[PATCH v2 4/4] mempool perf test: test random bulk sizes
Bulk requests to get or put objects in a mempool often vary in size.
A series of tests with pseudo random request sizes, to mitigate the
benefits of the CPU's dynamic branch predictor, was added.
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 102 ---
1 file changed, 95 insertions(+), 7 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index c252f1968a..40ecaa5049 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -33,6 +33,13 @@
* Mempool performance
* ===
*
+ *Each core get *n_keep* objects per bulk of a pseudorandom number
+ *between 1 and *n_max_bulk*.
+ *Objects are put back in the pool per bulk of a similar pseudorandom
number.
+ *Note: The very low entropy of the randomization algorithm is harmless,
because
+ * the sole purpose of randomization is to prevent the CPU's dynamic
branch
+ * predictor from enhancing the test results.
+ *
*Each core get *n_keep* objects per bulk of *n_get_bulk*. Then,
*objects are put back in the pool per bulk of *n_put_bulk*.
*
@@ -52,7 +59,12 @@
* - Two cores with user-owned cache
* - Max. cores with user-owned cache
*
- *- Bulk size (*n_get_bulk*, *n_put_bulk*)
+ *- Pseudorandom max bulk size (*n_max_bulk*)
+ *
+ * - Max bulk from CACHE_LINE_BURST to 256, and
RTE_MEMPOOL_CACHE_MAX_SIZE,
+ *where CACHE_LINE_BURST is the number of pointers fitting into one
CPU cache line.
+ *
+ *- Fixed bulk size (*n_get_bulk*, *n_put_bulk*)
*
* - Bulk get from 1 to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE
* - Bulk put from 1 to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE
@@ -93,6 +105,9 @@ static unsigned int external_cache_size =
RTE_MEMPOOL_CACHE_MAX_SIZE;
static RTE_ATOMIC(uint32_t) synchro;
+/* max random number of objects in one bulk operation (get and put) */
+static unsigned int n_max_bulk;
+
/* number of objects in one bulk operation (get or put) */
static unsigned int n_get_bulk;
static unsigned int n_put_bulk;
@@ -159,6 +174,50 @@ test_loop(struct rte_mempool *mp, struct rte_mempool_cache
*cache,
return 0;
}
+static __rte_always_inline int
+test_loop_random(struct rte_mempool *mp, struct rte_mempool_cache *cache,
+ unsigned int x_keep, unsigned int x_max_bulk)
+{
+ alignas(RTE_CACHE_LINE_SIZE) void *obj_table[MAX_KEEP];
+ unsigned int idx;
+ unsigned int i;
+ unsigned int r = 0;
+ unsigned int x_bulk;
+ int ret;
+
+ for (i = 0; likely(i < (N / x_keep)); i++) {
+ /* get x_keep objects by bulk of random [1 .. x_max_bulk] */
+ for (idx = 0; idx < x_keep; idx += x_bulk, r++) {
+ /* Generate a pseudorandom number [1 .. x_max_bulk]. */
+ x_bulk = ((r ^ (r >> 2) ^ (r << 3)) & (x_max_bulk - 1))
+ 1;
+ if (unlikely(idx + x_bulk > x_keep))
+ x_bulk = x_keep - idx;
+ ret = rte_mempool_generic_get(mp,
+ &obj_table[idx],
+ x_bulk,
+ cache);
+ if (unlikely(ret < 0)) {
+ rte_mempool_dump(stdout, mp);
+ return ret;
+ }
+ }
+
+ /* put the objects back by bulk of random [1 .. x_max_bulk] */
+ for (idx = 0; idx < x_keep; idx += x_bulk, r++) {
+ /* Generate a pseudorandom number [1 .. x_max_bulk]. */
+ x_bulk = ((r ^ (r >> 2) ^ (r << 3)) & (x_max_bulk - 1))
+ 1;
+ if (unlikely(idx + x_bulk > x_keep))
+ x_bulk = x_keep - idx;
+ rte_mempool_generic_put(mp,
+ &obj_table[idx],
+ x_bulk,
+ cache);
+ }
+ }
+
+ return 0;
+}
+
static int
per_lcore_mempool_test(void *arg)
{
@@ -181,9 +240,9 @@ per_lcore_mempool_test(void *arg)
}
/* n_get_bulk and n_put_bulk must be divisors of n_keep */
- if (((n_keep / n_get_bulk) * n_get_bulk) != n_keep)
+ if (n_max_bulk == 0 && (((n_keep / n_get_bulk) * n_get_bulk) != n_keep))
GOTO_ERR(ret, out);
- if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)
+ if (n_max_bulk == 0 && (((n_keep / n_put_bulk) * n_put_bulk) != n_keep))
GOTO_ERR(ret, out);
/* for constant n, n_get_bulk and n_put_bulk must be the same */
if (use_constant_values && n_put_bulk != n_get_bulk)
@@ -200,7 +259,9 @@ per_lcore_mempool_test(void *arg)
start_cycles = rte_get_
[PATCH v3 0/4] mempool perf test: test random bulk sizes
Bulk requests to get or put objects in a mempool often vary in size. A series of tests with pseudo random request sizes, to mitigate the benefits of the CPU's dynamic branch predictor, was added. Also, various other minor changes: - Improved the output formatting for readability. - Added test for the "default" mempool with cache. - Skip the tests for the "default" mempool, if it happens to use the same driver (i.e. operations) as already tested. - Replaced bare use of "unsigned" with "unsigned int", to make checkpatches happy. v3: * Fix code style for checking variable is set or zero. (Andrew Rybchenko) v2: * Split patch into individual patches. (Andrew Rybchenko) Morten Brørup (4): mempool perf test: replace bare unsigned with unsigned int mempool perf test: test default mempool with cache mempool perf test: improve output readability mempool perf test: test random bulk sizes app/test/test_mempool_perf.c | 222 +++ 1 file changed, 173 insertions(+), 49 deletions(-) -- 2.43.0
[PATCH v2 1/4] mempool perf test: replace bare unsigned with unsigned int
Updated old code using bare "unsigned" with "unsigned int".
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 24
1 file changed, 12 insertions(+), 12 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index 4dd74ef75a..d4271a5ef9 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -89,16 +89,16 @@
} while (0)
static int use_external_cache;
-static unsigned external_cache_size = RTE_MEMPOOL_CACHE_MAX_SIZE;
+static unsigned int external_cache_size = RTE_MEMPOOL_CACHE_MAX_SIZE;
static RTE_ATOMIC(uint32_t) synchro;
/* number of objects in one bulk operation (get or put) */
-static unsigned n_get_bulk;
-static unsigned n_put_bulk;
+static unsigned int n_get_bulk;
+static unsigned int n_put_bulk;
/* number of objects retrieved from mempool before putting them back */
-static unsigned n_keep;
+static unsigned int n_keep;
/* true if we want to test with constant n_get_bulk and n_put_bulk */
static int use_constant_values;
@@ -118,7 +118,7 @@ static struct mempool_test_stats stats[RTE_MAX_LCORE];
*/
static void
my_obj_init(struct rte_mempool *mp, __rte_unused void *arg,
- void *obj, unsigned i)
+ void *obj, unsigned int i)
{
uint32_t *objnum = obj;
memset(obj, 0, mp->elt_size);
@@ -163,7 +163,7 @@ static int
per_lcore_mempool_test(void *arg)
{
struct rte_mempool *mp = arg;
- unsigned lcore_id = rte_lcore_id();
+ unsigned int lcore_id = rte_lcore_id();
int ret = 0;
uint64_t start_cycles, end_cycles;
uint64_t time_diff = 0, hz = rte_get_timer_hz();
@@ -246,10 +246,10 @@ per_lcore_mempool_test(void *arg)
static int
launch_cores(struct rte_mempool *mp, unsigned int cores)
{
- unsigned lcore_id;
+ unsigned int lcore_id;
uint64_t rate;
int ret;
- unsigned cores_save = cores;
+ unsigned int cores_save = cores;
double hz = rte_get_timer_hz();
rte_atomic_store_explicit(&synchro, 0, rte_memory_order_relaxed);
@@ -260,7 +260,7 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u "
"n_put_bulk=%u n_keep=%u constant_n=%u ",
use_external_cache ?
- external_cache_size : (unsigned) mp->cache_size,
+ external_cache_size : (unsigned int) mp->cache_size,
cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);
if (rte_mempool_avail_count(mp) != MEMPOOL_SIZE) {
@@ -315,9 +315,9 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int
cores, int external_cac
unsigned int bulk_tab_put[] = { 1, 4, CACHE_LINE_BURST, 32, 64, 128,
256,
RTE_MEMPOOL_CACHE_MAX_SIZE, 0 };
unsigned int keep_tab[] = { 32, 128, 512, 2048, 8192, 32768, 0 };
- unsigned *get_bulk_ptr;
- unsigned *put_bulk_ptr;
- unsigned *keep_ptr;
+ unsigned int *get_bulk_ptr;
+ unsigned int *put_bulk_ptr;
+ unsigned int *keep_ptr;
int ret;
for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
--
2.43.0
[PATCH v3 4/4] mempool perf test: test random bulk sizes
Bulk requests to get or put objects in a mempool often vary in size.
A series of tests with pseudo random request sizes, to mitigate the
benefits of the CPU's dynamic branch predictor, was added.
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 102 ---
1 file changed, 95 insertions(+), 7 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index c252f1968a..40ecaa5049 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -33,6 +33,13 @@
* Mempool performance
* ===
*
+ *Each core get *n_keep* objects per bulk of a pseudorandom number
+ *between 1 and *n_max_bulk*.
+ *Objects are put back in the pool per bulk of a similar pseudorandom
number.
+ *Note: The very low entropy of the randomization algorithm is harmless,
because
+ * the sole purpose of randomization is to prevent the CPU's dynamic
branch
+ * predictor from enhancing the test results.
+ *
*Each core get *n_keep* objects per bulk of *n_get_bulk*. Then,
*objects are put back in the pool per bulk of *n_put_bulk*.
*
@@ -52,7 +59,12 @@
* - Two cores with user-owned cache
* - Max. cores with user-owned cache
*
- *- Bulk size (*n_get_bulk*, *n_put_bulk*)
+ *- Pseudorandom max bulk size (*n_max_bulk*)
+ *
+ * - Max bulk from CACHE_LINE_BURST to 256, and
RTE_MEMPOOL_CACHE_MAX_SIZE,
+ *where CACHE_LINE_BURST is the number of pointers fitting into one
CPU cache line.
