On Tue, May 02 2017, Javier González wrote:

quoted

On 2 May 2017, at 05.42, NeilBrown [off-list ref] wrote:

pblk_submit_read() uses bio_clone_bioset() but doesn't change the
io_vec, so bio_clone_fast() is a better choice.

It also uses fs_bio_set which is intended for filesystems.  Using it
in a device driver can deadlock.
So allocate a new bioset, and and use bio_clone_fast().

Signed-off-by: NeilBrown <redacted>
---
drivers/lightnvm/pblk-init.c |   12 +++++++++++-
drivers/lightnvm/pblk-read.c |    2 +-
drivers/lightnvm/pblk.h      |    1 +
3 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/lightnvm/pblk-init.c b/drivers/lightnvm/pblk-init.c
index b3fec8ec55b8..aaefbccce30e 100644
--- a/drivers/lightnvm/pblk-init.c
+++ b/drivers/lightnvm/pblk-init.c

@@ -23,6 +23,7 @@

static struct kmem_cache *pblk_blk_ws_cache, *pblk_rec_cache, *pblk_r_rq_cache,
					*pblk_w_rq_cache, *pblk_line_meta_cache;
static DECLARE_RWSEM(pblk_lock);
+struct bio_set *pblk_bio_set;

static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
			  struct bio *bio)

@@ -946,11 +947,20 @@ static struct nvm_tgt_type tt_pblk = {

static int __init pblk_module_init(void)
{
-	return nvm_register_tgt_type(&tt_pblk);
+	int ret;
+
+	pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
+	if (!pblk_bio_set)
+		return -ENOMEM;
+	ret = nvm_register_tgt_type(&tt_pblk);
+	if (ret)
+		bioset_free(pblk_bio_set);
+	return ret;
}

static void pblk_module_exit(void)
{
+	bioset_free(pblk_bio_set);
	nvm_unregister_tgt_type(&tt_pblk);
}

diff --git a/drivers/lightnvm/pblk-read.c b/drivers/lightnvm/pblk-read.c
index 4a12f14d78c6..8ee0c50a7a71 100644
--- a/drivers/lightnvm/pblk-read.c
+++ b/drivers/lightnvm/pblk-read.c

@@ -342,7 +342,7 @@ int pblk_submit_read(struct pblk *pblk, struct bio *bio)

		struct pblk_r_ctx *r_ctx = nvm_rq_to_pdu(rqd);

		/* Clone read bio to deal with read errors internally */
-		int_bio = bio_clone_bioset(bio, GFP_KERNEL, fs_bio_set);
+		int_bio = bio_clone_fast(bio, GFP_KERNEL, pblk_bio_set);
		if (!int_bio) {
			pr_err("pblk: could not clone read bio\n");
			return NVM_IO_ERR;

diff --git a/drivers/lightnvm/pblk.h b/drivers/lightnvm/pblk.h
index 99f3186b5288..95b665f23925 100644
--- a/drivers/lightnvm/pblk.h
+++ b/drivers/lightnvm/pblk.h

@@ -702,6 +702,7 @@ void pblk_write_should_kick(struct pblk *pblk);

/*
 * pblk read path
 */
+extern struct bio_set *pblk_bio_set;
int pblk_submit_read(struct pblk *pblk, struct bio *bio);
int pblk_submit_read_gc(struct pblk *pblk, u64 *lba_list, void *data,
			unsigned int nr_secs, unsigned int *secs_to_gc,

Hi Neil,

Looks good. Thanks for fixing this. I did not know that bio_clone_bioset
was not supposed to be used on drivers.

Prior to my patchset, using bio_clone_bioset() wasn't wrong in drivers,
though it was a waste when bio_clone_fast() would to just as well as is
more efficient.
After my patchset, using it can be problematic.  I'm wondering what I
should do to encourage those problems to be more visible so that if
people to us it, they'll get a warning or something.

Thanks,
NeilBrown

Reviewed-by: Javier González <redacted>
Tested-by: Javier González <redacted>