> From: Ming Lei <[email protected]> > Sent: Wednesday, May 29, 2019 7:57 PM > On Thu, May 23, 2019 at 11:18 AM Dexuan Cui <[email protected]> wrote: > > ... > > I checked some PCI HBA drivers, and they use scsi_block_requests(), but > > as I described above, I don't know how setting a flag can prevent another > > CPU from submitting I/O requests. > > > scsi_device_quiesce() has been called by scsi_dev_type_suspend() to prevent > any non-pm request from entering queue.
Oh, yes, you're right! I think I overlooked this function. So it looks all the sdevs are suspended and resumed automatically in drivers/scsi/scsi_pm.c, and the low level SCSI adapter driver only needs to suspend/resume the state of the adapter. But why do we need to call scsi_block_requests() in some low level SCSI adapter drivers's .suspend? e.g. in mpt3sas_driver's scsih_suspend(): if no new I/O requests can be submitted from the sdevs in the hibernation scenario, can you please explain why we need the scsi_block_requests() here? > Or do you still see IO requests coming during hibernation? No. I just didn't realize how scsi_bus_pm_ops works. Thanks for your reply! Thanks, -- Dexuan
