Re: [PATCH] mm/hugetlb: fix copy_hugetlb_page_range() to use ->pt_share_count

[jane.chu@oracle.com]

On 9/9/2025 6:14 PM, Harry Yoo wrote:
> On Tue, Sep 09, 2025 at 12:43:57PM -0600, Jane Chu wrote:
>> commit 59d9094df3d79 introduced ->pt_share_count dedicated to
> 
> nit: the format should be:
>    commit <sha1> ("summary")
> ?
> 
>> hugetlb PMD share count tracking, but omitted fixing
>> copy_hugetlb_page_range(), leaving the function relying on
>> page_count() for tracking that no longer works.
>>
>> When lazy page table copy for hugetlb is disabled (commit bcd51a3c679d),
> 
> same here.
> 
>> fork()'ing with hugetlb PMD sharing quickly lockup -
>>
>> [  239.446559] watchdog: BUG: soft lockup - CPU#75 stuck for 27s!
>> [  239.446611] RIP: 0010:native_queued_spin_lock_slowpath+0x7e/0x2e0
>> [  239.446631] Call Trace:
>> [  239.446633]  <TASK>
>> [  239.446636]  _raw_spin_lock+0x3f/0x60
>> [  239.446639]  copy_hugetlb_page_range+0x258/0xb50
>> [  239.446645]  copy_page_range+0x22b/0x2c0
>> [  239.446651]  dup_mmap+0x3e2/0x770
>> [  239.446654]  dup_mm.constprop.0+0x5e/0x230
>> [  239.446657]  copy_process+0xd17/0x1760
>> [  239.446660]  kernel_clone+0xc0/0x3e0
>> [  239.446661]  __do_sys_clone+0x65/0xa0
>> [  239.446664]  do_syscall_64+0x82/0x930
>> [  239.446668]  ? count_memcg_events+0xd2/0x190
>> [  239.446671]  ? syscall_trace_enter+0x14e/0x1f0
>> [  239.446676]  ? syscall_exit_work+0x118/0x150
>> [  239.446677]  ? arch_exit_to_user_mode_prepare.constprop.0+0x9/0xb0
>> [  239.446681]  ? clear_bhb_loop+0x30/0x80
>> [  239.446684]  ? clear_bhb_loop+0x30/0x80
>> [  239.446686]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
>>
>> There are two options to resolve the potential latent issue:
>>    1. remove the PMD sharing awareness from copy_hugetlb_page_range(),
>>    2. fix it.
>> This patch opts for the second option.
>>
>> Fixes: 59d9094df3d79 ("mm: hugetlb: independent PMD page table shared
>> count")
> 
> nit: we don't add newline even when Fixes: tag is line is over 75 characters?
>   
>> Signed-off-by: Jane Chu <jane.chu@oracle.com>
>> ---
> 
> The change in general looks good to me,
> Reviewed-by: Harry Yoo <harry.yoo@oracle.com>
> 
> +Cc Jann Horn who backported the commit 59d9094df3d79.

Thanks Harry!  Will fix the commit message and send v2.>
> Elaborating a little bit why it doesn't need to be backported:
> TL;DR: One needs to backport commit 3aa4ed8040e15 to pre-v6.0 kernels,
> or revert commit bcd51a3c679d to trigger this.
> 
> commit 59d9094df3d79 ("mm: hugetlb: independent PMD page table shared count")
> is introduced in 6.13 and backported to 5.10.y, 5.15.y, 6.1.y, 6.6.y, 6.12.y.
> 
> As lazy page table copy is enabled in v6.0 by commit bcd51a3c679d
> ("hugetlb: lazy page table copies in fork()"), the bug is not triggered
> because shared hugetlb VMAs go through lazy page table copy logic
> (vma_needs_copy() returns false) or they can't share page tables
> (uffd_disable_huge_pmd_share() returns false).
> 
> They shouldn't have anon_vma, VM_PFNMAP, VM_MIXEDMAP and UFFD-WP VMA
> cannot share page tables - so either vma_needs_copy() return false, or
> page tables cannot be shared.
> 
> And before commit 3aa4ed8040e15 ("mm/hugetlb: make detecting shared pte
> more reliable") introduced in v6.1, copy_hugetlb_page_range() doesn't check
> refcount to determine whether the page table is shared.

Yes, it's the combination of
   v6.1: 3aa4ed8040e15 mm/hugetlb: make detecting shared pte more reliable
   v6.13: 59d9094df3d79 mm: hugetlb: independent PMD page table shared count
without
   v6.0: bcd51a3c679d hugetlb: lazy page table copies in fork()
that is problematic.

Since the problematic combination doesn't exist in any upstream release,
no need to backport.  The patch is for a potential latent bug.

thanks,
-jane
   >
>>   mm/hugetlb.c | 13 ++++---------
>>   1 file changed, 4 insertions(+), 9 deletions(-)
>>
>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
>> index 753f99b4c718..8ca5b4f7805f 100644
>> --- a/mm/hugetlb.c
>> +++ b/mm/hugetlb.c
>> @@ -5594,18 +5594,13 @@ int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
>>   			break;
>>   		}
>>   
>> -		/*
>> -		 * If the pagetables are shared don't copy or take references.
>> -		 *
>> -		 * dst_pte == src_pte is the common case of src/dest sharing.
>> -		 * However, src could have 'unshared' and dst shares with
>> -		 * another vma. So page_count of ptep page is checked instead
>> -		 * to reliably determine whether pte is shared.
>> -		 */
>> -		if (page_count(virt_to_page(dst_pte)) > 1) {
>> +#ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
>> +		/* If the pagetables are shared don't copy or take references. */
>> +		if (ptdesc_pmd_pts_count(virt_to_ptdesc(dst_pte)) > 0) {
>>   			addr |= last_addr_mask;
>>   			continue;
>>   		}
>> +#endif
>>   
>>   		dst_ptl = huge_pte_lock(h, dst, dst_pte);
>>   		src_ptl = huge_pte_lockptr(h, src, src_pte);
>> -- 
>> 2.43.5
>>
>