RE: [RFC PATCH] Let device drivers disable msi on shutdown

[Keith Busch <keith.busch@intel.com>]

On Tue, 24 Jun 2014, Elliott, Robert (Server Storage) wrote:
> 1. That will cover the .shutdown function used by mptfc.c, mptspi.c,
> and mptscsih.c, but mptsas.c uses mptsas_shutdown rather than
> mptscsih_shutdown.  It doesn't call pci_disable_msi either.

Missed that; thanks.

> 2. mptscsih_suspend is another caller of mptscsih_shutdown
> (although the latter does nothing right now).  This is done
> before calling mpt_suspend, which writes to some chip registers
> to disable interrupts before calling pci_disable_msi.  Adding a
> pci_disable_msi call before those writes might not be safe.

Clearly more paths into this function than I investigated.

> 3. That mptscsih_suspend call chain will result in calling
> pci_disable_msi twice, which might trigger this in
> pci_disable_msi -> pci_msi_shutdown:
>        BUG_ON(list_empty(&dev->msi_list));

Ah, but pci_disable_msi will not even invoke pci_msi_shutdown:

         if (!pci_msi_enable || !dev || !dev->msi_enabled)
 	                return;

The same check is done in pci_msi_shutdown. If that check wasn't there,
every driver would hit that BUG_ON since most every other pci driver
disables their interrupts in their shutdown path before returning, and
then pci-driver calls it again! MPT appears to be the exception.