"alloc_tag was not set" when running mm/ksft_hmm.sh

"alloc_tag was not set" when running mm/ksft_hmm.sh

[Zenghui Yu]

Hi all,

Running mm/ksft_hmm.sh triggers the following splat:

 ------------[ cut here ]------------
 alloc_tag was not set
 WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0,
CPU#5: hmm-tests/2020
 Modules linked in: test_hmm rfkill drm backlight fuse
 CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted
7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
 Hardware name: QEMU QEMU Virtual Machine, BIOS
edk2-stable202408-prebuilt.qemu.org 08/13/2024
 pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
 pc : ___free_pages+0x2a0/0x2d0
 lr : ___free_pages+0x2a0/0x2d0
 sp : ffff80008345b530
 x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
 x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
 x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
 x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
 x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
 x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
 x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
 x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
 x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
 x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
 Call trace:
  ___free_pages+0x2a0/0x2d0 (P)
  __free_pages+0x14/0x20
  dmirror_devmem_free+0x13c/0x158 [test_hmm]
  free_zone_device_folio+0x144/0x1e4
  __folio_put+0x124/0x130
  free_folio_and_swap_cache+0xa8/0xcc
  __folio_split+0x664/0x7fc
  split_folio_to_list+0x50/0x5c
  migrate_vma_split_folio+0x13c/0x25c
  migrate_vma_collect_pmd+0xed4/0xf68
  walk_pgd_range+0x598/0x9a0
  __walk_page_range+0x90/0x1a0
  walk_page_range_mm_unsafe+0x194/0x20c
  walk_page_range+0x20/0x2c
  migrate_vma_setup+0x18c/0x224
  dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
  do_swap_page+0x1458/0x185c
  __handle_mm_fault+0x85c/0x1ba0
  handle_mm_fault+0xb0/0x290
  do_page_fault+0x1f8/0x6f8
  do_translation_fault+0x60/0x6c
  do_mem_abort+0x44/0x94
  el0_da+0x30/0xdc
  el0t_64_sync_handler+0xd0/0xe4
  el0t_64_sync+0x198/0x19c
 ---[ end trace 0000000000000000 ]---
 lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has
16744448 allocated at module unload


It was tested on kernel built with arm64's virt.config and

+CONFIG_ZONE_DEVICE=y
+CONFIG_DEVICE_PRIVATE=y
+CONFIG_TEST_HMM=m
+CONFIG_MEM_ALLOC_PROFILING=y
+CONFIG_MEM_ALLOC_PROFILING_DEBUG=y

Thanks,
Zenghui

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[David Hildenbrand (Arm)]

On 5/6/26 17:42, Zenghui Yu wrote:
> Hi all,
> 
> Running mm/ksft_hmm.sh triggers the following splat:
> 
>  ------------[ cut here ]------------
>  alloc_tag was not set
>  WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0,
> CPU#5: hmm-tests/2020
>  Modules linked in: test_hmm rfkill drm backlight fuse
>  CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted
> 7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
>  Hardware name: QEMU QEMU Virtual Machine, BIOS
> edk2-stable202408-prebuilt.qemu.org 08/13/2024
>  pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
>  pc : ___free_pages+0x2a0/0x2d0
>  lr : ___free_pages+0x2a0/0x2d0
>  sp : ffff80008345b530
>  x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
>  x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
>  x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
>  x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
>  x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
>  x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
>  x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
>  x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
>  x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
>  x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
>  Call trace:
>   ___free_pages+0x2a0/0x2d0 (P)
>   __free_pages+0x14/0x20
>   dmirror_devmem_free+0x13c/0x158 [test_hmm]
>   free_zone_device_folio+0x144/0x1e4
>   __folio_put+0x124/0x130
>   free_folio_and_swap_cache+0xa8/0xcc
>   __folio_split+0x664/0x7fc
>   split_folio_to_list+0x50/0x5c
>   migrate_vma_split_folio+0x13c/0x25c
>   migrate_vma_collect_pmd+0xed4/0xf68
>   walk_pgd_range+0x598/0x9a0
>   __walk_page_range+0x90/0x1a0
>   walk_page_range_mm_unsafe+0x194/0x20c
>   walk_page_range+0x20/0x2c
>   migrate_vma_setup+0x18c/0x224
>   dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
>   do_swap_page+0x1458/0x185c
>   __handle_mm_fault+0x85c/0x1ba0
>   handle_mm_fault+0xb0/0x290
>   do_page_fault+0x1f8/0x6f8
>   do_translation_fault+0x60/0x6c
>   do_mem_abort+0x44/0x94
>   el0_da+0x30/0xdc
>   el0t_64_sync_handler+0xd0/0xe4
>   el0t_64_sync+0x198/0x19c
>  ---[ end trace 0000000000000000 ]---
>  lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has
> 16744448 allocated at module unload
> 
> 
> It was tested on kernel built with arm64's virt.config and
> 
> +CONFIG_ZONE_DEVICE=y
> +CONFIG_DEVICE_PRIVATE=y
> +CONFIG_TEST_HMM=m
> +CONFIG_MEM_ALLOC_PROFILING=y
> +CONFIG_MEM_ALLOC_PROFILING_DEBUG=y

I assume there is a weird interaction between alloc tags and simulated
ZONE_DEVICE memory in test_hmm.c

@Alistair, any idea?

-- 
Cheers,

David

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Alistair Popple]

