hw/nvme: Use ioeventfd to handle doorbell updates

Add property "ioeventfd" which is enabled by default. When this is
enabled, updates on the doorbell registers will cause KVM to signal
an event to the QEMU main loop to handle the doorbell updates.
Therefore, instead of letting the vcpu thread run both guest VM and
IO emulation, we now use the main loop thread to do IO emulation and
thus the vcpu thread has more cycles for the guest VM.

Since ioeventfd does not tell us the exact value that is written, it is
only useful when shadow doorbell buffer is enabled, where we check
for the value in the shadow doorbell buffer when we get the doorbell
update event.

IOPS comparison on Linux 5.19-rc2: (Unit: KIOPS)

qd           1   4  16  64
qemu        35 121 176 153
ioeventfd   41 133 258 313

Changes since v3:
 - Do not deregister ioeventfd when it was not enabled on a SQ/CQ

Signed-off-by: Jinhao Fan <fanjinhao21s@ict.ac.cn>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme/ctrl.c

@@ -1400,8 +1400,15 @@ static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
 
     QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
     QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);
+
+    if (req->sq->ioeventfd_enabled) {
+        /* Post CQE directly since we are in main loop thread */
+        nvme_post_cqes(cq);
+    } else {
+        /* Schedule the timer to post CQE later since we are in vcpu thread */
         timer_mod(cq->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + 500);
     }
+}
 
 static void nvme_process_aers(void *opaque)
 {
@@ -4226,10 +4233,82 @@ static uint16_t nvme_io_cmd(NvmeCtrl *n, NvmeRequest *req)
     return NVME_INVALID_OPCODE | NVME_DNR;
 }
 
+static void nvme_cq_notifier(EventNotifier *e)
+{
+    NvmeCQueue *cq = container_of(e, NvmeCQueue, notifier);
+    NvmeCtrl *n = cq->ctrl;
+
+    event_notifier_test_and_clear(&cq->notifier);
+
+    nvme_update_cq_head(cq);
+
+    if (cq->tail == cq->head) {
+        if (cq->irq_enabled) {
+            n->cq_pending--;
+        }
+
+        nvme_irq_deassert(n, cq);
+    }
+
+    nvme_post_cqes(cq);
+}
+
+static int nvme_init_cq_ioeventfd(NvmeCQueue *cq)
+{
+    NvmeCtrl *n = cq->ctrl;
+    uint16_t offset = (cq->cqid << 3) + (1 << 2);
+    int ret;
+
+    ret = event_notifier_init(&cq->notifier, 0);
+    if (ret < 0) {
+        return ret;
+    }
+
+    event_notifier_set_handler(&cq->notifier, nvme_cq_notifier);
+    memory_region_add_eventfd(&n->iomem,
+                              0x1000 + offset, 4, false, 0, &cq->notifier);
+
+    return 0;
+}
+
+static void nvme_sq_notifier(EventNotifier *e)
+{
+    NvmeSQueue *sq = container_of(e, NvmeSQueue, notifier);
+
+    event_notifier_test_and_clear(&sq->notifier);
+
+    nvme_process_sq(sq);
+}
+
+static int nvme_init_sq_ioeventfd(NvmeSQueue *sq)
+{
+    NvmeCtrl *n = sq->ctrl;
+    uint16_t offset = sq->sqid << 3;
+    int ret;
+
+    ret = event_notifier_init(&sq->notifier, 0);
+    if (ret < 0) {
+        return ret;
+    }
+
+    event_notifier_set_handler(&sq->notifier, nvme_sq_notifier);
+    memory_region_add_eventfd(&n->iomem,
+                              0x1000 + offset, 4, false, 0, &sq->notifier);
+
+    return 0;
+}
+
 static void nvme_free_sq(NvmeSQueue *sq, NvmeCtrl *n)
 {
+    uint16_t offset = sq->sqid << 3;
+
     n->sq[sq->sqid] = NULL;
     timer_free(sq->timer);
+    if (sq->ioeventfd_enabled) {
+        memory_region_del_eventfd(&n->iomem,
+                                  0x1000 + offset, 4, false, 0, &sq->notifier);
+        event_notifier_cleanup(&sq->notifier);
+    }
     g_free(sq->io_req);
     if (sq->sqid) {
         g_free(sq);
@@ -4302,6 +4381,12 @@ static void nvme_init_sq(NvmeSQueue *sq, NvmeCtrl *n, uint64_t dma_addr,
     if (n->dbbuf_enabled) {
         sq->db_addr = n->dbbuf_dbs + (sqid << 3);
         sq->ei_addr = n->dbbuf_eis + (sqid << 3);
+
+        if (n->params.ioeventfd && sq->sqid != 0) {
+            if (!nvme_init_sq_ioeventfd(sq)) {
+                sq->ioeventfd_enabled = true;
+            }
+        }
     }
 
