Hi,
We found two issues in the code during suspend:
1. Andy Shevchenko found that, the save restore of pci config space
might cause potential issue. Current code uses
pci_read_config_dword() to read pci config header. However
hardware is not obliged to react correctly when trying to read
two/three 'adjacent' pci config registers with one dword read.
Q1: Should we save/restore the pci config space header according
to the PCI spec strictly(pci_read_config_dword() for 32bit, while
pci_read_config_word()
for 16bits, etc)?
2. The pci config space of some problematic devices(or due to firmware
bug) might become inaccessible after resumed from S3(suspend to mem)
on VM.
Q2: Should we do sanity check on pci config space before saving them?
Say, invoke pci_dev_is_present() before suspend, if the pci config space is
not sane, bypass the config space saving process, because there's no need
to save invalid pci config space.
Comments would be appreciated.
--
thanks,
Chenyu
