Re: [PATCH] RDMA/rxe: Replace page_address() with kmap_local_page()

["Yang, Xiao/杨 晓" <yangx.jy@fujitsu.com>]

On 2022/11/11 11:52, Li, Zhijian/李 智坚 wrote:
> Xiao,
> 
> What a coincidence!! i aslo had a similar patches do the same thing.
> which would be more easy to be reviewed, i will post it as well. You are
> free to use any of this or ignore it.

Hi Zhijian,

Thanks a lot for your reference.

Li Zhijian (5):
   RDMA/rxe: Remove rxe_phys_buf.size
   -> It has been included in my first patch[1].
   RDMA/rxe: use iova_to_vaddr to transform iova for rxe_mr_copy
   -> I don't want to do the check (e.g. mr_check_range()) in loops.
   RDMA/rxe: iova_to_vaddr cleanup
   RDMA/rxe: refactor iova_to_vaddr
   RDMA/rxe: Rename iova_to_vaddr to rxe_map_iova
   -> I think these three are similar to my second patch[2].

[1]: 
https://lore.kernel.org/linux-rdma/1668153085-15-1-git-send-email-yangx.jy@fujitsu.com/T/#t
[2]: 
https://lore.kernel.org/linux-rdma/1668153085-15-2-git-send-email-yangx.jy@fujitsu.com/T/#u

