[dpdk-dev] [PATCH RFC] eal: change default per socket memory allocation

[David Marchand]

From: Didier Pallard <didier.pallard at 6wind.com>

Currently, if there is more memory in hugepages than the amount
requested by dpdk application, the memory is allocated by taking as much
memory as possible from each socket, starting from first one.
For example if a system is configured with 8 GB in 2 sockets (4 GB per
socket), and dpdk is requesting only 4GB of memory, all memory will be
taken in socket 0 (that have exactly 4GB of free hugepages) even if some
cores are configured on socket 1, and there are free hugepages on socket
1...

Change this behaviour to allocate memory on all sockets where some cores
are configured, spreading the memory amongst sockets using following
ratio per socket:
N° of cores configured on the socket / Total number of configured cores
* requested memory

This algorithm is used when memory amount is specified globally using
-m option. Per socket memory allocation can always be done using
--socket-mem option.

Signed-off-by: Didier Pallard <didier.pallard at 6wind.com>
---
 lib/librte_eal/bsdapp/eal/eal.c          |   24 ++++++++++++++++++++++++
 lib/librte_eal/linuxapp/eal/eal.c        |   24 ++++++++++++++++++++++++
 lib/librte_eal/linuxapp/eal/eal_memory.c |    8 ++++----
 3 files changed, 52 insertions(+), 4 deletions(-)

diff --git a/lib/librte_eal/bsdapp/eal/eal.c b/lib/librte_eal/bsdapp/eal/eal.c
index 5c181b3..b6b5f20 100644
--- a/lib/librte_eal/bsdapp/eal/eal.c
+++ b/lib/librte_eal/bsdapp/eal/eal.c
@@ -869,6 +869,30 @@ rte_eal_init(int argc, char **argv)
 			internal_config.memory = eal_get_hugepage_mem_size();
 	}
 
+	/* Automatically spread requested memory amongst detected sockets according */
+	/* to number of cores from cpu mask present on each socket */
+	if (internal_config.no_hugetlbfs == 0 &&
+	    internal_config.process_type != RTE_PROC_SECONDARY &&
+	    internal_config.xen_dom0_support == 0 &&
+	    internal_config.force_sockets == 0) {
+		int cpu_per_socket[RTE_MAX_NUMA_NODES];
+		unsigned lcore_id, socket_id;
+
+		/* Compute number of cores per socket */
+		memset(cpu_per_socket, 0, sizeof(cpu_per_socket));
+		RTE_LCORE_FOREACH(lcore_id) {
+			cpu_per_socket[rte_lcore_to_socket_id(lcore_id)]++;
+		}
+
+		/* Set memory amount per socket; round up to be sure that sum of all */
+		/* sockets allocation is greater than requested memory size */
+		for (socket_id=0 ; socket_id<RTE_MAX_NUMA_NODES ; socket_id++) {
+			internal_config.socket_mem[socket_id] = (internal_config.memory *
+			    cpu_per_socket[socket_id] + rte_lcore_count() - 1) /
+			    rte_lcore_count();
+		}
+	}
+
 	if (internal_config.vmware_tsc_map == 1) {
 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
 		rte_cycles_vmware_tsc_map = 1;
diff --git a/lib/librte_eal/linuxapp/eal/eal.c b/lib/librte_eal/linuxapp/eal/eal.c
index de082ab..37701ec 100644
--- a/lib/librte_eal/linuxapp/eal/eal.c
+++ b/lib/librte_eal/linuxapp/eal/eal.c
@@ -1035,6 +1035,30 @@ rte_eal_init(int argc, char **argv)
 			internal_config.memory = eal_get_hugepage_mem_size();
 	}
 
+	/* Automatically spread requested memory amongst detected sockets according */
+	/* to number of cores from cpu mask present on each socket */
+	if (internal_config.no_hugetlbfs == 0 &&
+	    internal_config.process_type != RTE_PROC_SECONDARY &&
+	    internal_config.xen_dom0_support == 0 &&
+	    internal_config.force_sockets == 0) {
+		int cpu_per_socket[RTE_MAX_NUMA_NODES];
+		unsigned lcore_id, socket_id;
+
+		/* Compute number of cores per socket */
+		memset(cpu_per_socket, 0, sizeof(cpu_per_socket));
+		RTE_LCORE_FOREACH(lcore_id) {
+			cpu_per_socket[rte_lcore_to_socket_id(lcore_id)]++;
+		}
+
+		/* Set memory amount per socket; round up to be sure that sum of all */
+		/* sockets allocation is greater than requested memory size */
+		for (socket_id=0 ; socket_id<RTE_MAX_NUMA_NODES ; socket_id++) {
+			internal_config.socket_mem[socket_id] = (internal_config.memory *
+			    cpu_per_socket[socket_id] + rte_lcore_count() - 1) /
+			    rte_lcore_count();
+		}
+	}
+
 	if (internal_config.vmware_tsc_map == 1) {
 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
 		rte_cycles_vmware_tsc_map = 1;
diff --git a/lib/librte_eal/linuxapp/eal/eal_memory.c b/lib/librte_eal/linuxapp/eal/eal_memory.c
index 73a6394..a9e6fa1 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memory.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memory.c
@@ -876,16 +876,16 @@ calc_num_pages_per_socket(uint64_t * memory,
 	unsigned requested, available;
 	int total_num_pages = 0;
 	uint64_t remaining_mem, cur_mem;
-	uint64_t total_mem = internal_config.memory;
+	int64_t total_mem = (int64_t) internal_config.memory;
 
 	if (num_hp_info == 0)
 		return -1;
 
-	for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_mem != 0; socket++) {
+	for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_mem > 0; socket++) {
 		/* if specific memory amounts per socket weren't requested */
-		if (internal_config.force_sockets == 0) {
+		if (internal_config.force_sockets == 0 && memory[socket] == 0) {
 			/* take whatever is available */
-			memory[socket] = RTE_MIN(get_socket_mem_size(socket),
+			memory[socket] = RTE_MIN((int64_t) get_socket_mem_size(socket),
 					total_mem);
 		}
 		/* skips if the memory on specific socket wasn't requested */
-- 
1.7.10.4