eal: allow setting random number generator seed

RFC Superseded Stephen Hemminger 5 participants 8 messages Sep 11 – Sep 14

Share & tools

Stephen Hemminger

author

2026-09-11 22:58 UTC #0 lore raw

RFC 1/1 eal: allow setting random number generator seed

Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to overide the automatic random seed on DPDK startup.

This is not a security problem since rte_rand() is documented as not being cryptographically secure.

Signed-off-by Stephen Hemminger <[email protected]>

---
 lib/eal/common/rte_random.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

lib/eal/common/rte_random.c +12 -1

index 576a32a46c..cdad60049b 100644
@@ -247,7 +247,18 @@ eal_rand_init(void)
 
 	RTE_LCORE_VAR_ALLOC(rand_state);
 
-	seed = __rte_random_initial_seed();
+	const char *env = getenv("DPDK_RANDOM_SEED");
+	if (env != NULL && *env != '\0') {
+		char *end;
+
+		errno = 0;
+		seed = strtoull(env, &end, 0);
+		if (errno != 0 || *end != '\0')
+			rte_exit(EXIT_FAILURE,
+				 "invalid DPDK_RANDOM_SEED: %s\n", env);
+	} else {
+		seed = __rte_random_initial_seed();
+	}
 
 	rte_srand(seed);
 }

Signature · 1 line

2.53.0

Morten Brørup

Reviewed-by on v1 1/1

2026-09-12 06:27 UTC #1 lore raw

Quoted 40 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Saturday, 12 September 2026 00.59
Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to overide the automatic random seed on DPDK startup.
This is not a security problem since rte_rand() is documented as not being cryptographically secure.
Signed-off-by Stephen Hemminger <[email protected]>
---
 lib/eal/common/rte_random.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
lib/eal/common/rte_random.c +12 -2
index 576a32a46c..cdad60049b 100644
@@ -247,7 +247,18 @@ eal_rand_init(void)
 
 	RTE_LCORE_VAR_ALLOC(rand_state);
 
-	seed = __rte_random_initial_seed();
+	const char *env = getenv("DPDK_RANDOM_SEED");
+	if (env != NULL && *env != '\0') {
+		char *end;
+
+		errno = 0;
+		seed = strtoull(env, &end, 0);
+		if (errno != 0 || *end != '\0')
+			rte_exit(EXIT_FAILURE,
+				 "invalid DPDK_RANDOM_SEED: %s\n", env);
+	} else {
+		seed = __rte_random_initial_seed();
+	}
 
 	rte_srand(seed);
 }
--

Quoted 1 line

2.53.0

Might be useful.

Reviewed-by Morten Brørup <[email protected]>

Stephen Hemminger

author

2026-09-12 16:17 UTC #2 lore raw

PATCH v2 1/1 eal: allow setting random number generator seed

In reply to Stephen Hemminger

Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to override the automatic random seed on DPDK startup.

This is not a security problem since rte_rand() is documented as not being cryptographically secure.

Signed-off-by Stephen Hemminger <[email protected]>

Reviewed-by Morten Brørup <[email protected]>

---
v2 - fix header inclusion
- add docbook comment

 lib/eal/common/rte_random.c  | 14 +++++++++++++-
 lib/eal/include/rte_random.h |  8 +++++---
 2 files changed, 18 insertions(+), 4 deletions(-)

lib/eal/common/rte_random.c +13 -1

index 576a32a46c..3cc0e3ec5f 100644
@@ -8,6 +8,7 @@
 #endif
 #endif
 #include <unistd.h>
+#include <stdlib.h>
 
 #include <rte_bitops.h>
 #include <rte_branch_prediction.h>

@@ -247,7 +248,18 @@ eal_rand_init(void)
 
 	RTE_LCORE_VAR_ALLOC(rand_state);
 
-	seed = __rte_random_initial_seed();
+	const char *env = getenv("DPDK_RANDOM_SEED");
+	if (env != NULL && *env != '\0') {
+		char *end;
+
+		errno = 0;
+		seed = strtoull(env, &end, 0);
+		if (errno != 0 || *end != '\0')
+			rte_exit(EXIT_FAILURE,
+				 "invalid DPDK_RANDOM_SEED: %s\n", env);
+	} else {
+		seed = __rte_random_initial_seed();
+	}
 
 	rte_srand(seed);
 }

lib/eal/include/rte_random.h +5 -3

index 15cbe6215a..bdd4001e78 100644
@@ -20,9 +20,11 @@ extern "C" {
 /**
  * Seed the pseudo-random generator.
  *
- * The generator is automatically seeded by the EAL init with a timer
- * value. It may need to be re-seeded by the user with a real random
- * value.
+ * The generator is automatically seeded by the EAL init with
+ * a system provided random entropy source. But for testing
+ * it can be useful to force a repeatable starting point by
+ * setting the initial seed. This can be done by setting
+ * the `DPDK_RANDOM_SEED` environment variable.
  *
  * This function is not multi-thread safe in regards to other
  * rte_srand() calls, nor is it in relation to concurrent rte_rand(),

Signature · 1 line

2.53.0

Mattias Rönnblom

2026-09-14 11:29 UTC #3 lore raw

Den 2026-09-12 kl. 18:16, skrev Stephen Hemminger:

Quoted 3 lines

Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to override the automatic random seed on DPDK startup.

