On 7/21/20 2:53 PM, Bjorn Helgaas wrote:
On Fri, Jul 17, 2020 at 01:20:22PM -0700, Kuppuswamy, Sathyanarayanan wrote:
Hi Bjorn,
On 7/17/20 12:56 PM, Bjorn Helgaas wrote:
From: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
pci_aer_clear_device_status() clears the error bits in the PCIe Device
Status Register (PCI_EXP_DEVSTA). Every PCIe device has this register,
regardless of whether it supports AER.
Since its not related to AER, can we move it out of AER driver ? May be
to pci.c ?
OK. pci.c is really a grab bag of random stuff, but it *is* true that
this doesn't really seem to belong in aer.c.
So I don't mind moving it to pci.c (does just before
pcie_clear_root_pme_status() seem like a reasonable spot?)
But looking at this again makes me wonder whether putting the
pcie_aer_is_native() test inside pcie_clear_device_status() is the
right thing. It seems like that test fits better with the AER code,
i.e., in the *callers* of pcie_clear_device_status().
Yes. pcie_aer_is_native() should be left in AER driver.
It would mean repeating the test, since we call it twice, but it seems
like it might match up with the spec better. And I have a slight
aversion to functions that can silently return without doing what it
looks like they're supposed to do.
I can fix this all up if that seems right. Or let me know if you have
alternate ideas.
Agree with your approach. Its ok to add separate check for
pcie_aer_is_native().
Bjorn
--
Sathyanarayanan Kuppuswamy
Linux Kernel Developer
