DPDK-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
* [dpdk-dev v1] net/sxe2: replace private mempool cache bypass with rte_mbuf_raw_free_bulk
@ 2026-08-20 15:50 Kai Ji
  2026-08-21 18:03 ` Stephen Hemminger
  0 siblings, 1 reply; 2+ messages in thread
From: Kai Ji @ 2026-08-20 15:50 UTC (permalink / raw)
  To: dev; +Cc: Kai Ji, Bruce Richardson, Konstantin Ananyev, Jie Liu

The AVX-512 TX completion path directly manipulated the mempool cache
internals (cache->objs, cache->len, cache->flushthresh) instead of using
the mempool API.  This pattern is the same private bypass that existed in
the Intel common TX library before it was removed by commit 062d6fe5d0e4
("net/intel: do not bypass mbuf lib for buffer fast-free") for the same
reason: it omits mbuf instrumentation (history marking) and contains
dead flush code that accesses cache->objs[cache->size], which is one
past the end of the array when cache_size == RTE_MEMPOOL_CACHE_MAX_SIZE.

Replace with a single rte_mbuf_raw_free_bulk() call, matching the Intel
common library.  The MBUF_FAST_FREE offload guarantee (single pool,
refcnt == 1) makes this correct and the compiler inlines the bulk-free
call to eliminate the overhead difference.

Signed-off-by: Kai Ji <kai.ji@intel.com>
---
 drivers/net/sxe2/sxe2_txrx_vec_avx512.c | 37 ++-----------------------
 1 file changed, 3 insertions(+), 34 deletions(-)

diff --git a/drivers/net/sxe2/sxe2_txrx_vec_avx512.c b/drivers/net/sxe2/sxe2_txrx_vec_avx512.c
index a830c7a33b..9f992a9ddf 100644
--- a/drivers/net/sxe2/sxe2_txrx_vec_avx512.c
+++ b/drivers/net/sxe2/sxe2_txrx_vec_avx512.c
@@ -18,9 +18,6 @@ static __rte_always_inline int32_t sxe2_tx_bufs_free_vec_avx512(struct sxe2_tx_q
 	struct rte_mbuf *mbuf;
 	struct rte_mbuf *mbuf_free_arr[SXE2_TX_FREE_BUFFER_SIZE_MAX_VEC];
 	struct rte_mempool *mp;
-	struct rte_mempool_cache *cache;
-	void **cache_objs;
-	uint32_t copied;
 	uint32_t i;
 	int32_t ret;
 	uint16_t rs_thresh;
@@ -41,37 +38,9 @@ static __rte_always_inline int32_t sxe2_tx_bufs_free_vec_avx512(struct sxe2_tx_q
 	if ((txq->offloads & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE) &&
 			(rs_thresh & 31) == 0) {
 		mp = buffer[0].mbuf->pool;
-		cache = rte_mempool_default_cache(mp, rte_lcore_id());
-
-		if (cache == NULL || cache->len)
-			goto normal;
-
-		if (rs_thresh > RTE_MEMPOOL_CACHE_MAX_SIZE) {
-			(void)rte_mempool_ops_enqueue_bulk(mp, (void *)buffer, rs_thresh);
-			goto done;
-		}
-		cache_objs = &cache->objs[cache->len];
-
-		copied = 0;
-		while (copied < rs_thresh) {
-			const __m512i objs0 = _mm512_loadu_si512(&buffer[copied]);
-			const __m512i objs1 = _mm512_loadu_si512(&buffer[copied + 8]);
-			const __m512i objs2 = _mm512_loadu_si512(&buffer[copied + 16]);
-			const __m512i objs3 = _mm512_loadu_si512(&buffer[copied + 24]);
-
-			_mm512_storeu_si512(&cache_objs[copied], objs0);
-			_mm512_storeu_si512(&cache_objs[copied + 8], objs1);
-			_mm512_storeu_si512(&cache_objs[copied + 16], objs2);
-			_mm512_storeu_si512(&cache_objs[copied + 24], objs3);
-			copied += 32;
-		}
-		cache->len += rs_thresh;
-
-		if (cache->len >= cache->flushthresh) {
-			(void)rte_mempool_ops_enqueue_bulk(mp,
-					&cache->objs[cache->size], cache->len - cache->size);
-			cache->len = cache->size;
-		}
+		static_assert(sizeof(buffer[0]) == sizeof(struct rte_mbuf *),
+			"sxe2_tx_buffer_vec must be pointer-sized for bulk free cast");
+		rte_mbuf_raw_free_bulk(mp, (void *)buffer, rs_thresh);
 		goto done;
 	}
 
