diff options
author | Raghu Vatsavayi <rvatsavayi@caviumnetworks.com> | 2016-06-14 16:54:46 -0700 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2016-06-15 21:44:31 -0700 |
commit | 96ae48b7faa72260c9d7203f5ad6250e149fb085 (patch) | |
tree | 75e168613cb902dd853e25e3c18f041f25615a3b /drivers/net/ethernet | |
parent | fcd2b5e36ca91eae329d33a50977ee976122a3b9 (diff) | |
download | linux-96ae48b7faa72260c9d7203f5ad6250e149fb085.tar.bz2 |
liquidio:RX queue alloc changes
This patch is to allocate rx queue's memory based on numa node and also use
page based buffers for rx traffic improvements.
Signed-off-by: Derek Chickles <derek.chickles@caviumnetworks.com>
Signed-off-by: Satanand Burla <satananda.burla@caviumnetworks.com>
Signed-off-by: Felix Manlunas <felix.manlunas@caviumnetworks.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'drivers/net/ethernet')
-rw-r--r-- | drivers/net/ethernet/cavium/liquidio/octeon_device.c | 27 | ||||
-rw-r--r-- | drivers/net/ethernet/cavium/liquidio/octeon_droq.c | 35 | ||||
-rw-r--r-- | drivers/net/ethernet/cavium/liquidio/octeon_main.h | 23 |
3 files changed, 52 insertions, 33 deletions
diff --git a/drivers/net/ethernet/cavium/liquidio/octeon_device.c b/drivers/net/ethernet/cavium/liquidio/octeon_device.c index 967fe4dfd573..c06807d9548e 100644 --- a/drivers/net/ethernet/cavium/liquidio/octeon_device.c +++ b/drivers/net/ethernet/cavium/liquidio/octeon_device.c @@ -783,14 +783,15 @@ int octeon_setup_instr_queues(struct octeon_device *oct) int octeon_setup_output_queues(struct octeon_device *oct) { - u32 i, num_oqs = 0; + u32 num_oqs = 0; u32 num_descs = 0; u32 desc_size = 0; + u32 oq_no = 0; + int numa_node = cpu_to_node(oq_no % num_online_cpus()); + num_oqs = 1; /* this causes queue 0 to be default queue */ if (OCTEON_CN6XXX(oct)) { - /* CFG_GET_OQ_MAX_BASE_Q(CHIP_FIELD(oct, cn6xxx, conf)); */ - num_oqs = 1; num_descs = CFG_GET_NUM_DEF_RX_DESCS(CHIP_FIELD(oct, cn6xxx, conf)); desc_size = @@ -798,19 +799,15 @@ int octeon_setup_output_queues(struct octeon_device *oct) } oct->num_oqs = 0; + oct->droq[0] = vmalloc_node(sizeof(*oct->droq[0]), numa_node); + if (!oct->droq[0]) + oct->droq[0] = vmalloc(sizeof(*oct->droq[0])); + if (!oct->droq[0]) + return 1; - for (i = 0; i < num_oqs; i++) { - oct->droq[i] = vmalloc(sizeof(*oct->droq[i])); - if (!oct->droq[i]) - return 1; - - memset(oct->droq[i], 0, sizeof(struct octeon_droq)); - - if (octeon_init_droq(oct, i, num_descs, desc_size, NULL)) - return 1; - - oct->num_oqs++; - } + if (octeon_init_droq(oct, oq_no, num_descs, desc_size, NULL)) + return 1; + oct->num_oqs++; return 0; } diff --git a/drivers/net/ethernet/cavium/liquidio/octeon_droq.c b/drivers/net/ethernet/cavium/liquidio/octeon_droq.c index 174072b3740b..1f648dc4f96e 100644 --- a/drivers/net/ethernet/cavium/liquidio/octeon_droq.c +++ b/drivers/net/ethernet/cavium/liquidio/octeon_droq.c @@ -242,6 +242,8 @@ int octeon_init_droq(struct octeon_device *oct, struct octeon_droq *droq; u32 desc_ring_size = 0, c_num_descs = 0, c_buf_size = 0; u32 c_pkts_per_intr = 0, c_refill_threshold = 0; + int orig_node = dev_to_node(&oct->pci_dev->dev); + int numa_node = cpu_to_node(q_no % num_online_cpus()); dev_dbg(&oct->pci_dev->dev, "%s[%d]\n", __func__, q_no); @@ -261,15 +263,23 @@ int octeon_init_droq(struct octeon_device *oct, struct octeon_config *conf6x = CHIP_FIELD(oct, cn6xxx, conf); c_pkts_per_intr = (u32)CFG_GET_OQ_PKTS_PER_INTR(conf6x); - c_refill_threshold = (u32)CFG_GET_OQ_REFILL_THRESHOLD(conf6x); + c_refill_threshold = + (u32)CFG_GET_OQ_REFILL_THRESHOLD(conf6x); + } else { + return 1; } droq->max_count = c_num_descs; droq->buffer_size = c_buf_size; desc_ring_size = droq->max_count * OCT_DROQ_DESC_SIZE; + set_dev_node(&oct->pci_dev->dev, numa_node); droq->desc_ring = lio_dma_alloc(oct, desc_ring_size, (dma_addr_t *)&droq->desc_ring_dma); + set_dev_node(&oct->pci_dev->dev, orig_node); + if (!droq->desc_ring) + droq->desc_ring = lio_dma_alloc(oct, desc_ring_size, + (dma_addr_t *)&droq->desc_ring_dma); if (!droq->desc_ring) { dev_err(&oct->pci_dev->dev, @@ -283,12 +293,11 @@ int octeon_init_droq(struct octeon_device *oct, droq->max_count); droq->info_list = - cnnic_alloc_aligned_dma(oct->pci_dev, - (droq->max_count * OCT_DROQ_INFO_SIZE), - &droq->info_alloc_size, - &droq->info_base_addr, - &droq->info_list_dma); - + cnnic_numa_alloc_aligned_dma((droq->max_count * + OCT_DROQ_INFO_SIZE), + &droq->info_alloc_size, + &droq->info_base_addr, + numa_node); if (!droq->info_list) { dev_err(&oct->pci_dev->dev, "Cannot allocate memory for info list.\n"); lio_dma_free(oct, (droq->max_count * OCT_DROQ_DESC_SIZE), @@ -297,7 +306,12 @@ int octeon_init_droq(struct octeon_device *oct, } droq->recv_buf_list = (struct octeon_recv_buffer *) - vmalloc(droq->max_count * + vmalloc_node(droq->max_count * + OCT_DROQ_RECVBUF_SIZE, + numa_node); + if (!droq->recv_buf_list) + droq->recv_buf_list = (struct octeon_recv_buffer *) + vmalloc(droq->max_count * OCT_DROQ_RECVBUF_SIZE); if (!droq->recv_buf_list) { dev_err(&oct->pci_dev->dev, "Output queue recv buf list alloc failed\n"); @@ -949,6 +963,7 @@ int octeon_create_droq(struct octeon_device *oct, u32 desc_size, void *app_ctx) { struct octeon_droq *droq; + int numa_node = cpu_to_node(q_no % num_online_cpus()); if (oct->droq[q_no]) { dev_dbg(&oct->pci_dev->dev, "Droq already in use. Cannot create droq %d again\n", @@ -957,7 +972,9 @@ int octeon_create_droq(struct octeon_device *oct, } /* Allocate the DS for the new droq. */ - droq = vmalloc(sizeof(*droq)); + droq = vmalloc_node(sizeof(*droq), numa_node); + if (!droq) + droq = vmalloc(sizeof(*droq)); if (!droq) goto create_droq_fail; memset(droq, 0, sizeof(struct octeon_droq)); diff --git a/drivers/net/ethernet/cavium/liquidio/octeon_main.h b/drivers/net/ethernet/cavium/liquidio/octeon_main.h index cbd081981180..0ff3efc67b84 100644 --- a/drivers/net/ethernet/cavium/liquidio/octeon_main.h +++ b/drivers/net/ethernet/cavium/liquidio/octeon_main.h @@ -126,22 +126,27 @@ static inline int octeon_map_pci_barx(struct octeon_device *oct, } static inline void * -cnnic_alloc_aligned_dma(struct pci_dev *pci_dev, - u32 size, - u32 *alloc_size, - size_t *orig_ptr, - size_t *dma_addr __attribute__((unused))) +cnnic_numa_alloc_aligned_dma(u32 size, + u32 *alloc_size, + size_t *orig_ptr, + int numa_node) { int retries = 0; void *ptr = NULL; #define OCTEON_MAX_ALLOC_RETRIES 1 do { - ptr = - (void *)__get_free_pages(GFP_KERNEL, - get_order(size)); + struct page *page = NULL; + + page = alloc_pages_node(numa_node, + GFP_KERNEL, + get_order(size)); + if (!page) + page = alloc_pages(GFP_KERNEL, + get_order(size)); + ptr = (void *)page_address(page); if ((unsigned long)ptr & 0x07) { - free_pages((unsigned long)ptr, get_order(size)); + __free_pages(page, get_order(size)); ptr = NULL; /* Increment the size required if the first * attempt failed. |