On 2026-05-08 at 21:53 +1000, "David Hildenbrand (Arm)" <david@kernel.org> wrote...
> On 5/6/26 17:42, Zenghui Yu wrote:
> > Hi all,
> > 
> > Running mm/ksft_hmm.sh triggers the following splat:
> > 
> >  ------------[ cut here ]------------
> >  alloc_tag was not set
> >  WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0,
> > CPU#5: hmm-tests/2020
> >  Modules linked in: test_hmm rfkill drm backlight fuse
> >  CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted
> > 7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
> >  Hardware name: QEMU QEMU Virtual Machine, BIOS
> > edk2-stable202408-prebuilt.qemu.org 08/13/2024
> >  pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
> >  pc : ___free_pages+0x2a0/0x2d0
> >  lr : ___free_pages+0x2a0/0x2d0
> >  sp : ffff80008345b530
> >  x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
> >  x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
> >  x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
> >  x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
> >  x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
> >  x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
> >  x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
> >  x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
> >  x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
> >  x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
> >  Call trace:
> >   ___free_pages+0x2a0/0x2d0 (P)
> >   __free_pages+0x14/0x20
> >   dmirror_devmem_free+0x13c/0x158 [test_hmm]
> >   free_zone_device_folio+0x144/0x1e4
> >   __folio_put+0x124/0x130
> >   free_folio_and_swap_cache+0xa8/0xcc
> >   __folio_split+0x664/0x7fc
> >   split_folio_to_list+0x50/0x5c
> >   migrate_vma_split_folio+0x13c/0x25c
> >   migrate_vma_collect_pmd+0xed4/0xf68
> >   walk_pgd_range+0x598/0x9a0
> >   __walk_page_range+0x90/0x1a0
> >   walk_page_range_mm_unsafe+0x194/0x20c
> >   walk_page_range+0x20/0x2c
> >   migrate_vma_setup+0x18c/0x224
> >   dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
> >   do_swap_page+0x1458/0x185c
> >   __handle_mm_fault+0x85c/0x1ba0
> >   handle_mm_fault+0xb0/0x290
> >   do_page_fault+0x1f8/0x6f8
> >   do_translation_fault+0x60/0x6c
> >   do_mem_abort+0x44/0x94
> >   el0_da+0x30/0xdc
> >   el0t_64_sync_handler+0xd0/0xe4
> >   el0t_64_sync+0x198/0x19c
> >  ---[ end trace 0000000000000000 ]---
> >  lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has
> > 16744448 allocated at module unload
> > 
> > 
> > It was tested on kernel built with arm64's virt.config and
> > 
> > +CONFIG_ZONE_DEVICE=y
> > +CONFIG_DEVICE_PRIVATE=y
> > +CONFIG_TEST_HMM=m
> > +CONFIG_MEM_ALLOC_PROFILING=y
> > +CONFIG_MEM_ALLOC_PROFILING_DEBUG=y
> 
> I assume there is a weird interaction between alloc tags and simulated
> ZONE_DEVICE memory in test_hmm.c
> 
> @Alistair, any idea?

I haven't managed to reproduce this but could imagine that would be the problem
- not sure page_ext is supported for ZONE_DEVICE PRIVATE pages.

I'm still travelling back from LSFMM so will take a closer look next week.
Thanks!

 - Alistair

> -- 
> Cheers,
> 
> David

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Zenghui Yu]

On 2026/5/8 19:53, David Hildenbrand (Arm) wrote:
> On 5/6/26 17:42, Zenghui Yu wrote:
> > Hi all,
> >
> > Running mm/ksft_hmm.sh triggers the following splat:
> >
> >  ------------[ cut here ]------------
> >  alloc_tag was not set
> >  WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0,
> > CPU#5: hmm-tests/2020
> >  Modules linked in: test_hmm rfkill drm backlight fuse
> >  CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted
> > 7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
> >  Hardware name: QEMU QEMU Virtual Machine, BIOS
> > edk2-stable202408-prebuilt.qemu.org 08/13/2024
> >  pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
> >  pc : ___free_pages+0x2a0/0x2d0
> >  lr : ___free_pages+0x2a0/0x2d0
> >  sp : ffff80008345b530
> >  x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
> >  x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
> >  x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
> >  x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
> >  x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
> >  x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
> >  x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
> >  x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
> >  x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
> >  x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
> >  Call trace:
> >   ___free_pages+0x2a0/0x2d0 (P)
> >   __free_pages+0x14/0x20
> >   dmirror_devmem_free+0x13c/0x158 [test_hmm]
> >   free_zone_device_folio+0x144/0x1e4
> >   __folio_put+0x124/0x130
> >   free_folio_and_swap_cache+0xa8/0xcc
> >   __folio_split+0x664/0x7fc
> >   split_folio_to_list+0x50/0x5c
> >   migrate_vma_split_folio+0x13c/0x25c
> >   migrate_vma_collect_pmd+0xed4/0xf68
> >   walk_pgd_range+0x598/0x9a0
> >   __walk_page_range+0x90/0x1a0
> >   walk_page_range_mm_unsafe+0x194/0x20c
> >   walk_page_range+0x20/0x2c
> >   migrate_vma_setup+0x18c/0x224
> >   dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
> >   do_swap_page+0x1458/0x185c
> >   __handle_mm_fault+0x85c/0x1ba0
> >   handle_mm_fault+0xb0/0x290
> >   do_page_fault+0x1f8/0x6f8
> >   do_translation_fault+0x60/0x6c
> >   do_mem_abort+0x44/0x94
> >   el0_da+0x30/0xdc
> >   el0t_64_sync_handler+0xd0/0xe4
> >   el0t_64_sync+0x198/0x19c
> >  ---[ end trace 0000000000000000 ]---
> >  lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has
> > 16744448 allocated at module unload
> >
> >
> > It was tested on kernel built with arm64's virt.config and
> >
> > +CONFIG_ZONE_DEVICE=y
> > +CONFIG_DEVICE_PRIVATE=y
> > +CONFIG_TEST_HMM=m
> > +CONFIG_MEM_ALLOC_PROFILING=y
> > +CONFIG_MEM_ALLOC_PROFILING_DEBUG=y
> 
> I assume there is a weird interaction between alloc tags and simulated
> ZONE_DEVICE memory in test_hmm.c

FYI this can be reproduced by running the migrate_partial_unmap_fault
test case.

TEST_F(hmm, migrate_partial_unmap_fault)
{
	buffer->mirror = malloc(TWOMEG);
	buffer->ptr = map;	// points to a THP

	/* Initialize buffer in system memory. */
	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
		ptr[i] = i;

	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);

	munmap(buffer->ptr, ONEMEG);

	/* Fault pages back to system memory and check them. */
	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
		if (i * sizeof(int) < 0 ||
		    i * sizeof(int) >= ONEMEG)
			ASSERT_EQ(ptr[i], i);	// triggers a fault ->


dmirror_devmem_fault()
	migrate_vma_setup()
		migrate_vma_collect_pmd()
			// !pte_present(pte) && folio_test_large(folio)
			migrate_vma_split_folio()
				split_folio()
					[...]