There is a way to seed the PRNG already, rte_srand(). Why the tests can't use this?

Provided the test results depend on something that use the PRNG before the test driver has had the opportunity to call rte_srand(), controlling the *initial* seed may be required.

I don't think this feature should be controlled by an environment variable. There is no precedent for that. If this functionality is deemed useful, it should be an EAL command line option, it seems to me.

Quoted 30 lines

This is not a security problem since rte_rand() is documented as not being cryptographically secure.
Signed-off-by Stephen Hemminger <[email protected]>
Reviewed-by Morten Brørup <[email protected]>
---
v2 - fix header inclusion
- add docbook comment
  lib/eal/common/rte_random.c  | 14 +++++++++++++-
  lib/eal/include/rte_random.h |  8 +++++---
  2 files changed, 18 insertions(+), 4 deletions(-)
lib/eal/common/rte_random.c +1 -0
index 576a32a46c..3cc0e3ec5f 100644
@@ -8,6 +8,7 @@
  #endif
  #endif
  #include <unistd.h>
+#include <stdlib.h>
 
  #include <rte_bitops.h>
  #include <rte_branch_prediction.h>
@@ -247,7 +248,18 @@ eal_rand_init(void)
 
  	RTE_LCORE_VAR_ALLOC(rand_state);
 

lib/eal/common/rte_random.c eal_rand_init(void) · line 259

Quoted 10 lines

lib/eal/common/rte_random.c +9 -1

-	seed = __rte_random_initial_seed();
+	const char *env = getenv("DPDK_RANDOM_SEED");
+	if (env != NULL && *env != '\0') {
+		char *end;
+
+		errno = 0;
+		seed = strtoull(env, &end, 0);
+		if (errno != 0 || *end != '\0')
+			rte_exit(EXIT_FAILURE,
+				 "invalid DPDK_RANDOM_SEED: %s\n", env);

Is rte_exit() the way to deal with errors here? Not to be used in DPDK libraries, if I recall correctly.

I would think logging an error would suffice. Or rte_eal_init_alert().

Quoted 25 lines

lib/eal/common/rte_random.c +3 -0

+	} else {
+		seed = __rte_random_initial_seed();
+	}
 
  	rte_srand(seed);
  }

lib/eal/include/rte_random.h +5 -3

index 15cbe6215a..bdd4001e78 100644
@@ -20,9 +20,11 @@ extern "C" {
  /**
   * Seed the pseudo-random generator.
   *
- * The generator is automatically seeded by the EAL init with a timer
- * value. It may need to be re-seeded by the user with a real random
- * value.
+ * The generator is automatically seeded by the EAL init with
+ * a system provided random entropy source. But for testing
+ * it can be useful to force a repeatable starting point by
+ * setting the initial seed. This can be done by setting
+ * the `DPDK_RANDOM_SEED` environment variable.
   *
   * This function is not multi-thread safe in regards to other
   * rte_srand() calls, nor is it in relation to concurrent rte_rand(),

Konstantin Ananyev

2026-09-14 12:04 UTC #4 lore raw

Quoted 10 lines

Den 2026-09-12 kl. 18:16, skrev Stephen Hemminger:
Quoted 3 lines
Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to override the automatic random seed on DPDK startup.
There is a way to seed the PRNG already, rte_srand(). Why the tests can't use this?

Quoted 7 lines

Provided the test results depend on something that use the PRNG before the test driver has had the opportunity to call rte_srand(), controlling the *initial* seed may be required.
I don't think this feature should be controlled by an environment variable. There is no precedent for that. If this functionality is deemed useful, it should be an EAL command line option, it seems to me.

Quoted 31 lines

Quoted 31 lines
This is not a security problem since rte_rand() is documented as not being cryptographically secure.
Signed-off-by Stephen Hemminger <[email protected]>
Reviewed-by Morten Brørup <[email protected]>
---
v2 - fix header inclusion
- add docbook comment
  lib/eal/common/rte_random.c  | 14 +++++++++++++-
  lib/eal/include/rte_random.h |  8 +++++---
  2 files changed, 18 insertions(+), 4 deletions(-)
lib/eal/common/rte_random.c +1 -0
index 576a32a46c..3cc0e3ec5f 100644
@@ -8,6 +8,7 @@
  #endif
  #endif
  #include <unistd.h>
+#include <stdlib.h>
 
  #include <rte_bitops.h>
  #include <rte_branch_prediction.h>
@@ -247,7 +248,18 @@ eal_rand_init(void)
 
  	RTE_LCORE_VAR_ALLOC(rand_state);
 

Quoted 13 lines

lib/eal/common/rte_random.c eal_rand_init(void) · line 259
Quoted 10 lines
lib/eal/common/rte_random.c +9 -1
-	seed = __rte_random_initial_seed();
+	const char *env = getenv("DPDK_RANDOM_SEED");
+	if (env != NULL && *env != '\0') {
+		char *end;
+
+		errno = 0;
+		seed = strtoull(env, &end, 0);
+		if (errno != 0 || *end != '\0')
+			rte_exit(EXIT_FAILURE,
+				 "invalid DPDK_RANDOM_SEED: %s\n", env);
Is rte_exit() the way to deal with errors here? Not to be used in DPDK libraries, if I recall correctly.

