@@ -379,40 +379,32 @@ void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
(pfvf->hw.cq_ecount_wait - 1));
}
-dma_addr_t otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
- gfp_t gfp)
+dma_addr_t _otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
{
dma_addr_t iova;
+ u8 *buf;
- /* Check if request can be accommodated in previous allocated page */
- if (pool->page && ((pool->page_offset + pool->rbsize) <=
- (PAGE_SIZE << pool->rbpage_order))) {
- pool->pageref++;
- goto ret;
- }
-
- otx2_get_page(pool);
-
- /* Allocate a new page */
- pool->page = alloc_pages(gfp | __GFP_COMP | __GFP_NOWARN,
- pool->rbpage_order);
- if (unlikely(!pool->page))
+ buf = napi_alloc_frag(pool->rbsize);
+ if (unlikely(!buf))
return -ENOMEM;
- pool->page_offset = 0;
-ret:
- iova = (u64)otx2_dma_map_page(pfvf, pool->page, pool->page_offset,
- pool->rbsize, DMA_FROM_DEVICE);
- if (!iova) {
- if (!pool->page_offset)
- __free_pages(pool->page, pool->rbpage_order);
- pool->page = NULL;
+ iova = dma_map_single(pfvf->dev, buf, pool->rbsize, DMA_FROM_DEVICE);
+ if (unlikely(dma_mapping_error(pfvf->dev, iova)))
return -ENOMEM;
- }
- pool->page_offset += pool->rbsize;
+
return iova;
}
+static dma_addr_t otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool)
+{
+ dma_addr_t addr;
+
+ local_bh_disable();
+ addr = _otx2_alloc_rbuf(pfvf, pool);
+ local_bh_enable();
+ return addr;
+}
+
void otx2_tx_timeout(struct net_device *netdev, unsigned int txq)
{
struct otx2_nic *pfvf = netdev_priv(netdev);
@@ -805,7 +797,7 @@ static void otx2_pool_refill_task(struct work_struct *work)
free_ptrs = cq->pool_ptrs;
while (cq->pool_ptrs) {
- bufptr = otx2_alloc_rbuf(pfvf, rbpool, GFP_KERNEL);
+ bufptr = otx2_alloc_rbuf(pfvf, rbpool);
if (bufptr <= 0) {
/* Schedule a WQ if we fails to free atleast half of the
* pointers else enable napi for this RQ.
@@ -1064,7 +1056,6 @@ static int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
return err;
pool->rbsize = buf_size;
- pool->rbpage_order = get_order(buf_size);
/* Initialize this pool's context via AF */
aq = otx2_mbox_alloc_msg_npa_aq_enq(&pfvf->mbox);
@@ -1152,13 +1143,12 @@ int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
return -ENOMEM;
for (ptr = 0; ptr < num_sqbs; ptr++) {
- bufptr = otx2_alloc_rbuf(pfvf, pool, GFP_KERNEL);
+ bufptr = otx2_alloc_rbuf(pfvf, pool);
if (bufptr <= 0)
return bufptr;
otx2_aura_freeptr(pfvf, pool_id, bufptr);
sq->sqb_ptrs[sq->sqb_count++] = (u64)bufptr;
}
- otx2_get_page(pool);
}
return 0;
@@ -1204,13 +1194,12 @@ int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
pool = &pfvf->qset.pool[pool_id];
for (ptr = 0; ptr < num_ptrs; ptr++) {
- bufptr = otx2_alloc_rbuf(pfvf, pool, GFP_KERNEL);
+ bufptr = otx2_alloc_rbuf(pfvf, pool);
if (bufptr <= 0)
return bufptr;
otx2_aura_freeptr(pfvf, pool_id,
bufptr + OTX2_HEAD_ROOM);
}
- otx2_get_page(pool);
}
return 0;
@@ -434,18 +434,6 @@ static inline void otx2_aura_freeptr(struct otx2_nic *pfvf,
otx2_get_regaddr(pfvf, NPA_LF_AURA_OP_FREE0));
}
-/* Update page ref count */
-static inline void otx2_get_page(struct otx2_pool *pool)
-{
- if (!pool->page)
- return;
-
- if (pool->pageref)
- page_ref_add(pool->page, pool->pageref);
- pool->pageref = 0;
- pool->page = NULL;
-}
-
static inline int otx2_get_pool_idx(struct otx2_nic *pfvf, int type, int idx)
{
if (type == AURA_NIX_SQ)
@@ -589,8 +577,7 @@ int otx2_txschq_config(struct otx2_nic *pfvf, int lvl);
int otx2_txsch_alloc(struct otx2_nic *pfvf);
int otx2_txschq_stop(struct otx2_nic *pfvf);
void otx2_sqb_flush(struct otx2_nic *pfvf);
-dma_addr_t otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
- gfp_t gfp);
+dma_addr_t _otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool);
int otx2_rxtx_enable(struct otx2_nic *pfvf, bool enable);
void otx2_ctx_disable(struct mbox *mbox, int type, bool npa);
int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable);
@@ -286,7 +286,7 @@ static int otx2_rx_napi_handler(struct otx2_nic *pfvf,
/* Refill pool with new buffers */
while (cq->pool_ptrs) {
- bufptr = otx2_alloc_rbuf(pfvf, cq->rbpool, GFP_ATOMIC);
+ bufptr = _otx2_alloc_rbuf(pfvf, cq->rbpool);
if (unlikely(bufptr <= 0)) {
struct refill_work *work;
struct delayed_work *dwork;
@@ -304,7 +304,6 @@ static int otx2_rx_napi_handler(struct otx2_nic *pfvf,
otx2_aura_freeptr(pfvf, cq->cq_idx, bufptr + OTX2_HEAD_ROOM);
cq->pool_ptrs--;
}
- otx2_get_page(cq->rbpool);
return processed_cqe;
}
@@ -113,11 +113,7 @@ struct otx2_cq_poll {
struct otx2_pool {
struct qmem *stack;
struct qmem *fc_addr;
- u8 rbpage_order;
u16 rbsize;
- u32 page_offset;
- u16 pageref;
- struct page *page;
};
struct otx2_cq_queue {
In the current codes, the octeontx2 uses its own method to allocate the pool buffers, but there are some issues in this implementation. 1. We have to run the otx2_get_page() for each allocation cycle and this is pretty error prone. As I can see there is no invocation of the otx2_get_page() in otx2_pool_refill_task(), this will leave the allocated pages have the wrong refcount and may be freed wrongly. 2. It wastes memory. For example, if we only receive one packet in a NAPI RX cycle, and then allocate a 2K buffer with otx2_alloc_rbuf() to refill the pool buffers and leave the remain area of the allocated page wasted. On a kernel with 64K page, 62K area is wasted. IMHO it is really unnecessary to implement our own method for the buffers allocate, we can reuse the napi_alloc_frag() to simplify our code. Signed-off-by: Kevin Hao <haokexin@gmail.com> --- .../marvell/octeontx2/nic/otx2_common.c | 51 ++++++++----------- .../marvell/octeontx2/nic/otx2_common.h | 15 +----- .../marvell/octeontx2/nic/otx2_txrx.c | 3 +- .../marvell/octeontx2/nic/otx2_txrx.h | 4 -- 4 files changed, 22 insertions(+), 51 deletions(-)