Re: [PATCH] habanalabs: Remove unused enum member DMA_SRAM_TO_SRAM

[Greg Kroah-Hartman]

On Sun, Jan 23, 2022 at 06:11:54PM +0200, Oded Gabbay wrote:
> On Sun, Jan 23, 2022 at 6:08 PM Greg Kroah-Hartman
> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Sun, Jan 23, 2022 at 06:01:53PM +0200, Oded Gabbay wrote:
> > > On Sun, Jan 23, 2022 at 5:04 PM Greg Kroah-Hartman
> > > <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Sun, Jan 23, 2022 at 04:43:15PM +0200, Oded Gabbay wrote:
> > > > > On Tue, Jan 11, 2022 at 1:47 PM Cai Huoqing <cai.huoqing@xxxxxxxxx> wrote:
> > > > > >
> > > > > > The driver don't support the SRAM-to-SRAM translation of DMA,
> > > > > > so remove 'DMA_SRAM_TO_SRAM'.
> > > > > >
> > > > > > Signed-off-by: Cai Huoqing <cai.huoqing@xxxxxxxxx>
> > > > > > ---
> > > > > >  drivers/misc/habanalabs/include/goya/goya_packets.h | 1 -
> > > > > >  1 file changed, 1 deletion(-)
> > > > > >
> > > > > > diff --git a/drivers/misc/habanalabs/include/goya/goya_packets.h b/drivers/misc/habanalabs/include/goya/goya_packets.h
> > > > > > index ef54bad20509..25fbebdc6143 100644
> > > > > > --- a/drivers/misc/habanalabs/include/goya/goya_packets.h
> > > > > > +++ b/drivers/misc/habanalabs/include/goya/goya_packets.h
> > > > > > @@ -36,7 +36,6 @@ enum goya_dma_direction {
> > > > > >         DMA_SRAM_TO_HOST,
> > > > > >         DMA_DRAM_TO_HOST,
> > > > > >         DMA_DRAM_TO_DRAM,
> > > > > > -       DMA_SRAM_TO_SRAM,
> > > > > >         DMA_ENUM_MAX
> > > > > >  };
> > > > > >
> > > > > > --
> > > > > > 2.25.1
> > > > > >
> > > > >
> > > > > This is a general spec file in our s/w stack, and therefore a change
> > > > > in it in the driver will cause our driver to be out of sync with our
> > > > > user-space stack. i.e. the value of DMA_ENUM_MAX will be different in
> > > > > the driver and in the user-space stack. I don't know if there will be
> > > > > any consequences but I prefer not to risk it.
> > > >
> > > > If this is synced to userspace, shouldn't it be in a uapi file with a
> > > > specific value associated with it?
> > > >
> > > > thanks,
> > > >
> > > > greg k-h
> > >
> > > Yes, it was a mistake from day 1. A mistake we didn't repeat in future ASICs.
> > > I take great care of putting anything that is synced between driver
> > > and userspace in our uapi file.
> > >
> > > Having said that, after almost 3 years of having this mistake, I feel
> > > it is not too disastrous to leave it as is.
> > > Our Goya s/w stack is pretty much stable and I don't want to make
> > > changes without good reason.
> >
> > You should just move this to a uapi file to show that it is something
> > that can't be changed.  Can't you do that without breaking anything on
> > the kernel side?
> On the kernel side there isn't any problem with moving it.
> What I'm afraid of, is that the userspace will now have a duplicate
> definition of this enum.
> Both in the original file, which is included in the user space stack
> from day 1, and from our uapi file.
> In short, I'm afraid of breaking its compilation.
> If you think that's acceptable, I don't have a problem moving it.
> 
> >
> > Otherwise maintaining this is going to be hard as no one has a clue that
> > this is a value that userspace uses.
> 
> I can document it with a BIG comment in the file.
> And as I said, Goya s/w stack is pretty much fixed and I don't believe
> is going to change anymore.

The trick is long-term maintaince for the next 20+ years of this code in
the kernel.  A huge comment would help, but even better to move this
where it needs to go in a uapi file so that it makes it obvious we can
not change this.

thanks,

greg k-h