-- 
2.43.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [dpdk-dev v1] net/sxe2: replace private mempool cache bypass with rte_mbuf_raw_free_bulk
  2026-08-20 15:50 [dpdk-dev v1] net/sxe2: replace private mempool cache bypass with rte_mbuf_raw_free_bulk Kai Ji
@ 2026-08-21 18:03 ` Stephen Hemminger
  0 siblings, 0 replies; 2+ messages in thread
From: Stephen Hemminger @ 2026-08-21 18:03 UTC (permalink / raw)
  To: Kai Ji; +Cc: dev, Bruce Richardson, Konstantin Ananyev, Jie Liu

On Thu, 20 Aug 2026 15:50:08 +0000
Kai Ji <kai.ji@intel.com> wrote:

> The AVX-512 TX completion path directly manipulated the mempool cache
> internals (cache->objs, cache->len, cache->flushthresh) instead of using
> the mempool API.  This pattern is the same private bypass that existed in
> the Intel common TX library before it was removed by commit 062d6fe5d0e4
> ("net/intel: do not bypass mbuf lib for buffer fast-free") for the same
> reason: it omits mbuf instrumentation (history marking) and contains
> dead flush code that accesses cache->objs[cache->size], which is one
> past the end of the array when cache_size == RTE_MEMPOOL_CACHE_MAX_SIZE.
> 
> Replace with a single rte_mbuf_raw_free_bulk() call, matching the Intel
> common library.  The MBUF_FAST_FREE offload guarantee (single pool,
> refcnt == 1) makes this correct and the compiler inlines the bulk-free
> call to eliminate the overhead difference.
> 
> Signed-off-by: Kai Ji <kai.ji@intel.com>
> ---

Looks good but still some leftovers to remove.

FAILED: drivers/net/sxe2/libsxe2_avx512_lib.a.p/sxe2_txrx_vec_avx512.c.o 
gcc -Idrivers/net/sxe2/libsxe2_avx512_lib.a.p -Idrivers/net/sxe2 -I../drivers/net/sxe2 -Idrivers/common/sxe2 -I../drivers/common/sxe2 -Ilib/ethdev -I../lib/ethdev -Ilib/eal/common -I../lib/eal/common -I. -I.. -Iconfig -I../config -Ilib/eal/include -I../lib/eal/include -Ilib/eal/linux/include -I../lib/eal/linux/include -Ilib/eal/x86/include -I../lib/eal/x86/include -I../kernel/linux -Ilib/eal -I../lib/eal -Ilib/kvargs -I../lib/kvargs -Ilib/log -I../lib/log -Ilib/metrics -I../lib/metrics -Ilib/telemetry -I../lib/telemetry -Ilib/argparse -I../lib/argparse -Ilib/net -I../lib/net -Ilib/mbuf -I../lib/mbuf -Ilib/mempool -I../lib/mempool -Ilib/ring -I../lib/ring -Ilib/meter -I../lib/meter -Ilib/hash -I../lib/hash -Ilib/rcu -I../lib/rcu -Ilib/security -I../lib/security -Ilib/cryptodev -I../lib/cryptodev -Idrivers/bus/pci -I../drivers/bus/pci -I../drivers/bus/pci/linux -Ilib/pci -I../lib/pci -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -Wextra -Werror -std=c11 -O3 -include rte_config.h -Wvla -Wcast-qual -Wdeprecated -Wformat -Wformat-nonliteral -Wformat-security -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wpointer-arith -Wshadow -Wsign-compare -Wstrict-prototypes -Wundef -Wwrite-strings -Wno-packed-not-aligned -Wno-missing-field-initializers -D_GNU_SOURCE -fPIC -march=native -mrtm -DALLOW_EXPERIMENTAL_API -DALLOW_INTERNAL_API -Wno-format-truncation -g -DCC_AVX512_SUPPORT -mavx512f -mavx512bw -march=skylake-avx512 -MD -MQ drivers/net/sxe2/libsxe2_avx512_lib.a.p/sxe2_txrx_vec_avx512.c.o -MF drivers/net/sxe2/libsxe2_avx512_lib.a.p/sxe2_txrx_vec_avx512.c.o.d -o drivers/net/sxe2/libsxe2_avx512_lib.a.p/sxe2_txrx_vec_avx512.c.o -c ../drivers/net/sxe2/sxe2_txrx_vec_avx512.c
../drivers/net/sxe2/sxe2_txrx_vec_avx512.c: In function ‘sxe2_tx_bufs_free_vec_avx512’:
../drivers/net/sxe2/sxe2_txrx_vec_avx512.c:47:1: error: label ‘normal’ defined but not used [-Werror=unused-label]
 normal:
 ^~~~~~
cc1: all warnings being treated as errors
[2434/3759] Generating drivers/rte_net_sxe2_ma

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-08-21 18:03 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-08-20 15:50 [dpdk-dev v1] net/sxe2: replace private mempool cache bypass with rte_mbuf_raw_free_bulk Kai Ji
2026-08-21 18:03 ` Stephen Hemminger

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox