Re: [PATCH v3 0/8] xen/arm: Add support for SMMUv3 driver

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Mon, 14 Dec 2020, Rahul Singh wrote:
> Hello Julien,Stefano
> 
> > On 11 Dec 2020, at 2:29 pm, Julien Grall <julien@xxxxxxx> wrote:
> > 
> > Hi Rahul,
> > 
> > On 10/12/2020 16:56, Rahul Singh wrote:
> >> This patch series is v3 of the work to add support for the SMMUv3 driver.
> >> Approach taken is to first merge the Linux copy of the SMMUv3 driver
> >> (tag v5.8.18) and then modify the driver to build on XEN.
> >> MSI and PCI ATS functionality are not supported. Code is not tested and
> >> compiled. Code is guarded by the flag CONFIG_PCI_ATS and CONFIG_MSI to 
> >> compile
> >> the driver.
> >> Code specific to Linux is removed from the driver to avoid dead code.
> >> Driver is currently supported as tech preview.
> >> Following functionality should be supported before driver is out for tech
> >> preview
> >> 1. Investigate the timing analysis of using spin lock in place of mutex 
> >> when
> >>    attaching a  device to SMMU.
> >> 2. Merged the latest Linux SMMUv3 driver code once atomic operation is
> >>    available in XEN.
> >> 3. PCI ATS and MSI interrupts should be supported.
> >> 4. Investigate side-effect of using tasklet in place of threaded IRQ and 
> >> fix
> >>    if any.
> > In your last e-mail, you wrote that you would investigate and then come 
> > back to use. It wasn't clear that this meant you will not deal with it in 
> > this series.
> > 
> 
> Yes, I will investigate the side-effect of using tasklet but not part of this 
> patch series. It will be great if we proceed on this patch series as it is 
> (using tasklet in place of threaded-IRQ).
> 
> >> 5. fallthorugh keyword should be supported.
> > 
> > This one should really be done now... It is not a complicated one...
> 
> Ok. I will implement in next version.
>  
> > 
> >> 6. Implement the ffsll function in bitops.h file.
> 
> While implementing the code I found out that Linux is using the built-in 
> function “__builtin_ffsll() “ for ffsll and there is no implementation 
> available in Linux for ffsll.
> If we implement the ffsll in XEN we will diverge from Linux. I am thinking of 
> adding the below code in "xen/include/asm-arm/bitops.h”. 
> Please suggest if it is okay?
> 
> static always_inline int ffsll(long long x)                                   
>         
> {                                                                             
>   
>     return __builtin_ffsll(x);                                                
>     
> }

I think that's OK if it builds with clang too.