[PATCH V3 0/2] scsi: core: avoid big pre-allocation for sg list

[PATCH V3 0/2] scsi: core: avoid big pre-allocation for sg list

[Ming Lei]

Hi,

Since supporting to blk-mq, big pre-allocation for sg list is introduced,
this way is very unfriendly wrt. memory consumption.

There were Red Hat internal reports that some scsi_debug based tests
can't be run any more because of too big pre-allocation.

Also lpfc users commplained that 1GB+ ram is pre-allocatd for single
HBA.

sg_alloc_table_chained() is improved to support variant size of 1st
pre-allocated SGL in the 1st patch as suggested by Christoph.

The other two patches try to address this issue by allocating sg list runtime,
meantime pre-allocating one or two inline sg entries for small IO. This
ways follows NVMe's approach wrt. sg list allocation.

V3:
	- improve sg_alloc_table_chained() to accept variant size of
	the 1st pre-allocated SGL
	- applies the improved sg API to address the big pre-allocation
	issue

V2:
	- move inline sg table initializetion into one helper
	- introduce new helper for getting inline sg
	- comment log fix

Ming Lei (3):
  lib/sg_pool.c: improve APIs for allocating sg pool
  scsi: core: avoid to pre-allocate big chunk for protection meta data
  scsi: core: avoid to pre-allocate big chunk for sg list

 drivers/scsi/scsi_lib.c     | 36 +++++++++++++++++++++++----------
 include/linux/scatterlist.h | 27 ++++++++++++++++++++-----
 lib/scatterlist.c           | 36 +++++++++++++++++++++------------
 lib/sg_pool.c               | 49 ++++++++++++++++++++++++++++++---------------
 4 files changed, 103 insertions(+), 45 deletions(-)

Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Ewan D. Milne <emilne@redhat.com>
Cc: Hannes Reinecke <hare@suse.com>
-- 
2.9.5

[PATCH V3 1/3] lib/sg_pool.c: improve APIs for allocating sg pool

[Ming Lei]

Now sg_alloc_table_chained() allows user to provide one preallocated
SGL, and returns simply if the requested number isn't bigger than size of
that SGL. This way is nice for inline small SGL to small IO request.

However, scattergather code only allows that size of the 1st preallocated
SGL is SG_CHUNK_SIZE(128), and this way isn't flexiable and useful, because
it may take too much memory(4KB) to pre-allocat one such size SGL for each
IO request, especially block layer always pre-allocates IO request structure.
Instead it is more friendly to pre-allocate one small size inline SGL just
for small IO.

Introduces __sg_alloc_table_chained() and __sg_free_table_chained() with
one extra parameter to specify size of the pre-allocated SGL, then the
'first_chunk' SGL can include any number of entries.

Both __sg_free_table() and __sg_alloc_table() supposes that each SGL has
same size except for the last one, changes code to allow both to accept
variant size for the 1st preallocated SGL.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Ewan D. Milne <emilne@redhat.com>
Cc: Hannes Reinecke <hare@suse.com>
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 include/linux/scatterlist.h | 27 ++++++++++++++++++++-----
 lib/scatterlist.c           | 36 +++++++++++++++++++++------------
 lib/sg_pool.c               | 49 ++++++++++++++++++++++++++++++---------------
 3 files changed, 78 insertions(+), 34 deletions(-)

diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h
index b4be960c7e5d..045d7aa81f2c 100644
--- a/include/linux/scatterlist.h
+++ b/include/linux/scatterlist.h
@@ -266,10 +266,11 @@ int sg_split(struct scatterlist *in, const int in_mapped_nents,
 typedef struct scatterlist *(sg_alloc_fn)(unsigned int, gfp_t);
 typedef void (sg_free_fn)(struct scatterlist *, unsigned int);
 