+ *
+ *- Fixed bulk size (*n_get_bulk*, *n_put_bulk*)
*
* - Bulk get from 1 to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE
* - Bulk put from 1 to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE
@@ -93,6 +105,9 @@ static unsigned int external_cache_size =
RTE_MEMPOOL_CACHE_MAX_SIZE;
static RTE_ATOMIC(uint32_t) synchro;
+/* max random number of objects in one bulk operation (get and put) */
+static unsigned int n_max_bulk;
+
/* number of objects in one bulk operation (get or put) */
static unsigned int n_get_bulk;
static unsigned int n_put_bulk;
@@ -159,6 +174,50 @@ test_loop(struct rte_mempool *mp, struct rte_mempool_cache
*cache,
return 0;
}
+static __rte_always_inline int
+test_loop_random(struct rte_mempool *mp, struct rte_mempool_cache *cache,
+ unsigned int x_keep, unsigned int x_max_bulk)
+{
+ alignas(RTE_CACHE_LINE_SIZE) void *obj_table[MAX_KEEP];
+ unsigned int idx;
+ unsigned int i;
+ unsigned int r = 0;
+ unsigned int x_bulk;
+ int ret;
+
+ for (i = 0; likely(i < (N / x_keep)); i++) {
+ /* get x_keep objects by bulk of random [1 .. x_max_bulk] */
+ for (idx = 0; idx < x_keep; idx += x_bulk, r++) {
+ /* Generate a pseudorandom number [1 .. x_max_bulk]. */
+ x_bulk = ((r ^ (r >> 2) ^ (r << 3)) & (x_max_bulk - 1))
+ 1;
+ if (unlikely(idx + x_bulk > x_keep))
+ x_bulk = x_keep - idx;
+ ret = rte_mempool_generic_get(mp,
+ &obj_table[idx],
+ x_bulk,
+ cache);
+ if (unlikely(ret < 0)) {
+ rte_mempool_dump(stdout, mp);
+ return ret;
+ }
+ }
+
+ /* put the objects back by bulk of random [1 .. x_max_bulk] */
+ for (idx = 0; idx < x_keep; idx += x_bulk, r++) {
+ /* Generate a pseudorandom number [1 .. x_max_bulk]. */
+ x_bulk = ((r ^ (r >> 2) ^ (r << 3)) & (x_max_bulk - 1))
+ 1;
+ if (unlikely(idx + x_bulk > x_keep))
+ x_bulk = x_keep - idx;
+ rte_mempool_generic_put(mp,
+ &obj_table[idx],
+ x_bulk,
+ cache);
+ }
+ }
+
+ return 0;
+}
+
static int
per_lcore_mempool_test(void *arg)
{
@@ -181,9 +240,9 @@ per_lcore_mempool_test(void *arg)
}
/* n_get_bulk and n_put_bulk must be divisors of n_keep */
- if (((n_keep / n_get_bulk) * n_get_bulk) != n_keep)
+ if (n_max_bulk == 0 && (((n_keep / n_get_bulk) * n_get_bulk) != n_keep))
GOTO_ERR(ret, out);
- if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)
+ if (n_max_bulk == 0 && (((n_keep / n_put_bulk) * n_put_bulk) != n_keep))
GOTO_ERR(ret, out);
/* for constant n, n_get_bulk and n_put_bulk must be the same */
if (use_constant_values && n_put_bulk != n_get_bulk)
@@ -200,7 +259,9 @@ per_lcore_mempool_test(void *arg)
start_cycles = rte_get_
[PATCH v3 3/4] mempool perf test: improve output readability
Improved the output parameter ordering and formatting for readability.
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 20 +++-
1 file changed, 11 insertions(+), 9 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index 3594d81888..c252f1968a 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -257,11 +257,13 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
/* reset stats */
memset(stats, 0, sizeof(stats));
- printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u "
- "n_put_bulk=%u n_keep=%u constant_n=%u ",
- use_external_cache ?
- external_cache_size : (unsigned int) mp->cache_size,
- cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);
+ printf("mempool_autotest cache=%u cores=%u n_keep=%5u ",
+ use_external_cache ? external_cache_size : (unsigned int)
mp->cache_size,
+ cores,
+ n_keep);
+ printf("n_get_bulk=%3u n_put_bulk=%3u constant_n=%u ",
+ n_get_bulk, n_put_bulk,
+ use_constant_values);
if (rte_mempool_avail_count(mp) != MEMPOOL_SIZE) {
printf("mempool is not full\n");
@@ -301,7 +303,7 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
rate += (double)stats[lcore_id].enq_count * hz /
(double)stats[lcore_id].duration_cycles;
- printf("rate_persec=%" PRIu64 "\n", rate);
+ printf("rate_persec=%10" PRIu64 "\n", rate);
return 0;
}
@@ -320,9 +322,9 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int
cores, int external_cac
unsigned int *keep_ptr;
int ret;
- for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
- for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr;
put_bulk_ptr++) {
- for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
+ for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
+ for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr;
get_bulk_ptr++) {
+ for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr;
put_bulk_ptr++) {
if (*keep_ptr < *get_bulk_ptr || *keep_ptr <
*put_bulk_ptr)
continue;
--
2.43.0
[PATCH v2 3/4] mempool perf test: improve output readability
Improved the output parameter ordering and formatting for readability.
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 20 +++-
1 file changed, 11 insertions(+), 9 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index 3594d81888..c252f1968a 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -257,11 +257,13 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
/* reset stats */
memset(stats, 0, sizeof(stats));
- printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u "
- "n_put_bulk=%u n_keep=%u constant_n=%u ",
- use_external_cache ?
- external_cache_size : (unsigned int) mp->cache_size,
- cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);
+ printf("mempool_autotest cache=%u cores=%u n_keep=%5u ",
+ use_external_cache ? external_cache_size : (unsigned int)
mp->cache_size,
+ cores,
+ n_keep);
+ printf("n_get_bulk=%3u n_put_bulk=%3u constant_n=%u ",
+ n_get_bulk, n_put_bulk,
+ use_constant_values);
if (rte_mempool_avail_count(mp) != MEMPOOL_SIZE) {
printf("mempool is not full\n");
@@ -301,7 +303,7 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
rate += (double)stats[lcore_id].enq_count * hz /
(double)stats[lcore_id].duration_cycles;
- printf("rate_persec=%" PRIu64 "\n", rate);
+ printf("rate_persec=%10" PRIu64 "\n", rate);
return 0;
}
@@ -320,9 +322,9 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int
cores, int external_cac
unsigned int *keep_ptr;
int ret;
- for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
- for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr;
put_bulk_ptr++) {
- for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
+ for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
+ for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr;
get_bulk_ptr++) {
+ for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr;
put_bulk_ptr++) {
if (*keep_ptr < *get_bulk_ptr || *keep_ptr <
*put_bulk_ptr)
continue;
--
2.43.0
[DPDK/cryptodev Bug 1686] [dpdk-22.11.8RC1] unit_tests_cryptodev_func/cryptodev_aesni_gcm_autotest: test failing-Segmentation fault
https://bugs.dpdk.org/show_bug.cgi?id=1686 Luca Boccassi (luca.bocca...@gmail.com) changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED CC||luca.bocca...@gmail.com Resolution|--- |FIXED --- Comment #1 from Luca Boccassi (luca.bocca...@gmail.com) --- Fixed in stable/22.11 -- You are receiving this mail because: You are the assignee for the bug.
Re: [PATCH] examples/flow_filtering: fix make clean
On Mon, 31 Mar 2025 at 13:45, Thomas Monjalon: > > 29/03/2025 20:53, Tanzeel Ahmed: > > make clean is unable to delete build directory > > because *.o files are not removed. Removing .o files is a simpler fix, as in the patch. > > The other way to fix this would be to add all the > > c files into SRCS-y. > > How adding all C files into SRCS-y would help? We can bypass the obj files generation and use the current clean as it is
RE: [PATCH] common/cnxk: add flag for enabling opaque mode
> -Original Message-
> From: Nawal Kishor
> Sent: Tuesday, March 4, 2025 10:55 AM
> To: dev@dpdk.org; Nithin Kumar Dabilpuram ;
> Kiran Kumar Kokkilagadda ; Sunil Kumar Kori
> ; Satha Koteswara Rao Kottidi
> ; Harman Kalra
> Cc: Jerin Jacob ; Ashwin Sekhar T K
> ; Nawal Kishor
> Subject: [PATCH] common/cnxk: add flag for enabling opaque mode
>
> Added flag that overrides the default natural alignment mode and uses opaque
> mode.
>
> Signed-off-by: Nawal Kishor
Updated the git commit as follows and applied to dpdk-next-net-mrvl/for-main.
Thanks
common/cnxk: support enabling opaque mode
Added flag that overrides the default natural alignment mode and uses
opaque mode.
Signed-off-by: Nawal Kishor
> ---
> drivers/common/cnxk/roc_npa.c | 3 +++
> drivers/common/cnxk/roc_npa.h | 3 ++-
> 2 files changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/common/cnxk/roc_npa.c b/drivers/common/cnxk/roc_npa.c
> index a33f9a8499..d5ebfbfc11 100644
> --- a/drivers/common/cnxk/roc_npa.c
> +++ b/drivers/common/cnxk/roc_npa.c
> @@ -669,6 +669,9 @@ roc_npa_pool_create(uint64_t *aura_handle, uint32_t
> block_size,
> pool = &defpool;
> }
>
> + if (flags & ROC_NPA_FORCE_OPAQUE_MODE_F)
> + pool->nat_align = 0;
> +
> rc = npa_aura_pool_pair_alloc(lf, block_size, block_count, aura, pool,
> aura_handle, flags);
> if (rc) {
> diff --git a/drivers/common/cnxk/roc_npa.h b/drivers/common/cnxk/roc_npa.h
> index 8525038810..853c0fed43 100644
> --- a/drivers/common/cnxk/roc_npa.h
> +++ b/drivers/common/cnxk/roc_npa.h
> @@ -767,7 +767,8 @@ int __roc_api roc_npa_dev_init(struct roc_npa
> *roc_npa); int __roc_api roc_npa_dev_fini(struct roc_npa *roc_npa);
>
> /* Flags to pool create */
> -#define ROC_NPA_ZERO_AURA_F BIT(0)
> +#define ROC_NPA_ZERO_AURA_F BIT(0)
> +#define ROC_NPA_FORCE_OPAQUE_MODE_F BIT(1)
>
> /* Enumerations */
> enum roc_npa_buf_type {
> --
> 2.34.1
[PATCH 1/2] node: add global node mbuf dynfield
This patch defines rte_node specific dynamic field structure
(rte_node_mbuf_dynfield_t)
rte_node_mbuf_dynfield_t structure holds two types of fields
- Persistent data fields which are preserved across graph walk.