Best Regards,
Xiao Yang

> 
> 
> Thanks
> Zhijian
> 
> 
> 
> 
> On 11/11/2022 11:19, Xiao Yang wrote:
>> 1) Use kmap_local_page() for new in-kernel memory protection schemes.
>> 2) Do some cleanup(e.g. remove struct rxe_phys_buf).
>>
>> Signed-off-by: Xiao Yang <yangx.jy@fujitsu.com>
>> ---
>>    drivers/infiniband/sw/rxe/rxe_loc.h   |  2 +
>>    drivers/infiniband/sw/rxe/rxe_mr.c    | 84 +++++++++++++--------------
>>    drivers/infiniband/sw/rxe/rxe_resp.c  |  1 +
>>    drivers/infiniband/sw/rxe/rxe_verbs.c |  6 +-
>>    drivers/infiniband/sw/rxe/rxe_verbs.h |  9 +--
>>    5 files changed, 45 insertions(+), 57 deletions(-)
>>
>> diff --git a/drivers/infiniband/sw/rxe/rxe_loc.h b/drivers/infiniband/sw/rxe/rxe_loc.h
>> index c2a5c8814a48..a63d29156a66 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_loc.h
>> +++ b/drivers/infiniband/sw/rxe/rxe_loc.h
>> @@ -68,6 +68,8 @@ void rxe_mr_init_dma(int access, struct rxe_mr *mr);
>>    int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
>>    		     int access, struct rxe_mr *mr);
>>    int rxe_mr_init_fast(int max_pages, struct rxe_mr *mr);
>> +void *rxe_map_to_vaddr(struct rxe_mr *mr, int map_index, int addr_index, size_t offset);
>> +void rxe_unmap_vaddr(struct rxe_mr *mr, void *vaddr);
>>    int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length,
>>    		enum rxe_mr_copy_dir dir);
>>    int copy_data(struct rxe_pd *pd, int access, struct rxe_dma_info *dma,
>> diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infiniband/sw/rxe/rxe_mr.c
>> index bc081002bddc..4246b7f34a29 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_mr.c
>> +++ b/drivers/infiniband/sw/rxe/rxe_mr.c
>> @@ -115,13 +115,10 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
>>    		     int access, struct rxe_mr *mr)
>>    {
>>    	struct rxe_map		**map;
>> -	struct rxe_phys_buf	*buf = NULL;
>>    	struct ib_umem		*umem;
>>    	struct sg_page_iter	sg_iter;
>>    	int			num_buf;
>> -	void			*vaddr;
>>    	int err;
>> -	int i;
>>    
>>    	umem = ib_umem_get(&rxe->ib_dev, start, length, access);
>>    	if (IS_ERR(umem)) {
>> @@ -144,32 +141,19 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
>>    
>>    	mr->page_shift = PAGE_SHIFT;
>>    	mr->page_mask = PAGE_SIZE - 1;
>> +	mr->ibmr.page_size = PAGE_SIZE;
>>    
>> -	num_buf			= 0;
>> +	num_buf = 0;
>>    	map = mr->map;
>>    	if (length > 0) {
>> -		buf = map[0]->buf;
>> -
>>    		for_each_sgtable_page (&umem->sgt_append.sgt, &sg_iter, 0) {
>>    			if (num_buf >= RXE_BUF_PER_MAP) {
>>    				map++;
>> -				buf = map[0]->buf;
>>    				num_buf = 0;
>>    			}
>>    
>> -			vaddr = page_address(sg_page_iter_page(&sg_iter));
>> -			if (!vaddr) {
>> -				pr_warn("%s: Unable to get virtual address\n",
>> -						__func__);
>> -				err = -ENOMEM;
>> -				goto err_cleanup_map;
>> -			}
>> -
>> -			buf->addr = (uintptr_t)vaddr;
>> -			buf->size = PAGE_SIZE;
>> +			map[0]->addrs[num_buf] = (uintptr_t)sg_page_iter_page(&sg_iter);
>>    			num_buf++;
>> -			buf++;
>> -
>>    		}
>>    	}
>>    
>> @@ -181,10 +165,6 @@ int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
>>    
>>    	return 0;
>>    
>> -err_cleanup_map:
>> -	for (i = 0; i < mr->num_map; i++)
>> -		kfree(mr->map[i]);
>> -	kfree(mr->map);
>>    err_release_umem:
>>    	ib_umem_release(umem);
>>    err_out:
>> @@ -216,9 +196,9 @@ static void lookup_iova(struct rxe_mr *mr, u64 iova, int *m_out, int *n_out,
>>    			size_t *offset_out)
>>    {
>>    	size_t offset = iova - mr->ibmr.iova + mr->offset;
>> +	u64 length = mr->ibmr.page_size;
>>    	int			map_index;
>> -	int			buf_index;
>> -	u64			length;
>> +	int			addr_index;
>>    
>>    	if (likely(mr->page_shift)) {
>>    		*offset_out = offset & mr->page_mask;
>> @@ -227,27 +207,46 @@ static void lookup_iova(struct rxe_mr *mr, u64 iova, int *m_out, int *n_out,
>>    		*m_out = offset >> mr->map_shift;
>>    	} else {
>>    		map_index = 0;
>> -		buf_index = 0;
>> -
>> -		length = mr->map[map_index]->buf[buf_index].size;
>> +		addr_index = 0;
>>    
>>    		while (offset >= length) {
>>    			offset -= length;
>> -			buf_index++;
>> +			addr_index++;
>>    
>> -			if (buf_index == RXE_BUF_PER_MAP) {
>> +			if (addr_index == RXE_BUF_PER_MAP) {
>>    				map_index++;
>> -				buf_index = 0;
>> +				addr_index = 0;
>>    			}
>> -			length = mr->map[map_index]->buf[buf_index].size;
>>    		}
>>    
>>    		*m_out = map_index;
