Re: [PULL 1/1] hw/nvme: fix endianness issue for shadow doorbells

[Klaus Jensen <its@irrelevant.dk>]

[-- Attachment #1: Type: text/plain, Size: 1421 bytes --]

On Jul 20 09:43, Peter Maydell wrote:
> On Wed, 19 Jul 2023 at 21:13, Michael Tokarev <mjt@tls.msk.ru> wrote:
> >
> > 19.07.2023 10:36, Klaus Jensen wrote:
> > pu(req->cmd.dptr.prp2);
> > > +    uint32_t v;
> >
> > >           if (sq) {
> > > +            v = cpu_to_le32(sq->tail);
> >
> > > -            pci_dma_write(pci, sq->db_addr, &sq->tail, sizeof(sq->tail));
> > > +            pci_dma_write(pci, sq->db_addr, &v, sizeof(sq->tail));
> >
> > This and similar cases hurts my eyes.
> >
> > Why we pass address of v here, but use sizeof(sq->tail) ?
> >
> > Yes, I know both in theory should be of the same size, but heck,
> > this is puzzling at best, and confusing in a regular case.
> >
> > Dunno how it slipped in the review, it instantly catched my eye
> > in a row of applied patches..
> >
> > Also, why v is computed a few lines before it is used, with
> > some expressions between the assignment and usage?
> >
> > How about the following patch:
> 
> If you're going to change this, better to take the approach
> Philippe suggested in review of using stl_le_pci_dma().
> 
> https://lore.kernel.org/qemu-devel/376e5e45-a3e7-0029-603a-b7ad9673fac4@linaro.org/
> 

Yup, that was my plan for next. But the original patch was already
verified on hardware and mutiple testes, so wanted to go with that for
the "fix".

But yes, I will refactor into the much nicer stl/ldl api.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]