mm: hugetlb: avoid soft lockup when mprotect to large memory area

PATCH Applied Yang Shi 4 participants 6 messages Sep 29, 2025 – Oct 1, 2025

Subsystems (2)

Memory management HugeTLB

lore

Share & tools

lore

Yang Shi

author

2025-09-29 20:24 UTC #0 lore raw

PATCH v2 1/1 mm: hugetlb: avoid soft lockup when mprotect to large memory area

When calling mprotect() to a large hugetlb memory area in our customer's workload (~300GB hugetlb memory), soft lockup was observed:

watchdog: BUG: soft lockup - CPU#98 stuck for 23s! [t2_new_sysv:126916]

CPU: 98 PID: 126916 Comm: t2_new_sysv Kdump: loaded Not tainted 6.17-rc7
Hardware name: GIGACOMPUTING R2A3-T40-AAV1/Jefferson CIO, BIOS 5.4.4.1 07/15/2025

pstate: 20400009 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : mte_clear_page_tags+0x14/0x24
lr : mte_sync_tags+0x1c0/0x240
sp : ffff80003150bb80
x29: ffff80003150bb80 x28: ffff00739e9705a8 x27: 0000ffd2d6a00000
x26: 0000ff8e4bc00000 x25: 00e80046cde00f45 x24: 0000000000022458
x23: 0000000000000000 x22: 0000000000000004 x21: 000000011b380000
x20: ffff000000000000 x19: 000000011b379f40 x18: 0000000000000000
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc875e0aa5e2c x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000 x5 : fffffc01ce7a5c00 x4 : 00000000046cde00 x3 : fffffc0000000000 x2 : 0000000000000004 x1 : 0000000000000040 x0 : ffff0046cde7c000

Call trace:
  mte_clear_page_tags+0x14/0x24
  set_huge_pte_at+0x25c/0x280
  hugetlb_change_protection+0x220/0x430
  change_protection+0x5c/0x8c
  mprotect_fixup+0x10c/0x294
  do_mprotect_pkey.constprop.0+0x2e0/0x3d4
  __arm64_sys_mprotect+0x24/0x44
  invoke_syscall+0x50/0x160
  el0_svc_common+0x48/0x144
  do_el0_svc+0x30/0xe0
  el0_svc+0x30/0xf0
  el0t_64_sync_handler+0xc4/0x148
  el0t_64_sync+0x1a4/0x1a8

Soft lockup is not triggered with THP or base page because there is cond_resched() called for each PMD size.

Although the soft lockup was triggered by MTE, it should be not MTE specific. The other processing which takes long time in the loop may trigger soft lockup too.

So add cond_resched() for hugetlb to avoid soft lockup.

Fixes 8f860591ffb2 ("[PATCH] Enable mprotect on huge pages")

Tested-by Carl Worth <[email protected]>

Reviewed-by Christoph Lameter (Ampere) <[email protected]>

Reviewed-by Catalin Marinas <[email protected]>

Acked-by David Hildenbrand <[email protected]>

Acked-by Oscar Salvador <[email protected]>

Reviewed-by Anshuman Khandual <[email protected]>

Signed-off-by Yang Shi <[email protected]>

---
v2: - Made the subject and commit message less MTE specific and fixed
the fixes tag.
- Collected all R-bs and A-bs.

 mm/hugetlb.c | 2 ++
 1 file changed, 2 insertions(+)

mm/hugetlb.c +2 -0

index cb5c4e79e0b8..fe6606d91b31 100644
@@ -7242,6 +7242,8 @@ long hugetlb_change_protection(struct vm_area_struct *vma,
 						psize);
 		}
 		spin_unlock(ptl);
+
+		cond_resched();
 	}
 	/*
 	 * Must flush TLB before releasing i_mmap_rwsem: x86's huge_pmd_unshare

Signature · 1 line

2.47.0

Dev Jain

Reviewed-by on v2 1/1

2025-09-30 05:26 UTC #1 lore raw

Quoted 67 lines · Yang Shi

When calling mprotect() to a large hugetlb memory area in our customer's workload (~300GB hugetlb memory), soft lockup was observed:
watchdog: BUG: soft lockup - CPU#98 stuck for 23s! [t2_new_sysv:126916]