__folio_split() {
	unmap_folio();

	__folio_freeze_and_split_unmapped() {
		__split_unmapped_folio();

		for (...) {
			zone_device_private_split_cb(.., new_folio);
			// -> dmirror_devmem_folio_split() which doesn't
			// set alloc tag for the backing system memory
			// page being split, i.e., rpage_tail
		}

		zone_device_private_split_cb(.., NULL);
	}

	remap_page();

	for (...)
		free_folio_and_swap_cache(new_folio);
		// -> dmirror_devmem_free()/__free_page() which warns if
		// the page being freed doesn't have alloc tag set, in
		// alloc_tag_sub_check().
}

The WARN disappears with the following diff. But I'm not sure if I've
missed more important points (which is likely to happen ;-) ).

diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
index ed1bdcf1f8ab..eefa2a739917 100644
--- a/lib/alloc_tag.c
+++ b/lib/alloc_tag.c
@@ -191,6 +191,7 @@ void pgalloc_tag_split(struct folio *folio, int
old_order, int new_order)
 		}
 	}
 }
+EXPORT_SYMBOL(pgalloc_tag_split);

 void pgalloc_tag_swap(struct folio *new, struct folio *old)
 {
diff --git a/lib/test_hmm.c b/lib/test_hmm.c
index 213504915737..3bec51828916 100644
--- a/lib/test_hmm.c
+++ b/lib/test_hmm.c
@@ -1713,6 +1713,7 @@ static void dmirror_devmem_folio_split(struct
folio *head, struct folio *tail)
 	rfolio = page_folio(rpage);

 	if (tail == NULL) {
+		pgalloc_tag_split(rfolio, folio_order(rfolio), 0);
 		folio_reset_order(rfolio);
 		rfolio->mapping = NULL;
 		folio_set_count(rfolio, 1);

Thanks,
Zenghui

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[David Hildenbrand (Arm)]

On 5/11/26 14:19, Zenghui Yu wrote:
> On 2026/5/8 19:53, David Hildenbrand (Arm) wrote:
>> On 5/6/26 17:42, Zenghui Yu wrote:
>>> Hi all,
>>>
>>> Running mm/ksft_hmm.sh triggers the following splat:
>>>
>>>  ------------[ cut here ]------------
>>>  alloc_tag was not set
>>>  WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0,
>>> CPU#5: hmm-tests/2020
>>>  Modules linked in: test_hmm rfkill drm backlight fuse
>>>  CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted
>>> 7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
>>>  Hardware name: QEMU QEMU Virtual Machine, BIOS
>>> edk2-stable202408-prebuilt.qemu.org 08/13/2024
>>>  pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
>>>  pc : ___free_pages+0x2a0/0x2d0
>>>  lr : ___free_pages+0x2a0/0x2d0
>>>  sp : ffff80008345b530
>>>  x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
>>>  x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
>>>  x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
>>>  x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
>>>  x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
>>>  x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
>>>  x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
>>>  x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
>>>  x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
>>>  x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
>>>  Call trace:
>>>   ___free_pages+0x2a0/0x2d0 (P)
>>>   __free_pages+0x14/0x20
>>>   dmirror_devmem_free+0x13c/0x158 [test_hmm]
>>>   free_zone_device_folio+0x144/0x1e4
>>>   __folio_put+0x124/0x130
>>>   free_folio_and_swap_cache+0xa8/0xcc
>>>   __folio_split+0x664/0x7fc
>>>   split_folio_to_list+0x50/0x5c
>>>   migrate_vma_split_folio+0x13c/0x25c
>>>   migrate_vma_collect_pmd+0xed4/0xf68
>>>   walk_pgd_range+0x598/0x9a0
>>>   __walk_page_range+0x90/0x1a0
>>>   walk_page_range_mm_unsafe+0x194/0x20c
>>>   walk_page_range+0x20/0x2c
>>>   migrate_vma_setup+0x18c/0x224
>>>   dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
>>>   do_swap_page+0x1458/0x185c
>>>   __handle_mm_fault+0x85c/0x1ba0
>>>   handle_mm_fault+0xb0/0x290
>>>   do_page_fault+0x1f8/0x6f8
>>>   do_translation_fault+0x60/0x6c
>>>   do_mem_abort+0x44/0x94
>>>   el0_da+0x30/0xdc
>>>   el0t_64_sync_handler+0xd0/0xe4
>>>   el0t_64_sync+0x198/0x19c
>>>  ---[ end trace 0000000000000000 ]---
>>>  lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has
>>> 16744448 allocated at module unload
>>>
>>>
>>> It was tested on kernel built with arm64's virt.config and
>>>
>>> +CONFIG_ZONE_DEVICE=y
>>> +CONFIG_DEVICE_PRIVATE=y
>>> +CONFIG_TEST_HMM=m
>>> +CONFIG_MEM_ALLOC_PROFILING=y
>>> +CONFIG_MEM_ALLOC_PROFILING_DEBUG=y
>>
>> I assume there is a weird interaction between alloc tags and simulated
>> ZONE_DEVICE memory in test_hmm.c
> 
> FYI this can be reproduced by running the migrate_partial_unmap_fault
> test case.
> 
> TEST_F(hmm, migrate_partial_unmap_fault)
> {
> 	buffer->mirror = malloc(TWOMEG);
> 	buffer->ptr = map;	// points to a THP
> 
> 	/* Initialize buffer in system memory. */
> 	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
> 		ptr[i] = i;
> 
> 	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
> 
> 	munmap(buffer->ptr, ONEMEG);
> 
> 	/* Fault pages back to system memory and check them. */
> 	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
> 		if (i * sizeof(int) < 0 ||
> 		    i * sizeof(int) >= ONEMEG)
> 			ASSERT_EQ(ptr[i], i);	// triggers a fault ->
> 
> 
> dmirror_devmem_fault()
> 	migrate_vma_setup()
> 		migrate_vma_collect_pmd()
> 			// !pte_present(pte) && folio_test_large(folio)
> 			migrate_vma_split_folio()
> 				split_folio()
> 					[...]
> 
> __folio_split() {
> 	unmap_folio();
> 
> 	__folio_freeze_and_split_unmapped() {
> 		__split_unmapped_folio();
> 
> 		for (...) {
> 			zone_device_private_split_cb(.., new_folio);
> 			// -> dmirror_devmem_folio_split() which doesn't
> 			// set alloc tag for the backing system memory
> 			// page being split, i.e., rpage_tail
> 		}
> 
> 		zone_device_private_split_cb(.., NULL);
> 	}
> 
> 	remap_page();
> 
> 	for (...)
> 		free_folio_and_swap_cache(new_folio);
> 		// -> dmirror_devmem_free()/__free_page() which warns if
> 		// the page being freed doesn't have alloc tag set, in
> 		// alloc_tag_sub_check().
> }
> 
> The WARN disappears with the following diff. But I'm not sure if I've
> missed more important points (which is likely to happen ;-) ).
> 
> diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> index ed1bdcf1f8ab..eefa2a739917 100644
> --- a/lib/alloc_tag.c
> +++ b/lib/alloc_tag.c
> @@ -191,6 +191,7 @@ void pgalloc_tag_split(struct folio *folio, int
> old_order, int new_order)
>  		}
>  	}
>  }
> +EXPORT_SYMBOL(pgalloc_tag_split);
> 
>  void pgalloc_tag_swap(struct folio *new, struct folio *old)
>  {
> diff --git a/lib/test_hmm.c b/lib/test_hmm.c
> index 213504915737..3bec51828916 100644
> --- a/lib/test_hmm.c
> +++ b/lib/test_hmm.c
> @@ -1713,6 +1713,7 @@ static void dmirror_devmem_folio_split(struct
> folio *head, struct folio *tail)
>  	rfolio = page_folio(rpage);
> 
>  	if (tail == NULL) {
> +		pgalloc_tag_split(rfolio, folio_order(rfolio), 0);
>  		folio_reset_order(rfolio);
>  		rfolio->mapping = NULL;
>  		folio_set_count(rfolio, 1);
> 
> Thanks,
> Zenghui


