Re: [PATCH v3 0/7] Improve PCI memory mapping API

[Niklas Cassel <cassel@kernel.org>]

On Fri, Oct 04, 2024 at 10:25:54PM +0900, Damien Le Moal wrote:
> On 10/4/24 22:13, Niklas Cassel wrote:
> > On Fri, Oct 04, 2024 at 02:07:35PM +0900, Damien Le Moal wrote:

(snip)

> > I think the cover letter is missing some text on how this series has been
> > tested.
> > 
> > In V2 I suggested adding a new option to pcitest.c, so that it doesn't
> > ensure that buffers are aligned. pci_test will currently use a 4k alignment
> > by default, and for some PCI device IDs and vendor IDs, it will ensure that
> > the buffers are aligned to something else. (E.g. for the PCI device ID used
> > by rk3588, buffers will be aligned to 64K.)
> > 
> > By adding an --no-alignment option to pci_test, we can ensure that this new
> > API is actually working.
> > 
> > Did you perhaps ifdef out all the alignment from pci_endpoint_test.c when
> > testing?
> 
> Yes I did. And I also extensively tested using the nvme epf function driver
> (coming soon !) which has very random PCI addresses for data buffers (e.g.
> BIOSes and GRUB are happy using on-stack totally unaligned buffers...).

I know that you did test using a nvme EPF :)

But for anyone reading the cover letter, it wasn't clear how this series
was tested, so it would have been nice if the information which you
provided above would have been part of the cover letter.


Kind regards,
Niklas