>> -		*n_out = buf_index;
>> +		*n_out = addr_index;
>>    		*offset_out = offset;
>>    	}
>>    }
>>    
>> +void *rxe_map_to_vaddr(struct rxe_mr *mr, int map_index, int addr_index, size_t offset)
>> +{
>> +	void *vaddr = NULL;
>> +
>> +	if (mr->ibmr.type == IB_MR_TYPE_USER) {
>> +		vaddr = kmap_local_page((struct page *)mr->map[map_index]->addrs[addr_index]);
>> +		if (vaddr == NULL) {
>> +			pr_warn("Failed to map page");
>> +			return NULL;
>> +		}
>> +	} else
>> +		vaddr = (void *)(uintptr_t)mr->map[map_index]->addrs[addr_index];
>> +
>> +	return vaddr + offset;
>> +}
>> +
>> +void rxe_unmap_vaddr(struct rxe_mr *mr, void *vaddr)
>> +{
>> +	if (mr->ibmr.type == IB_MR_TYPE_USER)
>> +		kunmap_local(vaddr);
>> +}
>> +
>>    void *iova_to_vaddr(struct rxe_mr *mr, u64 iova, int length)
>>    {
>>    	size_t offset;
>> @@ -273,13 +272,13 @@ void *iova_to_vaddr(struct rxe_mr *mr, u64 iova, int length)
>>    
>>    	lookup_iova(mr, iova, &m, &n, &offset);
>>    
>> -	if (offset + length > mr->map[m]->buf[n].size) {
>> +	if (offset + length > mr->ibmr.page_size) {
>>    		pr_warn("crosses page boundary\n");
>>    		addr = NULL;
>>    		goto out;
>>    	}
>>    
>> -	addr = (void *)(uintptr_t)mr->map[m]->buf[n].addr + offset;
>> +	addr = rxe_map_to_vaddr(mr, m, n, offset);
>>    
>>    out:
>>    	return addr;
>> @@ -294,8 +293,6 @@ int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length,
>>    	int			err;
>>    	int			bytes;
>>    	u8			*va;
>> -	struct rxe_map		**map;
>> -	struct rxe_phys_buf	*buf;
>>    	int			m;
>>    	int			i;
>>    	size_t			offset;
>> @@ -325,17 +322,14 @@ int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length,
>>    
>>    	lookup_iova(mr, iova, &m, &i, &offset);
>>    
>> -	map = mr->map + m;
>> -	buf	= map[0]->buf + i;
>> -
>>    	while (length > 0) {
>>    		u8 *src, *dest;
>>    
>> -		va	= (u8 *)(uintptr_t)buf->addr + offset;
>> +		va = (u8 *)rxe_map_to_vaddr(mr, m, i, offset);
>>    		src = (dir == RXE_TO_MR_OBJ) ? addr : va;
>>    		dest = (dir == RXE_TO_MR_OBJ) ? va : addr;
>>    
>> -		bytes	= buf->size - offset;
>> +		bytes = mr->ibmr.page_size - offset;
>>    
>>    		if (bytes > length)
>>    			bytes = length;
>> @@ -346,14 +340,14 @@ int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length,
>>    		addr	+= bytes;
>>    
>>    		offset	= 0;
>> -		buf++;
>>    		i++;
>>    
>>    		if (i == RXE_BUF_PER_MAP) {
>>    			i = 0;
>> -			map++;
>> -			buf = map[0]->buf;
>> +			m++;
>>    		}
>> +
>> +		rxe_unmap_vaddr(mr, va);
>>    	}
>>    
>>    	return 0;
>> diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c
>> index c32bc12cc82f..31f9ba11a921 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_resp.c
>> +++ b/drivers/infiniband/sw/rxe/rxe_resp.c
>> @@ -652,6 +652,7 @@ static enum resp_states atomic_reply(struct rxe_qp *qp,
>>    
>>    	ret = RESPST_ACKNOWLEDGE;
>>    out:
>> +	rxe_unmap_vaddr(mr, vaddr);
>>    	return ret;
>>    }
>>    
>> diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c
>> index bcdfdadaebbc..13e4d660cb02 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_verbs.c
>> +++ b/drivers/infiniband/sw/rxe/rxe_verbs.c
>> @@ -948,16 +948,12 @@ static int rxe_set_page(struct ib_mr *ibmr, u64 addr)
>>    {
>>    	struct rxe_mr *mr = to_rmr(ibmr);
>>    	struct rxe_map *map;
>> -	struct rxe_phys_buf *buf;
>>    
>>    	if (unlikely(mr->nbuf == mr->num_buf))
>>    		return -ENOMEM;
>>    
>>    	map = mr->map[mr->nbuf / RXE_BUF_PER_MAP];
>> -	buf = &map->buf[mr->nbuf % RXE_BUF_PER_MAP];
>> -
>> -	buf->addr = addr;
>> -	buf->size = ibmr->page_size;
>> +	map->addrs[mr->nbuf % RXE_BUF_PER_MAP] = addr;
>>    	mr->nbuf++;
>>    
>>    	return 0;
>> diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.h b/drivers/infiniband/sw/rxe/rxe_verbs.h
>> index 22a299b0a9f0..d136f02d5b56 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_verbs.h
>> +++ b/drivers/infiniband/sw/rxe/rxe_verbs.h
>> @@ -277,15 +277,10 @@ enum rxe_mr_lookup_type {
>>    	RXE_LOOKUP_REMOTE,
>>    };
>>    
>> -#define RXE_BUF_PER_MAP		(PAGE_SIZE / sizeof(struct rxe_phys_buf))
>> -
>> -struct rxe_phys_buf {
>> -	u64      addr;
>> -	u64      size;
>> -};
>> +#define RXE_BUF_PER_MAP		(PAGE_SIZE / sizeof(u64))
>>    
>>    struct rxe_map {
>> -	struct rxe_phys_buf	buf[RXE_BUF_PER_MAP];
>> +	u64 addrs[RXE_BUF_PER_MAP];
>>    };
>>    
>>    static inline int rkey_is_mw(u32 rkey)