zone_device_private_split_cb(), that ends up calling ->folio_split().

We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
__folio_freeze_and_split_unmapped() before calling
zone_device_private_split_cb() when iterating the folios.

The zone_device_private_split_cb(folio, NULL); is then called on the first folio
after looping over the other (new) folios.

I would assume that __folio_freeze_and_split_unmapped() would already do the
right thing?

Maybe the issue is the hard-coded folio_reset_order() in
dmirror_devmem_folio_split(), where we seem to assume that we split to an
order-0 folio?

-- 
Cheers,

David

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Zenghui Yu]

Hi David,

On 5/11/26 8:47 PM, David Hildenbrand (Arm) wrote:
> On 5/11/26 14:19, Zenghui Yu wrote:
> > On 2026/5/8 19:53, David Hildenbrand (Arm) wrote:
> > > On 5/6/26 17:42, Zenghui Yu wrote:
> > > > Hi all,
> > > >
> > > > Running mm/ksft_hmm.sh triggers the following splat:
> > > >
> > > >  ------------[ cut here ]------------
> > > >  alloc_tag was not set
> > > >  WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0, CPU#5: hmm-tests/2020
> > > >  Modules linked in: test_hmm rfkill drm backlight fuse
> > > >  CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted 7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
> > > >  Hardware name: QEMU QEMU Virtual Machine, BIOS edk2-stable202408-prebuilt.qemu.org 08/13/2024
> > > >  pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
> > > >  pc : ___free_pages+0x2a0/0x2d0
> > > >  lr : ___free_pages+0x2a0/0x2d0
> > > >  sp : ffff80008345b530
> > > >  x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
> > > >  x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
> > > >  x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
> > > >  x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
> > > >  x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
> > > >  x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
> > > >  x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
> > > >  x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
> > > >  x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
> > > >  x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
> > > >  Call trace:
> > > >   ___free_pages+0x2a0/0x2d0 (P)
> > > >   __free_pages+0x14/0x20
> > > >   dmirror_devmem_free+0x13c/0x158 [test_hmm]
> > > >   free_zone_device_folio+0x144/0x1e4
> > > >   __folio_put+0x124/0x130
> > > >   free_folio_and_swap_cache+0xa8/0xcc
> > > >   __folio_split+0x664/0x7fc
> > > >   split_folio_to_list+0x50/0x5c
> > > >   migrate_vma_split_folio+0x13c/0x25c
> > > >   migrate_vma_collect_pmd+0xed4/0xf68
> > > >   walk_pgd_range+0x598/0x9a0
> > > >   __walk_page_range+0x90/0x1a0
> > > >   walk_page_range_mm_unsafe+0x194/0x20c
> > > >   walk_page_range+0x20/0x2c
> > > >   migrate_vma_setup+0x18c/0x224
> > > >   dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
> > > >   do_swap_page+0x1458/0x185c
> > > >   __handle_mm_fault+0x85c/0x1ba0
> > > >   handle_mm_fault+0xb0/0x290
> > > >   do_page_fault+0x1f8/0x6f8
> > > >   do_translation_fault+0x60/0x6c
> > > >   do_mem_abort+0x44/0x94
> > > >   el0_da+0x30/0xdc
> > > >   el0t_64_sync_handler+0xd0/0xe4
> > > >   el0t_64_sync+0x198/0x19c
> > > >  ---[ end trace 0000000000000000 ]---
> > > >  lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has 16744448 allocated at module unload
> > > >
> > > >
> > > > It was tested on kernel built with arm64's virt.config and
> > > >
> > > > +CONFIG_ZONE_DEVICE=y
> > > > +CONFIG_DEVICE_PRIVATE=y
> > > > +CONFIG_TEST_HMM=m
> > > > +CONFIG_MEM_ALLOC_PROFILING=y
> > > > +CONFIG_MEM_ALLOC_PROFILING_DEBUG=y
> > >
> > > I assume there is a weird interaction between alloc tags and simulated
> > > ZONE_DEVICE memory in test_hmm.c
> >
> > FYI this can be reproduced by running the migrate_partial_unmap_fault
> > test case.
> >
> > TEST_F(hmm, migrate_partial_unmap_fault)
> > {
> > 	buffer->mirror = malloc(TWOMEG);
> > 	buffer->ptr = map;	// points to a THP
> >
> > 	/* Initialize buffer in system memory. */
> > 	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
> > 		ptr[i] = i;
> >
> > 	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
> >
> > 	munmap(buffer->ptr, ONEMEG);
> >
> > 	/* Fault pages back to system memory and check them. */
> > 	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
> > 		if (i * sizeof(int) < 0 ||
> > 		    i * sizeof(int) >= ONEMEG)
> > 			ASSERT_EQ(ptr[i], i);	// triggers a fault ->
> >
> >
> > dmirror_devmem_fault()
> > 	migrate_vma_setup()
> > 		migrate_vma_collect_pmd()
> > 			// !pte_present(pte) && folio_test_large(folio)
> > 			migrate_vma_split_folio()
> > 				split_folio()
> > 					[...]
> >
> > __folio_split() {
> > 	unmap_folio();
> >
> > 	__folio_freeze_and_split_unmapped() {
> > 		__split_unmapped_folio();
> >
> > 		for (...) {
> > 			zone_device_private_split_cb(.., new_folio);
> > 			// -> dmirror_devmem_folio_split() which doesn't
> > 			// set alloc tag for the backing system memory
> > 			// page being split, i.e., rpage_tail
> > 		}
> >
> > 		zone_device_private_split_cb(.., NULL);
> > 	}
> >
> > 	remap_page();
> >
> > 	for (...)
> > 		free_folio_and_swap_cache(new_folio);
> > 		// -> dmirror_devmem_free()/__free_page() which warns if
> > 		// the page being freed doesn't have alloc tag set, in
> > 		// alloc_tag_sub_check().
> > }
> >
> > The WARN disappears with the following diff. But I'm not sure if I've
> > missed more important points (which is likely to happen ;-) ).
> >
> > diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> > index ed1bdcf1f8ab..eefa2a739917 100644
> > --- a/lib/alloc_tag.c
> > +++ b/lib/alloc_tag.c
> > @@ -191,6 +191,7 @@ void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
> >  		}
> >  	}
> >  }
> > +EXPORT_SYMBOL(pgalloc_tag_split);
> >
> >  void pgalloc_tag_swap(struct folio *new, struct folio *old)
> >  {
> > diff --git a/lib/test_hmm.c b/lib/test_hmm.c
> > index 213504915737..3bec51828916 100644
> > --- a/lib/test_hmm.c
> > +++ b/lib/test_hmm.c
> > @@ -1713,6 +1713,7 @@ static void dmirror_devmem_folio_split(struct folio *head, struct folio *tail)
> >  	rfolio = page_folio(rpage);
> >
> >  	if (tail == NULL) {
> > +		pgalloc_tag_split(rfolio, folio_order(rfolio), 0);
> >  		folio_reset_order(rfolio);
> >  		rfolio->mapping = NULL;
> >  		folio_set_count(rfolio, 1);
> 
> 
> zone_device_private_split_cb(), that ends up calling ->folio_split().
> 
> We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
> __folio_freeze_and_split_unmapped() before calling
> zone_device_private_split_cb() when iterating the folios.

