Re: [PATCH] arm: cns3xxx: fix writing to wrong PCI registers after alignment

[khalasa@piap.pl (Krzysztof Hałasa)]

Hi,

Bjorn Helgaas <helgaas@kernel.org> writes:

> 802b7c06adc7 replaced cns3xxx_pci_write_config(), which always used
> __raw_writel() so it only did 32-bit accesses, with
> pci_generic_config_write(), which uses writeb/writew/writel()
> depending on the size.
>
> 802b7c06adc7 also converted cns3xxx_pci_read_config() from always
> using __raw_readl() (a 32-bit access) to using
> pci_generic_config_read32(), which also always does a 32-bit access.
>
> This makes me think the cnx3xxx hardware is only capable of 32-bit
> accesses, and this patch should change the driver to use
> pci_generic_config_write32() instead of pci_generic_config_write() in
> addition to the mapping fix above.

Hasn't it already been verified that the CNS3xxx can do 8-bit accesses
(and probably 16-bit ones as well), and that the docs don't mention any
such limitation?

> I don't think hardware that only supports 32-bit PCI writes can be
> quite spec-compliant (see the comments in
> pci_generic_config_write32()).  So (1) you may see an additional
> dmesg warning when you convert to use it, and (2) it's possible that
> there may still be instability related to the corruption caused by
> using a 32-bit write when an 8-bit write was intended.

Right. I think IDE/SATA controllers are affected, for example.
-- 
Krzysztof Halasa

Industrial Research Institute for Automation and Measurements PIAP
Al. Jerozolimskie 202, 02-486 Warsaw, Poland