On Tue, Jul 19, 2022 at 12:55 PM Stafford Horne <shorne@xxxxxxxxx> wrote:
> diff --git a/drivers/comedi/drivers/comedi_isadma.c b/drivers/comedi/drivers/comedi_isadma.c
> index 700982464c53..508421809128 100644
> --- a/drivers/comedi/drivers/comedi_isadma.c
> +++ b/drivers/comedi/drivers/comedi_isadma.c
> @@ -104,8 +104,10 @@ unsigned int comedi_isadma_poll(struct comedi_isadma *dma)
>
> flags = claim_dma_lock();
> clear_dma_ff(desc->chan);
> +#ifdef CONFIG_X86_32
> if (!isa_dma_bridge_buggy)
> disable_dma(desc->chan);
> +#endif
There is a logic mistake here: if we are on something other than x86-32,
this always needs to call the disable_dma()/enable_dma().
Not sure how to best express this in a readable way, something like this
would work:
#ifdef CONFIG_X86_32
if (!isa_dma_bridge_buggy)
#endif
disable_dma(desc->chan);
or possibly at the start of this file, a
#ifndef CONFIG_X86_32
#define isa_dma_bridge_buggy 0
#endif
Or we could try to keep the generic definition in a global header
like linux/isa-dma.h.
> --- a/sound/core/isadma.c
> +++ b/sound/core/isadma.c
> @@ -73,8 +73,10 @@ unsigned int snd_dma_pointer(unsigned long dma, unsigned int size)
>
> flags = claim_dma_lock();
> clear_dma_ff(dma);
> +#ifdef CONFIG_X86_32
> if (!isa_dma_bridge_buggy)
> disable_dma(dma);
> +#endif
> result = get_dma_residue(dma);
> /*
Same here.
Arnd