If I read the code correctly, pgalloc_tag_split() in
__split_unmapped_folio() deals with device private pages' alloc tag. But
what alloc_tag_sub_check() warns on are real system memory pages (device
page's backing page), which are allocated by
dmirror_devmem_alloc_page()/folio_page().

static void dmirror_devmem_folio_split(struct folio *head, struct folio
*tail)
{
	struct page *rpage = BACKING_PAGE(folio_page(head, 0));

Thanks,
Zenghui

> The zone_device_private_split_cb(folio, NULL); is then called on the first folio
> after looping over the other (new) folios.
> 
> I would assume that __folio_freeze_and_split_unmapped() would already do the
> right thing?
> 
> Maybe the issue is the hard-coded folio_reset_order() in
> dmirror_devmem_folio_split(), where we seem to assume that we split to an
> order-0 folio?
> 

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Zenghui Yu]

On 2026/5/12 0:38, Zenghui Yu wrote:
> Hi David,
> 
> On 5/11/26 8:47 PM, David Hildenbrand (Arm) wrote:
> >
> > We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
> > __folio_freeze_and_split_unmapped() before calling
> > zone_device_private_split_cb() when iterating the folios.
> 
> If I read the code correctly, pgalloc_tag_split() in
> __split_unmapped_folio() deals with device private pages' alloc tag. But
> what alloc_tag_sub_check() warns on are real system memory pages (device
> page's backing page), which are allocated by
> dmirror_devmem_alloc_page()/folio_page().

Sorry, I meant dmirror_devmem_alloc_page()/folio_alloc().

Thanks,
Zenghui

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Alistair Popple]

