Re: [PATCH v5 2/2] PCI: dwc: Add support for 64-bit MSI target address

[Robin Murphy]

On 2022-09-09 15:47, Christoph Hellwig wrote:
> On Fri, Sep 09, 2022 at 02:47:19PM +0100, Robin Murphy wrote:
> > On 2022-09-09 14:29, Christoph Hellwig wrote:
> > > On Thu, Aug 25, 2022 at 06:50:25PM +0000, Will McVicker wrote:
> > > > Since not all devices require a 32-bit MSI address, add support to the
> > > > PCIe host driver to allow setting the DMA mask to 64-bits if the 32-bit
> > > > allocation fails. This allows kernels to disable ZONE_DMA32 and bounce
> > > > buffering (swiotlb) without risking not being able to get a 32-bit address
> > > > during DMA allocation.
> > >
> > > Umm.  You can't just disable ZONE_DMA32.  Linux absolutely requires a
> > > 32-bit dma mask to work, it is in fact the implicit default.
> >
> > Eh, it's behind CONFIG_EXPERT, which makes it enough of a "I think I know
> > what I'm doing and accept responsibility for picking up the pieces if it
> > breaks" thing.
>
> Seem like indeed on arm64 there is a way to disable it.  The x86 model
> is to just select it unconditionally, which I think is the right way
> if we don't want to get into completely random failures.

IIRC there were reasons for wanting as much ZONE_NORMAL memory as 
possible; for the embedded folks who are already typically running with 
"swiotlb=noforce" to save memory because they know their hardware, we 
may as well let them have the footgun. Distros and other general-purpose 
configs should rightly not be going anywhere near this.

Cheers,
Robin.