     assert(n->cq[cqid]);
@@ -4605,8 +4690,15 @@ static uint16_t nvme_get_log(NvmeCtrl *n, NvmeRequest *req)
 
 static void nvme_free_cq(NvmeCQueue *cq, NvmeCtrl *n)
 {
+    uint16_t offset = (cq->cqid << 3) + (1 << 2);
+
     n->cq[cq->cqid] = NULL;
     timer_free(cq->timer);
+    if (cq->ioeventfd_enabled) {
+        memory_region_del_eventfd(&n->iomem,
+                                  0x1000 + offset, 4, false, 0, &cq->notifier);
+        event_notifier_cleanup(&cq->notifier);
+    }
     if (msix_enabled(&n->parent_obj)) {
         msix_vector_unuse(&n->parent_obj, cq->vector);
     }
@@ -4665,6 +4757,12 @@ static void nvme_init_cq(NvmeCQueue *cq, NvmeCtrl *n, uint64_t dma_addr,
     if (n->dbbuf_enabled) {
         cq->db_addr = n->dbbuf_dbs + (cqid << 3) + (1 << 2);
         cq->ei_addr = n->dbbuf_eis + (cqid << 3) + (1 << 2);
+
+        if (n->params.ioeventfd && cqid != 0) {
+            if (!nvme_init_cq_ioeventfd(cq)) {
+                cq->ioeventfd_enabled = true;
+            }
+        }
     }
     n->cq[cqid] = cq;
     cq->timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, nvme_post_cqes, cq);
@@ -6039,6 +6137,12 @@ static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
             sq->ei_addr = eis_addr + (i << 3);
             pci_dma_write(&n->parent_obj, sq->db_addr, &sq->tail,
                     sizeof(sq->tail));
+
+            if (n->params.ioeventfd && sq->sqid != 0) {
+                if (!nvme_init_sq_ioeventfd(sq)) {
+                    sq->ioeventfd_enabled = true;
+                }
+            }
         }
 
         if (cq) {
@@ -6047,6 +6151,12 @@ static uint16_t nvme_dbbuf_config(NvmeCtrl *n, const NvmeRequest *req)
             cq->ei_addr = eis_addr + (i << 3) + (1 << 2);
             pci_dma_write(&n->parent_obj, cq->db_addr, &cq->head,
                     sizeof(cq->head));
+
+            if (n->params.ioeventfd && cq->cqid != 0) {
+                if (!nvme_init_cq_ioeventfd(cq)) {
+                    cq->ioeventfd_enabled = true;
+                }
+            }
         }
     }
 
@@ -7554,6 +7664,7 @@ static Property nvme_props[] = {
     DEFINE_PROP_UINT8("vsl", NvmeCtrl, params.vsl, 7),
     DEFINE_PROP_BOOL("use-intel-id", NvmeCtrl, params.use_intel_id, false),
     DEFINE_PROP_BOOL("legacy-cmb", NvmeCtrl, params.legacy_cmb, false),
+    DEFINE_PROP_BOOL("ioeventfd", NvmeCtrl, params.ioeventfd, true),
     DEFINE_PROP_UINT8("zoned.zasl", NvmeCtrl, params.zasl, 0),
     DEFINE_PROP_BOOL("zoned.auto_transition", NvmeCtrl,
                      params.auto_transition_zones, true),

hw/nvme/nvme.h

@@ -376,6 +376,8 @@ typedef struct NvmeSQueue {
     uint64_t    db_addr;
     uint64_t    ei_addr;
     QEMUTimer   *timer;
+    EventNotifier notifier;
+    bool        ioeventfd_enabled;
     NvmeRequest *io_req;
     QTAILQ_HEAD(, NvmeRequest) req_list;
     QTAILQ_HEAD(, NvmeRequest) out_req_list;
@@ -395,6 +397,8 @@ typedef struct NvmeCQueue {
     uint64_t    db_addr;
     uint64_t    ei_addr;
     QEMUTimer   *timer;
+    EventNotifier notifier;
+    bool        ioeventfd_enabled;
     QTAILQ_HEAD(, NvmeSQueue) sq_list;
     QTAILQ_HEAD(, NvmeRequest) req_list;
 } NvmeCQueue;
@@ -417,6 +421,7 @@ typedef struct NvmeParams {
     uint8_t  zasl;
     bool     auto_transition_zones;
     bool     legacy_cmb;
+    bool     ioeventfd;
     uint8_t  sriov_max_vfs;
     uint16_t sriov_vq_flexible;
     uint16_t sriov_vi_flexible;