Currently size of persistent data fields is zero.
- Overloadable data fields which are used by any two adjacent nodes.
Same fields can be repurposed by any other adjacent nodes
This dynfield can be also be used by out-of-tree nodes.
Signed-off-by: Nitin Saxena
---
doc/api/doxy-api-index.md | 3 +-
doc/guides/rel_notes/release_25_07.rst | 6 ++
lib/node/meson.build | 2 +
lib/node/node_mbuf_dynfield.c | 56 +++
lib/node/rte_node_mbuf_dynfield.h | 130 +
lib/node/version.map | 3 +
6 files changed, 199 insertions(+), 1 deletion(-)
create mode 100644 lib/node/node_mbuf_dynfield.c
create mode 100644 lib/node/rte_node_mbuf_dynfield.h
diff --git a/doc/api/doxy-api-index.md b/doc/api/doxy-api-index.md
index b2fc24b3e4..6b93b3cd97 100644
--- a/doc/api/doxy-api-index.md
+++ b/doc/api/doxy-api-index.md
@@ -219,7 +219,8 @@ The public API headers are grouped by topics:
[ip4_node](@ref rte_node_ip4_api.h),
[ip6_node](@ref rte_node_ip6_api.h),
[udp4_input_node](@ref rte_node_udp4_input_api.h)
-
+ * graph_nodes_mbuf:
+[node_mbuf_dynfield](@ref rte_node_mbuf_dynfield.h)
- **basic**:
[bitops](@ref rte_bitops.h),
[approx fraction](@ref rte_approx.h),
diff --git a/doc/guides/rel_notes/release_25_07.rst
b/doc/guides/rel_notes/release_25_07.rst
index cd1025aac0..72d5d88ff3 100644
--- a/doc/guides/rel_notes/release_25_07.rst
+++ b/doc/guides/rel_notes/release_25_07.rst
@@ -55,6 +55,12 @@ New Features
Also, make sure to start the actual text at the margin.
===
+* **Added rte_node specific global mbuf dynamic field.**
+
+ Instead each node registering mbuf dynamic field for its own purpose, a
+ global structure is added which can be used/overloaded by all nodes
+ (including out-of-tree nodes). This minimizes footprint of node specific mbuf
+ dynamic field.
Removed Items
-
diff --git a/lib/node/meson.build b/lib/node/meson.build
index 0bed97a96c..4330d0450b 100644
--- a/lib/node/meson.build
+++ b/lib/node/meson.build
@@ -8,6 +8,7 @@ if is_windows
endif
sources = files(
+'node_mbuf_dynfield.c',
'ethdev_ctrl.c',
'ethdev_rx.c',
'ethdev_tx.c',
@@ -30,6 +31,7 @@ headers = files(
'rte_node_ip4_api.h',
'rte_node_ip6_api.h',
'rte_node_udp4_input_api.h',
+'rte_node_mbuf_dynfield.h'
)
# Strict-aliasing rules are violated by uint8_t[] to context size casts.
diff --git a/lib/node/node_mbuf_dynfield.c b/lib/node/node_mbuf_dynfield.c
new file mode 100644
index 00..e42631f432
--- /dev/null
+++ b/lib/node/node_mbuf_dynfield.c
@@ -0,0 +1,56 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(C) 2025 Marvell International Ltd.
+ */
+#include
+#include
+#include
+#include
+#include
+
+#define NODE_MBUF_DYNFIELD_MEMZONE_NAME "__rte_node_mbuf_dynfield"
+
+struct node_mbuf_dynfield_mz {
+ int dynfield_offset;
+};
+
+static const struct rte_mbuf_dynfield node_mbuf_dynfield_desc = {
+ .name = "rte_node_mbuf_dynfield",
+ .size = sizeof(rte_node_mbuf_dynfield_t),
+ .align = alignof(rte_node_mbuf_dynfield_t),
+};
+
+int rte_node_mbuf_dynfield_register(void)
+{
+ struct node_mbuf_dynfield_mz *f = NULL;
+ const struct rte_memzone *mz = NULL;
+ int dyn_offset;
+
+ RTE_BUILD_BUG_ON(sizeof(rte_node_mbuf_dynfield_t) <
RTE_NODE_MBUF_DYNFIELD_SIZE);
+ RTE_BUILD_BUG_ON(sizeof(rte_node_mbuf_overload_fields_t) <
+RTE_NODE_MBUF_OVERLOADABLE_FIELDS_SIZE);
+
+ mz = rte_memzone_lookup(NODE_MBUF_DYNFIELD_MEMZONE_NAME);
+
+ if (!mz) {
+ mz = rte_memzone_reserve(NODE_MBUF_DYNFIELD_MEMZONE_NAME,
+sizeof(struct node_mbuf_dynfield_mz),
+SOCKET_ID_ANY, 0);
+ if (!mz) {
+ node_err("node_mbuf_dyn", "memzone reserve failed for
node mbuf field");
+ return -1;
+ }
+ dyn_offset =
rte_mbuf_dynfield_register(&node_mbuf_dynfield_desc);
+ if (dyn_offset < 0) {
+ node_err("node_mbuf_dyn", "rte_mbuf_dynfield_register
failed");
+ return -1;
+ }
+ f = (struct node_mbuf_dynfield_mz *)mz->addr;
+ f->dynfield_offset = dyn_offset;
+
+ node_dbg("node_mbuf_dyn", "memzone: %s of size 0x%zx at offset:
%d",
+mz->name, mz->len, f->dynfield_offset);
+ } else {
+ f = (struct node_mbuf_dynfield_mz *)mz->addr;
+ }
+ return f->dynfield_offset;
[PATCH 0/2] node: add mbuf dynamic field for nodes
Currently each rte_node registers separate mbuf dynamic fields for their own purpose. This leads to wastage of mbuf space as once mbuf get passed a particular node, the registered dynamic field(by that node) is no longer used. This patch series adds a global/common mbuf dynamic field which is reusable by all the nodes(including out-of-tree nodes). This helps to repurpose same mbuf dynamic field for other nodes. It contains two types of fields: (a) persistent (b) overloadable. While persistent fields are those which does not often changes during a graph walk such as rx/tx interface, buffer flags etc. Currently there are no persistent fields added but they can be added later Overloadable fields are those which can be overloaded by two adjacent nodes. Overloadable fields can be repurposed by other two adjacent nodes. This patch series also updates ip4/ip6 lookup/rewrite nodes to use overlaodable mbuf dynamic fields. Nitin Saxena (2): node: add global node mbuf dynfield node: use node mbuf dynfield in ip4 nodes doc/api/doxy-api-index.md | 3 +- doc/guides/rel_notes/release_25_07.rst | 6 ++ lib/node/ip4_lookup.c | 14 +-- lib/node/ip4_rewrite.c | 15 ++- lib/node/ip6_lookup.c | 15 ++- lib/node/ip6_rewrite.c | 14 +-- lib/node/meson.build | 2 + lib/node/node_mbuf_dynfield.c | 56 ++ lib/node/node_private.h| 40 +-- lib/node/rte_node_mbuf_dynfield.h | 139 + lib/node/version.map | 3 + 11 files changed, 233 insertions(+), 74 deletions(-) create mode 100644 lib/node/node_mbuf_dynfield.c create mode 100644 lib/node/rte_node_mbuf_dynfield.h -- 2.43.0
[PATCH] net/nfp: fix one coredump caused by RSS hash key
Some testpmd application commands will use 'rss_conf->rss_key'
parameter with a NULL value to call 'rte_eth_dev_rss_hash_conf_get()'
API, and NFP PMD will coredump at these situations.
Fix this by add a check about the 'rss_conf->rss_key' pointer.
Fixes: 934e4c60fbff ("nfp: add RSS")
Cc: sta...@dpdk.org
Signed-off-by: Chaoyong He
Reviewed-by: Long Wu
Reviewed-by: Peng Zhang
---
drivers/net/nfp/nfp_net_common.c | 8 +---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/net/nfp/nfp_net_common.c b/drivers/net/nfp/nfp_net_common.c
index aaa515bac2..600fcd5c0e 100644
--- a/drivers/net/nfp/nfp_net_common.c
+++ b/drivers/net/nfp/nfp_net_common.c
@@ -2062,9 +2062,11 @@ nfp_net_rss_hash_conf_get(struct rte_eth_dev *dev,
rss_conf->rss_key_len = nn_cfg_readl(hw, NFP_NET_CFG_RSS_KEY_SZ);
/* Reading the key byte a byte */
- for (i = 0; i < rss_conf->rss_key_len; i++) {
- key = nn_cfg_readb(hw, NFP_NET_CFG_RSS_KEY + i);
- memcpy(&rss_conf->rss_key[i], &key, 1);
+ if (rss_conf->rss_key != NULL) {
+ for (i = 0; i < rss_conf->rss_key_len; i++) {
+ key = nn_cfg_readb(hw, NFP_NET_CFG_RSS_KEY + i);
+ memcpy(&rss_conf->rss_key[i], &key, 1);
+ }
}
return 0;
--
2.43.5
Re: DPDK for rust
Hello, I pulled the raw library bindings out of DPDK source into a dedicated crate and cleaned up compilation warnings. Also I started a testpmd like application - runpmd. At this stage it can receive buffers and insert flows with very limited items and actions coverage. The total patches are big - over 69000 lines. Is there another way to share these patches, except the mailing list ? Regards, Gregory
Re: [PATCH 02/10] common/sfc_efx: use common base code build handling
On 3/31/25 19:09, Bruce Richardson wrote: Use the base code build handling logic in the drivers/meson.build file, rather than re-implementing it in the driver itself. Signed-off-by: Bruce Richardson Acked-by: Andrew Rybchenko
Re: [PATCH v6 4/8] build: generate symbol maps
28/03/2025 11:52, David Marchand:
> +scriptname, link_mode, abi_version_file, output, *files = sys.argv
You should add a comment to show the usage of the script.
