Re: [REGRESSION] changes to driver_override parsing broke DPDK script

[Greg KH <gregkh@linuxfoundation.org>]

On Wed, Aug 10, 2022 at 08:54:36AM +0300, Krzysztof Kozlowski wrote:
> On 09/08/2022 22:21, Bjorn Helgaas wrote:
> > [+cc regressions list]
> > 
> > 23d99baf9d72 appeared in v5.19-rc1.
> > 
> > On Tue, Aug 09, 2022 at 11:29:43AM -0700, Stephen Hemminger wrote:
> >> This commit broke the driver override script in DPDK.
> >> This is an API/ABI breakage, please revert or fix the commit.
> >>
> >> Report of problem:
> >> http://mails.dpdk.org/archives/dev/2022-August/247794.html
> 
> Thanks for the report. I'll take a look.
> 
> >>
> >>
> >> commit 23d99baf9d729ca30b2fb6798a7b403a37bfb800
> >> Author: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> >> Date:   Tue Apr 19 13:34:28 2022 +0200
> >>
> >>     PCI: Use driver_set_override() instead of open-coding
> >>     
> >>     Use a helper to set driver_override to the reduce amount of duplicated
> >>     code.  Make the driver_override field const char, because it is not
> >>     modified by the core and it matches other subsystems.
> >>     
> >>     Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
> >>     Acked-by: Bjorn Helgaas <bhelgaas@google.com>
> >>     Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> >>     Link: https://lore.kernel.org/r/20220419113435.246203-6-krzysztof.kozlowski@linaro.org
> >>     Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> >>
> >>
> >> The script is sending single nul character to remove override
> >> and that no longer works.
> 
> The sysfs API clearly states:
> "and
>  may be cleared with an empty string (echo > driver_override)."
> Documentation/ABI/testing/sysfs-bus-pci
> 
> Sending other data and expecting the same result is not conforming to
> API. Therefore we have usual example of some undocumented behavior which
> user-space started relying on and instead using API, user-space expect
> that undocumented behavior to be back.
> 
> Yay! I wonder what is the point to even describe the ABI if user-space
> can simply ignore it?

One can argue that a string of just '\0' is an "empty string" and we
should be able to properly handle this in the kernel.  Heck,
"\0\0\0\0\0\0" is also an "empty string", right?

I don't have an issue with fixing the kernel up here, it should be able
to handle this.

thanks,

greg k-h