Reinstate zvol_taskq to fix aio on zvol

Commit 37f9dac removed the zvol_taskq for processing zvol requests.
This was removed as part of switching to make_request_fn and was
motivated by a concern at the time over dispatch latency.

However, this also made all bio request synchronous, and caused
serious performance issues as the bio submitter would wait for
every bio it submitted, effectively making the IO depth 1.

This patch reinstate zvol_taskq, and to make sure overlapped I/Os
are ordered properly, we take range lock in zvol_request, and pass
it along with bio to the I/O functions zvol_{write,discard,read}.

In order to facilitate benchmarks a zvol_request_sync module
option was added to switch between sync and async request handling.
For the moment, the default behavior is synchronous but this is
likely to change pending additional testing.

Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Chunwei Chen <[email protected]>
Closes #5824

1 files changed, 25 insertions, 0 deletions

diff --git a/man/man5/zfs-module-parameters.5 b/man/man5/zfs-module-parameters.5
index f85c28bf8..77f1af6d5 100644
--- a/man/man5/zfs-module-parameters.5
+++ b/man/man5/zfs-module-parameters.5
@@ -2034,6 +2034,31 @@ Default value: \fB131,072\fR.
 .sp
 .ne 2
 .na
+\fBzvol_request_sync\fR (uint)
+.ad
+.RS 12n
+When processing I/O requests for a zvol submit them synchronously.  This
+effectively limits the queue depth to 1 for each I/O submitter.  When set
+to 0 requests are handled asynchronously by a thread pool.  The number of
+requests which can be handled concurrently is controller by \fBzvol_threads\fR.
+.sp
+Default value: \fB1\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBzvol_threads\fR (uint)
+.ad
+.RS 12n
+Max number of threads which can handle zvol I/O requests concurrently.
+.sp
+Default value: \fB32\fR.
+.RE
+
+.sp
+.ne 2
+.na
 \fBzfs_qat_disable\fR (int)
 .ad
 .RS 12n