[...]
> +if link_mode == 'mslinker':
> +with open(output, "w") as outfile:
> +print(f"EXPORTS", file=outfile)
> +for key in (abi, 'EXPERIMENTAL', 'INTERNAL'):
> +if key not in symbols:
> +continue
> +for symbol in sorted(symbols[key].keys()):
> +print(f"\t{symbol}", file=outfile)
> +del symbols[key]
> +else:
Maybe add a comment after the else to make explicit it is GNU format.
[...]
> +for file in sys.argv[1:]:
This script requires a comment with the usage as well
to make clear what we expect as arguments.
> +with open(file, encoding="utf-8") as f:
> +for ln in f.readlines():
> +if file_header_regexp.match(ln):
> +if file_header_regexp.match(ln).group(2) == "lib":
> +lib = '/'.join(file_header_regexp.match(ln).group(2, 3))
> +elif file_header_regexp.match(ln).group(3) == "intel":
> +lib = '/'.join(file_header_regexp.match(ln).group(2, 3,
> 4))
> +else:
> +lib = '/'.join(file_header_regexp.match(ln).group(2, 3))
[...]
> +#define RTE_EXPORT_EXPERIMENTAL_SYMBOL(a, ver)
> +#define RTE_EXPORT_INTERNAL_SYMBOL(a)
> +#define RTE_EXPORT_SYMBOL(a)
I would prefer an explicit variable name instead of "a".
[PATCH v3 1/4] mempool perf test: replace bare unsigned with unsigned int
Updated old code using bare "unsigned" with "unsigned int".
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 24
1 file changed, 12 insertions(+), 12 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index 4dd74ef75a..d4271a5ef9 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -89,16 +89,16 @@
} while (0)
static int use_external_cache;
-static unsigned external_cache_size = RTE_MEMPOOL_CACHE_MAX_SIZE;
+static unsigned int external_cache_size = RTE_MEMPOOL_CACHE_MAX_SIZE;
static RTE_ATOMIC(uint32_t) synchro;
/* number of objects in one bulk operation (get or put) */
-static unsigned n_get_bulk;
-static unsigned n_put_bulk;
+static unsigned int n_get_bulk;
+static unsigned int n_put_bulk;
/* number of objects retrieved from mempool before putting them back */
-static unsigned n_keep;
+static unsigned int n_keep;
/* true if we want to test with constant n_get_bulk and n_put_bulk */
static int use_constant_values;
@@ -118,7 +118,7 @@ static struct mempool_test_stats stats[RTE_MAX_LCORE];
*/
static void
my_obj_init(struct rte_mempool *mp, __rte_unused void *arg,
- void *obj, unsigned i)
+ void *obj, unsigned int i)
{
uint32_t *objnum = obj;
memset(obj, 0, mp->elt_size);
@@ -163,7 +163,7 @@ static int
per_lcore_mempool_test(void *arg)
{
struct rte_mempool *mp = arg;
- unsigned lcore_id = rte_lcore_id();
+ unsigned int lcore_id = rte_lcore_id();
int ret = 0;
uint64_t start_cycles, end_cycles;
uint64_t time_diff = 0, hz = rte_get_timer_hz();
@@ -246,10 +246,10 @@ per_lcore_mempool_test(void *arg)
static int
launch_cores(struct rte_mempool *mp, unsigned int cores)
{
- unsigned lcore_id;
+ unsigned int lcore_id;
uint64_t rate;
int ret;
- unsigned cores_save = cores;
+ unsigned int cores_save = cores;
double hz = rte_get_timer_hz();
rte_atomic_store_explicit(&synchro, 0, rte_memory_order_relaxed);
@@ -260,7 +260,7 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u "
"n_put_bulk=%u n_keep=%u constant_n=%u ",
use_external_cache ?
- external_cache_size : (unsigned) mp->cache_size,
+ external_cache_size : (unsigned int) mp->cache_size,
cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);
if (rte_mempool_avail_count(mp) != MEMPOOL_SIZE) {
@@ -315,9 +315,9 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int
cores, int external_cac
unsigned int bulk_tab_put[] = { 1, 4, CACHE_LINE_BURST, 32, 64, 128,
256,
RTE_MEMPOOL_CACHE_MAX_SIZE, 0 };
unsigned int keep_tab[] = { 32, 128, 512, 2048, 8192, 32768, 0 };
- unsigned *get_bulk_ptr;
- unsigned *put_bulk_ptr;
- unsigned *keep_ptr;
+ unsigned int *get_bulk_ptr;
+ unsigned int *put_bulk_ptr;
+ unsigned int *keep_ptr;
int ret;
for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
--
2.43.0
[PATCH v3 2/4] mempool perf test: test default mempool with cache
Added test for the "default" mempool with cache.
Skip the tests for the "default" mempool, if it happens to use the same
driver (i.e. operations) as already tested.
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 84 +---
1 file changed, 59 insertions(+), 25 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index d4271a5ef9..3594d81888 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -354,7 +354,10 @@ do_all_mempool_perf_tests(unsigned int cores)
{
struct rte_mempool *mp_cache = NULL;
struct rte_mempool *mp_nocache = NULL;
- struct rte_mempool *default_pool = NULL;
+ struct rte_mempool *default_pool_cache = NULL;
+ struct rte_mempool *default_pool_nocache = NULL;
+ const char *mp_cache_ops;
+ const char *mp_nocache_ops;
const char *default_pool_ops;
int ret = -1;
@@ -368,6 +371,7 @@ do_all_mempool_perf_tests(unsigned int cores)
printf("cannot allocate mempool (without cache)\n");
goto err;
}
+ mp_nocache_ops = rte_mempool_get_ops(mp_nocache->ops_index)->name;
/* create a mempool (with cache) */
mp_cache = rte_mempool_create("perf_test_cache", MEMPOOL_SIZE,
@@ -380,47 +384,76 @@ do_all_mempool_perf_tests(unsigned int cores)
printf("cannot allocate mempool (with cache)\n");
goto err;
}
+ mp_cache_ops = rte_mempool_get_ops(mp_cache->ops_index)->name;
default_pool_ops = rte_mbuf_best_mempool_ops();
- /* Create a mempool based on Default handler */
- default_pool = rte_mempool_create_empty("default_pool",
- MEMPOOL_SIZE,
- MEMPOOL_ELT_SIZE,
- 0, 0,
- SOCKET_ID_ANY, 0);
-
- if (default_pool == NULL) {
- printf("cannot allocate %s mempool\n", default_pool_ops);
+
+ /* Create a mempool (without cache) based on Default handler */
+ default_pool_nocache = rte_mempool_create_empty("default_pool_nocache",
+ MEMPOOL_SIZE,
+ MEMPOOL_ELT_SIZE,
+ 0, 0,
+ SOCKET_ID_ANY, 0);
+ if (default_pool_nocache == NULL) {
+ printf("cannot allocate %s mempool (without cache)\n",
default_pool_ops);
goto err;
}
-
- if (rte_mempool_set_ops_byname(default_pool, default_pool_ops, NULL)
- < 0) {
+ if (rte_mempool_set_ops_byname(default_pool_nocache, default_pool_ops,
NULL) < 0) {
printf("cannot set %s handler\n", default_pool_ops);
goto err;
}
-
- if (rte_mempool_populate_default(default_pool) < 0) {
+ if (rte_mempool_populate_default(default_pool_nocache) < 0) {
printf("cannot populate %s mempool\n", default_pool_ops);
goto err;
}
+ rte_mempool_obj_iter(default_pool_nocache, my_obj_init, NULL);
+
+ /* Create a mempool (with cache) based on Default handler */
+ default_pool_cache = rte_mempool_create_empty("default_pool_cache",
+ MEMPOOL_SIZE,
+ MEMPOOL_ELT_SIZE,
+ RTE_MEMPOOL_CACHE_MAX_SIZE, 0,
+ SOCKET_ID_ANY, 0);
+ if (default_pool_cache == NULL) {
+ printf("cannot allocate %s mempool (with cache)\n",
default_pool_ops);
+ goto err;
+ }
+ if (rte_mempool_set_ops_byname(default_pool_cache, default_pool_ops,
NULL) < 0) {
+ printf("cannot set %s handler\n", default_pool_ops);
+ goto err;
+ }
+ if (rte_mempool_populate_default(default_pool_cache) < 0) {
+ printf("cannot populate %s mempool\n", default_pool_ops);
+ goto err;
+ }
+ rte_mempool_obj_iter(default_pool_cache, my_obj_init, NULL);
- rte_mempool_obj_iter(default_pool, my_obj_init, NULL);
-
- printf("start performance test (without cache)\n");
+ printf("start performance test (using %s, without cache)\n",
+ mp_nocache_ops);
if (do_one_mempool_test(mp_nocache, cores, 0) < 0)
goto err;
- printf("start performance test for %s (without cache)\n",
- default_pool_ops);
- if (do_one_mempool_test(default_pool, cores, 0) < 0)
- goto err;
+ if (strcmp(default_pool_ops, mp_nocache_ops) != 0) {
+ printf("start performance test for %s (without cache)\n",
+ default_pool_ops);
+ if (do_one_mempool_test(default_pool_nocache, cores, 0) < 0)
+ goto err;
+ }
- printf("s
[PATCH v2 0/4] mempool perf test: test random bulk sizes
Bulk requests to get or put objects in a mempool often vary in size. A series of tests with pseudo random request sizes, to mitigate the benefits of the CPU's dynamic branch predictor, was added. Also, various other minor changes: - Improved the output formatting for readability. - Added test for the "default" mempool with cache. - Skip the tests for the "default" mempool, if it happens to use the same driver (i.e. operations) as already tested. - Replaced bare use of "unsigned" with "unsigned int", to make checkpatches happy. v3: * Fix code style for checking variable is set or zero. (Andrew Rybchenko) v2: * Split patch into individual patches. (Andrew Rybchenko) Morten Brørup (4): mempool perf test: replace bare unsigned with unsigned int mempool perf test: test default mempool with cache mempool perf test: improve output readability mempool perf test: test random bulk sizes app/test/test_mempool_perf.c | 222 +++ 1 file changed, 173 insertions(+), 49 deletions(-) -- 2.43.0
[PATCH v2 2/4] mempool perf test: test default mempool with cache
Added test for the "default" mempool with cache.
