On Wed, Jan 10, 2024 at 12:52:50PM -0800, Bart Van Assche wrote: > On 1/10/24 12:49, Niklas Cassel wrote: > > However, I'm worried that applying that libata patch will simply hide > > an actual problem in SCSI, which might lead to someone else stumbling > > on this SCSI bug in the future. > > > > Thoughts? > > Since the hang is caused by submitting a SCSI pass-through command, I'm > not sure this issue can be called a SCSI core bug. Aren't users on their > own who mix SCSI pass-through commands with commands submitted by the sd > driver? I tend to agree with you, but at the same time, it used to work, so in one way it is a regression... (I can hear Linus's voice saying "Never break user space!"... in my head.) And since there seems to be at least one script mixing hdparm with sysfs delete: https://www.thinkwiki.org/wiki/How_to_hotswap_Ultrabay_devices#Script_for_Ultrabay_eject there are probably others as well. We can work around it in libata. I'm just worried if there are any SCSI drivers using scsi_eh_scmd_add() together with scsi_eh_flush_done_q() that will hit this in the future. Perhaps some libsas driver? Kind regards, Niklas
