DPDK-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: Bruce Richardson <bruce.richardson@intel.com>
To: dev@dpdk.org
Cc: techboard@dpdk.org, Bruce Richardson <bruce.richardson@intel.com>
Subject: [RFC PATCH 14/44] eal: cleanup CPU init function
Date: Wed, 29 Apr 2026 17:58:06 +0100	[thread overview]
Message-ID: <20260429165845.2136843-15-bruce.richardson@intel.com> (raw)
In-Reply-To: <20260429165845.2136843-1-bruce.richardson@intel.com>

The CPU init function did extra work in zeroing the runtime config for
lcores, something that was being done a second time in configuring the
argparsing stage when preparing to set up the correct cpu affinities for
each lcore. Therefore we can remove that unnecessary initialization, and
have the function not make any changes to "runtime_state" but only the
"platform_info".

In the process we can make some other cleanups too:
* remove the limit on the lcore_to_socket_id array, and dynamically
  allocate it to the correct number of present cores.
* with the runtime lcore init gone, we can move the assignment to the
  lcore_to_socket_id into the first loop, allowing us to remove the
  second loop entirely.
* the log message about "skipping" lcores was incorrect for modern DPDK,
  since we no longer have a hard-link between physical core numbers and
  lcore ids. Therefore just report what cores are detected on what
  sockets.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/eal/common/eal_common_lcore.c | 83 +++++++------------------------
 1 file changed, 18 insertions(+), 65 deletions(-)

diff --git a/lib/eal/common/eal_common_lcore.c b/lib/eal/common/eal_common_lcore.c
index ca5106c623..808fffbb6c 100644
--- a/lib/eal/common/eal_common_lcore.c
+++ b/lib/eal/common/eal_common_lcore.c
@@ -167,18 +167,9 @@ socket_id_cmp(const void *a, const void *b)
 int
 rte_eal_cpu_init(void)
 {
-	/* pointer to global configuration */
 	struct rte_config *config = rte_eal_get_configuration();
 	struct eal_platform_info *platform_info = eal_get_platform_info();
-	struct eal_runtime_state *runtime_state = eal_get_runtime_state();
-	unsigned lcore_id;
-	unsigned count = 0;
-	unsigned int socket_id, prev_socket_id;
-#if CPU_SETSIZE > RTE_MAX_LCORE
-	int lcore_to_socket_id[CPU_SETSIZE] = {0};
-#else
-	int lcore_to_socket_id[RTE_MAX_LCORE] = {0};
-#endif
+	int *lcore_to_socket_id;
 
 	/* allocate cpu_info for all CPUs visible to the OS */
 	platform_info->cpu_count = eal_cpu_max();
@@ -188,6 +179,12 @@ rte_eal_cpu_init(void)
 		EAL_LOG(ERR, "Cannot allocate cpu_info array");
 		return -1;
 	}
+	lcore_to_socket_id = calloc(platform_info->cpu_count, sizeof(*lcore_to_socket_id));
+	if (lcore_to_socket_id == NULL) {
+		EAL_LOG(ERR, "Cannot allocate lcore_to_socket_id array");
+		free(platform_info->cpu_info);
+		return -1;
+	}
 
 	/* populate cpu_info with hardware topology for all detected CPUs */
 	for (size_t cpu_id = 0; cpu_id < platform_info->cpu_count; cpu_id++) {
@@ -196,68 +193,23 @@ rte_eal_cpu_init(void)
 		platform_info->cpu_info[cpu_id].detected = true;
 		platform_info->cpu_info[cpu_id].numa_id = eal_cpu_socket_id(cpu_id);
 		platform_info->cpu_info[cpu_id].core_id = eal_cpu_core_id(cpu_id);
-	}
+		/* store numa id for later processing to determine all unique numa nodes */
+		lcore_to_socket_id[cpu_id] = platform_info->cpu_info[cpu_id].numa_id;
 
-	/*
-	 * Parse the maximum set of logical cores, detect the subset of running
-	 * ones and enable them by default.
-	 */
-	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
-		runtime_state->lcore_cfg[lcore_id].core_index = count;
-
-		/* init cpuset for per lcore config */
-		CPU_ZERO(&runtime_state->lcore_cfg[lcore_id].cpuset);
-		runtime_state->lcore_cfg[lcore_id].first_cpu = UINT16_MAX;
-
-		if (eal_cpu_detected(lcore_id) == 0) {
-			config->lcore_role[lcore_id] = ROLE_OFF;
-			runtime_state->lcore_cfg[lcore_id].core_index = -1;
-			continue;
-		}
-
-		/* find socket first */
-		socket_id = platform_info->cpu_info[lcore_id].numa_id;
-		lcore_to_socket_id[lcore_id] = socket_id;
-
-		/* By default, lcore 1:1 map to cpu id */
-		CPU_SET(lcore_id, &runtime_state->lcore_cfg[lcore_id].cpuset);
-		runtime_state->lcore_cfg[lcore_id].first_cpu = lcore_id;
-
-		/* By default, each detected core is enabled */
-		config->lcore_role[lcore_id] = ROLE_RTE;
-		EAL_LOG(DEBUG, "Detected lcore %u as "
-				"core %u on NUMA node %u",
-				lcore_id,
-				platform_info->cpu_info[lcore_id].core_id,
-				platform_info->cpu_info[lcore_id].numa_id);
-		count++;
+		EAL_LOG(DEBUG, "Detected CPU %zu as core %u on NUMA node %u",
+				cpu_id,
+				platform_info->cpu_info[cpu_id].core_id,
+				platform_info->cpu_info[cpu_id].numa_id);
 	}
-	for (; lcore_id < CPU_SETSIZE; lcore_id++) {
-		if (eal_cpu_detected(lcore_id) == 0)
-			continue;
-		if (unlikely(lcore_id >= platform_info->cpu_count))
-			break;
-		lcore_to_socket_id[lcore_id] = platform_info->cpu_info[lcore_id].numa_id;
-		EAL_LOG(DEBUG, "Skipped lcore %u as core %u on NUMA node %u",
-			lcore_id, platform_info->cpu_info[lcore_id].core_id,
-			platform_info->cpu_info[lcore_id].numa_id);
-	}
-
-	/* Set the count of enabled logical cores of the EAL configuration */
-	config->lcore_count = count;
-	EAL_LOG(DEBUG,
-			"Maximum logical cores by configuration: %u",
-			RTE_MAX_LCORE);
-	EAL_LOG(INFO, "Detected CPU lcores: %u", config->lcore_count);
 