Skip the tests for the "default" mempool, if it happens to use the same
driver (i.e. operations) as already tested.
Signed-off-by: Morten Brørup
Acked-by: Andrew Rybchenko
---
app/test/test_mempool_perf.c | 84 +---
1 file changed, 59 insertions(+), 25 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index d4271a5ef9..3594d81888 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -354,7 +354,10 @@ do_all_mempool_perf_tests(unsigned int cores)
{
struct rte_mempool *mp_cache = NULL;
struct rte_mempool *mp_nocache = NULL;
- struct rte_mempool *default_pool = NULL;
+ struct rte_mempool *default_pool_cache = NULL;
+ struct rte_mempool *default_pool_nocache = NULL;
+ const char *mp_cache_ops;
+ const char *mp_nocache_ops;
const char *default_pool_ops;
int ret = -1;
@@ -368,6 +371,7 @@ do_all_mempool_perf_tests(unsigned int cores)
printf("cannot allocate mempool (without cache)\n");
goto err;
}
+ mp_nocache_ops = rte_mempool_get_ops(mp_nocache->ops_index)->name;
/* create a mempool (with cache) */
mp_cache = rte_mempool_create("perf_test_cache", MEMPOOL_SIZE,
@@ -380,47 +384,76 @@ do_all_mempool_perf_tests(unsigned int cores)
printf("cannot allocate mempool (with cache)\n");
goto err;
}
+ mp_cache_ops = rte_mempool_get_ops(mp_cache->ops_index)->name;
default_pool_ops = rte_mbuf_best_mempool_ops();
- /* Create a mempool based on Default handler */
- default_pool = rte_mempool_create_empty("default_pool",
- MEMPOOL_SIZE,
- MEMPOOL_ELT_SIZE,
- 0, 0,
- SOCKET_ID_ANY, 0);
-
- if (default_pool == NULL) {
- printf("cannot allocate %s mempool\n", default_pool_ops);
+
+ /* Create a mempool (without cache) based on Default handler */
+ default_pool_nocache = rte_mempool_create_empty("default_pool_nocache",
+ MEMPOOL_SIZE,
+ MEMPOOL_ELT_SIZE,
+ 0, 0,
+ SOCKET_ID_ANY, 0);
+ if (default_pool_nocache == NULL) {
+ printf("cannot allocate %s mempool (without cache)\n",
default_pool_ops);
goto err;
}
-
- if (rte_mempool_set_ops_byname(default_pool, default_pool_ops, NULL)
- < 0) {
+ if (rte_mempool_set_ops_byname(default_pool_nocache, default_pool_ops,
NULL) < 0) {
printf("cannot set %s handler\n", default_pool_ops);
goto err;
}
-
- if (rte_mempool_populate_default(default_pool) < 0) {
+ if (rte_mempool_populate_default(default_pool_nocache) < 0) {
printf("cannot populate %s mempool\n", default_pool_ops);
goto err;
}
+ rte_mempool_obj_iter(default_pool_nocache, my_obj_init, NULL);
+
+ /* Create a mempool (with cache) based on Default handler */
+ default_pool_cache = rte_mempool_create_empty("default_pool_cache",
+ MEMPOOL_SIZE,
+ MEMPOOL_ELT_SIZE,
+ RTE_MEMPOOL_CACHE_MAX_SIZE, 0,
+ SOCKET_ID_ANY, 0);
+ if (default_pool_cache == NULL) {
+ printf("cannot allocate %s mempool (with cache)\n",
default_pool_ops);
+ goto err;
+ }
+ if (rte_mempool_set_ops_byname(default_pool_cache, default_pool_ops,
NULL) < 0) {
+ printf("cannot set %s handler\n", default_pool_ops);
+ goto err;
+ }
+ if (rte_mempool_populate_default(default_pool_cache) < 0) {
+ printf("cannot populate %s mempool\n", default_pool_ops);
+ goto err;
+ }
+ rte_mempool_obj_iter(default_pool_cache, my_obj_init, NULL);
- rte_mempool_obj_iter(default_pool, my_obj_init, NULL);
-
- printf("start performance test (without cache)\n");
+ printf("start performance test (using %s, without cache)\n",
+ mp_nocache_ops);
if (do_one_mempool_test(mp_nocache, cores, 0) < 0)
goto err;
- printf("start performance test for %s (without cache)\n",
- default_pool_ops);
- if (do_one_mempool_test(default_pool, cores, 0) < 0)
- goto err;
+ if (strcmp(default_pool_ops, mp_nocache_ops) != 0) {
+ printf("start performance test for %s (without cache)\n",
+ default_pool_ops);
+ if (do_one_mempool_test(default_pool_nocache, cores, 0) < 0)
+ goto err;
+ }
- printf("s
[PATCH v2 0/4] null PMD optimizations and fixes
While reviewing use of null device for testing, noticed several things this driver is doing which are unnecessary. v2 - fix spelling and patch gliches add byte count Stephen Hemminger (4): net/null: fix packet copy net/null: Tx optimizations net/null: optimize Rx net/null: count all queues drivers/net/null/rte_eth_null.c | 149 1 file changed, 74 insertions(+), 75 deletions(-) -- 2.47.2
Re: [PATCH] doc: reword contributor's guide for grammar/clarity
On Tue, 1 Apr 2025 16:45:45 -0700 Nandini Persad wrote: > I'm reviewing the Contributor's Guidelines to be more > clear and concise where necessary. > > Signed-off-by: Nandini Persad > --- > doc/guides/contributing/coding_style.rst | 105 +++ > doc/guides/contributing/design.rst | 9 +- > doc/guides/contributing/new_library.rst | 31 --- > 3 files changed, 77 insertions(+), 68 deletions(-) > > diff --git a/doc/guides/contributing/coding_style.rst > b/doc/guides/contributing/coding_style.rst > index 1ebc79ca3c..c465bd0f1f 100644 > --- a/doc/guides/contributing/coding_style.rst > +++ b/doc/guides/contributing/coding_style.rst > @@ -15,14 +15,13 @@ It is based on the Linux Kernel coding guidelines and the > FreeBSD 7.2 Kernel Dev > General Guidelines > -- > > -The rules and guidelines given in this document cannot cover every > situation, so the following general guidelines should be used as a fallback: > +This document's rules and guidelines cannot cover every scenario, so the > following general principles > +should be used as a fallback. > > -* The code style should be consistent within each individual file. > -* In the case of creating new files, the style should be consistent within > each file in a given directory or module. > -* The primary reason for coding standards is to increase code readability > and comprehensibility, therefore always use whatever option will make the > code easiest to read. > - > -Line length is recommended to be not more than 80 characters, including > comments. > -[Tab stop size should be assumed to be 8-characters wide]. > +* Maintain a consistent coding style within each file. > +* When creating new files, ensure consistency with other files in the same > directory or module. > +* Prioritize readability and clarity. Choose the style that makes the code > easiest to read. > +* Keep line length within 80 characters, including comments. Assume a tab > stop size of 8 characters. DPDK has updated to allow up to 100 characters (happened in last year or so).
[PATCH v2 4/4] net/null: count all queues
If RTE_ETHDEV_QUEUE_STAT_CNTRS is less than the number of queues
in a device, the device should still count packets for all queues.
Add byte counters.
Remove the igb_ prefix which was inherited from other driver.
