[Qemu-devel] [PATCH for-2.5 0/3] scsi: preparations for thread-safe virtio-scsi dataplane

[Qemu-devel] [PATCH for-2.5 0/3] scsi: preparations for thread-safe virtio-scsi dataplane

[Paolo Bonzini]

As a preparation for making virtio-scsi dataplane really thread-safe,
this series fixes a bug in rerror/werror=stop (patch 1) and cleans up
the code so that AIO callbacks are clearly identified.  These will need
to acquire/release the AioContext, so it's important not to miss any
of them.

Paolo

Paolo Bonzini (3):
  scsi: create restart bottom half in the right AioContext
  scsi-disk: identify AIO callbacks more clearly
  scsi-generic: identify AIO callbacks more clearly

 hw/scsi/scsi-bus.c     |  3 +-
 hw/scsi/scsi-disk.c    | 91 +++++++++++++++++++++++++++++++++-----------------
 hw/scsi/scsi-generic.c | 66 ++++++++++++++++++++++--------------
 3 files changed, 104 insertions(+), 56 deletions(-)

-- 
2.4.3

[Qemu-devel] [PATCH 1/3] scsi: create restart bottom half in the right AioContext

[Paolo Bonzini]

This matches commit 4407c1c (virtio-blk: Schedule BH in the right context,
2014-06-17), which did the same thing for virtio-blk.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/scsi-bus.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
index f0ae462..ffac8f4 100644
--- a/hw/scsi/scsi-bus.c
+++ b/hw/scsi/scsi-bus.c
@@ -136,7 +136,8 @@ static void scsi_dma_restart_cb(void *opaque, int running, RunState state)
         return;
     }
     if (!s->bh) {
-        s->bh = qemu_bh_new(scsi_dma_restart_bh, s);
+        AioContext *ctx = blk_get_aio_context(s->conf.blk);
+        s->bh = aio_bh_new(ctx, scsi_dma_restart_bh, s);
         qemu_bh_schedule(s->bh);
     }
 }
-- 
2.4.3

[Qemu-devel] [PATCH 2/3] scsi-disk: identify AIO callbacks more clearly

[Paolo Bonzini]

Functions that are not callbacks should assert that aiocb is NULL and
have a non-opaque argument (usually a pointer to SCSIDiskReq).

AIO callbacks should assert that aiocb is not NULL and take care of
calling block_acct done.  They also have an opaque argument.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/scsi-disk.c | 91 +++++++++++++++++++++++++++++++++++------------------
 1 file changed, 61 insertions(+), 30 deletions(-)

diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index 64f0694..8fb0f43 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -217,6 +217,8 @@ static void scsi_write_do_fua(SCSIDiskReq *r)
 {
     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
 
+    assert(r->req.aiocb == NULL);
+
     if (r->req.io_canceled) {
         scsi_req_cancel_complete(&r->req);
         goto done;
@@ -235,15 +237,10 @@ done:
     scsi_req_unref(&r->req);
 }
 
-static void scsi_dma_complete_noio(void *opaque, int ret)
+static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
 {
-    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
-    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+    assert(r->req.aiocb == NULL);
 
-    if (r->req.aiocb != NULL) {
-        r->req.aiocb = NULL;
-        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
-    }
     if (r->req.io_canceled) {
         scsi_req_cancel_complete(&r->req);
         goto done;
@@ -271,9 +268,13 @@ done:
 static void scsi_dma_complete(void *opaque, int ret)
 {
     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
 
     assert(r->req.aiocb != NULL);
-    scsi_dma_complete_noio(opaque, ret);
+    r->req.aiocb = NULL;
+
+    block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    scsi_dma_complete_noio(r, ret);
 }
 
 static void scsi_read_complete(void * opaque, int ret)
@@ -308,16 +309,13 @@ done:
 }
 
 /* Actually issue a read to the block device.  */
-static void scsi_do_read(void *opaque, int ret)
+static void scsi_do_read(SCSIDiskReq *r, int ret)
 {
-    SCSIDiskReq *r = opaque;
     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
     uint32_t n;
 
