From: Vitaly Chikunov <vt@xxxxxxxxxxxx> Sent: Friday, February 4, 2022 8:06 PM
>
> On Fri, Feb 04, 2022 at 04:10:49PM +0000, Michael Kelley (LINUX) wrote:
> > From: Vitaly Chikunov <vt@xxxxxxxxxxxx> Sent: Friday, February 4, 2022 5:05 AM
> > >
> > > Clang 12.0.1 cannot understand that value 64 is excluded from the shift
> > > at compile time (for use in global context) resulting in build error:
> > >
> > > drivers/hv/vmbus_drv.c:2082:29: error: shift count >= width of type [-Werror,-
> > > Wshift-count-overflow]
> > > static u64 vmbus_dma_mask = DMA_BIT_MASK(64);
> > > ^~~~~~~~~~~~~~~~
> > > ./include/linux/dma-mapping.h:76:54: note: expanded from macro 'DMA_BIT_MASK'
> > > #define DMA_BIT_MASK(n) (((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
> > > ^ ~~~
> > >
> > > Avoid using DMA_BIT_MASK macro for that corner case.
> > >
> > > Cc: Tianyu Lan <Tianyu.Lan@xxxxxxxxxxxxx>
> > > Cc: Michael Kelley <mikelley@xxxxxxxxxxxxx>
> > > Cc: Long Li <longli@xxxxxxxxxxxxx>
> > > Cc: Wei Liu <wei.liu@xxxxxxxxxx>
> > > Signed-off-by: Vitaly Chikunov <vt@xxxxxxxxxxxx>
> > > ---
> > > drivers/hv/vmbus_drv.c | 3 ++-
> > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
> > > index 17bf55fe3169..a1306ca15d3f 100644
> > > --- a/drivers/hv/vmbus_drv.c
> > > +++ b/drivers/hv/vmbus_drv.c
> > > @@ -2079,7 +2079,8 @@ struct hv_device *vmbus_device_create(const guid_t *type,
> > > return child_device_obj;
> > > }
> > >
> > > -static u64 vmbus_dma_mask = DMA_BIT_MASK(64);
> > > +/* Use ~0ULL instead of DMA_BIT_MASK(64) to work around a bug in Clang. */
> > > +static u64 vmbus_dma_mask = ~0ULL;
> > > /*
> > > * vmbus_device_register - Register the child device
> > > */
> > > --
> > > 2.33.0
> >
> > Instead of the hack approach, does the following code rearrangement solve
> > the problem by eliminating the use of DMA_BIT_MASK(64) in a static initializer?
> > I don't have Clang handy to try it.
>
> Yes this compiles well on Clang.
>
> Vitaly,
>
Do you want to submit a patch with my approach, or should I do it? I'm
happy to do it, but I don't want to pre-empt what you have already
started without your OK.
Michael
> >
> > This approach also more closely follows the pattern used in other device types.
> >
> > Michael
> >
> >
> > diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
> > index 17bf55f..0d96634 100644
> > --- a/drivers/hv/vmbus_drv.c
> > +++ b/drivers/hv/vmbus_drv.c
> > @@ -2079,7 +2079,6 @@ struct hv_device *vmbus_device_create(const guid_t
> *type,
> > return child_device_obj;
> > }
> >
> > -static u64 vmbus_dma_mask = DMA_BIT_MASK(64);
> > /*
> > * vmbus_device_register - Register the child device
> > */
> > @@ -2120,8 +2119,9 @@ int vmbus_device_register(struct hv_device
> *child_device_obj)
> > }
> > hv_debug_add_dev_dir(child_device_obj);
> >
> > - child_device_obj->device.dma_mask = &vmbus_dma_mask;
> > child_device_obj->device.dma_parms = &child_device_obj->dma_parms;
> > + child_device_obj->device.dma_mask = &child_device_obj->dma_mask;
> > + dma_set_mask(&child_device_obj->device, DMA_BIT_MASK(64));
> > return 0;
> >
> > err_kset_unregister:
> > diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
> > index f565a89..fe2e017 100644
> > --- a/include/linux/hyperv.h
> > +++ b/include/linux/hyperv.h
> > @@ -1262,6 +1262,7 @@ struct hv_device {
> > struct vmbus_channel *channel;
> > struct kset *channels_kset;
> > struct device_dma_parameters dma_parms;
> > + u64 dma_mask;
> >
> > /* place holder to keep track of the dir for hv device in debugfs */
> > struct dentry *debug_dir;
