On Wed, 28 Jan 2026, "Kumar, Kaushlendra" <[email protected]> wrote: > On Wed, 28 Jan 2026, Jani Nikula <[email protected]> wrote: >> On Wed, 28 Jan 2026, Kaushlendra Kumar <[email protected]> wrote: >>> + count = min_t(size_t, GEN7_L3LOG_SIZE - offset, count); >> >> This may make count not be a multiple of sizeof(u32) again. Note how >> offset is treated below. > > Thank you for catching this! > > You're absolutely right about the alignment concern. While technically > in this specific case (GEN7_L3LOG_SIZE = 0x80, and offset is already > validated to be u32-aligned by l3_access_valid()), the subtraction > (GEN7_L3LOG_SIZE - offset) would always yield a u32-aligned result. > > Anything I am missing here?
No, I missed that l3_access_valid() ensures offset is already aligned. The whole thing is a bit too convoluted for my liking, but the whole thing matches i915_l3_read() with this patch, and there's no value in iterating this further. Reviewed-by: Jani Nikula <[email protected]> -- Jani Nikula, Intel