On 2026-05-12 at 02:38 +1000, Zenghui Yu <zenghui.yu@linux.dev> wrote...
> Hi David,
> 
> On 5/11/26 8:47 PM, David Hildenbrand (Arm) wrote:
> > On 5/11/26 14:19, Zenghui Yu wrote:
> > > On 2026/5/8 19:53, David Hildenbrand (Arm) wrote:
> > > > On 5/6/26 17:42, Zenghui Yu wrote:
> > > > > Hi all,
> > > > >
> > > > > Running mm/ksft_hmm.sh triggers the following splat:
> > > > >
> > > > >  ------------[ cut here ]------------
> > > > >  alloc_tag was not set
> > > > >  WARNING: ./include/linux/alloc_tag.h:164 at ___free_pages+0x2a0/0x2d0, CPU#5: hmm-tests/2020
> > > > >  Modules linked in: test_hmm rfkill drm backlight fuse
> > > > >  CPU: 5 UID: 0 PID: 2020 Comm: hmm-tests Kdump: loaded Not tainted 7.1.0-rc2-00099-gadc1e5c6203c-dirty #285 PREEMPT
> > > > >  Hardware name: QEMU QEMU Virtual Machine, BIOS edk2-stable202408-prebuilt.qemu.org 08/13/2024
> > > > >  pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
> > > > >  pc : ___free_pages+0x2a0/0x2d0
> > > > >  lr : ___free_pages+0x2a0/0x2d0
> > > > >  sp : ffff80008345b530
> > > > >  x29: ffff80008345b530 x28: ffff80008345b700 x27: ffffffffbfff8040
> > > > >  x26: ffff0000c41cb360 x25: ffff0000c0c64008 x24: ffff800081aae400
> > > > >  x23: 05ffff0000000200 x22: 0000000000000000 x21: 0000000000000000
> > > > >  x20: fffffdffc5f20040 x19: 0000000000000000 x18: fffffffffffe7c78
> > > > >  x17: 0000000000000000 x16: 0000000000000000 x15: fffffffffffe7c98
> > > > >  x14: 00000000000001d1 x13: ffff8000818f3d58 x12: 0000000000000573
> > > > >  x11: fffffffffffe7c98 x10: ffff80008194bd58 x9 : 3ffffffffffff000
> > > > >  x8 : ffff8000818f3d58 x7 : ffff80008194bd58 x6 : 0000000000000000
> > > > >  x5 : ffff0001fedb1088 x4 : 0000000000000001 x3 : 0000000000000000
> > > > >  x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0000c7f58000
> > > > >  Call trace:
> > > > >   ___free_pages+0x2a0/0x2d0 (P)
> > > > >   __free_pages+0x14/0x20
> > > > >   dmirror_devmem_free+0x13c/0x158 [test_hmm]
> > > > >   free_zone_device_folio+0x144/0x1e4
> > > > >   __folio_put+0x124/0x130
> > > > >   free_folio_and_swap_cache+0xa8/0xcc
> > > > >   __folio_split+0x664/0x7fc
> > > > >   split_folio_to_list+0x50/0x5c
> > > > >   migrate_vma_split_folio+0x13c/0x25c
> > > > >   migrate_vma_collect_pmd+0xed4/0xf68
> > > > >   walk_pgd_range+0x598/0x9a0
> > > > >   __walk_page_range+0x90/0x1a0
> > > > >   walk_page_range_mm_unsafe+0x194/0x20c
> > > > >   walk_page_range+0x20/0x2c
> > > > >   migrate_vma_setup+0x18c/0x224
> > > > >   dmirror_devmem_fault+0x188/0x2b8 [test_hmm]
> > > > >   do_swap_page+0x1458/0x185c
> > > > >   __handle_mm_fault+0x85c/0x1ba0
> > > > >   handle_mm_fault+0xb0/0x290
> > > > >   do_page_fault+0x1f8/0x6f8
> > > > >   do_translation_fault+0x60/0x6c
> > > > >   do_mem_abort+0x44/0x94
> > > > >   el0_da+0x30/0xdc
> > > > >   el0t_64_sync_handler+0xd0/0xe4
> > > > >   el0t_64_sync+0x198/0x19c
> > > > >  ---[ end trace 0000000000000000 ]---
> > > > >  lib/test_hmm.c:705 module test_hmm func:dmirror_devmem_alloc_page has 16744448 allocated at module unload
> > > > >
> > > > >
> > > > > It was tested on kernel built with arm64's virt.config and
> > > > >
> > > > > +CONFIG_ZONE_DEVICE=y
> > > > > +CONFIG_DEVICE_PRIVATE=y
> > > > > +CONFIG_TEST_HMM=m
> > > > > +CONFIG_MEM_ALLOC_PROFILING=y
> > > > > +CONFIG_MEM_ALLOC_PROFILING_DEBUG=y
> > > >
> > > > I assume there is a weird interaction between alloc tags and simulated
> > > > ZONE_DEVICE memory in test_hmm.c
> > >
> > > FYI this can be reproduced by running the migrate_partial_unmap_fault
> > > test case.

Thanks. I have reproduced it now that my fingers are skinnier.

> > > TEST_F(hmm, migrate_partial_unmap_fault)
> > > {
> > > 	buffer->mirror = malloc(TWOMEG);
> > > 	buffer->ptr = map;	// points to a THP
> > >
> > > 	/* Initialize buffer in system memory. */
> > > 	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
> > > 		ptr[i] = i;
> > >
> > > 	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
> > >
> > > 	munmap(buffer->ptr, ONEMEG);
> > >
> > > 	/* Fault pages back to system memory and check them. */
> > > 	for (i = 0, ptr = buffer->ptr; i < TWOMEG / sizeof(*ptr); ++i)
> > > 		if (i * sizeof(int) < 0 ||
> > > 		    i * sizeof(int) >= ONEMEG)
> > > 			ASSERT_EQ(ptr[i], i);	// triggers a fault ->
> > >
> > >
> > > dmirror_devmem_fault()
> > > 	migrate_vma_setup()
> > > 		migrate_vma_collect_pmd()
> > > 			// !pte_present(pte) && folio_test_large(folio)
> > > 			migrate_vma_split_folio()
> > > 				split_folio()
> > > 					[...]
> > >
> > > __folio_split() {
> > > 	unmap_folio();
> > >
> > > 	__folio_freeze_and_split_unmapped() {
> > > 		__split_unmapped_folio();
> > >
> > > 		for (...) {
> > > 			zone_device_private_split_cb(.., new_folio);
> > > 			// -> dmirror_devmem_folio_split() which doesn't
> > > 			// set alloc tag for the backing system memory
> > > 			// page being split, i.e., rpage_tail
> > > 		}
> > >
> > > 		zone_device_private_split_cb(.., NULL);
> > > 	}
> > >
> > > 	remap_page();
> > >
> > > 	for (...)
> > > 		free_folio_and_swap_cache(new_folio);
> > > 		// -> dmirror_devmem_free()/__free_page() which warns if
> > > 		// the page being freed doesn't have alloc tag set, in
> > > 		// alloc_tag_sub_check().
> > > }
> > >
> > > The WARN disappears with the following diff. But I'm not sure if I've
> > > missed more important points (which is likely to happen ;-) ).
> > >
> > > diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> > > index ed1bdcf1f8ab..eefa2a739917 100644
> > > --- a/lib/alloc_tag.c
> > > +++ b/lib/alloc_tag.c
> > > @@ -191,6 +191,7 @@ void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
> > >  		}
> > >  	}
> > >  }
> > > +EXPORT_SYMBOL(pgalloc_tag_split);
> > >
> > >  void pgalloc_tag_swap(struct folio *new, struct folio *old)
> > >  {
> > > diff --git a/lib/test_hmm.c b/lib/test_hmm.c
> > > index 213504915737..3bec51828916 100644
> > > --- a/lib/test_hmm.c
> > > +++ b/lib/test_hmm.c
> > > @@ -1713,6 +1713,7 @@ static void dmirror_devmem_folio_split(struct folio *head, struct folio *tail)
> > >  	rfolio = page_folio(rpage);
> > >
> > >  	if (tail == NULL) {
> > > +		pgalloc_tag_split(rfolio, folio_order(rfolio), 0);
> > >  		folio_reset_order(rfolio);
> > >  		rfolio->mapping = NULL;
> > >  		folio_set_count(rfolio, 1);
> > 
> > 
> > zone_device_private_split_cb(), that ends up calling ->folio_split().
> > 
> > We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
> > __folio_freeze_and_split_unmapped() before calling
> > zone_device_private_split_cb() when iterating the folios.
> 
> If I read the code correctly, pgalloc_tag_split() in
> __split_unmapped_folio() deals with device private pages' alloc tag. But
> what alloc_tag_sub_check() warns on are real system memory pages (device
> page's backing page), which are allocated by
> dmirror_devmem_alloc_page()/folio_page().
> 
> static void dmirror_devmem_folio_split(struct folio *head, struct folio
> *tail)
> {
> 	struct page *rpage = BACKING_PAGE(folio_page(head, 0));
> 
> Thanks,
> Zenghui
> 
> > The zone_device_private_split_cb(folio, NULL); is then called on the first folio
> > after looping over the other (new) folios.
> >
> > I would assume that __folio_freeze_and_split_unmapped() would already do the
> > right thing?