 	/* sort all socket id's in ascending order */
-	qsort(lcore_to_socket_id, RTE_DIM(lcore_to_socket_id),
+	qsort(lcore_to_socket_id, platform_info->cpu_count,
 			sizeof(lcore_to_socket_id[0]), socket_id_cmp);
 
-	prev_socket_id = -1;
+	int prev_socket_id = -1;
 	config->numa_node_count = 0;
-	for (lcore_id = 0; lcore_id < RTE_DIM(lcore_to_socket_id); lcore_id++) {
-		socket_id = lcore_to_socket_id[lcore_id];
+	for (size_t cpu_id = 0; cpu_id < platform_info->cpu_count; cpu_id++) {
+		int socket_id = lcore_to_socket_id[cpu_id];
 		if (socket_id != prev_socket_id)
 			config->numa_nodes[config->numa_node_count++] =	socket_id;
 		prev_socket_id = socket_id;
@@ -266,6 +218,7 @@ rte_eal_cpu_init(void)
 	}
 	EAL_LOG(INFO, "Detected NUMA nodes: %u", config->numa_node_count);
 
+	free(lcore_to_socket_id);
 	return 0;
 }
 
-- 
2.51.0


  parent reply	other threads:[~2026-04-29 17:01 UTC|newest]

Thread overview: 50+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-04-29 16:57 [RFC PATCH 00/44] Allow intitializing EAL without argc/argv Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 01/44] eal: define new functionally distinct config structs Bruce Richardson
2026-04-29 19:03   ` Stephen Hemminger
2026-04-30  7:56     ` Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 02/44] eal: move memory request fields to user config Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 03/44] eal: move NUMA " Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 04/44] eal: move hugepage policy " Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 05/44] eal: move process " Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 06/44] eal: move advanced user config options to user cfg struct Bruce Richardson
2026-04-29 16:57 ` [RFC PATCH 07/44] eal: move hugepage size info to platform info struct Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 08/44] telemetry: make cpuset init parameter const Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 09/44] eal: move runtime state to appropriate structure Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 10/44] eal: record details of all cpus in platform info Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 11/44] eal: use platform info for lcore lookups Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 12/44] eal: add RTE_CPU_FFS macro Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 13/44] eal: store lcore configuration in runtime data Bruce Richardson
2026-04-29 16:58 ` Bruce Richardson [this message]
2026-04-29 16:58 ` [RFC PATCH 15/44] eal: move numa node information to platform info struct Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 16/44] eal: move lcore role and count to runtime state Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 17/44] eal: make lcore role a field in lcore config struct Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 18/44] eal: move main lcore setting to runtime " Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 19/44] eal: move iova mode and process type to runtime cfg Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 20/44] eal: move memory config pointer to runtime state struct Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 21/44] eal: remove rte_config structure Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 22/44] eal: separate runtime state update from arg parsing Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 23/44] eal: move devopt_list staging list into user_cfg Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 24/44] eal: separate plugin paths from loaded plugin objects Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 25/44] eal: simplify internal driver path iteration APIs Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 26/44] eal: move trace config into user config struct Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 27/44] eal: record service cores in " Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 28/44] eal: store user-provided lcore info " Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 29/44] eal: clarify docs on params taking lcore IDs Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 30/44] eal: remove internal config reset function Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 31/44] eal: move functions setting runtime state Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 32/44] eal: initialize platform info on first use Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 33/44] eal: remove duplicated scan of sysfs for hugepage details Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 34/44] eal: add utilities for working with user config struct Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 35/44] eal: split EAL init into two stages Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 36/44] eal: provide hooks for init with externally supplied config Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 37/44] eal_cfg: add new library to programmatically init DPDK Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 38/44] eal_cfg: configure defaults for easier testing and use Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 39/44] app/test: enable testing init using EAL config lib Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 40/44] eal_cfg: add basic setters and getters Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 41/44] eal_cfg: add hugepage memory configuration Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 42/44] eal_cfg: support configuring lcores Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 43/44] eal_cfg: support device and driver lists Bruce Richardson
2026-04-29 16:58 ` [RFC PATCH 44/44] eal_cfg: add APIs for configuring remaining init settings Bruce Richardson
2026-04-29 21:40 ` [RFC PATCH 00/44] Allow intitializing EAL without argc/argv Stephen Hemminger
2026-04-29 22:04 ` Stephen Hemminger
2026-04-30  8:00   ` Bruce Richardson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260429165845.2136843-15-bruce.richardson@intel.com \
    --to=bruce.richardson@intel.com \
    --cc=dev@dpdk.org \
    --cc=techboard@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox