Brett Russ wrote:2) Isn't it wrong for the IRQ disable at the chip to occur *after*
free_irq() is called to disconnect the handler (independent of
question 1...since this is the case currently)? Granted, all of the
ports have gone through scsi_remove_host() but theoretically there
still is a possibility the chip could interrupt.
If I'm wrong on both counts I'll blame it on need for sleep... :-)
Moving AHCI away from ata_pci_remove_one() was actually intentional. This gives the driver a bit more freedom: legacy region handling and ->host_stop() became unnecessary. Also, I was concerned that ata_pci_remove_one() might grow into a one-size-fits-all unmaintainable behemoth.
Short term, if one were being obsessive, a potential cleanup could be to make common the two loops in ahci_remove_one()/ata_pci_remove_one().
Long term, libata driver API should become more like the register_foo()/unregister_foo() interfaces you see elsewhere in the kernel. That direction has the potential to shake up the current code path through ata_pci_remove_one().
So... your patch, while technically correct, is going in the opposite direction to where I want to go :)