Well you know what they say about assumptions :) Although in this case
__folio_freeze_and_split_unmapped() isn't called on the backing page anyway
(it's called to split the ZONE_DEVICE page, not the page simulating device
memory). The problem is we're not splitting the tag associated with the backing
page for the simulated memory.

I came up with the below fix last night, but I suspect it will quite reasonably
get NACKED on the basis of the symbol export so was looking at other solutions.

The simulated memory should just be used like a bare physical address range. So
there really is no reason for the backing page simulating device memory to be
allocated as a higher order folio. Using the struct page to store some metadata
for the simulated device is convenient though to avoid creating a test-specific
data structure for this. So I am looking at going back to allocating the
simulated backing memory as always order-0 pages in the test which is what it
was prior to the introduction of large device pages, but that was causing a
crash I'm yet to debug.

 - Alistair

---

diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
index ed1bdcf1f8ab..8828cfcbab43 100644
--- a/lib/alloc_tag.c
+++ b/lib/alloc_tag.c
@@ -191,6 +191,7 @@ void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
                }
        }
 }
+EXPORT_SYMBOL_GPL(pgalloc_tag_split);
 
 void pgalloc_tag_swap(struct folio *new, struct folio *old)
 {
diff --git a/lib/test_hmm.c b/lib/test_hmm.c
index 213504915737..977f080de6f3 100644
--- a/lib/test_hmm.c
+++ b/lib/test_hmm.c
@@ -29,6 +29,7 @@
 #include <linux/rmap.h>
 #include <linux/mmu_notifier.h>
 #include <linux/migrate.h>
+#include <linux/pgalloc_tag.h>
 
 #include "test_hmm_uapi.h"
 
@@ -1713,6 +1714,16 @@ static void dmirror_devmem_folio_split(struct folio *head, struct folio *tail)
        rfolio = page_folio(rpage);
 
        if (tail == NULL) {
+               pgalloc_tag_split(rfolio, folio_order(rfolio), 0);
                folio_reset_order(rfolio);
                rfolio->mapping = NULL;
                folio_set_count(rfolio, 1);

> > Maybe the issue is the hard-coded folio_reset_order() in
> > dmirror_devmem_folio_split(), where we seem to assume that we split to an
> > order-0 folio?

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Alistair Popple]

On 2026-05-12 at 11:05 +1000, Zenghui Yu <yuzenghui@huawei.com> wrote...
> On 2026/5/12 0:38, Zenghui Yu wrote:
> > Hi David,
> > 
> > On 5/11/26 8:47 PM, David Hildenbrand (Arm) wrote:
> > >
> > > We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
> > > __folio_freeze_and_split_unmapped() before calling
> > > zone_device_private_split_cb() when iterating the folios.
> > 
> > If I read the code correctly, pgalloc_tag_split() in
> > __split_unmapped_folio() deals with device private pages' alloc tag. But
> > what alloc_tag_sub_check() warns on are real system memory pages (device
> > page's backing page), which are allocated by
> > dmirror_devmem_alloc_page()/folio_page().
> 
> Sorry, I meant dmirror_devmem_alloc_page()/folio_alloc().

Sorry, this crossed my other reply. The above analysis is correct - we don't
call pgalloc_tag_split() for the backing folio. Doing that fixes the problem,
but I'm not sure it's the cleanest way to fix things - arguably we shouldn't be
allocating compound folios for the backing store anyway.

 - Alistair

> Thanks,
> Zenghui

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[David Hildenbrand (Arm)]

On 5/12/26 03:28, Alistair Popple wrote:
> On 2026-05-12 at 02:38 +1000, Zenghui Yu <zenghui.yu@linux.dev> wrote...
>> Hi David,
>>
>> On 5/11/26 8:47 PM, David Hildenbrand (Arm) wrote:
> 
> Thanks. I have reproduced it now that my fingers are skinnier.
> 
>>>
>>>
>>> zone_device_private_split_cb(), that ends up calling ->folio_split().
>>>
>>> We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
>>> __folio_freeze_and_split_unmapped() before calling
>>> zone_device_private_split_cb() when iterating the folios.
>>
>> If I read the code correctly, pgalloc_tag_split() in
>> __split_unmapped_folio() deals with device private pages' alloc tag. But
>> what alloc_tag_sub_check() warns on are real system memory pages (device
>> page's backing page), which are allocated by
>> dmirror_devmem_alloc_page()/folio_page().
>>
>> static void dmirror_devmem_folio_split(struct folio *head, struct folio
>> *tail)
>> {
>> 	struct page *rpage = BACKING_PAGE(folio_page(head, 0));
>>
>> Thanks,
>> Zenghui
>>
>>> The zone_device_private_split_cb(folio, NULL); is then called on the first folio
>>> after looping over the other (new) folios.
>>>
>>> I would assume that __folio_freeze_and_split_unmapped() would already do the
>>> right thing?
> 
> Well you know what they say about assumptions :) Although in this case
> __folio_freeze_and_split_unmapped() isn't called on the backing page anyway
> (it's called to split the ZONE_DEVICE page, not the page simulating device
> memory). 

Now my brain hurts :)

> The problem is we're not splitting the tag associated with the backing
> page for the simulated memory.
> 
> I came up with the below fix last night, but I suspect it will quite reasonably
> get NACKED on the basis of the symbol export so was looking at other solutions.

I think there are other problems ...