CPU: 98 PID: 126916 Comm: t2_new_sysv Kdump: loaded Not tainted 6.17-rc7
Hardware name: GIGACOMPUTING R2A3-T40-AAV1/Jefferson CIO, BIOS 5.4.4.1 07/15/2025
pstate: 20400009 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : mte_clear_page_tags+0x14/0x24
lr : mte_sync_tags+0x1c0/0x240
sp : ffff80003150bb80
x29: ffff80003150bb80 x28: ffff00739e9705a8 x27: 0000ffd2d6a00000
x26: 0000ff8e4bc00000 x25: 00e80046cde00f45 x24: 0000000000022458
x23: 0000000000000000 x22: 0000000000000004 x21: 000000011b380000
x20: ffff000000000000 x19: 000000011b379f40 x18: 0000000000000000
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc875e0aa5e2c x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000 x5 : fffffc01ce7a5c00 x4 : 00000000046cde00 x3 : fffffc0000000000 x2 : 0000000000000004 x1 : 0000000000000040 x0 : ffff0046cde7c000
Call trace:
   mte_clear_page_tags+0x14/0x24
   set_huge_pte_at+0x25c/0x280
   hugetlb_change_protection+0x220/0x430
   change_protection+0x5c/0x8c
   mprotect_fixup+0x10c/0x294
   do_mprotect_pkey.constprop.0+0x2e0/0x3d4
   __arm64_sys_mprotect+0x24/0x44
   invoke_syscall+0x50/0x160
   el0_svc_common+0x48/0x144
   do_el0_svc+0x30/0xe0
   el0_svc+0x30/0xf0
   el0t_64_sync_handler+0xc4/0x148
   el0t_64_sync+0x1a4/0x1a8
Soft lockup is not triggered with THP or base page because there is cond_resched() called for each PMD size.
Although the soft lockup was triggered by MTE, it should be not MTE specific. The other processing which takes long time in the loop may trigger soft lockup too.
So add cond_resched() for hugetlb to avoid soft lockup.
Fixes 8f860591ffb2 ("[PATCH] Enable mprotect on huge pages")
Tested-by Carl Worth <[email protected]>
Reviewed-by Christoph Lameter (Ampere) <[email protected]>
Reviewed-by Catalin Marinas <[email protected]>
Acked-by David Hildenbrand <[email protected]>
Acked-by Oscar Salvador <[email protected]>
Reviewed-by Anshuman Khandual <[email protected]>
Signed-off-by Yang Shi <[email protected]>
---
v2: - Made the subject and commit message less MTE specific and fixed
the fixes tag.
- Collected all R-bs and A-bs.
  mm/hugetlb.c | 2 ++
  1 file changed, 2 insertions(+)
mm/hugetlb.c +0 -0
index cb5c4e79e0b8..fe6606d91b31 100644

mm/hugetlb.c long hugetlb_change_protection(struct vm_area_struct *vma, · line 7249

Quoted 8 lines

mm/hugetlb.c +2 -0

@@ -7242,6 +7242,8 @@ long hugetlb_change_protection(struct vm_area_struct *vma,
  						psize);
  		}
  		spin_unlock(ptl);
+
+		cond_resched();
  	}
  	/*
  	 * Must flush TLB before releasing i_mmap_rwsem: x86's huge_pmd_unshare

Reviewed-by Dev Jain <[email protected]>

Does it make sense to also do cond_resched() in the huge_pmd_unshare() branch? That also amounts to clearing a page. And I can see for example, zap_huge_pmd() and change_huge_pmd() consume a cond_resched().

Yang Shi

author

2025-09-30 18:08 UTC #2 lore raw

Quoted 64 lines · Dev Jain

Quoted 62 lines · Yang Shi
When calling mprotect() to a large hugetlb memory area in our customer's workload (~300GB hugetlb memory), soft lockup was observed:
watchdog: BUG: soft lockup - CPU#98 stuck for 23s! [t2_new_sysv:126916]

