Damien,

quoted

quoted

How about set larger valid value between sdkp->max_xfer_blocks,
and sdkp->opt_xfer_blocks to rw_max?

Again, if your device reports an opt_xfer_blocks value that is too
small for its own good, that is a problem with this device.

Correct. It is very much intentional that we do not default to issuing
the largest commands supported by the physical hardware.

If the device is not reporting an optimal transfer size, and the block
layer default is too small, the solution is to adjust max_sectors_kb in
sysfs (by adding a udev rule, for instance).

Sorry, I said wrong.
As others mentioned, it's device problem if opt_xfer_blocks is wrong.
kernel modification is not needed for it.

I want to discuss using max_xfer_blocks instead of opt_xfer_blocks as a optional.
For example, device reports opt_xfer_blocks is 512KB and 1MB as a
max_xfer_blocks too. Currently rw_max is set with 512KB only.
I want a option to select max_xfer_blocks for rw_max.
Are there any historical problem when rw_max is set with max_xfer_blocks?

How about below approach if max_xfer_blocks can be set to rw_max?
It's using queue flag as a option. Do you have good idea to suggest?

static bool sd_validate_max_xfer_size(struct scsi_disk *sdkp,
				      unsigned int dev_max)
{
	struct request_queue *q = sdkp->disk->queue;

	if (!blk_queue_allow_sd_rw_max(q))
		return false;

	if (sdkp->max_xfer_blocks == 0)
		return false;

        ......

	return true;
}

static int sd_revalidate_disk(struct gendisk *disk)
{
	......

	if (sd_validate_max_xfer_size(sdkp, dev_max)) {
		q->limits.io_opt = logical_to_bytes(sdp, sdkp->max_xfer_blocks);
		rw_max = logical_to_sectors(sdp, sdkp->max_xfer_blocks);
	} else if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
		q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
		rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
	} else {
	
	......
}

-- 
Martin K. Petersen	Oracle Linux Engineering