> 
> The simulated memory should just be used like a bare physical address range. So
> there really is no reason for the backing page simulating device memory to be
> allocated as a higher order folio. Using the struct page to store some metadata
> for the simulated device is convenient though to avoid creating a test-specific
> data structure for this. So I am looking at going back to allocating the
> simulated backing memory as always order-0 pages in the test which is what it
> was prior to the introduction of large device pages, but that was causing a
> crash I'm yet to debug.
> 

... such as doing a folio_page(folio_alloc()), followed by a __free_pages().

Why are we even allocating folios here and manually splitting them?

Looking at dmirror_devmem_folio_split(), aren't we using folios here for
something that ... is not a folio?

Likely we really shouldn't be using folios here ... :)

-- 
Cheers,

David

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[Alistair Popple]

On 2026-05-12 at 16:47 +1000, "David Hildenbrand (Arm)" <david@kernel.org> wrote...
> On 5/12/26 03:28, Alistair Popple wrote:
> > On 2026-05-12 at 02:38 +1000, Zenghui Yu <zenghui.yu@linux.dev> wrote...
> >> Hi David,
> >>
> >> On 5/11/26 8:47 PM, David Hildenbrand (Arm) wrote:
> > 
> > Thanks. I have reproduced it now that my fingers are skinnier.
> > 
> >>>
> >>>
> >>> zone_device_private_split_cb(), that ends up calling ->folio_split().
> >>>
> >>> We do have a call to pgalloc_tag_split() in __split_unmapped_folio(), invoked in
> >>> __folio_freeze_and_split_unmapped() before calling
> >>> zone_device_private_split_cb() when iterating the folios.
> >>
> >> If I read the code correctly, pgalloc_tag_split() in
> >> __split_unmapped_folio() deals with device private pages' alloc tag. But
> >> what alloc_tag_sub_check() warns on are real system memory pages (device
> >> page's backing page), which are allocated by
> >> dmirror_devmem_alloc_page()/folio_page().
> >>
> >> static void dmirror_devmem_folio_split(struct folio *head, struct folio
> >> *tail)
> >> {
> >> 	struct page *rpage = BACKING_PAGE(folio_page(head, 0));
> >>
> >> Thanks,
> >> Zenghui
> >>
> >>> The zone_device_private_split_cb(folio, NULL); is then called on the first folio
> >>> after looping over the other (new) folios.
> >>>
> >>> I would assume that __folio_freeze_and_split_unmapped() would already do the
> >>> right thing?
> > 
> > Well you know what they say about assumptions :) Although in this case
> > __folio_freeze_and_split_unmapped() isn't called on the backing page anyway
> > (it's called to split the ZONE_DEVICE page, not the page simulating device
> > memory). 
> 
> Now my brain hurts :)

I have never liked this bit of the HMM selftests. It has always made my brain
hurt.

> > The problem is we're not splitting the tag associated with the backing
> > page for the simulated memory.
> > 
> > I came up with the below fix last night, but I suspect it will quite reasonably
> > get NACKED on the basis of the symbol export so was looking at other solutions.
> 
> I think there are other problems ...
> 
> > 
> > The simulated memory should just be used like a bare physical address range. So
> > there really is no reason for the backing page simulating device memory to be
> > allocated as a higher order folio. Using the struct page to store some metadata
> > for the simulated device is convenient though to avoid creating a test-specific
> > data structure for this. So I am looking at going back to allocating the
> > simulated backing memory as always order-0 pages in the test which is what it
> > was prior to the introduction of large device pages, but that was causing a
> > crash I'm yet to debug.
> > 
> 
> ... such as doing a folio_page(folio_alloc()), followed by a __free_pages().
> 
> Why are we even allocating folios here and manually splitting them?
> 
> Looking at dmirror_devmem_folio_split(), aren't we using folios here for
> something that ... is not a folio?
> 
> Likely we really shouldn't be using folios here ... :)

Exactly my point, just more succinct :)

I just need to make it work without doing that.

 - Alistair

> -- 
> Cheers,
> 
> David

Re: "alloc_tag was not set" when running mm/ksft_hmm.sh

[David Hildenbrand (Arm)]

On 5/12/26 09:46, Alistair Popple wrote:
> On 2026-05-12 at 16:47 +1000, "David Hildenbrand (Arm)" <david@kernel.org> wrote...
>> On 5/12/26 03:28, Alistair Popple wrote:
>>> On 2026-05-12 at 02:38 +1000, Zenghui Yu <zenghui.yu@linux.dev> wrote...
>>>
>>> Thanks. I have reproduced it now that my fingers are skinnier.
>>>
>>>
>>> Well you know what they say about assumptions :) Although in this case
>>> __folio_freeze_and_split_unmapped() isn't called on the backing page anyway
>>> (it's called to split the ZONE_DEVICE page, not the page simulating device
>>> memory). 
>>
>> Now my brain hurts :)
> 
> I have never liked this bit of the HMM selftests. It has always made my brain
> hurt.
> 
>>> The problem is we're not splitting the tag associated with the backing
>>> page for the simulated memory.
>>>
>>> I came up with the below fix last night, but I suspect it will quite reasonably
>>> get NACKED on the basis of the symbol export so was looking at other solutions.
>>
>> I think there are other problems ...
>>
>>>
>>> The simulated memory should just be used like a bare physical address range. So
>>> there really is no reason for the backing page simulating device memory to be
>>> allocated as a higher order folio. Using the struct page to store some metadata
>>> for the simulated device is convenient though to avoid creating a test-specific
>>> data structure for this. So I am looking at going back to allocating the
>>> simulated backing memory as always order-0 pages in the test which is what it
>>> was prior to the introduction of large device pages, but that was causing a
>>> crash I'm yet to debug.
>>>
>>
>> ... such as doing a folio_page(folio_alloc()), followed by a __free_pages().
>>
>> Why are we even allocating folios here and manually splitting them?
>>
>> Looking at dmirror_devmem_folio_split(), aren't we using folios here for
>> something that ... is not a folio?
>>
>> Likely we really shouldn't be using folios here ... :)
> 
> Exactly my point, just more succinct :)
> 
> I just need to make it work without doing that.

Ok, great. You could also allocate a larger buddy page and just split it
manually through split_page.

split_page will even do the pgalloc_tag_split :)

-- 
Cheers,

David