PCI PM: clear IO and MEM when disabling a device
PATCH 8/9 PCI PM: clear IO and MEM when disabling a device
This patch modifies pci_disable_device() to clear IO and MEM from the COMMAND PCI config register. This is required before entering D3, but also probably a good general practice for system suspend.
--- pci.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/pci/pci.c +6 -2
@@ -312,8 +312,12 @@u16 pci_command;pci_read_config_word (dev,PCI_COMMAND , &pci_command); -if (pci_command &PCI_COMMAND_MASTER ) { - pci_command &= ~PCI_COMMAND_MASTER ; +if (pci_command & (PCI_COMMAND_MASTER | +PCI_COMMAND_IO | +PCI_COMMAND_MEMORY )) { + pci_command &= ~ (PCI_COMMAND_MASTER | +PCI_COMMAND_IO | +PCI_COMMAND_MEMORY );pci_write_config_word (dev,PCI_COMMAND , pci_command); } dev->is_busmaster =0 ;
Quoted 4 lines · Adam Belay
This patch modifies
pci_disable_device()to clear IO and MEM from the COMMAND PCI config register. This is required before entering D3, but also probably a good general practice for system suspend.
I'd submitted a patch to do this about a month ago, see
http://marc.theaimsgroup.com/?l=linux-pci&m=114669552428309&w=2
That patch does the same thing, but in arch specific code (since that's where memory and IO decode get enabled today). This patch will cover the other arch's too, so I'm fine with this if this is the right thing to do. My original patch is already in Greg's tree and needs to be removed if this patch is accepted.
thanks,
Rajesh
Quoted 3 lines · Adam Belay
This patch modifies
pci_disable_device()to clear IO and MEM from the COMMAND PCI config register. This is required before entering D3, but also probably a good general practice for system suspend.
And will break a great deal of platforms. Don't do it :) The problem is, I think, mostly related to firmware issues.
Ben.
Quoted 9 lines · Benjamin Herrenschmidt
Quoted 3 lines · Adam Belay
This patch modifies
pci_disable_device()to clear IO and MEM from the COMMAND PCI config register. This is required before entering D3, but also probably a good general practice for system suspend.And will break a great deal of platforms. Don't do it :) The problem is, I think, mostly related to firmware issues.
Ben.
If the device must be enabled before passing control to firmware, then shouldn't its controlling driver never call pci_disable_device()? I'd like to have pci_disable_device() actually disable devices, as this is needed for D3.
Thanks,
Adam