CPU: 98 PID: 126916 Comm: t2_new_sysv Kdump: loaded Not tainted 6.17-rc7
Hardware name: GIGACOMPUTING R2A3-T40-AAV1/Jefferson CIO, BIOS 5.4.4.1 07/15/2025
pstate: 20400009 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : mte_clear_page_tags+0x14/0x24
lr : mte_sync_tags+0x1c0/0x240
sp : ffff80003150bb80
x29: ffff80003150bb80 x28: ffff00739e9705a8 x27: 0000ffd2d6a00000
x26: 0000ff8e4bc00000 x25: 00e80046cde00f45 x24: 0000000000022458
x23: 0000000000000000 x22: 0000000000000004 x21: 000000011b380000
x20: ffff000000000000 x19: 000000011b379f40 x18: 0000000000000000
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc875e0aa5e2c x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000 x5 : fffffc01ce7a5c00 x4 : 00000000046cde00 x3 : fffffc0000000000 x2 : 0000000000000004 x1 : 0000000000000040 x0 : ffff0046cde7c000
Call trace:
   mte_clear_page_tags+0x14/0x24
   set_huge_pte_at+0x25c/0x280
   hugetlb_change_protection+0x220/0x430
   change_protection+0x5c/0x8c
   mprotect_fixup+0x10c/0x294
   do_mprotect_pkey.constprop.0+0x2e0/0x3d4
   __arm64_sys_mprotect+0x24/0x44
   invoke_syscall+0x50/0x160
   el0_svc_common+0x48/0x144
   do_el0_svc+0x30/0xe0
   el0_svc+0x30/0xf0
   el0t_64_sync_handler+0xc4/0x148
   el0t_64_sync+0x1a4/0x1a8
Soft lockup is not triggered with THP or base page because there is cond_resched() called for each PMD size.
Although the soft lockup was triggered by MTE, it should be not MTE specific. The other processing which takes long time in the loop may trigger soft lockup too.
So add cond_resched() for hugetlb to avoid soft lockup.
Fixes 8f860591ffb2 ("[PATCH] Enable mprotect on huge pages")
Tested-by Carl Worth <[email protected]>
Reviewed-by Christoph Lameter (Ampere) <[email protected]>
Reviewed-by Catalin Marinas <[email protected]>
Acked-by David Hildenbrand <[email protected]>
Acked-by Oscar Salvador <[email protected]>
Reviewed-by Anshuman Khandual <[email protected]>
Signed-off-by Yang Shi <[email protected]>
---
v2: - Made the subject and commit message less MTE specific and fixed the fixes tag.
- Collected all R-bs and A-bs.
mm/hugetlb.c | 2 ++
  1 file changed, 2 insertions(+)

Quoted 16 lines

Quoted 13 lines

mm/hugetlb.c +2 -0

index cb5c4e79e0b8..fe6606d91b31 100644
@@ -7242,6 +7242,8 @@ long hugetlb_change_protection(struct 
 vm_area_struct *vma,
                           psize);
           }
           spin_unlock(ptl);
+
+        cond_resched();
       }
       /*

Quoted 1 line

* Must flush TLB before releasing i_mmap_rwsem: x86's huge_pmd_unshare

Reviewed-by Dev Jain <[email protected]>

Thank you.

Quoted 4 lines

Does it make sense to also do cond_resched() in the huge_pmd_unshare() branch? That also amounts to clearing a page. And I can see for example, zap_huge_pmd() and change_huge_pmd() consume a cond_resched().

Thanks for raising this. I did think about it. But I didn't convince myself because shared pmd should be not that common IMHO (If I'm wrong, please feel free to correct me). At least PMD can't be shared if the memory is tagged IIRC. So I'd like to keep the patch minimal for now and defer adding cond_resched() until it is hit by some real life workload.

Yang

Christoph Lameter (Ampere)

2025-09-30 18:43 UTC #3 lore raw

Quoted 11 lines · Yang Shi

Quoted 5 lines
Does it make sense to also do cond_resched() in the huge_pmd_unshare() branch?
That also amounts to clearing a page. And I can see for example, zap_huge_pmd()
and change_huge_pmd() consume a cond_resched().
Thanks for raising this. I did think about it. But I didn't convince myself because shared pmd should be not that common IMHO (If I'm wrong, please feel free to correct me). At least PMD can't be shared if the memory is tagged IIRC. So I'd like to keep the patch minimal for now and defer adding cond_resched() until it is hit by some real life workload.

It would be good to send out a second path that covers the other cases for discussion.

Dev Jain

2025-10-01 04:23 UTC #4 lore raw

In reply to Yang Shi

Quoted 84 lines · Yang Shi

Quoted 64 lines · Dev Jain
Quoted 62 lines · Yang Shi
When calling mprotect() to a large hugetlb memory area in our customer's workload (~300GB hugetlb memory), soft lockup was observed:
watchdog: BUG: soft lockup - CPU#98 stuck for 23s! [t2_new_sysv:126916]

