diff options
Diffstat (limited to 'block/nvme.c')
-rw-r--r-- | block/nvme.c | 73 |
1 files changed, 41 insertions, 32 deletions
diff --git a/block/nvme.c b/block/nvme.c index f4f27b6..b48f6f2 100644 --- a/block/nvme.c +++ b/block/nvme.c @@ -31,7 +31,7 @@ #define NVME_SQ_ENTRY_BYTES 64 #define NVME_CQ_ENTRY_BYTES 16 #define NVME_QUEUE_SIZE 128 -#define NVME_BAR_SIZE 8192 +#define NVME_DOORBELL_SIZE 4096 /* * We have to leave one slot empty as that is the full queue case where @@ -81,15 +81,6 @@ typedef struct { QEMUBH *completion_bh; } NVMeQueuePair; -/* Memory mapped registers */ -typedef volatile struct { - NvmeBar ctrl; - struct { - uint32_t sq_tail; - uint32_t cq_head; - } doorbells[]; -} NVMeRegs; - #define INDEX_ADMIN 0 #define INDEX_IO(n) (1 + n) @@ -102,7 +93,11 @@ enum { struct BDRVNVMeState { AioContext *aio_context; QEMUVFIOState *vfio; - NVMeRegs *regs; + /* Memory mapped registers */ + volatile struct { + uint32_t sq_tail; + uint32_t cq_head; + } *doorbells; /* The submission/completion queue pairs. * [0]: admin queue. * [1..]: io queues. @@ -247,14 +242,14 @@ static NVMeQueuePair *nvme_create_queue_pair(BDRVNVMeState *s, error_propagate(errp, local_err); goto fail; } - q->sq.doorbell = &s->regs->doorbells[idx * s->doorbell_scale].sq_tail; + q->sq.doorbell = &s->doorbells[idx * s->doorbell_scale].sq_tail; nvme_init_queue(s, &q->cq, size, NVME_CQ_ENTRY_BYTES, &local_err); if (local_err) { error_propagate(errp, local_err); goto fail; } - q->cq.doorbell = &s->regs->doorbells[idx * s->doorbell_scale].cq_head; + q->cq.doorbell = &s->doorbells[idx * s->doorbell_scale].cq_head; return q; fail: @@ -694,6 +689,7 @@ static int nvme_init(BlockDriverState *bs, const char *device, int namespace, uint64_t timeout_ms; uint64_t deadline, now; Error *local_err = NULL; + volatile NvmeBar *regs = NULL; qemu_co_mutex_init(&s->dma_map_lock); qemu_co_queue_init(&s->dma_flush_queue); @@ -712,32 +708,32 @@ static int nvme_init(BlockDriverState *bs, const char *device, int namespace, goto out; } - s->regs = qemu_vfio_pci_map_bar(s->vfio, 0, 0, NVME_BAR_SIZE, errp); - if (!s->regs) { + regs = qemu_vfio_pci_map_bar(s->vfio, 0, 0, sizeof(NvmeBar), + PROT_READ | PROT_WRITE, errp); + if (!regs) { ret = -EINVAL; goto out; } - /* Perform initialize sequence as described in NVMe spec "7.6.1 * Initialization". */ - cap = le64_to_cpu(s->regs->ctrl.cap); - if (!(cap & (1ULL << 37))) { + cap = le64_to_cpu(regs->cap); + if (!NVME_CAP_CSS(cap)) { error_setg(errp, "Device doesn't support NVMe command set"); ret = -EINVAL; goto out; } - s->page_size = MAX(4096, 1 << (12 + ((cap >> 48) & 0xF))); - s->doorbell_scale = (4 << (((cap >> 32) & 0xF))) / sizeof(uint32_t); + s->page_size = MAX(4096, 1 << NVME_CAP_MPSMIN(cap)); + s->doorbell_scale = (4 << NVME_CAP_DSTRD(cap)) / sizeof(uint32_t); bs->bl.opt_mem_alignment = s->page_size; - timeout_ms = MIN(500 * ((cap >> 24) & 0xFF), 30000); + timeout_ms = MIN(500 * NVME_CAP_TO(cap), 30000); /* Reset device