On 05/01/2026 16:34, Marek Vasut wrote: > On 1/5/26 2:09 PM, Matt Coster wrote: >> On 06/11/2025 23:24, Marek Vasut wrote: >>> Fix support for build on 32bit systems. Include >>> linux/io-64-nonatomic-hi-lo.h >>> to provide non-atomic readq()/writeq()/ioread64()/iowrite64() accessors, and >>> use __ffs64() instead of plain ffs() on 64bit number SZ_1T. >>> >>> This allows this driver to bind on Renesas R-Car H2 which contains >>> Rogue G6400 BVNC 1.39.4.1 . >>> >>> Signed-off-by: Marek Vasut <[email protected]> >> >> Hi Marek, > > Hello Matt, > >> My apologies, this one appears to have slipped through the cracks on our >> end. > > No worries. > >>> +++ b/drivers/gpu/drm/imagination/Kconfig >>> @@ -3,7 +3,7 @@ >>> config DRM_POWERVR >>> tristate "Imagination Technologies PowerVR (Series 6 and later) & IMG >>> Graphics" >>> - depends on (ARM64 || RISCV && 64BIT) >>> + depends on ARM || ARM64 || RISCV >> >> This seems fine to me. Do you know any reason why the single change >> below might *not* be sufficient to support non-64-bit riscv? I can't >> think of any, I just wanted to double check this had been considered. > I do not have any 32bit RV to test this on, I only have 32bit ARM (R-Car H2).
I appreciate that you'd like to work on getting these older cores supported in the driver, but as it stands there's no real way to test this change beyond ensuring that it compiles. I've asked around internally and the consensus is that communicating with the GPU on a 32-bit platform requires more consideration than just using the shims provided by io-64-nonatomic-hi-lo.h to avoid introducing races and other potential security holes. My suggestion is that this patch be shelved for now and used as part of a larger series later which adds basic support for a core on a 32-bit platform. Cheers, Matt -- Matt Coster E: [email protected]
OpenPGP_signature.asc
Description: OpenPGP digital signature
