Re: [PATCH] dmaengine: switch from 'pci_' to 'dma_' API

[Christophe JAILLET]

Le 23/08/2021 à 21:47, Julia Lawall a écrit :
> On Mon, 23 Aug 2021, Christophe JAILLET wrote:
>
> > Le 23/08/2021 à 09:29, Andy Shevchenko a écrit :
> > > On Sun, Aug 22, 2021 at 02:40:22PM +0200, Christophe JAILLET wrote:
> > > > The wrappers in include/linux/pci-dma-compat.h should go away.
> > > >
> > > > The patch has been generated with the coccinelle script below.
> > > >
> > > > It has been hand modified to use 'dma_set_mask_and_coherent()' instead of
> > > > 'pci_set_dma_mask()/pci_set_consistent_dma_mask()' when applicable.
> > > > This is less verbose.
> > > >
> > > > It has been compile tested.
> > >
> > > > @@
> > > > expression e1, e2;
> > > > @@
> > > > -    pci_set_consistent_dma_mask(e1, e2)
> > > > +    dma_set_coherent_mask(&e1->dev, e2)
> > >
> > > Can we, please, replace this long noise in the commit message with a link to
> > > a
> > > script in coccinelle data base?
> >
> > Hi,
> >
> > There is no script in the coccinelle data base up to now, and there is no
> > point in adding one now.
> > The goal of these patches is to remove a deprecated API, so when the job will
> > be finished, this script would be of no use and would be removed.
> >
> > However, I agree that the script as-is is noisy.
> >
> > I'll replace it with a link to a message already available in lore.
>
> You can perhaps include a script that represents a very typical case or
> the specific case that is relevant to the patch.
>
> julia
Hi julia,

I still have to post ~9 of such patches.
I'll try your proposal to only include relevant part of the script.
It should be do-able.

The most puzzling in the remaining files is:
   drivers/char/xillybus/xillybus_pcie.c

In this driver, 'xilly_pci_direction()' looks odd and somewhat useless 
(but I'll let it more or less as-is. It may catch corner cases)

What puzzles me is "struct xilly_mapping" where the "void *" seems to
hold different kinds of pointer depending of the context.

I could add some casting but would prefer to avoid this and have a 
cleaner solution.

I've not look in detail yet but my first attempt were not very conclusive.



"message/fusion" has not been Acked yet.
All patches have been posted but there are huge and I can easily 
understand that reviewing them is complex because of the GFP_KERNEL I 
introduce. Only one has been merged yet.
Moreover, according to -next logs, no single other patch has been merged 
since end of June.

CJ