CPU: 98 PID: 126916 Comm: t2_new_sysv Kdump: loaded Not tainted 6.17-rc7
Hardware name: GIGACOMPUTING R2A3-T40-AAV1/Jefferson CIO, BIOS 5.4.4.1 07/15/2025
pstate: 20400009 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : mte_clear_page_tags+0x14/0x24
lr : mte_sync_tags+0x1c0/0x240
sp : ffff80003150bb80
x29: ffff80003150bb80 x28: ffff00739e9705a8 x27: 0000ffd2d6a00000
x26: 0000ff8e4bc00000 x25: 00e80046cde00f45 x24: 0000000000022458
x23: 0000000000000000 x22: 0000000000000004 x21: 000000011b380000
x20: ffff000000000000 x19: 000000011b379f40 x18: 0000000000000000
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc875e0aa5e2c x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000 x5 : fffffc01ce7a5c00 x4 : 00000000046cde00 x3 : fffffc0000000000 x2 : 0000000000000004 x1 : 0000000000000040 x0 : ffff0046cde7c000
Call trace:
   mte_clear_page_tags+0x14/0x24
   set_huge_pte_at+0x25c/0x280
   hugetlb_change_protection+0x220/0x430
   change_protection+0x5c/0x8c
   mprotect_fixup+0x10c/0x294
   do_mprotect_pkey.constprop.0+0x2e0/0x3d4
   __arm64_sys_mprotect+0x24/0x44
   invoke_syscall+0x50/0x160
   el0_svc_common+0x48/0x144
   do_el0_svc+0x30/0xe0
   el0_svc+0x30/0xf0
   el0t_64_sync_handler+0xc4/0x148
   el0t_64_sync+0x1a4/0x1a8
Soft lockup is not triggered with THP or base page because there is cond_resched() called for each PMD size.
Although the soft lockup was triggered by MTE, it should be not MTE specific. The other processing which takes long time in the loop may trigger soft lockup too.
So add cond_resched() for hugetlb to avoid soft lockup.
Fixes 8f860591ffb2 ("[PATCH] Enable mprotect on huge pages")
Tested-by Carl Worth <[email protected]>
Reviewed-by Christoph Lameter (Ampere) <[email protected]>
Reviewed-by Catalin Marinas <[email protected]>
Acked-by David Hildenbrand <[email protected]>
Acked-by Oscar Salvador <[email protected]>
Reviewed-by Anshuman Khandual <[email protected]>
Signed-off-by Yang Shi <[email protected]>
---
v2: - Made the subject and commit message less MTE specific and fixed the fixes tag.
- Collected all R-bs and A-bs.
mm/hugetlb.c | 2 ++
  1 file changed, 2 insertions(+)
Quoted 16 lines
Quoted 13 lines
mm/hugetlb.c +2 -0
index cb5c4e79e0b8..fe6606d91b31 100644
@@ -7242,6 +7242,8 @@ long hugetlb_change_protection(struct 
 vm_area_struct *vma,
                           psize);
           }
           spin_unlock(ptl);
+
+        cond_resched();
       }
       /*
Quoted 1 line
* Must flush TLB before releasing i_mmap_rwsem: x86's huge_pmd_unshare
Reviewed-by Dev Jain <[email protected]>

Quoted 8 lines

Thank you.
Quoted 4 lines
Does it make sense to also do cond_resched() in the huge_pmd_unshare() branch? That also amounts to clearing a page. And I can see for example, zap_huge_pmd() and change_huge_pmd() consume a cond_resched().
Thanks for raising this. I did think about it. But I didn't convince myself because shared pmd should be not that common IMHO (If I'm wrong, please feel free to correct me). At least PMD can't be shared if the memory is tagged IIRC. So I'd like to keep the patch minimal for now and defer adding cond_resched() until it is hit by some real life workload.

If we have large swathes of hugetlb memory like in your workload, and it is MAP_SHARED, then there should be high chances of sharing the PMD. Although, I incorrectly

observed that we are clearing a page there - we are only clearing the pud entry which is 8 bytes. So yes a soft lockup should be highly unlikely. But since cond_resched()

is cheap (I assume this is the case since it is liberally sprinkled all over the codebase) I think we should be consistent. Probably not an immediate concern and not a matter

of this patch.

