Hi Malcolm,
On Mon, 2016-10-31 at 18:04 +1100, Malcolm Haak wrote:
> Hi Nicholas,
>
> I also patched rtslib (pull request on github) to work around not
> being able to resolve the /dev/* device to the HCL and back the other
> way.
>
> [root@tvtln01-eth SOURCES]# head /sys/kernel/config/target/core/pscsi_*/*/info
> ==> /sys/kernel/config/target/core/pscsi_0/changer1/info <==
> Status: ACTIVATED Max Queue Depth: 0 SectorSize: 0 HwMaxSectors: 0
> SCSI Device Bus Location: Channel ID: 0 Target ID: 0 LUN: 0 Host ID: 10
> Vendor: STK Model: L700 Rev: 0105
>
> ==> /sys/kernel/config/target/core/pscsi_1/tape1/info <==
> Status: ACTIVATED Max Queue Depth: 0 SectorSize: 0 HwMaxSectors: 0
> SCSI Device Bus Location: Channel ID: 0 Target ID: 1 LUN: 0 Host ID: 10
> Vendor: IBM Model: ULT3580-TD5 Rev: 0105
>
<SNIP>
>
> ==> /sys/kernel/config/target/core/pscsi_1/tape1/attrib/hw_block_size <==
> 0
>
> ==> /sys/kernel/config/target/core/pscsi_1/tape1/attrib/hw_max_sectors <==
> 0
>
> ==> /sys/kernel/config/target/core/pscsi_1/tape1/attrib/hw_queue_depth <==
> 32
...
>
> Here is my patch below. I went with <1 because I was too lazy to check
> if the value was signed or not, so it can probably be changed to an =
>
> It's not a big patch at all but it works quite well!
>
> *** a/drivers/target/target_core_device.c 2015-01-29
> 18:15:53.000000000 -0500
> --- b/drivers/target/target_core_device.c.mod 2016-10-30
> 03:21:41.955410797 -0400
> ***************
> *** 639,644 ****
> --- 639,647 ----
> * Limit max_sectors to a PAGE_SIZE aligned value for modern
> * transport_allocate_data_tasks() operation.
> */
> +
> + if (block_size < 1)
> + return 0;
> alignment = max(1ul, PAGE_SIZE / block_size);
> aligned_max_sectors = rounddown(max_sectors, alignment);
>
Thanks for posting.
Note PSCSI is (still) issuing MODE_SENSE to extract hw_sector_size at
'echo 1 > /sys/kernel/config/target/core/$PSCSI/$DEV/enable' time,
via pscsi_tape_read_blocksize() code.
pscsi_tape_read_blocksize() should be setting hw_sector_size = 1024 by
default when '0' blocksize is returned by MODE_SENSE. Looking again,
there is a bug wrt to MODE_SENSE failing + return without setting the
hw_sector_size = 1024 default. Perhaps that is what you've been hitting
with mvhtl..?
Looking further, I think hw_max_sectors is also being set to '0' for
TYPE_TAPE in pscsi_add_device_to_list(). This is likely where you ran
into problems with hw_max_sectors = PAGE_SIZE.
That said, here's a patch to address these two items. Let's see what
MODE_SENSE is returning.
At your earliest convenience, please verify.
diff --git a/drivers/target/target_core_pscsi.c b/drivers/target/target_core_pscsi.c
index 9125d93..33dc0fd 100644
--- a/drivers/target/target_core_pscsi.c
+++ b/drivers/target/target_core_pscsi.c
@@ -154,7 +154,7 @@ static void pscsi_tape_read_blocksize(struct se_device *dev,
buf = kzalloc(12, GFP_KERNEL);
if (!buf)
- return;
+ goto out_free;
memset(cdb, 0, MAX_COMMAND_SIZE);
cdb[0] = MODE_SENSE;
@@ -162,16 +162,20 @@ static void pscsi_tape_read_blocksize(struct se_device *dev,
ret = scsi_execute_req(sdev, cdb, DMA_FROM_DEVICE, buf, 12, NULL,
HZ, 1, NULL);
- if (ret)
+ if (ret) {
+ pr_warn("PSCSI TYPE_TAPE MODE_SENSE failed: %d\n", ret);
goto out_free;
+ }
/*
* If MODE_SENSE still returns zero, set the default value to 1024.
*/
sdev->sector_size = (buf[9] << 16) | (buf[10] << 8) | (buf[11]);
+ pr_info("PSCSI TYPE_TAPE detected hw sdev->sector_size: %d\n", sdev->sector_size);
+out_free:
if (!sdev->sector_size)
sdev->sector_size = 1024;
-out_free:
+
kfree(buf);
}
@@ -316,7 +320,7 @@ static int pscsi_add_device_to_list(struct se_device *dev,
dev->dev_attrib.hw_block_size = sd->sector_size;
dev->dev_attrib.hw_max_sectors =
- min_t(int, sd->host->max_sectors, queue_max_hw_sectors(q));
+ min_not_zero(sd->host->max_sectors, queue_max_hw_sectors(q));
dev->dev_attrib.hw_queue_depth = sd->queue_depth;
/*
@@ -339,8 +343,12 @@ static int pscsi_add_device_to_list(struct se_device *dev,
/*
* For TYPE_TAPE, attempt to determine blocksize with MODE_SENSE.
*/
- if (sd->type == TYPE_TAPE)
+ if (sd->type == TYPE_TAPE) {
pscsi_tape_read_blocksize(dev, sd);
+ dev->dev_attrib.hw_block_size = sd->sector_size;
+ pr_info("PSCSI: TYPE_TAPE setting hw_block_size: %u\n",
+ dev->dev_attrib.hw_block_size);
+ }
return 0;
}
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
