Re: [PATCH 7/7] misc: pci_endpoint_test: Add support for PCITEST_IRQ_TYPE_AUTO

[Niklas Cassel <cassel@kernel.org>]

On Tue, Mar 18, 2025 at 02:26:56PM +0530, Manivannan Sadhasivam wrote:
> > 
> > I guess one option would be to remove the
> > "pci_ep_ioctl(PCITEST_SET_IRQTYPE, 1);" calls from the test cases that you
> > added, and then let the test cases themselves set the proper irq_type in
> > the BAR register. But, wouldn't that be an API change? READ/WRITE/COPY
> > test ioctls have always respected the (a successful) PCITEST_SET_IRQTYPE,
> > now all of a sudden, they shouldn't?
> > 
> 
> This makes no difference IMO. The previous behavior which you explained above,
> ignored the result of 'pcitest -i 1'. And it was not user configurable. I think
> the original intention was to use MSI for tests if available, else use whatever
> the platform supports.
> 
> If you want to restore the original behavior, you should remove the ASSERT_EQ()
> from READ/WRITE/COPY tests first. Then to ensure that the tests make use of the
> supported IRQ type, you can have the logic in the READ/WRITE/COPY tests itself.
> If test->irq_type != PCITEST_IRQ_TYPE_UNDEFINED, then just use whatever the
> test->irq_type is. Otherwise, use whatever the platform supports.

I can submit a patch series that modifies PCITEST_{READ,WRITE,COPY} to always
figure out the IRQ type to use by themselves.

But you can't have the cake and eat it too.

Either PCITEST_{READ,WRITE,COPY} always ignores PCITEST_SET_IRQTYPE or
they don't always ignore PCITEST_SET_IRQTYPE.

Only ignoring it "if test->irq_type != PCITEST_IRQ_TYPE_UNDEFINED"
makes no sense IMO.


Kind regards,
Niklas