Use node based allocations for queues. consider the dma address alignment changes, while calculating the queue base address. added checks in cleanup functions. Minor changes to queue variable names Signed-off-by: Srikanth Jampala <Jampala.Srikanth@xxxxxxxxxx> Reviewed-by: Gadam Sreerama <sgadam@xxxxxxxxxx> --- drivers/crypto/cavium/nitrox/nitrox_dev.h | 55 ++++++------ drivers/crypto/cavium/nitrox/nitrox_hal.c | 2 +- drivers/crypto/cavium/nitrox/nitrox_isr.c | 2 +- drivers/crypto/cavium/nitrox/nitrox_lib.c | 90 +++++++++++--------- drivers/crypto/cavium/nitrox/nitrox_reqmgr.c | 24 +++--- 5 files changed, 94 insertions(+), 79 deletions(-) diff --git a/drivers/crypto/cavium/nitrox/nitrox_dev.h b/drivers/crypto/cavium/nitrox/nitrox_dev.h index abb5c8ac54e5..62fb75923bfe 100644 --- a/drivers/crypto/cavium/nitrox/nitrox_dev.h +++ b/drivers/crypto/cavium/nitrox/nitrox_dev.h @@ -9,44 +9,51 @@ #define VERSION_LEN 32 +/** + * struct nitrox_cmdq - NITROX command queue + * @cmd_qlock: command queue lock + * @resp_qlock: response queue lock + * @backlog_qlock: backlog queue lock + * @ndev: NITROX device + * @response_head: submitted request list + * @backlog_head: backlog queue + * @dbell_csr_addr: doorbell register address for this queue + * @base: command queue base address + * @dma: dma address of the base + * @pending_count: request pending at device + * @backlog_count: backlog request count + * @write_idx: next write index for the command + * @instr_size: command size + * @qno: command queue number + * @qsize: command queue size + * @unalign_base: unaligned base address + * @unalign_dma: unaligned dma address + */ struct nitrox_cmdq { - /* command queue lock */ - spinlock_t cmdq_lock; - /* response list lock */ - spinlock_t response_lock; - /* backlog list lock */ - spinlock_t backlog_lock; - - /* request submitted to chip, in progress */ + spinlock_t cmd_qlock; + spinlock_t resp_qlock; + spinlock_t backlog_qlock; + + struct nitrox_device *ndev; struct list_head response_head; - /* hw queue full, hold in backlog list */ struct list_head backlog_head; - /* doorbell address */ u8 __iomem *dbell_csr_addr; - /* base address of the queue */ - u8 *head; + u8 *base; + dma_addr_t dma; - struct nitrox_device *ndev; - /* flush pending backlog commands */ struct work_struct backlog_qflush; - /* requests posted waiting for completion */ atomic_t pending_count; - /* requests in backlog queues */ atomic_t backlog_count; int write_idx; - /* command size 32B/64B */ u8 instr_size; u8 qno; u32 qsize; - /* unaligned addresses */ - u8 *head_unaligned; - dma_addr_t dma_unaligned; - /* dma address of the base */ - dma_addr_t dma; + u8 *unalign_base; + dma_addr_t unalign_dma; }; /** @@ -152,7 +159,7 @@ enum vf_mode { * @nr_queues: Number of command queues * @mode: Device mode PF/VF * @ctx_pool: DMA pool for crypto context - * @pkt_cmdqs: SE Command queues + * @pkt_inq: Packet input rings * @msix: MSI-X information * @bh: post processing work * @hw: hardware information @@ -177,7 +184,7 @@ struct nitrox_device { enum vf_mode mode; struct dma_pool *ctx_pool; - struct nitrox_cmdq *pkt_cmdqs; + struct nitrox_cmdq *pkt_inq; struct nitrox_msix msix; struct nitrox_bh bh; diff --git a/drivers/crypto/cavium/nitrox/nitrox_hal.c b/drivers/crypto/cavium/nitrox/nitrox_hal.c index 4627a1976176..a9b82387cf53 100644 --- a/drivers/crypto/cavium/nitrox/nitrox_hal.c +++ b/drivers/crypto/cavium/nitrox/nitrox_hal.c @@ -119,7 +119,7 @@ void nitrox_config_pkt_input_rings(struct nitrox_device *ndev) int i; for (i = 0; i < ndev->nr_queues; i++) { - struct nitrox_cmdq *cmdq = &ndev->pkt_cmdqs[i]; + struct nitrox_cmdq *cmdq = &ndev->pkt_inq[i]; union nps_pkt_in_instr_rsize pkt_in_rsize; u64 offset; diff --git a/drivers/crypto/cavium/nitrox/nitrox_isr.c b/drivers/crypto/cavium/nitrox/nitrox_isr.c index 4f4334929f35..baeacacbddbe 100644 --- a/drivers/crypto/cavium/nitrox/nitrox_isr.c +++ b/drivers/crypto/cavium/nitrox/nitrox_isr.c @@ -335,7 +335,7 @@ static int nitrox_setup_pkt_slc_bh(struct nitrox_device *ndev) offset = NPS_PKT_SLC_CNTSX(i); /* pre calculate completion count address */ bh->completion_cnt_csr_addr = NITROX_CSR_ADDR(ndev, offset); - bh->cmdq = &ndev->pkt_cmdqs[i]; + bh->cmdq = &ndev->pkt_inq[i]; tasklet_init(&bh->resp_handler, pkt_slc_resp_handler, (unsigned long)bh); diff --git a/drivers/crypto/cavium/nitrox/nitrox_lib.c b/drivers/crypto/cavium/nitrox/nitrox_lib.c index 7a721c7c12a4..ed08aa9380ee 100644 --- a/drivers/crypto/cavium/nitrox/nitrox_lib.c +++ b/drivers/crypto/cavium/nitrox/nitrox_lib.c @@ -17,30 +17,27 @@ #define CRYPTO_CTX_SIZE 256 -/* command queue alignments */ -#define PKT_IN_ALIGN 16 +/* packet inuput ring alignments */ +#define PKTIN_Q_ALIGN_BYTES 16 -static int cmdq_common_init(struct nitrox_cmdq *cmdq) +static int nitrox_cmdq_init(struct nitrox_cmdq *cmdq, int align_bytes) { struct nitrox_device *ndev = cmdq->ndev; - u32 qsize; - - qsize = (ndev->qlen) * cmdq->instr_size; - cmdq->head_unaligned = dma_zalloc_coherent(DEV(ndev), - (qsize + PKT_IN_ALIGN), - &cmdq->dma_unaligned, - GFP_KERNEL); - if (!cmdq->head_unaligned) + + cmdq->qsize = (ndev->qlen * cmdq->instr_size) + align_bytes; + cmdq->unalign_base = dma_zalloc_coherent(DEV(ndev), cmdq->qsize, + &cmdq->unalign_dma, + GFP_KERNEL); + if (!cmdq->unalign_base) return -ENOMEM; - cmdq->head = PTR_ALIGN(cmdq->head_unaligned, PKT_IN_ALIGN); - cmdq->dma = PTR_ALIGN(cmdq->dma_unaligned, PKT_IN_ALIGN); - cmdq->qsize = (qsize + PKT_IN_ALIGN); + cmdq->dma = PTR_ALIGN(cmdq->unalign_dma, align_bytes); + cmdq->base = cmdq->unalign_base + (cmdq->dma - cmdq->unalign_dma); cmdq->write_idx = 0; - spin_lock_init(&cmdq->response_lock); - spin_lock_init(&cmdq->cmdq_lock); - spin_lock_init(&cmdq->backlog_lock); + spin_lock_init(&cmdq->cmd_qlock); + spin_lock_init(&cmdq->resp_qlock); + spin_lock_init(&cmdq->backlog_qlock); INIT_LIST_HEAD(&cmdq->response_head); INIT_LIST_HEAD(&cmdq->backlog_head); @@ -51,68 +48,79 @@ static int cmdq_common_init(struct nitrox_cmdq *cmdq) return 0; } -static void cmdq_common_cleanup(struct nitrox_cmdq *cmdq) +static void nitrox_cmdq_reset(struct nitrox_cmdq *cmdq) +{ + cmdq->write_idx = 0; + atomic_set(&cmdq->pending_count, 0); + atomic_set(&cmdq->backlog_count, 0); +} + +static void nitrox_cmdq_cleanup(struct nitrox_cmdq *cmdq) { struct nitrox_device *ndev = cmdq->ndev; + if (!cmdq->unalign_base) + return; + cancel_work_sync(&cmdq->backlog_qflush); dma_free_coherent(DEV(ndev), cmdq->qsize, - cmdq->head_unaligned, cmdq->dma_unaligned); - - atomic_set(&cmdq->pending_count, 0); - atomic_set(&cmdq->backlog_count, 0); + cmdq->unalign_base, cmdq->unalign_dma); + nitrox_cmdq_reset(cmdq); cmdq->dbell_csr_addr = NULL; - cmdq->head = NULL; + cmdq->unalign_base = NULL; + cmdq->base = NULL; + cmdq->unalign_dma = 0; cmdq->dma = 0; cmdq->qsize = 0; cmdq->instr_size = 0; } -static void nitrox_cleanup_pkt_cmdqs(struct nitrox_device *ndev) +static void nitrox_free_pktin_queues(struct nitrox_device *ndev) { int i; for (i = 0; i < ndev->nr_queues; i++) { - struct nitrox_cmdq *cmdq = &ndev->pkt_cmdqs[i]; + struct nitrox_cmdq *cmdq = &ndev->pkt_inq[i]; - cmdq_common_cleanup(cmdq); + nitrox_cmdq_cleanup(cmdq); } - kfree(ndev->pkt_cmdqs); - ndev->pkt_cmdqs = NULL; + kfree(ndev->pkt_inq); + ndev->pkt_inq = NULL; } -static int nitrox_init_pkt_cmdqs(struct nitrox_device *ndev) +static int nitrox_alloc_pktin_queues(struct nitrox_device *ndev) { - int i, err, size; + int i, err; - size = ndev->nr_queues * sizeof(struct nitrox_cmdq); - ndev->pkt_cmdqs = kzalloc(size, GFP_KERNEL); - if (!ndev->pkt_cmdqs) + ndev->pkt_inq = kcalloc_node(ndev->nr_queues, + sizeof(struct nitrox_cmdq), + GFP_KERNEL, ndev->node); + if (!ndev->pkt_inq) return -ENOMEM; for (i = 0; i < ndev->nr_queues; i++) { struct nitrox_cmdq *cmdq; u64 offset; - cmdq = &ndev->pkt_cmdqs[i]; + cmdq = &ndev->pkt_inq[i]; cmdq->ndev = ndev; cmdq->qno = i; cmdq->instr_size = sizeof(struct nps_pkt_instr); + /* packet input ring doorbell address */ offset = NPS_PKT_IN_INSTR_BAOFF_DBELLX(i); - /* SE ring doorbell address for this queue */ cmdq->dbell_csr_addr = NITROX_CSR_ADDR(ndev, offset); - err = cmdq_common_init(cmdq); + err = nitrox_cmdq_init(cmdq, PKTIN_Q_ALIGN_BYTES); if (err) - goto pkt_cmdq_fail; + goto pktq_fail; } return 0; -pkt_cmdq_fail: - nitrox_cleanup_pkt_cmdqs(ndev); +pktq_fail: + nitrox_free_pktin_queues(ndev); return err; } @@ -194,7 +202,7 @@ int nitrox_common_sw_init(struct nitrox_device *ndev) if (err) return err; - err = nitrox_init_pkt_cmdqs(ndev); + err = nitrox_alloc_pktin_queues(ndev); if (err) destroy_crypto_dma_pool(ndev); @@ -207,6 +215,6 @@ int nitrox_common_sw_init(struct nitrox_device *ndev) */ void nitrox_common_sw_cleanup(struct nitrox_device *ndev) { - nitrox_cleanup_pkt_cmdqs(ndev); + nitrox_free_pktin_queues(ndev); destroy_crypto_dma_pool(ndev); } diff --git a/drivers/crypto/cavium/nitrox/nitrox_reqmgr.c b/drivers/crypto/cavium/nitrox/nitrox_reqmgr.c index f325b48ba904..d606c7af4667 100644 --- a/drivers/crypto/cavium/nitrox/nitrox_reqmgr.c +++ b/drivers/crypto/cavium/nitrox/nitrox_reqmgr.c @@ -382,11 +382,11 @@ static inline void backlog_list_add(struct nitrox_softreq *sr, { INIT_LIST_HEAD(&sr->backlog); - spin_lock_bh(&cmdq->backlog_lock); + spin_lock_bh(&cmdq->backlog_qlock); list_add_tail(&sr->backlog, &cmdq->backlog_head); atomic_inc(&cmdq->backlog_count); atomic_set(&sr->status, REQ_BACKLOG); - spin_unlock_bh(&cmdq->backlog_lock); + spin_unlock_bh(&cmdq->backlog_qlock); } static inline void response_list_add(struct nitrox_softreq *sr, @@ -394,17 +394,17 @@ static inline void response_list_add(struct nitrox_softreq *sr, { INIT_LIST_HEAD(&sr->response); - spin_lock_bh(&cmdq->response_lock); + spin_lock_bh(&cmdq->resp_qlock); list_add_tail(&sr->response, &cmdq->response_head); - spin_unlock_bh(&cmdq->response_lock); + spin_unlock_bh(&cmdq->resp_qlock); } static inline void response_list_del(struct nitrox_softreq *sr, struct nitrox_cmdq *cmdq) { - spin_lock_bh(&cmdq->response_lock); + spin_lock_bh(&cmdq->resp_qlock); list_del(&sr->response); - spin_unlock_bh(&cmdq->response_lock); + spin_unlock_bh(&cmdq->resp_qlock); } static struct nitrox_softreq * @@ -439,11 +439,11 @@ static void post_se_instr(struct nitrox_softreq *sr, int idx; u8 *ent; - spin_lock_bh(&cmdq->cmdq_lock); + spin_lock_bh(&cmdq->cmd_qlock); idx = cmdq->write_idx; /* copy the instruction */ - ent = cmdq->head + (idx * cmdq->instr_size); + ent = cmdq->base + (idx * cmdq->instr_size); memcpy(ent, &sr->instr, cmdq->instr_size); atomic_set(&sr->status, REQ_POSTED); @@ -459,7 +459,7 @@ static void post_se_instr(struct nitrox_softreq *sr, cmdq->write_idx = incr_index(idx, 1, ndev->qlen); - spin_unlock_bh(&cmdq->cmdq_lock); + spin_unlock_bh(&cmdq->cmd_qlock); /* increment the posted command count */ atomic64_inc(&ndev->stats.posted); @@ -474,7 +474,7 @@ static int post_backlog_cmds(struct nitrox_cmdq *cmdq) if (!atomic_read(&cmdq->backlog_count)) return 0; - spin_lock_bh(&cmdq->backlog_lock); + spin_lock_bh(&cmdq->backlog_qlock); list_for_each_entry_safe(sr, tmp, &cmdq->backlog_head, backlog) { struct skcipher_request *skreq; @@ -497,7 +497,7 @@ static int post_backlog_cmds(struct nitrox_cmdq *cmdq) /* backlog requests are posted, wakeup with -EINPROGRESS */ skcipher_request_complete(skreq, -EINPROGRESS); } - spin_unlock_bh(&cmdq->backlog_lock); + spin_unlock_bh(&cmdq->backlog_qlock); return ret; } @@ -578,7 +578,7 @@ int nitrox_process_se_request(struct nitrox_device *ndev, /* select the queue */ qno = smp_processor_id() % ndev->nr_queues; - sr->cmdq = &ndev->pkt_cmdqs[qno]; + sr->cmdq = &ndev->pkt_inq[qno]; /* * 64-Byte Instruction Format -- 2.17.1