-    if (r->req.aiocb != NULL) {
-        r->req.aiocb = NULL;
-        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
-    }
+    assert (r->req.aiocb == NULL);
+
     if (r->req.io_canceled) {
         scsi_req_cancel_complete(&r->req);
         goto done;
@@ -349,6 +347,18 @@ done:
     scsi_req_unref(&r->req);
 }
 
+static void scsi_do_read_cb(void *opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    assert (r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+
+    block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    scsi_do_read(opaque, ret);
+}
+
 /* Read more data from scsi device into buffer.  */
 static void scsi_read_data(SCSIRequest *req)
 {
@@ -384,7 +394,7 @@ static void scsi_read_data(SCSIRequest *req)
     if (first && scsi_is_cmd_fua(&r->req.cmd)) {
         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
                          BLOCK_ACCT_FLUSH);
-        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read, r);
+        r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
     } else {
         scsi_do_read(r, 0);
     }
@@ -430,16 +440,12 @@ static int scsi_handle_rw_error(SCSIDiskReq *r, int error)
     return action != BLOCK_ERROR_ACTION_IGNORE;
 }
 
-static void scsi_write_complete(void * opaque, int ret)
+static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
 {
-    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
-    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
     uint32_t n;
 
-    if (r->req.aiocb != NULL) {
-        r->req.aiocb = NULL;
-        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
-    }
+    assert (r->req.aiocb == NULL);
+
     if (r->req.io_canceled) {
         scsi_req_cancel_complete(&r->req);
         goto done;
@@ -467,6 +473,18 @@ done:
     scsi_req_unref(&r->req);
 }
 
+static void scsi_write_complete(void * opaque, int ret)
+{
+    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
+    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
+
+    assert (r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+
+    block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
+    scsi_write_complete_noio(r, ret);
+}
+
 static void scsi_write_data(SCSIRequest *req)
 {
     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
@@ -480,18 +498,18 @@ static void scsi_write_data(SCSIRequest *req)
     scsi_req_ref(&r->req);
     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
         DPRINTF("Data transfer direction invalid\n");
-        scsi_write_complete(r, -EINVAL);
+        scsi_write_complete_noio(r, -EINVAL);
         return;
     }
 
     if (!r->req.sg && !r->qiov.size) {
         /* Called for the first time.  Ask the driver to send us more data.  */
         r->started = true;
-        scsi_write_complete(r, 0);
+        scsi_write_complete_noio(r, 0);
         return;
     }
     if (s->tray_open) {
-        scsi_write_complete(r, -ENOMEDIUM);
+        scsi_write_complete_noio(r, -ENOMEDIUM);
         return;
     }
 
@@ -500,7 +518,7 @@ static void scsi_write_data(SCSIRequest *req)
         if (r->req.sg) {
             scsi_dma_complete_noio(r, 0);
         } else {
-            scsi_write_complete(r, 0);
+            scsi_write_complete_noio(r, 0);
         }
         return;
     }
@@ -1557,15 +1575,17 @@ typedef struct UnmapCBData {
     int count;
 } UnmapCBData;
 
-static void scsi_unmap_complete(void *opaque, int ret)
+static void scsi_unmap_complete(void *opaque, int ret);
+
+static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
 {
-    UnmapCBData *data = opaque;
     SCSIDiskReq *r = data->r;
     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
     uint64_t sector_num;
     uint32_t nb_sectors;
 
-    r->req.aiocb = NULL;
+    assert(r->req.aiocb == NULL);
+
     if (r->req.io_canceled) {
         scsi_req_cancel_complete(&r->req);
         goto done;
@@ -1601,6 +1621,17 @@ done:
     g_free(data);
 }
 
+static void scsi_unmap_complete(void *opaque, int ret)
+{
+    UnmapCBData *data = opaque;
+    SCSIDiskReq *r = data->r;
+
+    assert(r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+
+    scsi_unmap_complete_noio(data, ret);
+}
+
 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
 {
     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
@@ -1638,7 +1669,7 @@ static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
 
     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
     scsi_req_ref(&r->req);
-    scsi_unmap_complete(data, 0);
+    scsi_unmap_complete_noio(data, 0);
     return;
 
 invalid_param_len:
-- 
2.4.3

Re: [Qemu-devel] [PATCH 2/3] scsi-disk: identify AIO callbacks more clearly

[Fam Zheng]

On Thu, 08/06 12:01, Paolo Bonzini wrote:
>  /* Actually issue a read to the block device.  */
> -static void scsi_do_read(void *opaque, int ret)
> +static void scsi_do_read(SCSIDiskReq *r, int ret)
>  {
> -    SCSIDiskReq *r = opaque;
>      SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
>      uint32_t n;
>  
> -    if (r->req.aiocb != NULL) {
> -        r->req.aiocb = NULL;
> -        block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
> -    }
> +    assert (r->req.aiocb == NULL);
> +
>      if (r->req.io_canceled) {
>          scsi_req_cancel_complete(&r->req);
>          goto done;
> @@ -349,6 +347,18 @@ done:
>      scsi_req_unref(&r->req);
>  }
>  
> +static void scsi_do_read_cb(void *opaque, int ret)
> +{
> +    SCSIDiskReq *r = (SCSIDiskReq *)opaque;
> +    SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
> +
> +    assert (r->req.aiocb != NULL);

Majority of asserts across the code base don't get whitespace in the middle,
but 4 out of 8 in this patch do :)

[Qemu-devel] [PATCH 3/3] scsi-generic: identify AIO callbacks more clearly

[Paolo Bonzini]

Functions that are not callbacks should assert that aiocb is NULL and
have a SCSIGenericReq argument.

AIO callbacks should assert that aiocb is not NULL.  They also have an
opaque argument.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/scsi-generic.c | 66 +++++++++++++++++++++++++++++++-------------------
 1 file changed, 41 insertions(+), 25 deletions(-)

diff --git a/hw/scsi/scsi-generic.c b/hw/scsi/scsi-generic.c
index e53470f..1b6350b 100644
--- a/hw/scsi/scsi-generic.c
+++ b/hw/scsi/scsi-generic.c
@@ -88,12 +88,12 @@ static void scsi_free_request(SCSIRequest *req)
 }
 
 /* Helper function for command completion.  */
-static void scsi_command_complete(void *opaque, int ret)
+static void scsi_command_complete_noio(SCSIGenericReq *r, int ret)
 {
     int status;
-    SCSIGenericReq *r = (SCSIGenericReq *)opaque;
 
-    r->req.aiocb = NULL;
+    assert(r->req.aiocb == NULL);
+
     if (r->req.io_canceled) {
         scsi_req_cancel_complete(&r->req);
         goto done;
@@ -142,6 +142,15 @@ done:
     scsi_req_unref(&r->req);
 }
 
+static void scsi_command_complete(void *opaque, int ret)
+{
+    SCSIGenericReq *r = (SCSIGenericReq *)opaque;
+
+    assert(r->req.aiocb != NULL);
+    r->req.aiocb = NULL;
+    scsi_command_complete_noio(r, ret);
+}
+
 static int execute_command(BlockBackend *blk,
                            SCSIGenericReq *r, int direction,
                            BlockCompletionFunc *complete)
@@ -172,33 +181,37 @@ static void scsi_read_complete(void * opaque, int ret)
     SCSIDevice *s = r->req.dev;
     int len;
 
+    assert(r->req.aiocb != NULL);
     r->req.aiocb = NULL;
+
     if (ret || r->req.io_canceled) {
-        scsi_command_complete(r, ret);
+        scsi_command_complete_noio(r, ret);
         return;
     }
+
     len = r->io_header.dxfer_len - r->io_header.resid;
     DPRINTF("Data ready tag=0x%x len=%d\n", r->req.tag, len);
 
     r->len = -1;
     if (len == 0) {
-        scsi_command_complete(r, 0);
-    } else {
-        /* Snoop READ CAPACITY output to set the blocksize.  */
-        if (r->req.cmd.buf[0] == READ_CAPACITY_10 &&
-            (ldl_be_p(&r->buf[0]) != 0xffffffffU || s->max_lba == 0)) {
-            s->blocksize = ldl_be_p(&r->buf[4]);
-            s->max_lba = ldl_be_p(&r->buf[0]) & 0xffffffffULL;
-        } else if (r->req.cmd.buf[0] == SERVICE_ACTION_IN_16 &&
-                   (r->req.cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
-            s->blocksize = ldl_be_p(&r->buf[8]);
-            s->max_lba = ldq_be_p(&r->buf[0]);
-        }
-        blk_set_guest_block_size(s->conf.blk, s->blocksize);
+        scsi_command_complete_noio(r, 0);
+        return;
+    }
 
