[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20241213122739.4050137-9-linyunsheng@huawei.com>
Date: Fri, 13 Dec 2024 20:27:39 +0800
From: Yunsheng Lin <linyunsheng@...wei.com>
To: <davem@...emloft.net>, <kuba@...nel.org>, <pabeni@...hat.com>
CC: <somnath.kotur@...adcom.com>, <liuyonglong@...wei.com>,
<fanghaiqing@...wei.com>, <zhangkun09@...wei.com>, Yunsheng Lin
<linyunsheng@...wei.com>, Robin Murphy <robin.murphy@....com>, Alexander
Duyck <alexander.duyck@...il.com>, IOMMU <iommu@...ts.linux.dev>, Jesper
Dangaard Brouer <hawk@...nel.org>, Ilias Apalodimas
<ilias.apalodimas@...aro.org>, Eric Dumazet <edumazet@...gle.com>, Simon
Horman <horms@...nel.org>, <netdev@...r.kernel.org>,
<linux-kernel@...r.kernel.org>
Subject: [PATCH RFCv5 8/8] page_pool: use list instead of array for alloc cache
As the alloc cache is always protected by NAPI context
protection, use encoded_next as a pointer to a next item
to avoid the using the array.
Testing shows there is about 3ns improvement for the
performance of 'time_bench_page_pool01_fast_path' test
case.
CC: Robin Murphy <robin.murphy@....com>
CC: Alexander Duyck <alexander.duyck@...il.com>
CC: IOMMU <iommu@...ts.linux.dev>
Signed-off-by: Yunsheng Lin <linyunsheng@...wei.com>
---
include/net/page_pool/types.h | 2 +-
net/core/page_pool.c | 59 +++++++++++++++++++++++++----------
2 files changed, 43 insertions(+), 18 deletions(-)
diff --git a/include/net/page_pool/types.h b/include/net/page_pool/types.h
index be50a25c4aa0..d942e1da5ff7 100644
--- a/include/net/page_pool/types.h
+++ b/include/net/page_pool/types.h
@@ -50,7 +50,7 @@
#define PP_ALLOC_CACHE_REFILL 64
struct pp_alloc_cache {
u32 count;
- netmem_ref cache[PP_ALLOC_CACHE_SIZE];
+ struct page_pool_item *list;
/* Keep batched refilled pages here to avoid doing the atomic operation
* for each page.
diff --git a/net/core/page_pool.c b/net/core/page_pool.c
index ab832bfa004b..aadbf62e9708 100644
--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -382,6 +382,27 @@ static netmem_ref page_pool_consume_ring(struct page_pool *pool)
return list->pp_netmem;
}
+static netmem_ref __page_pool_consume_alloc(struct page_pool *pool)
+{
+ struct page_pool_item *item = pool->alloc.list;
+
+ pool->alloc.list = page_pool_item_get_next(item);
+ pool->alloc.count--;
+
+ return item->pp_netmem;
+}
+
+static void __page_pool_recycle_in_alloc(struct page_pool *pool,
+ netmem_ref netmem)
+{
+ struct page_pool_item *item;
+
+ item = netmem_get_pp_item(netmem);
+ page_pool_item_set_next(item, pool->alloc.list);
+ pool->alloc.list = item;
+ pool->alloc.count++;
+}
+
static __always_inline void __page_pool_release_page_dma(struct page_pool *pool,
netmem_ref netmem,
bool destroyed)
@@ -673,10 +694,12 @@ static void __page_pool_return_page(struct page_pool *pool, netmem_ref netmem,
static noinline netmem_ref page_pool_refill_alloc_cache(struct page_pool *pool)
{
- struct page_pool_item *refill;
+ struct page_pool_item *refill, *alloc, *curr;
netmem_ref netmem;
int pref_nid; /* preferred NUMA node */
+ DEBUG_NET_WARN_ON_ONCE(pool->alloc.count || pool->alloc.list);
+
/* Quicker fallback, avoid locks when ring is empty */
refill = pool->alloc.refill;
if (unlikely(!refill && !READ_ONCE(pool->ring.list))) {
@@ -694,6 +717,7 @@ static noinline netmem_ref page_pool_refill_alloc_cache(struct page_pool *pool)
pref_nid = numa_mem_id(); /* will be zero like page_to_nid() */
#endif
+ alloc = NULL;
/* Refill alloc array, but only if NUMA match */
do {
if (unlikely(!refill)) {
@@ -702,10 +726,13 @@ static noinline netmem_ref page_pool_refill_alloc_cache(struct page_pool *pool)
break;
}
+ curr = refill;
netmem = refill->pp_netmem;
refill = page_pool_item_get_next(refill);
if (likely(netmem_is_pref_nid(netmem, pref_nid))) {
- pool->alloc.cache[pool->alloc.count++] = netmem;
+ page_pool_item_set_next(curr, alloc);
