This is a followup to the earlier discussion of MPT watchdog device removal calling directly into PCI core API: [1] http://thread.gmane.org/gmane.linux.scsi/80629 I've tested two safer, alternative methods of removing MPT hosts from the SCSI topology. Both involve wrapping the existing MPT .remove routine, ensuring mutual exclusion between regularly scheduled PCI device removal and the drivers' periodic watchdog thread. Both changes tested well against surprise PCI removal of LSI SAS 9211-8i HBAs while driving direct IO out to attached disks. The first version is straightforward, essentially just adding a common mutex and checking that the driver still cares about a given PCI device before removing. The second version is an attempt to detach only from the SCSI topology as soon as possible. Later PCI removal cleans up the rest of the resources. Neither patch maps cleanly to the MPT fusion driver as its watchdog thread resides in the mptbase module. The code currently uses PCI core API pci_remove_bus_device to route around the driver's module dependencies to call from mptbase to mptsas: Module Size Used by mptsas 62366 8 << PCI .remove mptscsih 38803 1 mptsas mptbase 99878 2 mptsas,mptscsih << watchdog The fusion driver has devised other means of calling from mptbase to mptscsih, for example, via the schedule_dead_ioc_flush_running_cmds function pointer. The two changes I explored were made with a relatively light hand, so I didn't know how best to proceed with a MPT fusion patch. Comments on either removal patch strategy welcome. It would also be great if we had documentation guiding the SCSI LLDs in how to safely and completely remove attached hosts in hotplug and defective HW scenarios. Regards, -- Joe -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