Signed-off-by: Stephen Hemminger
---
drivers/net/null/rte_eth_null.c | 57 -
1 file changed, 27 insertions(+), 30 deletions(-)
diff --git a/drivers/net/null/rte_eth_null.c b/drivers/net/null/rte_eth_null.c
index 40ce5c6ed2..8a9b74a03b 100644
--- a/drivers/net/null/rte_eth_null.c
+++ b/drivers/net/null/rte_eth_null.c
@@ -319,38 +319,39 @@ eth_dev_info(struct rte_eth_dev *dev,
}
static int
-eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
+eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
{
- unsigned int i, num_stats;
- unsigned long rx_total = 0, tx_total = 0;
- const struct pmd_internals *internal;
+ const struct pmd_internals *internal = dev->data->dev_private;
+ unsigned int i;
- if ((dev == NULL) || (igb_stats == NULL))
- return -EINVAL;
+ for (i = 0; i < dev->data->nb_rx_queues; i++) {
+ uint64_t pkts = internal->rx_null_queues[i].rx_pkts;
+ uint64_t bytes = internal->rx_null_queues[i].rx_bytes;
+
+ stats->ipackets += pkts;
+ stats->ibytes += bytes;
- internal = dev->data->dev_private;
- num_stats = RTE_MIN((unsigned int)RTE_ETHDEV_QUEUE_STAT_CNTRS,
- RTE_MIN(dev->data->nb_rx_queues,
- RTE_DIM(internal->rx_null_queues)));
- for (i = 0; i < num_stats; i++) {
- igb_stats->q_ipackets[i] =
- internal->rx_null_queues[i].rx_pkts;
- rx_total += igb_stats->q_ipackets[i];
+ if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
+ stats->q_ipackets[i] = pkts;
+ stats->q_ibytes[i] = bytes;
+ }
}
- num_stats = RTE_MIN((unsigned int)RTE_ETHDEV_QUEUE_STAT_CNTRS,
- RTE_MIN(dev->data->nb_tx_queues,
- RTE_DIM(internal->tx_null_queues)));
- for (i = 0; i < num_stats; i++) {
- uint64_t pkts =
rte_atomic_load_explicit(&internal->tx_null_queues[i].tx_pkts,
- rte_memory_order_relaxed);
+ for (i = 0; i < dev->data->nb_tx_queues; i++) {
+ const struct null_queue *q = &internal->tx_null_queues[i];
+ uint64_t pkts, bytes;
- igb_stats->q_opackets[i] = pkts;
- tx_total += pkts;
- }
+ pkts = rte_atomic_load_explicit(&q->tx_pkts,
rte_memory_order_relaxed);
+ bytes = rte_atomic_load_explicit(&q->tx_bytes,
rte_memory_order_relaxed);
+
+ stats->opackets = pkts;
+ stats->obytes = bytes;
- igb_stats->ipackets = rx_total;
- igb_stats->opackets = tx_total;
+ if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
+ stats->q_opackets[i] = pkts;
+ stats->q_obytes[i] = bytes;
+ }
+ }
return 0;
}
@@ -358,13 +359,9 @@ eth_stats_get(struct rte_eth_dev *dev, struct
rte_eth_stats *igb_stats)
static int
eth_stats_reset(struct rte_eth_dev *dev)
{
+ struct pmd_internals *internal = dev->data->dev_private;
unsigned int i;
- struct pmd_internals *internal;
-
- if (dev == NULL)
- return -EINVAL;
- internal = dev->data->dev_private;
for (i = 0; i < RTE_DIM(internal->rx_null_queues); i++) {
internal->rx_null_queues[i].rx_pkts = 0;
internal->rx_null_queues[i].rx_bytes = 0;
--
2.47.2
[PATCH] doc: reword contributor's guide for grammar/clarity
I'm reviewing the Contributor's Guidelines to be more clear and concise where necessary. Signed-off-by: Nandini Persad --- doc/guides/contributing/coding_style.rst | 105 +++ doc/guides/contributing/design.rst | 9 +- doc/guides/contributing/new_library.rst | 31 --- 3 files changed, 77 insertions(+), 68 deletions(-) diff --git a/doc/guides/contributing/coding_style.rst b/doc/guides/contributing/coding_style.rst index 1ebc79ca3c..c465bd0f1f 100644 --- a/doc/guides/contributing/coding_style.rst +++ b/doc/guides/contributing/coding_style.rst @@ -15,14 +15,13 @@ It is based on the Linux Kernel coding guidelines and the FreeBSD 7.2 Kernel Dev General Guidelines -- -The rules and guidelines given in this document cannot cover every situation, so the following general guidelines should be used as a fallback: +This document's rules and guidelines cannot cover every scenario, so the following general principles +should be used as a fallback. -* The code style should be consistent within each individual file. -* In the case of creating new files, the style should be consistent within each file in a given directory or module. -* The primary reason for coding standards is to increase code readability and comprehensibility, therefore always use whatever option will make the code easiest to read. - -Line length is recommended to be not more than 80 characters, including comments. -[Tab stop size should be assumed to be 8-characters wide]. +* Maintain a consistent coding style within each file. +* When creating new files, ensure consistency with other files in the same directory or module. +* Prioritize readability and clarity. Choose the style that makes the code easiest to read. +* Keep line length within 80 characters, including comments. Assume a tab stop size of 8 characters. .. note:: @@ -36,7 +35,7 @@ Usual Comments ~~ These comments should be used in normal cases. -To document a public API, a doxygen-like format must be used: refer to :ref:`doxygen_guidelines`. +To document a public API, a Doxygen-like format must be used. Refer to the :ref:`doxygen_guidelines`. .. code-block:: c @@ -110,15 +109,19 @@ Headers should be protected against multiple inclusion with the usual: Macros ~~ -Do not ``#define`` or declare names except with the standard DPDK prefix: ``RTE_``. -This is to ensure there are no collisions with definitions in the application itself. +Use only the standard DPDK prefix (RTE_) when defining or declaring names +to prevent conflicts with application definitions. + +Macro Naming: -The names of "unsafe" macros (ones that have side effects), and the names of macros for manifest constants, are all in uppercase. +* "Unsafe" macros (those with side effects) and macros for manifest constants must be in uppercase. +* Expression-like macros should either expand to a single token or be enclosed in outer parentheses. +* If a macro inlines a function, use the lowercase function name and an uppercase macro name. -The expansions of expression-like macros are either a single token or have outer parentheses. -If a macro is an inline expansion of a function, the function name is all in lowercase and the macro has the same name all in uppercase. -If the macro encapsulates a compound statement, enclose it in a do-while loop, so that it can be used safely in if statements. -Any final statement-terminating semicolon should be supplied by the macro invocation rather than the macro, to make parsing easier for pretty-printers and editors. +Encapsulation: + +* Macros that wrap compound statements should be enclosed in a do while loop to ensure safe use in "if" statements. +* The semicolon terminating a statement should be provided by the macro invocation, not the macro itself, to improve readability for formatters and editors. For example: @@ -138,38 +141,34 @@ Conditional Compilation .. note:: - Conditional compilation should be used only when absolutely necessary, - as it increases the number of target binaries that need to be built and tested. - See below for details of some utility macros/defines available - to allow ifdefs/macros to be replaced by C conditional in some cases. - -Some high-level guidelines on the use of conditional compilation: - -* If code can compile on all platforms/systems, - but cannot run on some due to lack of support, - then regular C conditionals, as described in the next section, - should be used instead of conditional compilation. -* If the code in question cannot compile on all systems, - but constitutes only a small fragment of a file, - then conditional compilation should be used, as described in this section. -* If the code for conditional compilation implements an interface in an OS - or platform-specific way, then create a file for each OS or platform - and select the appropriate file using the Meson build system. - In most cases, these
[PATCH v2 2/4] net/null: Tx optimizations
All the null device does is call pktmbuf_free, can be marked lockless.
The statitistics need to use atomic for this.
Add byte count statistics as well.
Use rte_pktmbuf_free_bulk instead of a loop.
And pktmbuf_free handles multi-segment packets without problems.
There is no reason for eth_null_tx to check arguments,
no other tx_burst function does.
Signed-off-by: Stephen Hemminger
---
drivers/net/null/rte_eth_null.c | 50 ++---
1 file changed, 27 insertions(+), 23 deletions(-)
diff --git a/drivers/net/null/rte_eth_null.c b/drivers/net/null/rte_eth_null.c
index 966748689f..b7cc90300d 100644
--- a/drivers/net/null/rte_eth_null.c
+++ b/drivers/net/null/rte_eth_null.c
@@ -39,6 +39,7 @@ struct null_queue {
RTE_ATOMIC(uint64_t) rx_pkts;
RTE_ATOMIC(uint64_t) tx_pkts;
+ RTE_ATOMIC(uint64_t) tx_bytes;
};
struct pmd_options {
@@ -145,19 +146,18 @@ eth_null_no_rx(void *q __rte_unused, struct rte_mbuf
**bufs __rte_unused,
static uint16_t
eth_null_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
{
- int i;
struct null_queue *h = q;
-
- if ((q == NULL) || (bufs == NULL))
- return 0;
+ unsigned int i;
+ uint64_t bytes = 0;
for (i = 0; i < nb_bufs; i++)
- rte_pktmbuf_free(bufs[i]);
+ bytes += rte_pktmbuf_pkt_len(bufs[i]);
- /* NOTE: review for potential ordering optimization */
- rte_atomic_fetch_add_explicit(&h->tx_pkts, i, rte_memory_order_seq_cst);
+ rte_pktmbuf_free_bulk(bufs, nb_bufs);
+ rte_atomic_fetch_add_explicit(&h->tx_pkts, nb_bufs,
rte_memory_order_relaxed);
+ rte_atomic_fetch_add_explicit(&h->tx_bytes, bytes,
rte_memory_order_relaxed);
- return i;
+ return nb_bufs;
}
static uint16_t
@@ -165,22 +165,19 @@ eth_null_copy_tx(void *q, struct rte_mbuf **bufs,
uint16_t nb_bufs)
{
struct null_queue *h = q;
unsigned int i;
-
- if ((q == NULL) || (bufs == NULL))
- return 0;
+ uint64_t bytes = 0;
for (i = 0; i < nb_bufs; i++) {
struct rte_mbuf *m = bufs[i];
size_t len = RTE_MIN(h->internals->packet_size, m->data_len);
rte_memcpy(h->dummy_packet, rte_pktmbuf_mtod(m, void *), len);
- rte_pktmbuf_free(bufs[i]);
+ bytes += m->pkt_len;
}
- /* NOTE: review for potential ordering optimization */
- rte_atomic_fetch_add_explicit(&h->tx_pkts, i, rte_memory_order_seq_cst);
-
- return i;
+ rte_atomic_fetch_add_explicit(&h->tx_pkts, nb_bufs,
rte_memory_order_relaxed);
+ rte_atomic_fetch_add_explicit(&h->tx_bytes, bytes,
rte_memory_order_relaxed);
+ return nb_bufs;
}
static int
@@ -314,6 +311,8 @@ eth_dev_info(struct rte_eth_dev *dev,
dev_info->max_rx_queues = RTE_DIM(internals->rx_null_queues);
dev_info->max_tx_queues = RTE_DIM(internals->tx_null_queues);
dev_info->min_rx_bufsize = 0;
+ dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
RTE_ETH_TX_OFFLOAD_MT_LOCKFREE;
+
dev_info->reta_size = internals->reta_size;
dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
dev_info->hash_key_size = sizeof(internals->rss_key);
@@ -346,10 +345,11 @@ eth_stats_get(struct rte_eth_dev *dev, struct
rte_eth_stats *igb_stats)
RTE_MIN(dev->data->nb_tx_queues,
RTE_DIM(internal->tx_null_queues)));
for (i = 0; i < num_stats; i++) {
- /* NOTE: review for atomic access */
- igb_stats->q_opackets[i] =
- internal->tx_null_queues[i].tx_pkts;
- tx_total += igb_stats->q_opackets[i];
+ uint64_t pkts =
rte_atomic_load_explicit(&internal->tx_null_queues[i].tx_pkts,
+ rte_memory_order_relaxed);
+
+ igb_stats->q_opackets[i] = pkts;
+ tx_total += pkts;
}
igb_stats->ipackets = rx_total;
@@ -371,9 +371,13 @@ eth_stats_reset(struct rte_eth_dev *dev)
for (i = 0; i < RTE_DIM(internal->rx_null_queues); i++)
/* NOTE: review for atomic access */
internal->rx_null_queues[i].rx_pkts = 0;
- for (i = 0; i < RTE_DIM(internal->tx_null_queues); i++)
- /* NOTE: review for atomic access */
- internal->tx_null_queues[i].tx_pkts = 0;
+
+ for (i = 0; i < RTE_DIM(internal->tx_null_queues); i++) {
+ struct null_queue *q = &internal->tx_null_queues[i];
+
+ rte_atomic_store_explicit(&q->tx_pkts, 0,
rte_memory_order_relaxed);
+ rte_atomic_store_explicit(&q->tx_bytes, 0,
rte_memory_order_relaxed);
+ }
return 0;
}
--
2.47.2
[PATCH v2 3/4] net/null: optimize Rx
No other rx_burst function checks args, remove it.