-void __sg_free_table(struct sg_table *, unsigned int, bool, sg_free_fn *);
+void __sg_free_table(struct sg_table *, unsigned int, unsigned int,
+		     sg_free_fn *);
 void sg_free_table(struct sg_table *);
 int __sg_alloc_table(struct sg_table *, unsigned int, unsigned int,
-		     struct scatterlist *, gfp_t, sg_alloc_fn *);
+		     struct scatterlist *, unsigned int, gfp_t, sg_alloc_fn *);
 int sg_alloc_table(struct sg_table *, unsigned int, gfp_t);
 int __sg_alloc_table_from_pages(struct sg_table *sgt, struct page **pages,
 				unsigned int n_pages, unsigned int offset,
@@ -331,9 +332,25 @@ size_t sg_zero_buffer(struct scatterlist *sgl, unsigned int nents,
 #endif
 
 #ifdef CONFIG_SG_POOL
-void sg_free_table_chained(struct sg_table *table, bool first_chunk);
-int sg_alloc_table_chained(struct sg_table *table, int nents,
-			   struct scatterlist *first_chunk);
+void __sg_free_table_chained(struct sg_table *table,
+			     unsigned nents_first_chunk);
+int __sg_alloc_table_chained(struct sg_table *table, int nents,
+			     struct scatterlist *first_chunk,
+			     unsigned nents_first_chunk);
+
+static inline void sg_free_table_chained(struct sg_table *table,
+					 bool first_chunk)
+{
+	__sg_free_table_chained(table, first_chunk ? SG_CHUNK_SIZE : 0);
+}
+
+static inline int sg_alloc_table_chained(struct sg_table *table,
+					 int nents,
+					 struct scatterlist *first_chunk)
+{
+	return __sg_alloc_table_chained(table, nents, first_chunk,
+					first_chunk ? SG_CHUNK_SIZE : 0);
+}
 #endif
 
 /*
diff --git a/lib/scatterlist.c b/lib/scatterlist.c
index 739dc9fe2c55..77ec8eec3fd0 100644
--- a/lib/scatterlist.c
+++ b/lib/scatterlist.c
@@ -181,7 +181,8 @@ static void sg_kfree(struct scatterlist *sg, unsigned int nents)
  * __sg_free_table - Free a previously mapped sg table
  * @table:	The sg table header to use
  * @max_ents:	The maximum number of entries per single scatterlist
- * @skip_first_chunk: don't free the (preallocated) first scatterlist chunk
+ * @nents_first_chunk: Number of entries int the (preallocated) first
+ * 	scatterlist chunk, 0 means no such preallocated first chunk
  * @free_fn:	Free function
  *
  *  Description:
@@ -191,9 +192,10 @@ static void sg_kfree(struct scatterlist *sg, unsigned int nents)
  *
  **/
 void __sg_free_table(struct sg_table *table, unsigned int max_ents,
-		     bool skip_first_chunk, sg_free_fn *free_fn)
+		     unsigned int nents_first_chunk, sg_free_fn *free_fn)
 {
 	struct scatterlist *sgl, *next;
+	unsigned curr_max_ents = nents_first_chunk ?: max_ents;
 
 	if (unlikely(!table->sgl))
 		return;
@@ -209,9 +211,9 @@ void __sg_free_table(struct sg_table *table, unsigned int max_ents,
 		 * sg_size is then one less than alloc size, since the last
 		 * element is the chain pointer.
 		 */
-		if (alloc_size > max_ents) {
-			next = sg_chain_ptr(&sgl[max_ents - 1]);
-			alloc_size = max_ents;
+		if (alloc_size > curr_max_ents) {
+			next = sg_chain_ptr(&sgl[curr_max_ents - 1]);
+			alloc_size = curr_max_ents;
 			sg_size = alloc_size - 1;
 		} else {
 			sg_size = alloc_size;
@@ -219,11 +221,12 @@ void __sg_free_table(struct sg_table *table, unsigned int max_ents,
 		}
 
 		table->orig_nents -= sg_size;
-		if (skip_first_chunk)
-			skip_first_chunk = false;
+		if (nents_first_chunk)
+			nents_first_chunk = 0;
 		else
 			free_fn(sgl, alloc_size);
 		sgl = next;
+		curr_max_ents = max_ents;
 	}
 
