[PATCH 0/1] IB/umem: use get_user_pages_fast() to pin DMA pages

[PATCH 0/1] IB/umem: use get_user_pages_fast() to pin DMA pages

[John Hubbard]

Hi Jason, Andrew,

Jason: Here is the change to get_user_pages_fast(), that you requested
during the review of "mm/gup: track dma-pinned pages: FOLL_PIN,
FOLL_LONGTERM" [1].

This is a stand-alone patch that applies to today's 5.4.0-rc6 linux.git.

If anyone could please provide any run-time testing (and of course,
Reviewed-by's), that would be huge. I've compiled and run it, but
bpftrace tells me (as I already knew) that my current anemic IB setup
is not actually exercising this code path.

Andrew: unless instructed otherwise, I plan to also re-post this as part
of the next (v3) version of [1], and hopefully have the whole series go
through your tree. That would avoid merge conflicts, and as I understand
it, there is no particular urgency for this patch, so we might as well
do it the easy way.


John Hubbard (1):
  IB/umem: use get_user_pages_fast() to pin DMA pages

 drivers/infiniband/core/umem.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

-- 
2.24.0

[PATCH 1/1] IB/umem: use get_user_pages_fast() to pin DMA pages

[John Hubbard]

And get rid of the mmap_sem calls, as part of that. Note
that get_user_pages_fast() will, if necessary, fall back to
__gup_longterm_unlocked(), which takes the mmap_sem as needed.

Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
---
 drivers/infiniband/core/umem.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index 24244a2f68cc..3d664a2539eb 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -271,16 +271,13 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
 	sg = umem->sg_head.sgl;
 
 	while (npages) {
-		down_read(&mm->mmap_sem);
-		ret = get_user_pages(cur_base,
-				     min_t(unsigned long, npages,
-					   PAGE_SIZE / sizeof (struct page *)),
-				     gup_flags | FOLL_LONGTERM,
-				     page_list, NULL);
-		if (ret < 0) {
-			up_read(&mm->mmap_sem);
+		ret = get_user_pages_fast(cur_base,
+					  min_t(unsigned long, npages,
+						PAGE_SIZE /
+						sizeof(struct page *)),
+					  gup_flags | FOLL_LONGTERM, page_list);
+		if (ret < 0)
 			goto umem_release;
-		}
 
 		cur_base += ret * PAGE_SIZE;
 		npages   -= ret;
@@ -288,8 +285,6 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
 		sg = ib_umem_add_sg_table(sg, page_list, ret,
 			dma_get_max_seg_size(context->device->dma_device),
 			&umem->sg_nents);
-
-		up_read(&mm->mmap_sem);
 	}
 
 	sg_mark_end(sg);
-- 
2.24.0

Re: [PATCH 1/1] IB/umem: use get_user_pages_fast() to pin DMA pages

[Ira Weiny]

On Fri, Nov 08, 2019 at 06:04:34PM -0800, John Hubbard wrote:
> And get rid of the mmap_sem calls, as part of that. Note
> that get_user_pages_fast() will, if necessary, fall back to
> __gup_longterm_unlocked(), which takes the mmap_sem as needed.
> 
> Cc: Jason Gunthorpe <jgg@ziepe.ca>
> Cc: Ira Weiny <ira.weiny@intel.com>

Reviewed-by: Ira Weiny <ira.weiny@intel.com>

> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
> ---
>  drivers/infiniband/core/umem.c | 17 ++++++-----------
>  1 file changed, 6 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
> index 24244a2f68cc..3d664a2539eb 100644
> --- a/drivers/infiniband/core/umem.c
> +++ b/drivers/infiniband/core/umem.c
> @@ -271,16 +271,13 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
>  	sg = umem->sg_head.sgl;
>  
>  	while (npages) {
> -		down_read(&mm->mmap_sem);
> -		ret = get_user_pages(cur_base,
> -				     min_t(unsigned long, npages,
> -					   PAGE_SIZE / sizeof (struct page *)),
> -				     gup_flags | FOLL_LONGTERM,
> -				     page_list, NULL);
> -		if (ret < 0) {
> -			up_read(&mm->mmap_sem);
> +		ret = get_user_pages_fast(cur_base,
> +					  min_t(unsigned long, npages,
> +						PAGE_SIZE /
> +						sizeof(struct page *)),
> +					  gup_flags | FOLL_LONGTERM, page_list);
> +		if (ret < 0)
>  			goto umem_release;
> -		}
>  
>  		cur_base += ret * PAGE_SIZE;
>  		npages   -= ret;
> @@ -288,8 +285,6 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
>  		sg = ib_umem_add_sg_table(sg, page_list, ret,
>  			dma_get_max_seg_size(context->device->dma_device),
>  			&umem->sg_nents);
> -
> -		up_read(&mm->mmap_sem);
>  	}
>  
>  	sg_mark_end(sg);
> -- 
> 2.24.0
> 

