RE: [PATCH 1/1] Drivers: hv: vmbus: Rework use of DMA_BIT_MASK(64)

[Michael Kelley (LINUX)]

From: Nathan Chancellor <nathan@xxxxxxxxxx> Sent: Sunday, February 6, 2022 5:38 PM
> 
> Hi Michael,
> 
> On Sun, Feb 06, 2022 at 11:36:56AM -0800, Michael Kelley wrote:
> > Using DMA_BIT_MASK(64) as an initializer for a global variable
> > causes problems with Clang 12.0.1. The compiler doesn't understand
> > that value 64 is excluded from the shift at compile time, resulting
> > in a build error.
> >
> > While this is a compiler problem, avoid the issue by setting up
> > the dma_mask memory as part of struct hv_device, and initialize
> > it using dma_set_mask().
> >
> > Reported-by: Nathan Chancellor <nathan@xxxxxxxxxx>
> > Reported-by: Vitaly Chikunov <vt@xxxxxxxxxxxx>
> > Reported-by: Jakub Kicinski <kuba@xxxxxxxxxx>
> > Fixes: 743b237c3a7b ("scsi: storvsc: Add Isolation VM support for storvsc driver")
> > Signed-off-by: Michael Kelley <mikelley@xxxxxxxxxxxxx>
> 
> Thanks a lot for working around this. I am hoping that this will be
> fixed in clang soon, as it is high priority on our list of issues to
> fix. Once it has been fixed, we should be able to undo this workaround
> in one way or another.

FWIW, the new code is as good a solution as the old code.  The new code
also follows some existing patterns, such as with struct platform_device.
As such, I don't think of this as a workaround that needs to be undone
in the future.

Michael

> 
> I can confirm the warning is resolved, which will allow us to build
> ARCH=arm64 and ARCH=x86_64 allmodconfig with -Werror on mainline with
> clang once another fix [1] is merged.
> 
> Reviewed-by: Nathan Chancellor <nathan@xxxxxxxxxx>
> Tested-by: Nathan Chancellor <nathan@xxxxxxxxxx>
>