to get a clean state. */ - s->regs->ctrl.cc = cpu_to_le32(le32_to_cpu(s->regs->ctrl.cc) & 0xFE); + regs->cc = cpu_to_le32(le32_to_cpu(regs->cc) & 0xFE); /* Wait for CSTS.RDY = 0. */ deadline = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + timeout_ms * SCALE_MS; - while (le32_to_cpu(s->regs->ctrl.csts) & 0x1) { + while (NVME_CSTS_RDY(le32_to_cpu(regs->csts))) { if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) > deadline) { error_setg(errp, "Timeout while waiting for device to reset (%" PRId64 " ms)", @@ -747,6 +743,13 @@ static int nvme_init(BlockDriverState *bs, const char *device, int namespace, } } + s->doorbells = qemu_vfio_pci_map_bar(s->vfio, 0, sizeof(NvmeBar), + NVME_DOORBELL_SIZE, PROT_WRITE, errp); + if (!s->doorbells) { + ret = -EINVAL; + goto out; + } + /* Set up admin queue. */ s->queues = g_new(NVMeQueuePair *, 1); s->queues[INDEX_ADMIN] = nvme_create_queue_pair(s, aio_context, 0, @@ -758,18 +761,19 @@ static int nvme_init(BlockDriverState *bs, const char *device, int namespace, } s->nr_queues = 1; QEMU_BUILD_BUG_ON(NVME_QUEUE_SIZE & 0xF000); - s->regs->ctrl.aqa = cpu_to_le32((NVME_QUEUE_SIZE << 16) | NVME_QUEUE_SIZE); - s->regs->ctrl.asq = cpu_to_le64(s->queues[INDEX_ADMIN]->sq.iova); - s->regs->ctrl.acq = cpu_to_le64(s->queues[INDEX_ADMIN]->cq.iova); + regs->aqa = cpu_to_le32((NVME_QUEUE_SIZE << AQA_ACQS_SHIFT) | + (NVME_QUEUE_SIZE << AQA_ASQS_SHIFT)); + regs->asq = cpu_to_le64(s->queues[INDEX_ADMIN]->sq.iova); + regs->acq = cpu_to_le64(s->queues[INDEX_ADMIN]->cq.iova); /* After setting up all control registers we can enable device now. */ - s->regs->ctrl.cc = cpu_to_le32((ctz32(NVME_CQ_ENTRY_BYTES) << 20) | - (ctz32(NVME_SQ_ENTRY_BYTES) << 16) | - 0x1); + regs->cc = cpu_to_le32((ctz32(NVME_CQ_ENTRY_BYTES) << CC_IOCQES_SHIFT) | + (ctz32(NVME_SQ_ENTRY_BYTES) << CC_IOSQES_SHIFT) | + CC_EN_MASK); /* Wait for CSTS.RDY = 1. */ now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); - deadline = now + timeout_ms * 1000000; - while (!(le32_to_cpu(s->regs->ctrl.csts) & 0x1)) { + deadline = now + timeout_ms * SCALE_MS; + while (!NVME_CSTS_RDY(le32_to_cpu(regs->csts))) { if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) > deadline) { error_setg(errp, "Timeout while waiting for device to start (%" PRId64 " ms)", @@ -800,6 +804,10 @@ static int nvme_init(BlockDriverState *bs, const char *device, int namespace, ret = -EIO; } out: + if (regs) { + qemu_vfio_pci_unmap_bar(s->vfio, 0, (void *)regs, 0, sizeof(NvmeBar)); + } + /* Cleaning up is done in nvme_file_open() upon error. */ return ret; } @@ -872,7 +880,8 @@ static void nvme_close(BlockDriverState *bs) &s->irq_notifier[MSIX_SHARED_IRQ_IDX], false, NULL, NULL); event_notifier_cleanup(&s->irq_notifier[MSIX_SHARED_IRQ_IDX]); - qemu_vfio_pci_unmap_bar(s->vfio, 0, (void *)s->regs, 0, NVME_BAR_SIZE); + qemu_vfio_pci_unmap_bar(s->vfio, 0, (void *)s->doorbells, + sizeof(NvmeBar), NVME_DOORBELL_SIZE); qemu_vfio_close(s->vfio); g_free(s->device); |