Since rx_burst can only safely be called by a single thread
at a time, there is no need for atomic operations on statistics.
Add byte count statistics.
Signed-off-by: Stephen Hemminger
---
drivers/net/null/rte_eth_null.c | 38 -
1 file changed, 18 insertions(+), 20 deletions(-)
diff --git a/drivers/net/null/rte_eth_null.c b/drivers/net/null/rte_eth_null.c
index b7cc90300d..40ce5c6ed2 100644
--- a/drivers/net/null/rte_eth_null.c
+++ b/drivers/net/null/rte_eth_null.c
@@ -37,7 +37,9 @@ struct null_queue {
struct rte_mempool *mb_pool;
void *dummy_packet;
- RTE_ATOMIC(uint64_t) rx_pkts;
+ uint64_t rx_pkts;
+ uint64_t rx_bytes;
+
RTE_ATOMIC(uint64_t) tx_pkts;
RTE_ATOMIC(uint64_t) tx_bytes;
};
@@ -85,12 +87,10 @@ RTE_LOG_REGISTER_DEFAULT(eth_null_logtype, NOTICE);
static uint16_t
eth_null_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
{
- int i;
+ unsigned int i;
struct null_queue *h = q;
unsigned int packet_size;
-
- if ((q == NULL) || (bufs == NULL))
- return 0;
+ uint64_t bytes = 0;
packet_size = h->internals->packet_size;
if (rte_pktmbuf_alloc_bulk(h->mb_pool, bufs, nb_bufs) != 0)
@@ -99,24 +99,22 @@ eth_null_rx(void *q, struct rte_mbuf **bufs, uint16_t
nb_bufs)
for (i = 0; i < nb_bufs; i++) {
bufs[i]->data_len = (uint16_t)packet_size;
bufs[i]->pkt_len = packet_size;
+ bytes += packet_size;
bufs[i]->port = h->internals->port_id;
}
- /* NOTE: review for potential ordering optimization */
- rte_atomic_fetch_add_explicit(&h->rx_pkts, i, rte_memory_order_seq_cst);
-
- return i;
+ h->rx_pkts += nb_bufs;
+ h->rx_bytes += bytes;
+ return nb_bufs;
}
static uint16_t
eth_null_copy_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
{
- int i;
+ unsigned int i;
struct null_queue *h = q;
unsigned int packet_size;
-
- if ((q == NULL) || (bufs == NULL))
- return 0;
+ uint64_t bytes = 0;
packet_size = h->internals->packet_size;
if (rte_pktmbuf_alloc_bulk(h->mb_pool, bufs, nb_bufs) != 0)
@@ -127,13 +125,13 @@ eth_null_copy_rx(void *q, struct rte_mbuf **bufs,
uint16_t nb_bufs)
packet_size);
bufs[i]->data_len = (uint16_t)packet_size;
bufs[i]->pkt_len = packet_size;
+ bytes += packet_size;
bufs[i]->port = h->internals->port_id;
}
- /* NOTE: review for potential ordering optimization */
- rte_atomic_fetch_add_explicit(&h->rx_pkts, i, rte_memory_order_seq_cst);
-
- return i;
+ h->rx_pkts += nb_bufs;
+ h->rx_bytes += bytes;
+ return nb_bufs;
}
static uint16_t
@@ -335,7 +333,6 @@ eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats
*igb_stats)
RTE_MIN(dev->data->nb_rx_queues,
RTE_DIM(internal->rx_null_queues)));
for (i = 0; i < num_stats; i++) {
- /* NOTE: review for atomic access */
igb_stats->q_ipackets[i] =
internal->rx_null_queues[i].rx_pkts;
rx_total += igb_stats->q_ipackets[i];
@@ -368,9 +365,10 @@ eth_stats_reset(struct rte_eth_dev *dev)
return -EINVAL;
internal = dev->data->dev_private;
- for (i = 0; i < RTE_DIM(internal->rx_null_queues); i++)
- /* NOTE: review for atomic access */
+ for (i = 0; i < RTE_DIM(internal->rx_null_queues); i++) {
internal->rx_null_queues[i].rx_pkts = 0;
+ internal->rx_null_queues[i].rx_bytes = 0;
+ }
for (i = 0; i < RTE_DIM(internal->tx_null_queues); i++) {
struct null_queue *q = &internal->tx_null_queues[i];
--
2.47.2
[PATCH] doc: remove known issues
I have uploaded all these known issues into Bugzilla, so they are not needed here anymore. Signed-off-by: --- doc/guides/rel_notes/known_issues.rst | 875 -- 1 file changed, 875 deletions(-) delete mode 100644 doc/guides/rel_notes/known_issues.rst diff --git a/doc/guides/rel_notes/known_issues.rst b/doc/guides/rel_notes/known_issues.rst deleted file mode 100644 index 73c72ba484..00 --- a/doc/guides/rel_notes/known_issues.rst +++ /dev/null @@ -1,875 +0,0 @@ -.. SPDX-License-Identifier: BSD-3-Clause -Copyright(c) 2010-2014 Intel Corporation. - -Known Issues and Limitations in Legacy Releases -=== - -This section describes known issues with the DPDK software that aren't covered in the version specific release -notes sections. - - -Unit Test for Link Bonding may fail at test_tlb_tx_burst() --- - -**Description**: - Unit tests will fail in ``test_tlb_tx_burst()`` function with error for uneven distribution of packets. - -**Implication**: - Unit test link_bonding_autotest will fail. - -**Resolution/Workaround**: - There is no workaround available. - -**Affected Environment/Platform**: - Fedora 20. - -**Driver/Module**: - Link Bonding. - - -Pause Frame Forwarding does not work properly on igb - - -**Description**: - For igb devices rte_eth_flow_ctrl_set does not work as expected. - Pause frames are always forwarded on igb, regardless of the ``RFCE``, ``MPMCF`` and ``DPF`` registers. - -**Implication**: - Pause frames will never be rejected by the host on 1G NICs and they will always be forwarded. - -**Resolution/Workaround**: - There is no workaround available. - -**Affected Environment/Platform**: - All. - -**Driver/Module**: - Poll Mode Driver (PMD). - - -In packets provided by the PMD, some flags are missing --- - -**Description**: - In packets provided by the PMD, some flags are missing. - The application does not have access to information provided by the hardware - (packet is broadcast, packet is multicast, packet is IPv4 and so on). - -**Implication**: - The ``ol_flags`` field in the ``rte_mbuf`` structure is not correct and should not be used. - -**Resolution/Workaround**: - The application has to parse the Ethernet header itself to get the information, which is slower. - -**Affected Environment/Platform**: - All. - -**Driver/Module**: - Poll Mode Driver (PMD). - -The rte_malloc library is not fully implemented - -**Description**: - The ``rte_malloc`` library is not fully implemented. - -**Implication**: - All debugging features of rte_malloc library described in architecture documentation are not yet implemented. - -**Resolution/Workaround**: - No workaround available. - -**Affected Environment/Platform**: - All. - -**Driver/Module**: - ``rte_malloc``. - - -HPET reading is slow - - -**Description**: - Reading the HPET chip is slow. - -**Implication**: - An application that calls ``rte_get_hpet_cycles()`` or ``rte_timer_manage()`` runs slower. - -**Resolution/Workaround**: - The application should not call these functions too often in the main loop. - An alternative is to use the TSC register through ``rte_rdtsc()`` which is faster, - but specific to an lcore and is a cycle reference, not a time reference. - -**Affected Environment/Platform**: - All. - -**Driver/Module**: - Environment Abstraction Layer (EAL). - - -HPET timers do not work on the Osage customer reference platform - - -**Description**: - HPET timers do not work on the Osage customer reference platform which includes an Intel® Xeon® processor 5500 - series processor) using the released BIOS from Intel. - -**Implication**: - On Osage boards, the implementation of the ``rte_delay_us()`` function must be changed to not use the HPET timer. - -**Resolution/Workaround**: - This can be addressed by building the system with ``RTE_LIBEAL_USE_HPET`` unset - or by using the ``--no-hpet`` EAL option. - -**Affected Environment/Platform**: - The Osage customer reference platform. - Other vendor platforms with Intel® Xeon® processor 5500 series processors should - work correctly, provided the BIOS supports HPET. - -**Driver/Module**: - ``lib/eal/include/rte_cycles.h`` - - -Not all variants of supported NIC types have been used in testing -- - -**Description**: - The supported network interface cards can come in a number of variants with different device ID's. - Not all of these variants have been tested with the DPDK. - - The NIC device identifiers used during testing: - - * Intel® Ethernet Controller XL710 for 40GbE QSFP
[PATCH v2 1/4] net/null: fix packet copy
If doing copy on transmit, can potentially copy past the data
in the mbuf. Change to only copy data from that segment.
