This commit improves handling of packets where the allocated memory
is less than 64 bytes. In the DPDK datapath this never matters, as
an mbuf always pre-allocates enough space, however this can occur in
test environments such as the dummy netdev. The fix is required to
ensure ASAN enabled builds don't error on testing this, hence the
fix is valuable.
The solution implemented uses a mask-to-zero if the available buffer
size is less than 64 bytes, and a branch for which type of load is used.
Fixes: 250ceddcc2d0 ("dpif-netdev/mfex: Add AVX512 based optimized miniflow
extract")
Reported-by: Ilya Maximets <[email protected]>
Signed-off-by: Harry van Haaren <[email protected]>
---
lib/dpif-netdev-extract-avx512.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/lib/dpif-netdev-extract-avx512.c b/lib/dpif-netdev-extract-avx512.c
index e060ab14a..d23349482 100644
--- a/lib/dpif-netdev-extract-avx512.c
+++ b/lib/dpif-netdev-extract-avx512.c
@@ -488,7 +488,14 @@ mfex_avx512_process(struct dp_packet_batch *packets,
/* Load packet data and probe with AVX512 mask & compare. */
const uint8_t *pkt = dp_packet_data(packet);
- __m512i v_pkt0 = _mm512_loadu_si512(pkt);
+ __m512i v_pkt0;
+ if (size >= 64) {
+ v_pkt0 = _mm512_loadu_si512(pkt);
+ } else {
+ uint64_t load_kmask = (1ULL << size) - 1;
+ v_pkt0 = _mm512_maskz_loadu_epi8(load_kmask, pkt);
+ }
+
__m512i v_pkt0_masked = _mm512_and_si512(v_pkt0, v_mask);
__mmask64 k_cmp = _mm512_cmpeq_epi8_mask(v_pkt0_masked, v_vals);
if (k_cmp != UINT64_MAX) {
--
2.25.1
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