Quoted 4 lines

Yang

David Hildenbrand

2025-10-01 08:32 UTC #5 lore raw

Quoted 110 lines · Dev Jain

Quoted 97 lines · Yang Shi
Quoted 76 lines · Dev Jain
Quoted 74 lines · Yang Shi
When calling mprotect() to a large hugetlb memory area in our customer's
workload (~300GB hugetlb memory), soft lockup was observed:
watchdog: BUG: soft lockup - CPU#98 stuck for 23s! [t2_new_sysv:126916]

CPU: 98 PID: 126916 Comm: t2_new_sysv Kdump: loaded Not tainted
6.17-rc7
Hardware name: GIGACOMPUTING R2A3-T40-AAV1/Jefferson CIO, BIOS
5.4.4.1 07/15/2025
pstate: 20400009 (nzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : mte_clear_page_tags+0x14/0x24
lr : mte_sync_tags+0x1c0/0x240
sp : ffff80003150bb80
x29: ffff80003150bb80 x28: ffff00739e9705a8 x27: 0000ffd2d6a00000
x26: 0000ff8e4bc00000 x25: 00e80046cde00f45 x24: 0000000000022458
x23: 0000000000000000 x22: 0000000000000004 x21: 000000011b380000
x20: ffff000000000000 x19: 000000011b379f40 x18: 0000000000000000
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc875e0aa5e2c x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000 x5 : fffffc01ce7a5c00 x4 : 00000000046cde00 x3 : fffffc0000000000 x2 : 0000000000000004 x1 : 0000000000000040 x0 : ffff0046cde7c000
Call trace:
    mte_clear_page_tags+0x14/0x24
    set_huge_pte_at+0x25c/0x280
    hugetlb_change_protection+0x220/0x430
    change_protection+0x5c/0x8c
    mprotect_fixup+0x10c/0x294
    do_mprotect_pkey.constprop.0+0x2e0/0x3d4
    __arm64_sys_mprotect+0x24/0x44
    invoke_syscall+0x50/0x160
    el0_svc_common+0x48/0x144
    do_el0_svc+0x30/0xe0
    el0_svc+0x30/0xf0
    el0t_64_sync_handler+0xc4/0x148
    el0t_64_sync+0x1a4/0x1a8
Soft lockup is not triggered with THP or base page because there is cond_resched() called for each PMD size.
Although the soft lockup was triggered by MTE, it should be not MTE specific. The other processing which takes long time in the loop may trigger soft lockup too.
So add cond_resched() for hugetlb to avoid soft lockup.
Fixes 8f860591ffb2 ("[PATCH] Enable mprotect on huge pages")
Tested-by Carl Worth <[email protected]>
Reviewed-by Christoph Lameter (Ampere) <[email protected]>
Reviewed-by Catalin Marinas <[email protected]>
Acked-by David Hildenbrand <[email protected]>
Acked-by Oscar Salvador <[email protected]>
Reviewed-by Anshuman Khandual <[email protected]>
Signed-off-by Yang Shi <[email protected]>
---
v2: - Made the subject and commit message less MTE specific and fixed
the fixes tag.
- Collected all R-bs and A-bs.
   mm/hugetlb.c | 2 ++
   1 file changed, 2 insertions(+)
mm/hugetlb.c +0 -0
index cb5c4e79e0b8..fe6606d91b31 100644
@@ -7242,6 +7242,8 @@ long hugetlb_change_protection(struct
 vm_area_struct *vma,
                           psize);
           }
           spin_unlock(ptl);
Quoted 8 lines
Quoted 6 lines
mm/hugetlb.c +2 -0
+
+        cond_resched();
       }
       /*
* Must flush TLB before releasing i_mmap_rwsem: x86's
huge_pmd_unshare
Reviewed-by Dev Jain <[email protected]>
Thank you.
Quoted 6 lines
Does it make sense to also do cond_resched() in the huge_pmd_unshare() branch?
That also amounts to clearing a page. And I can see for example, zap_huge_pmd()
and change_huge_pmd() consume a cond_resched().
Quoted 6 lines
Thanks for raising this. I did think about it. But I didn't convince myself because shared pmd should be not that common IMHO (If I'm wrong, please feel free to correct me). At least PMD can't be shared if the memory is tagged IIRC. So I'd like to keep the patch minimal for now and defer adding cond_resched() until it is hit by some real life workload.
If we have large swathes of hugetlb memory like in your workload, and it is MAP_SHARED, then there should be high chances of sharing the PMD. Although, I incorrectly

Quoted 7 lines

observed that we are clearing a page there - we are only clearing the pud entry which is 8 bytes. So yes a soft lockup should be highly unlikely. But since cond_resched()
is cheap (I assume this is the case since it is liberally sprinkled all over the codebase) I think we should be consistent. Probably not an immediate concern and not a matter

Right, that's one of the cases where we might just want to wait either until is is reported or until hugetlb is finally removed in a couple of decades ;)

Signature · 3 lines

Cheers

David / dhildenb