Re: [PATCH] staging: comedi: s626: Remove pci-dma-compat wrapper APIs.

[Ian Abbott]

On 11/07/2020 14:38, Christophe JAILLET wrote:
> Le 11/07/2020 à 14:35, Suraj Upadhyay a écrit :
> > The legacy API wrappers in include/linux/pci-dma-compat.h
> > should go away as it creates unnecessary midlayering
> > for include/linux/dma-mapping.h APIs, instead use dma-mapping.h
> > APIs directly.
> >
> > The patch has been generated with the coccinelle script below
> > and compile-tested.
> >
> > [...]
> > @@ expression E1, E2, E3; @@
> > - pci_alloc_consistent(E1, E2, E3)
> > + dma_alloc_coherent(&E1->dev, E2, E3, GFP_ATOMIC)
> >
> > @@ expression E1, E2, E3; @@
> > - pci_zalloc_consistent(E1, E2, E3)
> > + dma_alloc_coherent(&E1->dev, E2, E3, GFP_ATOMIC)
>
> This is the tricky part of this kind of cleanup, see below.
>
> GFP_ATOMIC can't be wrong because it is was exactly what was done with 
> the pci_ function.
> However, most of the time, it can safely be replaced by GFP_KERNEL which 
> gives more opportunities to the memory allocator.
>
> > [...]
> > diff --git a/drivers/staging/comedi/drivers/s626.c 
> > b/drivers/staging/comedi/drivers/s626.c
> > index 084a8e7b9fc2..c159416662fd 100644
> > --- a/drivers/staging/comedi/drivers/s626.c
> > +++ b/drivers/staging/comedi/drivers/s626.c
> > @@ -2130,13 +2130,15 @@ static int s626_allocate_dma_buffers(struct 
> > comedi_device *dev)
> >       void *addr;
> >       dma_addr_t appdma;
> > -    addr = pci_alloc_consistent(pcidev, S626_DMABUF_SIZE, &appdma);
> > +    addr = dma_alloc_coherent(&pcidev->dev, S626_DMABUF_SIZE, &appdma,
> > +                  GFP_ATOMIC);
> >       if (!addr)
> >           return -ENOMEM;
> >       devpriv->ana_buf.logical_base = addr;
> >       devpriv->ana_buf.physical_base = appdma;
> > -    addr = pci_alloc_consistent(pcidev, S626_DMABUF_SIZE, &appdma);
> > +    addr = dma_alloc_coherent(&pcidev->dev, S626_DMABUF_SIZE, &appdma,
> > +                  GFP_ATOMIC);
> >       if (!addr)
> >           return -ENOMEM;
> >       devpriv->rps_buf.logical_base = addr;
> 's626_allocate_dma_buffers()' is only called from 's626_auto_attach()'.
> In this function, around the call to 's626_allocate_dma_buffers()', you 
> can see:
>    - a few lines before, a call to 'comedi_alloc_devpriv()'
>
>    - a few lines after, a call to 'comedi_alloc_subdevices()'
>
> These 2 functions make some memory allocation using GFP_KERNEL.
>
> So it is likely that using GFP_KERNEL in your proposal is also valid.

Indeed, GFP_KERNEL is perfectly fine here.  It could be done as a 
follow-up patch, or done in a v2 of this patch with a mention in the 
patch description.

--
-=( Ian Abbott <abbotti@xxxxxxxxx> || Web: www.mev.co.uk )=-
-=( MEV Ltd. is a company registered in England & Wales. )=-
-=( Registered number: 02862268.  Registered address:    )=-
-=( 15 West Park Road, Bramhall, STOCKPORT, SK7 3JZ, UK. )=-