Re: [RFC PATCH v4 6/8] hugetlb: add vma based lock for pmd sharing synchronization

[Miaohe Lin]

On 2022/7/30 2:00, Mike Kravetz wrote:
> On 07/29/22 10:55, Miaohe Lin wrote:
>> On 2022/7/7 4:23, Mike Kravetz wrote:
>>> Allocate a rw semaphore and hang off vm_private_data for
>>> synchronization use by vmas that could be involved in pmd sharing.  Only
>>> add infrastructure for the new lock here.  Actual use will be added in
>>> subsequent patch.
>>>
>>> Signed-off-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
>>> ---
>>>  include/linux/hugetlb.h |  36 +++++++++-
>>>  kernel/fork.c           |   6 +-
>>>  mm/hugetlb.c            | 150 ++++++++++++++++++++++++++++++++++++----
>>>  mm/rmap.c               |   8 ++-
>>>  4 files changed, 178 insertions(+), 22 deletions(-)
>>>
>>
>> <snip>
>>
>>>  
>>>  /* Forward declaration */
>>>  static int hugetlb_acct_memory(struct hstate *h, long delta);
>>> +static bool vma_pmd_shareable(struct vm_area_struct *vma);
>>>  
>>>  static inline bool subpool_is_free(struct hugepage_subpool *spool)
>>>  {
>>> @@ -904,6 +905,89 @@ resv_map_set_hugetlb_cgroup_uncharge_info(struct resv_map *resv_map,
>>>  #endif
>>>  }
>>>  
>>> +static bool __vma_shareable_flags_pmd(struct vm_area_struct *vma)
>>> +{
>>> +	return vma->vm_flags & (VM_MAYSHARE | VM_SHARED) &&
>>
>> Should me make __vma_aligned_range_pmd_shareable check (VM_MAYSHARE | VM_SHARED) like above
>> instead of VM_MAYSHARE to make code more consistent?
>>
> 
> I 'think' we want them to be different.  Note this subtle code and
> explanation in __unmap_hugepage_range_final().
> 
> 	/*
> 	 * Clear this flag so that x86's huge_pmd_share page_table_shareable
> 	 * test will fail on a vma being torn down, and not grab a page table
> 	 * on its way out.  We're lucky that the flag has such an appropriate
> 	 * name, and can in fact be safely cleared here. We could clear it
> 	 * before the __unmap_hugepage_range above, but all that's necessary
> 	 * is to clear it before releasing the i_mmap_rwsem. This works
> 	 * because in the context this is called, the VMA is about to be
> 	 * destroyed and the i_mmap_rwsem is held.
> 	 */
> 	vma->vm_flags &= ~VM_MAYSHARE;
> 
> So, when making a decision to share or not we need to only check VM_MAYSHARE.
> When making decisions about about the vma_lock, we need to check both.  In most
> cases, just VM_MAYSHARE would be sufficient but we need to handle this case
> where VM_SHARED and !VM_MAYSHARE.  Mostly in the unmap/cleanup cases.

Many thanks for your explanation. :)

> 
>>> +		vma->vm_private_data;
>>> +}
>>> +
>>> +void hugetlb_vma_lock_read(struct vm_area_struct *vma)
>>> +{
>>> +	if (__vma_shareable_flags_pmd(vma))
>>> +		down_read((struct rw_semaphore *)vma->vm_private_data);
>>> +}
>>> +
>>> +void hugetlb_vma_unlock_read(struct vm_area_struct *vma)
>>> +{
>>> +	if (__vma_shareable_flags_pmd(vma))
>>> +		up_read((struct rw_semaphore *)vma->vm_private_data);
>>> +}
>>> +
>>> +void hugetlb_vma_lock_write(struct vm_area_struct *vma)
>>> +{
>>> +	if (__vma_shareable_flags_pmd(vma))
>>> +		down_write((struct rw_semaphore *)vma->vm_private_data);
>>> +}
>>> +
>>> +void hugetlb_vma_unlock_write(struct vm_area_struct *vma)
>>> +{
>>> +	if (__vma_shareable_flags_pmd(vma))
>>> +		up_write((struct rw_semaphore *)vma->vm_private_data);
>>> +}
>>> +
>>> +int hugetlb_vma_trylock_write(struct vm_area_struct *vma)
>>> +{
>>> +	if (!__vma_shareable_flags_pmd(vma))
>>> +		return 1;
>>> +
>>> +	return down_write_trylock((struct rw_semaphore *)vma->vm_private_data);
>>> +}
>>> +
>>> +void hugetlb_vma_assert_locked(struct vm_area_struct *vma)
>>> +{
>>> +	if (__vma_shareable_flags_pmd(vma))
>>> +		lockdep_assert_held((struct rw_semaphore *)
>>> +				vma->vm_private_data);
>>> +}
>>> +
>>> +static void hugetlb_free_vma_lock(struct vm_area_struct *vma)
>>> +{
>>> +	/* Only present in sharable vmas */
>>> +	if (!vma || !(vma->vm_flags & (VM_MAYSHARE | VM_SHARED)))

Does above check need a comment? VM_SHARED is check here but not in below hugetlb_alloc_vma_lock.

Thanks.

>>> +		return;
>>> +
>>> +	if (vma->vm_private_data) {
>>> +		kfree(vma->vm_private_data);
>>> +		vma->vm_private_data = NULL;
>>> +	}
>>> +}
>>> +
>>> +static void hugetlb_alloc_vma_lock(struct vm_area_struct *vma)
>>> +{
>>> +	struct rw_semaphore *vma_sema;
>>> +
>>> +	/* Only establish in (flags) sharable vmas */
>>> +	if (!vma || !(vma->vm_flags & (VM_MAYSHARE | VM_SHARED)))
> 
> Based on my explanation above, I think this should only check VM_MAYSHARE.
> 
>>> +		return;
>>> +> +	if (!vma_pmd_shareable(vma)) {
>>> +		vma->vm_private_data = NULL;
>>> +		return;
>>> +	}
>>> +
>>> +	vma_sema = kmalloc(sizeof(*vma_sema), GFP_KERNEL);
>>> +	if (!vma_sema) {
>>> +		/*
>>> +		 * If we can not allocate semaphore, then vma can not
>>> +		 * participate in pmd sharing.
>>> +		 */
>>> +		vma->vm_private_data = NULL;
>>> +	} else {
>>> +		init_rwsem(vma_sema);
>>> +		vma->vm_private_data = vma_sema;
>>> +	}
>>
>> This code is really subtle. If it's called from hugetlb_vm_op_open during fork after
>> hugetlb_dup_vma_private is done, there should already be a kmalloc-ed vma_sema for this
>> vma (because hugetlb_alloc_vma_lock is also called by hugetlb_dup_vma_private). So we
>> can't simply change the value of vm_private_data here or vma_sema will be leaked ?
> 
> Yes, I believe your analysis is correct.
> 
>>                                                                                    But
>> when hugetlb_alloc_vma_lock is called from hugetlb_reserve_pages, it should work fine.
>> Or am I miss something?
> 
> You are right.  This is an issue in the current code.  I will address in
> the next version.
> 
> Thanks for all your comments on this series!
>