Brian King wrote: > Kilian CAVALOTTI wrote: > > >>What I can't really understand is how a device could fail on a READ CAPACITY >>16 if plugged on an HBA, but work if plugged on another one... > > > Looks to me like the aic7xxx driver does not change max_cmd_len from > the default of 12, so any 16 byte cdb gets failed back by scsi core. Indeed, the DID_ABORT on the READ CAPACITY(16) suggests that the command is not making it out to the scsi bus. Strange that in lk 2.6 the scsi subsystem still defaults to a maximum of 12 byte scsi commands, unless overridden by the LLD. Perhaps Kilian could try the following patch (borrowed from scsi_debug). The patch is against lk 2.6.15-rc1 but should by widely applicable (unless the aic7... series really does have a 12 byte limit). I will report back if this patch has any adverse impact on "sg_readcap -16" which should report: "illegal request/invalid command operation code" on my hardware. Doug Gilbert
--- linux/drivers/scsi/aic7xxx/aic7xxx_osm.c 2005-11-15 13:15:26.000000000 +1000 +++ linux/drivers/scsi/aic7xxx/aic7xxx_osm.c2615rc1big 2005-11-19 08:45:08.000000000 +1000 @@ -637,6 +637,9 @@ if (bootverbose) sdev_printk(KERN_INFO, sdev, "Slave Configure\n"); + if (sdev->host->max_cmd_len < MAX_COMMAND_SIZE) + sdev->host->max_cmd_len = MAX_COMMAND_SIZE; + ahc_linux_device_queue_depth(sdev); /* Initial Domain Validation */ --- linux/drivers/scsi/aic7xxx/aic79xx_osm.c 2005-11-15 13:15:26.000000000 +1000 +++ linux/drivers/scsi/aic7xxx/aic79xx_osm.c2615rc1big 2005-11-19 08:45:29.000000000 +1000 @@ -559,6 +559,9 @@ if (bootverbose) sdev_printk(KERN_INFO, sdev, "Slave Configure\n"); + if (sdev->host->max_cmd_len < MAX_COMMAND_SIZE) + sdev->host->max_cmd_len = MAX_COMMAND_SIZE; + ahd_linux_device_queue_depth(sdev); /* Initial Domain Validation */
