vmas+0x7c/0x208
 do_vmi_align_munmap+0x180/0x1a8
 do_vmi_munmap+0xac/0x188
 __vm_munmap+0xe0/0x1e0
 __arm64_sys_munmap+0x20/0x38
 invoke_syscall+0x48/0x104
 el0_svc_common.constprop.0+0x40/0xe0
 do_el0_svc+0x1c/0x28
 el0_svc+0x4c/0x16c
 el0t_64_sync_handler+0x10c/0x140
 el0t_64_sync+0x198/0x19c
irq event stamp: 281312
hardirqs last  enabled at (281311): [<ffffbf24c780fd04>] bad_range+0x164/0x1c0
hardirqs last disabled at (281312): [<ffffbf24c89c4550>] el1_dbg+0x24/0x98
softirqs last  enabled at (281054): [<ffffbf24c752d99c>] handle_softirqs+0x4cc/0x518
softirqs last disabled at (281019): [<ffffbf24c7450694>] __do_softirq+0x14/0x20
---[ end trace 0000000000000000 ]---

-- 
Catalin