 	table->sgl = NULL;
@@ -246,6 +249,8 @@ EXPORT_SYMBOL(sg_free_table);
  * @table:	The sg table header to use
  * @nents:	Number of entries in sg list
  * @max_ents:	The maximum number of entries the allocator returns per call
+ * @nents_first_chunk: Number of entries int the (preallocated) first
+ * 	scatterlist chunk, 0 means no such preallocated chunk provided by user
  * @gfp_mask:	GFP allocation mask
  * @alloc_fn:	Allocator to use
  *
@@ -262,10 +267,13 @@ EXPORT_SYMBOL(sg_free_table);
  **/
 int __sg_alloc_table(struct sg_table *table, unsigned int nents,
 		     unsigned int max_ents, struct scatterlist *first_chunk,
-		     gfp_t gfp_mask, sg_alloc_fn *alloc_fn)
+		     unsigned int nents_first_chunk, gfp_t gfp_mask,
+		     sg_alloc_fn *alloc_fn)
 {
 	struct scatterlist *sg, *prv;
 	unsigned int left;
+	unsigned curr_max_ents = nents_first_chunk ?: max_ents;
+	unsigned prv_max_ents;
 
 	memset(table, 0, sizeof(*table));
 
@@ -281,8 +289,8 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents,
 	do {
 		unsigned int sg_size, alloc_size = left;
 
-		if (alloc_size > max_ents) {
-			alloc_size = max_ents;
+		if (alloc_size > curr_max_ents) {
+			alloc_size = curr_max_ents;
 			sg_size = alloc_size - 1;
 		} else
 			sg_size = alloc_size;
@@ -316,7 +324,7 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents,
 		 * If this is not the first mapping, chain previous part.
 		 */
 		if (prv)
-			sg_chain(prv, max_ents, sg);
+			sg_chain(prv, prv_max_ents, sg);
 		else
 			table->sgl = sg;
 
@@ -327,6 +335,8 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents,
 			sg_mark_end(&sg[sg_size - 1]);
 
 		prv = sg;
+		prv_max_ents = curr_max_ents;
+		curr_max_ents = max_ents;
 	} while (left);
 
 	return 0;
@@ -349,9 +359,9 @@ int sg_alloc_table(struct sg_table *table, unsigned int nents, gfp_t gfp_mask)
 	int ret;
 
 	ret = __sg_alloc_table(table, nents, SG_MAX_SINGLE_ALLOC,
-			       NULL, gfp_mask, sg_kmalloc);
+			       NULL, 0, gfp_mask, sg_kmalloc);
 	if (unlikely(ret))
-		__sg_free_table(table, SG_MAX_SINGLE_ALLOC, false, sg_kfree);
+		__sg_free_table(table, SG_MAX_SINGLE_ALLOC, 0, sg_kfree);
 
 	return ret;
 }
diff --git a/lib/sg_pool.c b/lib/sg_pool.c
index d1c1e6388eaa..8026e210a25a 100644
--- a/lib/sg_pool.c
+++ b/lib/sg_pool.c
@@ -67,56 +67,73 @@ static struct scatterlist *sg_pool_alloc(unsigned int nents, gfp_t gfp_mask)
 }
 
 /**
- * sg_free_table_chained - Free a previously mapped sg table
+ * __sg_free_table_chained - Free a previously mapped sg table
  * @table:	The sg table header to use
- * @first_chunk: was first_chunk not NULL in sg_alloc_table_chained?
+ * @nents_first_chunk: size of the first_chunk SGL passed to
+ *		__sg_alloc_table_chained
  *
  *  Description:
  *    Free an sg table previously allocated and setup with
- *    sg_alloc_table_chained().
+ *    __sg_alloc_table_chained().
+ *
+ *    @nents_first_chunk has to be same with that same parameter passed
+ *    to __sg_alloc_table_chained().
  *
  **/
-void sg_free_table_chained(struct sg_table *table, bool first_chunk)
+void __sg_free_table_chained(struct sg_table *table,
+		unsigned nents_first_chunk)
 {
-	if (first_chunk && table->orig_nents <= SG_CHUNK_SIZE)
+	if (table->orig_nents <= nents_first_chunk)
 		return;
-	__sg_free_table(table, SG_CHUNK_SIZE, first_chunk, sg_pool_free);
+
+	if (nents_first_chunk == 1)
+		nents_first_chunk = 0;
+
+	__sg_free_table(table, SG_CHUNK_SIZE, nents_first_chunk, sg_pool_free);
 }
-EXPORT_SYMBOL_GPL(sg_free_table_chained);
+EXPORT_SYMBOL_GPL(__sg_free_table_chained);
 
 /**
- * sg_alloc_table_chained - Allocate and chain SGLs in an sg table
+ * __sg_alloc_table_chained - Allocate and chain SGLs in an sg table
  * @table:	The sg table header to use
  * @nents:	Number of entries in sg list
  * @first_chunk: first SGL
+ * @nents_first_chunk: number of the SGL of @first_chunk
  *
  *  Description:
  *    Allocate and chain SGLs in an sg table. If @nents@ is larger than
- *    SG_CHUNK_SIZE a chained sg table will be setup.
+ *    @nents_first_chunk a chained sg table will be setup.
  *
  **/
-int sg_alloc_table_chained(struct sg_table *table, int nents,
-		struct scatterlist *first_chunk)
+int __sg_alloc_table_chained(struct sg_table *table, int nents,
+		struct scatterlist *first_chunk, unsigned nents_first_chunk)
 {
 	int ret;
 
 	BUG_ON(!nents);
 
-	if (first_chunk) {
-		if (nents <= SG_CHUNK_SIZE) {
+	if (first_chunk && nents_first_chunk) {
+		if (nents <= nents_first_chunk) {
 			table->nents = table->orig_nents = nents;
 			sg_init_table(table->sgl, nents);
 			return 0;
 		}
 	}
 
+	/* User supposes that the 1st SGL includes real entry */
+	if (nents_first_chunk == 1) {
+		first_chunk = NULL;
+		nents_first_chunk = 0;
+	}
+
 	ret = __sg_alloc_table(table, nents, SG_CHUNK_SIZE,
-			       first_chunk, GFP_ATOMIC, sg_pool_alloc);
+			       first_chunk, nents_first_chunk,
+			       GFP_ATOMIC, sg_pool_alloc);
 	if (unlikely(ret))
-		sg_free_table_chained(table, (bool)first_chunk);
+		__sg_free_table_chained(table, nents_first_chunk);
 	return ret;
 }
-EXPORT_SYMBOL_GPL(sg_alloc_table_chained);
+EXPORT_SYMBOL_GPL(__sg_alloc_table_chained);
 
 static __init int sg_pool_init(void)
 {
-- 
2.9.5

Re: [PATCH V3 1/3] lib/sg_pool.c: improve APIs for allocating sg pool

[Christoph Hellwig]

On Fri, Apr 26, 2019 at 08:53:44AM +0800, Ming Lei wrote:
> Introduces __sg_alloc_table_chained() and __sg_free_table_chained() with
> one extra parameter to specify size of the pre-allocated SGL, then the
> 'first_chunk' SGL can include any number of entries.

Can you just update the existing sg_alloc_table_chained and
sg_free_table_chained chained instead?  We only have a handful users
of them in total, and most of them should eventually use your new
semantics anyway.

Re: [PATCH V3 1/3] lib/sg_pool.c: improve APIs for allocating sg pool

[Ming Lei]

On Fri, Apr 26, 2019 at 04:49:28PM +0200, Christoph Hellwig wrote:
> On Fri, Apr 26, 2019 at 08:53:44AM +0800, Ming Lei wrote:
> > Introduces __sg_alloc_table_chained() and __sg_free_table_chained() with
> > one extra parameter to specify size of the pre-allocated SGL, then the
> > 'first_chunk' SGL can include any number of entries.
> 
> Can you just update the existing sg_alloc_table_chained and
> sg_free_table_chained chained instead?  We only have a handful users
> of them in total, and most of them should eventually use your new
> semantics anyway.

Yeah, I'd like to do that too.

Not do that in V3 just because both nvme and network change are
involved.

Will do this way if our maintainers don't object.

Thanks,
Ming

Re: [PATCH V3 1/3] lib/sg_pool.c: improve APIs for allocating sg pool

[Christoph Hellwig]

On Sat, Apr 27, 2019 at 06:23:39AM +0800, Ming Lei wrote:
> On Fri, Apr 26, 2019 at 04:49:28PM +0200, Christoph Hellwig wrote:
> > On Fri, Apr 26, 2019 at 08:53:44AM +0800, Ming Lei wrote:
> > > Introduces __sg_alloc_table_chained() and __sg_free_table_chained() with
> > > one extra parameter to specify size of the pre-allocated SGL, then the
> > > 'first_chunk' SGL can include any number of entries.
> > 
> > Can you just update the existing sg_alloc_table_chained and
> > sg_free_table_chained chained instead?  We only have a handful users
> > of them in total, and most of them should eventually use your new
> > semantics anyway.
> 
> Yeah, I'd like to do that too.
> 
> Not do that in V3 just because both nvme and network change are
> involved.
> 
> Will do this way if our maintainers don't object.

sunrpc is maintained by the network maintainers, and the rdma code that
is calling this in particular has been herded by Chuck Lever.  Add them
to Cc, I bet they won't object to the trivial paramter change.  In fact
I wouldn't be surprised if they'd be interested in the smaller inline
SGLs as well for next merge window.

[PATCH V3 2/3] scsi: core: avoid to pre-allocate big chunk for protection meta data

[Ming Lei]

Now scsi_mq_setup_tags() pre-allocates a big buffer for protection
sg entries, and the buffer size is scsi_mq_sgl_size().

This way isn't correct, scsi_mq_sgl_size() is used to pre-allocate
sg entries for IO data. And the protection data buffer is much less,
for example, one 512byte sector needs 8byte protection data, and
the max sector number for one request is 2560(BLK_DEF_MAX_SECTORS),
so the max protection data size is just 20k.

The usual case is that one bio builds one single bip segment. Attribute
to bio split, bio merge is seldom done for big IO, and it is only done
in case of small bios. And protection data segment number is usually
same with bio count in the request, so the number won't be very big,
and allocating from slab is fast enough.

Reduce to pre-allocate one sg entry for protection data, and switch
to runtime allocation in case that the protection data segment number
is bigger than 1. Then we can save huge pre-alocation, for example,
500+MB is saved on single lpfc HBA.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Ewan D. Milne <emilne@redhat.com>
Cc: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 drivers/scsi/scsi_lib.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 07dfc17d4824..989539de78c6 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -39,6 +39,12 @@
 #include "scsi_priv.h"
 #include "scsi_logging.h"
 
+/*
+ * Size of integrity metadata is usually small, 1 inline sg should
+ * cover normal cases.
+ */
+#define  SCSI_INLINE_PROT_SG_CNT  1
+
 static struct kmem_cache *scsi_sdb_cache;
 static struct kmem_cache *scsi_sense_cache;
 static struct kmem_cache *scsi_sense_isadma_cache;
@@ -558,7 +564,8 @@ static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
 	if (cmd->sdb.table.nents)
 		sg_free_table_chained(&cmd->sdb.table, true);
 	if (scsi_prot_sg_count(cmd))
-		sg_free_table_chained(&cmd->prot_sdb->table, true);
+		__sg_free_table_chained(&cmd->prot_sdb->table,
+				SCSI_INLINE_PROT_SG_CNT);
 }
 
 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
@@ -1045,8 +1052,9 @@ blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
 
 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
 
-		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
-				prot_sdb->table.sgl)) {
+		if (__sg_alloc_table_chained(&prot_sdb->table, ivecs,
+					prot_sdb->table.sgl,
+					SCSI_INLINE_PROT_SG_CNT)) {
 			ret = BLK_STS_RESOURCE;
 			goto out_free_sgtables;
 		}
@@ -1846,7 +1854,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost)
 	sgl_size = scsi_mq_sgl_size(shost);
 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
 	if (scsi_host_get_prot(shost))
-		cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
+		cmd_size += sizeof(struct scsi_data_buffer) +
+			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
 
 	memset(&shost->tag_set, 0, sizeof(shost->tag_set));
 	shost->tag_set.ops = &scsi_mq_ops;
-- 
2.9.5

