Re: [for-next PATCH 1/1] RDMA/rxe: sgt_append from ib_umem_get is not highmem

[Yanjun Zhu <yanjun.zhu@xxxxxxxxx>]

在 2022/11/19 18:30, lizhijian@xxxxxxxxxxx 写道:
> There was a thread that tries to refactor iova_to_vaddr[1][2], where
> page_address will be drop. if so, your changes will be no longer needed.

My commit is to indicate that your commit is not good enough.

And you should make all the related commits in a patch series.

Zhu Yanjun

> [1]https://lore.kernel.org/lkml/a7decec2-77d9-db4c-ff66-ff597da8bc71@xxxxxxxxxxx/T/
> [2]https://www.spinics.net/lists/linux-rdma/msg114053.html
>
> Thanks
> Zhijian
>
>
> On 20/11/2022 09:29, Zhu Yanjun wrote:
> > From: Zhu Yanjun <yanjun.zhu@xxxxxxxxx>
> >
> > In ib_umem_get, sgt_append is allocated from the function
> > sg_alloc_append_table_from_pages. And it is not from highmem.
> >
> > As such, the return value of page_address will not be NULL.
> >
> > Signed-off-by: Zhu Yanjun <yanjun.zhu@xxxxxxxxx>
> > ---
> >    drivers/infiniband/sw/rxe/rxe_mr.c | 9 ++-------
> >    1 file changed, 2 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infiniband/sw/rxe/rxe_mr.c
> > index b1423000e4bc..3f33a4cdef24 100644
> > --- a/drivers/infiniband/sw/rxe/rxe_mr.c
> > +++ b/drivers/infiniband/sw/rxe/rxe_mr.c
> > @@ -119,7 +119,6 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
> >    	struct ib_umem		*umem;
> >    	struct sg_page_iter	sg_iter;
> >    	int			num_buf;
> > -	void			*vaddr;
> >    	int err;
> >    
> >    	umem = ib_umem_get(&rxe->ib_dev, start, length, access);
> > @@ -149,6 +148,8 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
> >    		buf = map[0]->buf;
> >    
> >    		for_each_sgtable_page (&umem->sgt_append.sgt, &sg_iter, 0) {
> > +			void *vaddr;
> > +
> >    			if (num_buf >= RXE_BUF_PER_MAP) {
> >    				map++;
> >    				buf = map[0]->buf;
> > @@ -156,16 +157,10 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
> >    			}
> >    
> >    			vaddr = page_address(sg_page_iter_page(&sg_iter));
> > -			if (!vaddr) {
> > -				rxe_dbg_mr(mr, "Unable to get virtual address\n");
> > -				err = -ENOMEM;
> > -				goto err_release_umem;
> > -			}
> >    			buf->addr = (uintptr_t)vaddr;
> >    			buf->size = PAGE_SIZE;
> >    			num_buf++;
> >    			buf++;
> > -
> >    		}
> >    	}
> >