Re: [PATCH 1/1] IB/umem: use get_user_pages_fast() to pin DMA pages

[John Hubbard]

On 11/10/19 7:21 PM, Ira Weiny wrote:
> On Fri, Nov 08, 2019 at 06:04:34PM -0800, John Hubbard wrote:
>> And get rid of the mmap_sem calls, as part of that. Note
>> that get_user_pages_fast() will, if necessary, fall back to
>> __gup_longterm_unlocked(), which takes the mmap_sem as needed.
>>
>> Cc: Jason Gunthorpe <jgg@ziepe.ca>
>> Cc: Ira Weiny <ira.weiny@intel.com>
> 
> Reviewed-by: Ira Weiny <ira.weiny@intel.com>
> 

Thanks for the review, Ira! This will show up shortly, in the v3
series of "mm/gup: track dma-pinned pages: FOLL_PIN, FOLL_LONGTERM".


thanks,
-- 
John Hubbard
NVIDIA

>> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
>> ---
>>  drivers/infiniband/core/umem.c | 17 ++++++-----------
>>  1 file changed, 6 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
>> index 24244a2f68cc..3d664a2539eb 100644
>> --- a/drivers/infiniband/core/umem.c
>> +++ b/drivers/infiniband/core/umem.c
>> @@ -271,16 +271,13 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
>>  	sg = umem->sg_head.sgl;
>>  
>>  	while (npages) {
>> -		down_read(&mm->mmap_sem);
>> -		ret = get_user_pages(cur_base,
>> -				     min_t(unsigned long, npages,
>> -					   PAGE_SIZE / sizeof (struct page *)),
>> -				     gup_flags | FOLL_LONGTERM,
>> -				     page_list, NULL);
>> -		if (ret < 0) {
>> -			up_read(&mm->mmap_sem);
>> +		ret = get_user_pages_fast(cur_base,
>> +					  min_t(unsigned long, npages,
>> +						PAGE_SIZE /
>> +						sizeof(struct page *)),
>> +					  gup_flags | FOLL_LONGTERM, page_list);
>> +		if (ret < 0)
>>  			goto umem_release;
>> -		}
>>  
>>  		cur_base += ret * PAGE_SIZE;
>>  		npages   -= ret;
>> @@ -288,8 +285,6 @@ struct ib_umem *ib_umem_get(struct ib_udata *udata, unsigned long addr,
>>  		sg = ib_umem_add_sg_table(sg, page_list, ret,
>>  			dma_get_max_seg_size(context->device->dma_device),
>>  			&umem->sg_nents);
>> -
>> -		up_read(&mm->mmap_sem);
>>  	}
>>  
>>  	sg_mark_end(sg);
>> -- 
>> 2.24.0
>>
> 

Re: [PATCH 1/1] IB/umem: use get_user_pages_fast() to pin DMA pages

[Jason Gunthorpe]

On Fri, Nov 08, 2019 at 06:04:34PM -0800, John Hubbard wrote:
> And get rid of the mmap_sem calls, as part of that. Note
> that get_user_pages_fast() will, if necessary, fall back to
> __gup_longterm_unlocked(), which takes the mmap_sem as needed.
> 
> Cc: Jason Gunthorpe <jgg@ziepe.ca>
> Cc: Ira Weiny <ira.weiny@intel.com>
> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
> ---
>  drivers/infiniband/core/umem.c | 17 ++++++-----------
>  1 file changed, 6 insertions(+), 11 deletions(-)

Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>

This can go through Andrew's tree if he takes this larger series,
otherwise I'll grab it

Thanks,
Jason