dm verity: Add optional "try_verify_in_tasklet" feature

Using tasklets for disk verification can reduce IO latency. When there
are accelerated hash instructions it is often better to compute the
hash immediately using a tasklet rather than deferring verification to
a work-queue. This reduces time spent waiting to schedule work-queue
jobs, but requires spending slightly more time in interrupt context.

If the dm-bufio cache does not have the required hashes we fallback to
the work-queue implementation. FEC is only possible using work-queue
because code to support the FEC feature may sleep.

The following shows a speed comparison of random reads on a dm-verity
device. The dm-verity device uses a 1G ramdisk for data and a 1G
ramdisk for hashes. One test was run using tasklets and one test was
run using the existing work-queue solution. Both tests were run when
the dm-bufio cache was hot. The tasklet implementation performs
significantly better since there is no time spent waiting for
work-queue jobs to be scheduled.

   READ: bw=181MiB/s (190MB/s), 181MiB/s-181MiB/s (190MB/s-190MB/s),
   io=512MiB (537MB), run=2827-2827msec
   READ: bw=23.6MiB/s (24.8MB/s), 23.6MiB/s-23.6MiB/s (24.8MB/s-24.8MB/s),
   io=512MiB (537MB), run=21688-21688msec

Signed-off-by: Nathan Huckleberry <nhuck@google.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>

1 files changed, 5 insertions, 1 deletions

diff --git a/drivers/md/dm-verity.h b/drivers/md/dm-verity.h
index 4e769d13473a..fb92b22a6404 100644
--- a/drivers/md/dm-verity.h
+++ b/drivers/md/dm-verity.h
@@ -13,6 +13,7 @@
 
 #include <linux/dm-bufio.h>
 #include <linux/device-mapper.h>
+#include <linux/interrupt.h>
 #include <crypto/hash.h>
 
 #define DM_VERITY_MAX_LEVELS		63
@@ -51,9 +52,10 @@ struct dm_verity {
 	unsigned char hash_per_block_bits;	/* log2(hashes in hash block) */
 	unsigned char levels;	/* the number of tree levels */
 	unsigned char version;
+	bool hash_failed:1;	/* set if hash of any block failed */
+	bool use_tasklet:1;	/* try to verify in tasklet before work-queue */
 	unsigned digest_size;	/* digest size for the current hash algorithm */
 	unsigned int ahash_reqsize;/* the size of temporary space for crypto */
-	int hash_failed;	/* set to 1 if hash of any block failed */
 	enum verity_mode mode;	/* mode for handling verification errors */
 	unsigned corrupted_errs;/* Number of errors for corrupted blocks */
 
@@ -76,10 +78,12 @@ struct dm_verity_io {
 
 	sector_t block;
 	unsigned n_blocks;
+	bool in_tasklet;
 
 	struct bvec_iter iter;
 
 	struct work_struct work;
+	struct tasklet_struct tasklet;
 
 	/*
 	 * Three variably-size fields follow this struct: