On 20/05/2021 16:57, Bart Van Assche wrote:
On 5/19/21 7:31 AM, John Garry wrote:
Function sdev_store_queue_depth() enforces that the sdev queue depth cannot
exceed Shost.can_queue.
The sdev initial value comes from shost cmd_per_lun.
However, the LLDD may still set cmd_per_lun > can_queue, which leads to an
initial sdev queue depth greater than can_queue.
Such an issue was reported in [0], which caused a hang. That has since
been fixed in commit fc09acb7de31 ("scsi: scsi_debug: Fix cmd_per_lun,
set to max_queue").
Stop this possibly happening for other drivers by capping
shost.cmd_per_lun at shost.can_queue.
[0] https://lore.kernel.org/linux-scsi/YHaez6iN2HHYxYOh@T590/
Signed-off-by: John Garry <john.garry@xxxxxxxxxx>
---
Earlier patch was in https://lore.kernel.org/linux-scsi/1618848384-204144-1-git-send-email-john.garry@xxxxxxxxxx/
diff --git a/drivers/scsi/hosts.c b/drivers/scsi/hosts.c
index ba72bd4202a2..624e2582c3df 100644
--- a/drivers/scsi/hosts.c
+++ b/drivers/scsi/hosts.c
@@ -220,6 +220,9 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
goto fail;
}
+ shost->cmd_per_lun = min_t(short, shost->cmd_per_lun,
+ shost->can_queue);
+
error = scsi_init_sense_cache(shost);
if (error)
goto fail;
Hi Bart,
In SCSI header files a mix of int, short and unsigned int is used for
cmd_per_lun and can_queue. How about changing the types of these two
member variables in include/scsi/*h into u16?
I don't mind doing that, but is there any requirement for can_queue to
not be limited to 16b?
It seems intentional that can_queue is int and cmd_per_lun is short.
As an aside, if short is 16b, it does not even seem to have efficient
packing on Scsi_host today (although we can move things around):
int can_queue;
short cmd_per_lun;
short unsigned int sg_tablesize;
short unsigned int sg_prot_tablesize;
/* 16b hole */
unsigned int max_sectors;
Anyway:
Reviewed-by: Bart Van Assche <bvanassche@xxxxxxx>
thanks!
.