Fixes: c743e50c475f ("null: new poll mode driver")
Cc: muk...@igel.co.jp
Cc: sta...@dpdk.org
Signed-off-by: Stephen Hemminger
---
drivers/net/null/rte_eth_null.c | 12 ++--
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/drivers/net/null/rte_eth_null.c b/drivers/net/null/rte_eth_null.c
index 6764cf2ec1..966748689f 100644
--- a/drivers/net/null/rte_eth_null.c
+++ b/drivers/net/null/rte_eth_null.c
@@ -35,7 +35,7 @@ struct null_queue {
struct pmd_internals *internals;
struct rte_mempool *mb_pool;
- struct rte_mbuf *dummy_packet;
+ void *dummy_packet;
RTE_ATOMIC(uint64_t) rx_pkts;
RTE_ATOMIC(uint64_t) tx_pkts;
@@ -163,17 +163,17 @@ eth_null_tx(void *q, struct rte_mbuf **bufs, uint16_t
nb_bufs)
static uint16_t
eth_null_copy_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
{
- int i;
struct null_queue *h = q;
- unsigned int packet_size;
+ unsigned int i;
if ((q == NULL) || (bufs == NULL))
return 0;
- packet_size = h->internals->packet_size;
for (i = 0; i < nb_bufs; i++) {
- rte_memcpy(h->dummy_packet, rte_pktmbuf_mtod(bufs[i], void *),
- packet_size);
+ struct rte_mbuf *m = bufs[i];
+ size_t len = RTE_MIN(h->internals->packet_size, m->data_len);
+
+ rte_memcpy(h->dummy_packet, rte_pktmbuf_mtod(m, void *), len);
rte_pktmbuf_free(bufs[i]);
}
--
2.47.2
Re: [PATCH] dts: add Scapy asynchronous sniffer
On Tue, Mar 18, 2025 at 1:41 PM Luca Vizzarro wrote: > > Currently packet capturing in Scapy is used by running Scapy's own > class AsyncSniffer in the same shell as packet sending. Because there is > some start up time involved in the AsyncSniffer, doing this every single > time a test case requires it can become slow and unreliable if sniffer > was not ready yet at the time of sending. > > Add a multi-threaded implementation for sniffing and capturing packets > on-demand using Scapy. The sniffer lives in its own dedicated shell, > which is setup for sniffing at the start up of the test run. Packets are > only captured when requested by the tests through the use of Event > signals. > > Signed-off-by: Luca Vizzarro > Reviewed-by: Paul Szczepanek Reviewed-by: Dean Marx
Re: [PATCH v6 3/8] buildtools: display version when listing symbols
On Tue, Apr 1, 2025 at 6:11 PM Thomas Monjalon wrote: > > 28/03/2025 11:52, David Marchand: > > Display the version when a symbol was introduced. > > This is needed for scripting the conversion from static to dynamically > > generated version maps. > > It is also useful when listing experimental symbols. > > I feel the title needs a bit more context, maybe adding "map"? > buildtools: display version when listing symbols from map > buildtools: display symbols version from map The second form lgtm. > Do we still need this script after this series? It is still used in check-symbols.sh, to catch symbols present in binaries that are not marked as experimental / internal. I don't think there is a way to replace this check. > Should we remove some scripts? I tried to remove as many as I could, but maybe I overlooked some. -- David Marchand
[PATCH] app/testpmd: fix hash key update problem
There has logic problem in 'port_rss_hash_key_update()', the user input
will be overwritten by the call to 'rte_eth_dev_rss_hash_conf_get()',
so the RSS functions will not get update as expected.
--
testpmd> show port 0 rss-hash key
RSS functions:
ipv4 ipv6
RSS key:
6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B7
3BBEAC01FA
testpmd> port config 0 rss-hash-key ipv6-tcp 6D5A56DA255B0EC24167253D
43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
testpmd> show port 0 rss-hash key
RSS functions:
ipv4 ipv6
RSS key:
6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B7
3BBEAC01FA
testpmd>
--
Fixes: 8205e241b2b0 ("app/testpmd: add missing type to RSS hash commands")
Cc: nelio.laranje...@6wind.com
Cc: sta...@dpdk.org
Signed-off-by: Chaoyong He
Reviewed-by: Long Wu
Reviewed-by: Peng Zhang
---
app/test-pmd/config.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/app/test-pmd/config.c b/app/test-pmd/config.c
index e89af21cec..c32e011fa6 100644
--- a/app/test-pmd/config.c
+++ b/app/test-pmd/config.c
@@ -4806,11 +4806,12 @@ port_rss_hash_key_update(portid_t port_id, char
rss_type[], uint8_t *hash_key,
rss_conf.rss_key = NULL;
rss_conf.rss_key_len = 0;
- rss_conf.rss_hf = str_to_rsstypes(rss_type);
+ rss_conf.rss_hf = 0;
diag = rte_eth_dev_rss_hash_conf_get(port_id, &rss_conf);
if (diag == 0) {
rss_conf.rss_key = hash_key;
rss_conf.rss_key_len = hash_key_len;
+ rss_conf.rss_hf = str_to_rsstypes(rss_type);
diag = rte_eth_dev_rss_hash_update(port_id, &rss_conf);
}
if (diag == 0)
--
2.43.5
RE: [EXTERNAL] [PATCH] eventdev: fix dereferencing null atomic locks pointer in test-eventdev
> -Original Message-
> From: Luka Jankovic
> Sent: Wednesday, March 26, 2025 3:40 PM
> To: luka.janko...@ericsson.com
> Cc: Jerin Jacob ; dev@dpdk.org
> Subject: [EXTERNAL] [PATCH] eventdev: fix dereferencing null atomic locks
> pointer in test-eventdev
>
> Update atomic_init_locks to immediately return if memory allocation fails.
> Atomic queue and atq tests updated to handle atomic locks being null. Coverity
> issue: 457876 Fixes: 9d619f82321b ("app/eventdev: introduce atomic tests")
> Signed-off-by:
> Update atomic_init_locks to immediately return if memory allocation fails.
> Atomic queue and atq tests updated to handle atomic locks being null.
>
> Coverity issue: 457876
> Fixes: 9d619f82321b ("app/eventdev: introduce atomic tests")
>
> Signed-off-by: Luka Jankovic
Please check the git history of the file to know the starting tag.
Updated the git commit as follows and applied to
dpdk-next-net-eventdev/for-main. Thanks
app/eventdev: fix dereferencing null
Update atomic_init_locks to immediately return if memory allocation
fails.
Atomic queue and ATQ tests updated to handle atomic locks being NULL
Coverity issue: 457876
Fixes: 9d619f82321b ("app/eventdev: introduce atomic tests")
Cc: sta...@dpdk.org
Signed-off-by: Luka Jankovic
RE: [PATCH 3/3] net/cnxk: update IP header of reassembled packets
> -Original Message- > From: Rahul Bhansali > Sent: Friday, March 21, 2025 3:34 PM > To: dev@dpdk.org; Nithin Kumar Dabilpuram ; > Kiran Kumar Kokkilagadda ; Sunil Kumar Kori > ; Satha Koteswara Rao Kottidi > ; Harman Kalra > Cc: Jerin Jacob ; Rahul Bhansali > Subject: [PATCH 3/3] net/cnxk: update IP header of reassembled packets > > Updates IP header data of reassembled packets, > 1) IPv4 length, MF bit and checksum correction > 2) IPv6 length update and fragment extension header remove Also, corrects > data length of multisegment decrypted packets. > > Signed-off-by: Rahul Bhansali https://mails.dpdk.org/archives/test-report/2025-March/863905.html - fix this issue without performance impact.
Re: [PATCH v2 0/3] allow easier use of high lcore-ids
On Mon, Mar 24, 2025 at 05:30:26PM +, Bruce Richardson wrote: > Traditionally, DPDK has had a direct mapping of internal lcore-ids, to > the actual core numbers in use. With higher core count servers becoming > more prevalent the issue becomes one of increasing memory footprint when > using such a scheme, due to the need to have all arrays dimensioned for > all cores on the system, whether or not those cores are in use by the > app. > > Therefore, the decision was made in the past to not expand the > build-time RTE_MAX_LCORE value beyond 128. Instead, it was recommended > that users use the "--lcores" EAL parameter to take the high-numbered > cores they wish to use and map them to lcore-ids within the 0 - 128 > range. While this works, this is a little clunky as it means that > instead of just passing, for example, "-l 130-139", the user must > instead pass "--lcores 0@130,1@131,2@132,3@133," > > This patchset attempts to simplify the situation by adding a new flag to > do this mapping automatically. To use cores 130-139 and map them to ids > 0-9 internally, the EAL args now become: "-l 130-139 --map-lcore-ids", > or using the shorter "-M" version of the flag: "-Ml 130-139". > > Adding this new parameter required some rework of the existing arg > parsing code, because in current DPDK the args are parsed and checked in > the order they appear on the commandline. This means that using the > example above, the core parameter 130-139 will be rejected immediately > before the "map-lcore-ids" parameter is seen. To work around this, the > core (and service core) parameters are not parsed when seen, instead > they are only saved off and parsed after all arguments are parsed. The > "-l" and "-c" parameters are converted into "--lcores" arguments, so all > assigning of lcore ids is done there in all cases. > > RFC->v2: > * converted printf to DEBUG log > * added "-M" as shorter version of flag > * added documentation > * renamed internal API that was changed to avoid any potential hidden > runtime issues. > > Bruce Richardson (3): > eal: centralize core parameter parsing > eal: convert core masks and lists to core sets > eal: allow automatic mapping of high lcore ids > Ping for review. At a high level, does this feature seem useful to users? /Bruce
Re: [PATCH 1/2] node: add global node mbuf dynfield
On Tue, 1 Apr 2025 09:50:46 +0530
Nitin Saxena wrote:
> +int rte_node_mbuf_dynfield_register(void)
> +{
> + struct node_mbuf_dynfield_mz *f = NULL;
> + const struct rte_memzone *mz = NULL;
> + int dyn_offset;
> +
> + RTE_BUILD_BUG_ON(sizeof(rte_node_mbuf_dynfield_t) <
> RTE_NODE_MBUF_DYNFIELD_SIZE);
> + RTE_BUILD_BUG_ON(sizeof(rte_node_mbuf_overload_fields_t) <
> + RTE_NODE_MBUF_OVERLOADABLE_FIELDS_SIZE);
> +
> + mz = rte_memzone_lookup(NODE_MBUF_DYNFIELD_MEMZONE_NAME);
Seems wasteful to have a whole memzone for this, the data is small.
Is there a reason it could not just be a global variable like timestamp.
I would prefer this was a clone of timestamp code, and put in
rte_mbuf_dynfield.c
Re: [PATCH v6 4/8] build: generate symbol maps
28/03/2025 11:52, David Marchand: > Rather than maintain a file in parallel of the code, symbols to be > exported can be marked with a token RTE_EXPORT_*SYMBOL. > > From those marks, the build framework generates map files only for > symbols actually compiled (which means that the WINDOWS_NO_EXPORT hack > becomes unnecessary). > > The build framework directly creates a map file in the format that the > linker expects (rather than converting from GNU linker to MSVC linker). At this point of the series, the build is broken. I suppose it is a tradeoff to avoid squashing 2 big patches together. This is the error I hit: drivers/meson.build:275:37: ERROR: Unknown variable "sources_avx2".
