Re: [PATCH 4/5] mm/hmm: hmm_vma_fault() doesn't always call hmm_range_unregister()

[Ralph Campbell <rcampbell@nvidia.com>]

On 5/12/19 8:07 AM, Jerome Glisse wrote:
> On Tue, May 07, 2019 at 11:12:14AM -0700, Ralph Campbell wrote:
>>
>> On 5/7/19 6:15 AM, Souptick Joarder wrote:
>>> On Tue, May 7, 2019 at 5:00 AM <rcampbell@nvidia.com> wrote:
>>>>
>>>> From: Ralph Campbell <rcampbell@nvidia.com>
>>>>
>>>> The helper function hmm_vma_fault() calls hmm_range_register() but is
>>>> missing a call to hmm_range_unregister() in one of the error paths.
>>>> This leads to a reference count leak and ultimately a memory leak on
>>>> struct hmm.
>>>>
>>>> Always call hmm_range_unregister() if hmm_range_register() succeeded.
>>>
>>> How about * Call hmm_range_unregister() in error path if
>>> hmm_range_register() succeeded* ?
>>
>> Sure, sounds good.
>> I'll include that in v2.
> 
> NAK for the patch see below why
> 
>>
>>>>
>>>> Signed-off-by: Ralph Campbell <rcampbell@nvidia.com>
>>>> Cc: John Hubbard <jhubbard@nvidia.com>
>>>> Cc: Ira Weiny <ira.weiny@intel.com>
>>>> Cc: Dan Williams <dan.j.williams@intel.com>
>>>> Cc: Arnd Bergmann <arnd@arndb.de>
>>>> Cc: Balbir Singh <bsingharora@gmail.com>
>>>> Cc: Dan Carpenter <dan.carpenter@oracle.com>
>>>> Cc: Matthew Wilcox <willy@infradead.org>
>>>> Cc: Souptick Joarder <jrdr.linux@gmail.com>
>>>> Cc: Andrew Morton <akpm@linux-foundation.org>
>>>> ---
>>>>    include/linux/hmm.h | 3 ++-
>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/include/linux/hmm.h b/include/linux/hmm.h
>>>> index 35a429621e1e..fa0671d67269 100644
>>>> --- a/include/linux/hmm.h
>>>> +++ b/include/linux/hmm.h
>>>> @@ -559,6 +559,7 @@ static inline int hmm_vma_fault(struct hmm_range *range, bool block)
>>>>                   return (int)ret;
>>>>
>>>>           if (!hmm_range_wait_until_valid(range, HMM_RANGE_DEFAULT_TIMEOUT)) {
>>>> +               hmm_range_unregister(range);
>>>>                   /*
>>>>                    * The mmap_sem was taken by driver we release it here and
>>>>                    * returns -EAGAIN which correspond to mmap_sem have been
>>>> @@ -570,13 +571,13 @@ static inline int hmm_vma_fault(struct hmm_range *range, bool block)
>>>>
>>>>           ret = hmm_range_fault(range, block);
>>>>           if (ret <= 0) {
>>>> +               hmm_range_unregister(range);
>>>
>>> what is the reason to moved it up ?
>>
>> I moved it up because the normal calling pattern is:
>>      down_read(&mm->mmap_sem)
>>      hmm_vma_fault()
>>          hmm_range_register()
>>          hmm_range_fault()
>>          hmm_range_unregister()
>>      up_read(&mm->mmap_sem)
>>
>> I don't think it is a bug to unlock mmap_sem and then unregister,
>> it is just more consistent nesting.
> 
> So this is not the usage pattern with HMM usage pattern is:
> 
> hmm_range_register()
> hmm_range_fault()
> hmm_range_unregister()
> 
> The hmm_vma_fault() is gonne so this patch here break thing.
> 
> See https://cgit.freedesktop.org/~glisse/linux/log/?h=hmm-5.2-v3

The patch series is on top of v5.1-rc6-mmotm-2019-04-25-16-30.
hmm_vma_fault() is defined there and in your hmm-5.2-v3 branch as
a backward compatibility transition function in include/linux/hmm.h.
So I agree the new API is to use hmm_range_register(), etc.
This is intended to cover the transition period.
Note that hmm_vma_fault() is being called from
drivers/gpu/drm/nouveau/nouveau_svm.c in both trees.