[PATCH RFC 3/4] hw/block/nvme: convert flush to aiocb

[Klaus Jensen]

From: Klaus Jensen <k.jensen@samsung.com>

Convert flush from ad-hoc multi aio tracking to use standard QEMU AIOCB
processing.

Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
---
 hw/block/nvme.c | 143 +++++++++++++++++++++++++-----------------------
 1 file changed, 74 insertions(+), 69 deletions(-)

diff --git a/hw/block/nvme.c b/hw/block/nvme.c
index 48a1abe52787..773b41527c79 100644
--- a/hw/block/nvme.c
+++ b/hw/block/nvme.c
@@ -1459,41 +1459,6 @@ static void nvme_rw_cb(void *opaque, int ret)
     nvme_enqueue_req_completion(nvme_cq(req), req);
 }
 
-struct nvme_aio_flush_ctx {
-    NvmeRequest     *req;
-    NvmeNamespace   *ns;
-    BlockAcctCookie acct;
-};
-
-static void nvme_aio_flush_cb(void *opaque, int ret)
-{
-    struct nvme_aio_flush_ctx *ctx = opaque;
-    NvmeRequest *req = ctx->req;
-    uintptr_t *num_flushes = (uintptr_t *)&req->opaque;
-
-    BlockBackend *blk = ctx->ns->blkconf.blk;
-    BlockAcctCookie *acct = &ctx->acct;
-    BlockAcctStats *stats = blk_get_stats(blk);
-
-    trace_pci_nvme_aio_flush_cb(nvme_cid(req), blk_name(blk));
-
-    if (!ret) {
-        block_acct_done(stats, acct);
-    } else {
-        block_acct_failed(stats, acct);
-        nvme_aio_err(req, ret);
-    }
-
-    (*num_flushes)--;
-    g_free(ctx);
-
-    if (*num_flushes) {
-        return;
-    }
-
-    nvme_enqueue_req_completion(nvme_cq(req), req);
-}
-
 static void nvme_misc_cb(void *opaque, int ret)
 {
     NvmeRequest *req = opaque;
@@ -2055,13 +2020,74 @@ static uint16_t nvme_compare(NvmeCtrl *n, NvmeRequest 
*req)
     return NVME_NO_COMPLETE;
 }
 
+typedef struct NvmeFlushAIOCB {
+    BlockAIOCB common;
+    BlockAIOCB *aiocb;
+    NvmeRequest *req;
+    QEMUBH *bh;
+    int ret;
+
+    int nsid;
+} NvmeFlushAIOCB;
+
+static void nvme_flush_cancel(BlockAIOCB *acb)
+{
+    NvmeFlushAIOCB *iocb = container_of(acb, NvmeFlushAIOCB, common);
+    NvmeCtrl *n = nvme_ctrl(iocb->req);
+
+    iocb->nsid = n->num_namespaces + 1;
+    iocb->ret = -ECANCELED;
+
+    if (iocb->aiocb) {
+        blk_aio_cancel_async(iocb->aiocb);
+        iocb->aiocb = NULL;
+    }
+}
+
+static const AIOCBInfo nvme_flush_aiocb_info = {
+    .aiocb_size = sizeof(NvmeFlushAIOCB),
+    .cancel_async = nvme_flush_cancel,
+};
+
+static void nvme_flush_bh(void *opaque)
+{
+    NvmeFlushAIOCB *iocb = opaque;
+
+    iocb->common.cb(iocb->common.opaque, iocb->ret);
+    qemu_bh_delete(iocb->bh);
+    iocb->bh = NULL;
+    qemu_aio_unref(iocb);
+}
+
+static void nvme_flush_aio_cb(void *opaque, int ret)
+{
+    NvmeFlushAIOCB *iocb = opaque;
+    NvmeRequest *req = iocb->req;
+    NvmeCtrl *n = nvme_ctrl(req);
+
+    if (ret < 0) {
+        iocb->ret = ret;
+        goto done;
+    }
+
+    while (iocb->nsid <= n->num_namespaces) {
+        NvmeNamespace *ns = nvme_ns(n, iocb->nsid++);
+        if (ns) {
+            iocb->aiocb = blk_aio_flush(ns->blkconf.blk, nvme_flush_aio_cb,
+                                        iocb);
+            return;
+        }
+    }
+
+done:
+    iocb->aiocb = NULL;
+    qemu_bh_schedule(iocb->bh);
+}
+
 static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest *req)
 {
+    NvmeFlushAIOCB *iocb;
     uint32_t nsid = le32_to_cpu(req->cmd.nsid);
-    uintptr_t *num_flushes = (uintptr_t *)&req->opaque;
-    uint16_t status;
-    struct nvme_aio_flush_ctx *ctx;
-    NvmeNamespace *ns;
 
     trace_pci_nvme_flush(nvme_cid(req), nsid);
 
@@ -2071,42 +2097,21 @@ static uint16_t nvme_flush(NvmeCtrl *n, NvmeRequest 
*req)
             return NVME_INVALID_FIELD | NVME_DNR;
         }
 
-        block_acct_start(blk_get_stats(req->ns->blkconf.blk), &req->acct, 0,
-                         BLOCK_ACCT_FLUSH);
-        req->aiocb = blk_aio_flush(req->ns->blkconf.blk, nvme_rw_cb, req);
+        req->aiocb = blk_aio_flush(req->ns->blkconf.blk, nvme_misc_cb, req);
         return NVME_NO_COMPLETE;
     }
 
-    /* 1-initialize; see comment in nvme_dsm */
-    *num_flushes = 1;
+    iocb = qemu_aio_get(&nvme_flush_aiocb_info, NULL, nvme_misc_cb, req);
 
-    for (int i = 1; i <= n->num_namespaces; i++) {
-        ns = nvme_ns(n, i);
-        if (!ns) {
-            continue;
-        }
+    iocb->req = req;
+    iocb->bh = qemu_bh_new(nvme_flush_bh, iocb);
+    iocb->ret = 0;
+    iocb->nsid = 1;
 
-        ctx = g_new(struct nvme_aio_flush_ctx, 1);
-        ctx->req = req;
-        ctx->ns = ns;
+    nvme_flush_aio_cb(iocb, 0);
+    req->aiocb = &iocb->common;
 
-        (*num_flushes)++;
-
-        block_acct_start(blk_get_stats(ns->blkconf.blk), &ctx->acct, 0,
-                         BLOCK_ACCT_FLUSH);
-        blk_aio_flush(ns->blkconf.blk, nvme_aio_flush_cb, ctx);
-    }
-
-    /* account for the 1-initialization */
-    (*num_flushes)--;
-
-    if (*num_flushes) {
-        status = NVME_NO_COMPLETE;
-    } else {
-        status = req->status;
-    }
-
-    return status;
+    return NVME_NO_COMPLETE;
 }
 
 static uint16_t nvme_read(NvmeCtrl *n, NvmeRequest *req)
-- 
2.30.1