Again, +1
Whole patch looks to me like a strange hack that completely ignores DPDK coding practices. My vote is NACK.

Quoted 28 lines

I would think logging an error would suffice. Or rte_eal_init_alert().

Quoted 25 lines

lib/eal/common/rte_random.c +3 -0

+	} else {
+		seed = __rte_random_initial_seed();
+	}
 
  	rte_srand(seed);
  }

lib/eal/include/rte_random.h +5 -3

index 15cbe6215a..bdd4001e78 100644
@@ -20,9 +20,11 @@ extern "C" {
  /**
   * Seed the pseudo-random generator.
   *
- * The generator is automatically seeded by the EAL init with a timer
- * value. It may need to be re-seeded by the user with a real random
- * value.
+ * The generator is automatically seeded by the EAL init with
+ * a system provided random entropy source. But for testing
+ * it can be useful to force a repeatable starting point by
+ * setting the initial seed. This can be done by setting
+ * the `DPDK_RANDOM_SEED` environment variable.
   *
   * This function is not multi-thread safe in regards to other
   * rte_srand() calls, nor is it in relation to concurrent rte_rand(),

Stephen Hemminger

author

2026-09-14 16:47 UTC #5 lore raw

Quoted 10 lines · On Mon, 14 Sep 2026 12:04:16 +0000 Konstantin Ananyev…

Quoted 8 lines
Provided the test results depend on something that use the PRNG before the test driver has had the opportunity to call rte_srand(), controlling the *initial* seed may be required.
I don't think this feature should be controlled by an environment variable. There is no precedent for that. If this functionality is deemed useful, it should be an EAL command line option, it seems to me.
+1

I thought about that but getting command line options into EAL when using meson test runner is awkward (not impossible though).

Stephen Hemminger

author

2026-09-14 16:47 UTC #6 lore raw

In reply to Mattias Rönnblom

Quoted 8 lines · On Mon, 14 Sep 2026 13:29:22 +0200 Mattias Rönnblom <[email protected]>…

Den 2026-09-12 kl. 18:16, skrev Stephen Hemminger:
Quoted 3 lines
Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to override the automatic random seed on DPDK startup.
There is a way to seed the PRNG already, rte_srand(). Why the tests can't use this?

Quoted 7 lines

Provided the test results depend on something that use the PRNG before the test driver has had the opportunity to call rte_srand(), controlling the *initial* seed may be required.
I don't think this feature should be controlled by an environment variable. There is no precedent for that. If this functionality is deemed useful, it should be an EAL command line option, it seems to me.

There are several tests already using rte_srand() so that is one option. The choice of environment variable was to match the existing test harness infrastructure that uses MALLOC_PERTURB to pre-seed malloc.

Marat Khalili

Acked-by on v1 1/1

2026-09-14 09:05 UTC #7 lore raw

In reply to Stephen Hemminger

Quoted 44 lines

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: Friday 11 September 2026 23:59
To: [email protected]
Cc: Stephen Hemminger <[email protected]>; Mattias Rönnblom <[email protected]>
Subject: [RFC] eal: allow setting random number generator seed
Many performance tests use rte_rand() and the random number can perturb the results. Add an ability to overide the automatic random seed on DPDK startup.
This is not a security problem since rte_rand() is documented as not being cryptographically secure.
Signed-off-by Stephen Hemminger <[email protected]>
---
 lib/eal/common/rte_random.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
lib/eal/common/rte_random.c +12 -2
index 576a32a46c..cdad60049b 100644
@@ -247,7 +247,18 @@ eal_rand_init(void)
 
 	RTE_LCORE_VAR_ALLOC(rand_state);
 
-	seed = __rte_random_initial_seed();
+	const char *env = getenv("DPDK_RANDOM_SEED");
+	if (env != NULL && *env != '\0') {
+		char *end;
+
+		errno = 0;
+		seed = strtoull(env, &end, 0);
+		if (errno != 0 || *end != '\0')
+			rte_exit(EXIT_FAILURE,
+				 "invalid DPDK_RANDOM_SEED: %s\n", env);
+	} else {
+		seed = __rte_random_initial_seed();
+	}
 
 	rte_srand(seed);
 }
--

Quoted 1 line

2.53.0

Lgtm, although as a minor nit env is maybe not the best name here.

Regardless,

Acked-by Marat Khalili <[email protected]>