-        scsi_req_data(&r->req, len);
-        scsi_req_unref(&r->req);
+    /* Snoop READ CAPACITY output to set the blocksize.  */
+    if (r->req.cmd.buf[0] == READ_CAPACITY_10 &&
+        (ldl_be_p(&r->buf[0]) != 0xffffffffU || s->max_lba == 0)) {
+        s->blocksize = ldl_be_p(&r->buf[4]);
+        s->max_lba = ldl_be_p(&r->buf[0]) & 0xffffffffULL;
+    } else if (r->req.cmd.buf[0] == SERVICE_ACTION_IN_16 &&
+               (r->req.cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
+        s->blocksize = ldl_be_p(&r->buf[8]);
+        s->max_lba = ldq_be_p(&r->buf[0]);
     }
+    blk_set_guest_block_size(s->conf.blk, s->blocksize);
+
+    scsi_req_data(&r->req, len);
+    scsi_req_unref(&r->req);
 }
 
 /* Read more data from scsi device into buffer.  */
@@ -213,14 +226,14 @@ static void scsi_read_data(SCSIRequest *req)
     /* The request is used as the AIO opaque value, so add a ref.  */
     scsi_req_ref(&r->req);
     if (r->len == -1) {
-        scsi_command_complete(r, 0);
+        scsi_command_complete_noio(r, 0);
         return;
     }
 
     ret = execute_command(s->conf.blk, r, SG_DXFER_FROM_DEV,
                           scsi_read_complete);
     if (ret < 0) {
-        scsi_command_complete(r, ret);
+        scsi_command_complete_noio(r, ret);
     }
 }
 
@@ -230,9 +243,12 @@ static void scsi_write_complete(void * opaque, int ret)
     SCSIDevice *s = r->req.dev;
 
     DPRINTF("scsi_write_complete() ret = %d\n", ret);
+
+    assert(r->req.aiocb != NULL);
     r->req.aiocb = NULL;
+
     if (ret || r->req.io_canceled) {
-        scsi_command_complete(r, ret);
+        scsi_command_complete_noio(r, ret);
         return;
     }
 
@@ -242,7 +258,7 @@ static void scsi_write_complete(void * opaque, int ret)
         DPRINTF("block size %d\n", s->blocksize);
     }
 
-    scsi_command_complete(r, ret);
+    scsi_command_complete_noio(r, ret);
 }
 
 /* Write data to a scsi device.  Returns nonzero on failure.
@@ -264,7 +280,7 @@ static void scsi_write_data(SCSIRequest *req)
     scsi_req_ref(&r->req);
     ret = execute_command(s->conf.blk, r, SG_DXFER_TO_DEV, scsi_write_complete);
     if (ret < 0) {
-        scsi_command_complete(r, ret);
+        scsi_command_complete_noio(r, ret);
     }
 }
 
@@ -306,7 +322,7 @@ static int32_t scsi_send_command(SCSIRequest *req, uint8_t *cmd)
         ret = execute_command(s->conf.blk, r, SG_DXFER_NONE,
                               scsi_command_complete);
         if (ret < 0) {
-            scsi_command_complete(r, ret);
+            scsi_command_complete_noio(r, ret);
             return 0;
         }
         return 0;
-- 
2.4.3

Re: [Qemu-devel] [PATCH for-2.5 0/3] scsi: preparations for thread-safe virtio-scsi dataplane

[Fam Zheng]

On Thu, 08/06 12:01, Paolo Bonzini wrote:
> As a preparation for making virtio-scsi dataplane really thread-safe,
> this series fixes a bug in rerror/werror=stop (patch 1) and cleans up
> the code so that AIO callbacks are clearly identified.  These will need
> to acquire/release the AioContext, so it's important not to miss any
> of them.

Except the minor whitespace problem in patch 2,

Reviewed-by: Fam Zheng <famz@redhat.com>