+ pool->alloc.count++;
+ alloc = curr;
} else {
/* NUMA mismatch;
* (1) release 1 page to page-allocator and
@@ -725,7 +752,8 @@ static noinline netmem_ref page_pool_refill_alloc_cache(struct page_pool *pool)
/* Return last page */
if (likely(pool->alloc.count > 0)) {
atomic_sub(pool->alloc.count, &pool->ring.count);
- netmem = pool->alloc.cache[--pool->alloc.count];
+ pool->alloc.list = page_pool_item_get_next(alloc);
+ pool->alloc.count--;
alloc_stat_inc(pool, refill);
}
@@ -740,7 +768,7 @@ static netmem_ref __page_pool_get_cached(struct page_pool *pool)
/* Caller MUST guarantee safe non-concurrent access, e.g. softirq */
if (likely(pool->alloc.count)) {
/* Fast-path */
- netmem = pool->alloc.cache[--pool->alloc.count];
+ netmem = __page_pool_consume_alloc(pool);
alloc_stat_inc(pool, fast);
} else {
netmem = page_pool_refill_alloc_cache(pool);
@@ -859,6 +887,7 @@ static struct page *__page_pool_alloc_page_order(struct page_pool *pool,
static noinline netmem_ref __page_pool_alloc_pages_slow(struct page_pool *pool,
gfp_t gfp)
{
+ netmem_ref netmems[PP_ALLOC_CACHE_REFILL] = {0};
const int bulk = PP_ALLOC_CACHE_REFILL;
unsigned int pp_order = pool->p.order;
bool dma_map = pool->dma_map;
@@ -869,16 +898,12 @@ static noinline netmem_ref __page_pool_alloc_pages_slow(struct page_pool *pool,
if (unlikely(pp_order))
return page_to_netmem(__page_pool_alloc_page_order(pool, gfp));
- /* Unnecessary as alloc cache is empty, but guarantees zero count */
- if (unlikely(pool->alloc.count > 0))
- return pool->alloc.cache[--pool->alloc.count];
-
- /* Mark empty alloc.cache slots "empty" for alloc_pages_bulk_array */
- memset(&pool->alloc.cache, 0, sizeof(void *) * bulk);
+ /* alloc cache should be empty */
+ DEBUG_NET_WARN_ON_ONCE(pool->alloc.count || pool->alloc.list);
nr_pages = alloc_pages_bulk_array_node(gfp,
pool->p.nid, bulk,
- (struct page **)pool->alloc.cache);
+ (struct page **)netmems);
if (unlikely(!nr_pages))
return 0;
@@ -886,7 +911,7 @@ static noinline netmem_ref __page_pool_alloc_pages_slow(struct page_pool *pool,
* page element have not been (possibly) DMA mapped.
*/
for (i = 0; i < nr_pages; i++) {
- netmem = pool->alloc.cache[i];
+ netmem = netmems[i];
if (unlikely(!page_pool_set_pp_info(pool, netmem))) {
put_page(netmem_to_page(netmem));
@@ -899,7 +924,7 @@ static noinline netmem_ref __page_pool_alloc_pages_slow(struct page_pool *pool,
continue;
}
- pool->alloc.cache[pool->alloc.count++] = netmem;
+ __page_pool_recycle_in_alloc(pool, netmem);
/* Track how many pages are held 'in-flight' */
pool->pages_state_hold_cnt++;
trace_page_pool_state_hold(pool, netmem,
@@ -908,7 +933,7 @@ static noinline netmem_ref __page_pool_alloc_pages_slow(struct page_pool *pool,
/* Return last page */
if (likely(pool->alloc.count > 0)) {
- netmem = pool->alloc.cache[--pool->alloc.count];
+ netmem = __page_pool_consume_alloc(pool);
alloc_stat_inc(pool, slow);
} else {
netmem = 0;
@@ -1078,7 +1103,7 @@ static bool page_pool_recycle_in_cache(netmem_ref netmem,
}
/* Caller MUST have verified/know (page_ref_count(page) == 1) */
- pool->alloc.cache[pool->alloc.count++] = netmem;
+ __page_pool_recycle_in_alloc(pool, netmem);
recycle_stat_inc(pool, cached);
return true;
}
@@ -1392,7 +1417,7 @@ static void page_pool_empty_alloc_cache_once(struct page_pool *pool)
* call concurrently.
*/
while (pool->alloc.count) {
- netmem = pool->alloc.cache[--pool->alloc.count];
+ netmem = __page_pool_consume_alloc(pool);
page_pool_return_page(pool, netmem);
}
@@ -1532,7 +1557,7 @@ void page_pool_update_nid(struct page_pool *pool, int new_nid)
/* Flush pool alloc cache, as refill will check NUMA node */
while (pool->alloc.count) {
- netmem = pool->alloc.cache[--pool->alloc.count];
+ netmem = __page_pool_consume_alloc(pool);
__page_pool_return_page(pool, netmem, false);
}
}
--
2.33.0
Powered by blists - more mailing lists