Re: [PATCH V3 2/3] scsi: core: avoid to pre-allocate big chunk for protection meta data

[Christoph Hellwig]

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH V3 3/3] scsi: core: avoid to pre-allocate big chunk for sg list

[Ming Lei]

Now scsi_mq_setup_tags() pre-allocates a big buffer for IO sg list,
and the buffer size is scsi_mq_sgl_size() which depends on smaller
value between shost->sg_tablesize and SG_CHUNK_SIZE.

Modern HBA's DMA is often capable of deadling with very big segment
number, so scsi_mq_sgl_size() is often big. Suppose the max sg number
of SG_CHUNK_SIZE is taken, scsi_mq_sgl_size() will be 4KB.

Then if one HBA has lots of queues, and each hw queue's depth is
high, pre-allocation for sg list can consume huge memory.
For example of lpfc, nr_hw_queues can be 70, each queue's depth
can be 3781, so the pre-allocation for data sg list is 70*3781*2k
=517MB for single HBA.

There is Red Hat internal report that scsi_debug based tests can't
be run any more since legacy io path is killed because too big
pre-allocation.

So switch to runtime allocation for sg list, meantime pre-allocate 2
inline sg entries. This way has been applied to NVMe PCI for a while,
so it should be fine for SCSI too. Also runtime sg entries allocation
has verified and run always in the original legacy io path.

Not see performance effect in my big BS test on scsi_debug.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Ewan D. Milne <emilne@redhat.com>
Cc: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 drivers/scsi/scsi_lib.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 989539de78c6..b701fc65da76 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -45,6 +45,8 @@
  */
 #define  SCSI_INLINE_PROT_SG_CNT  1
 
+#define  SCSI_INLINE_SG_CNT  2
+
 static struct kmem_cache *scsi_sdb_cache;
 static struct kmem_cache *scsi_sense_cache;
 static struct kmem_cache *scsi_sense_isadma_cache;
@@ -562,7 +564,8 @@ static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
 {
 	if (cmd->sdb.table.nents)
-		sg_free_table_chained(&cmd->sdb.table, true);
+		__sg_free_table_chained(&cmd->sdb.table,
+				SCSI_INLINE_SG_CNT);
 	if (scsi_prot_sg_count(cmd))
 		__sg_free_table_chained(&cmd->prot_sdb->table,
 				SCSI_INLINE_PROT_SG_CNT);
@@ -998,8 +1001,10 @@ static blk_status_t scsi_init_sgtable(struct request *req,
 	/*
 	 * If sg table allocation fails, requeue request later.
 	 */
-	if (unlikely(sg_alloc_table_chained(&sdb->table,
-			blk_rq_nr_phys_segments(req), sdb->table.sgl)))
+	if (unlikely(__sg_alloc_table_chained(&sdb->table,
+					blk_rq_nr_phys_segments(req),
+					sdb->table.sgl,
+					SCSI_INLINE_SG_CNT)))
 		return BLK_STS_RESOURCE;
 
 	/* 
@@ -1565,9 +1570,9 @@ static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
 }
 
 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
-static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost)
+static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
 {
-	return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) *
+	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
 		sizeof(struct scatterlist);
 }
 
@@ -1757,7 +1762,7 @@ static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
 	if (scsi_host_get_prot(shost)) {
 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
 			shost->hostt->cmd_size;
-		cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost);
+		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
 	}
 
 	return 0;
@@ -1851,7 +1856,7 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost)
 {
 	unsigned int cmd_size, sgl_size;
 
-	sgl_size = scsi_mq_sgl_size(shost);
+	sgl_size = scsi_mq_inline_sgl_size(shost);
 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
 	if (scsi_host_get_prot(shost))
 		cmd_size += sizeof(struct scsi_data_buffer) +
-- 
2.9.5

Re: [PATCH V3 3/3] scsi: core: avoid to pre-allocate big chunk for sg list

[Christoph Hellwig]

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>