On Thu, 4 Dec 2025 at 19:34, Jim MacArthur <[email protected]> wrote:
>
> If both frame and element count are 65535, which appears valid from my
> reading of the OMAP5912 documentation, then some of the calculations
> will overflow the 32-bit signed integer range and produce a negative
> min_elems value.
>
> Raised by #3204 (https://gitlab.com/qemu-project/qemu/-/issues/3204).
>
> Signed-off-by: Jim MacArthur <[email protected]>
> ---
>  hw/dma/omap_dma.c | 32 +++++++++++++++++++++-----------
>  1 file changed, 21 insertions(+), 11 deletions(-)
>
> diff --git a/hw/dma/omap_dma.c b/hw/dma/omap_dma.c
> index 101f91f4a3..93e6503ff9 100644
> --- a/hw/dma/omap_dma.c
> +++ b/hw/dma/omap_dma.c
> @@ -504,9 +504,19 @@ static void omap_dma_transfer_setup(struct soc_dma_ch_s 
> *dma)
>      struct omap_dma_channel_s *ch = dma->opaque;
>      struct omap_dma_s *s = dma->dma->opaque;
>      int frames, min_elems, elements[__omap_dma_intr_last];
> +    uint64_t frames64, frame64, elements64, element64;
>
>      a = &ch->active_set;
>
> +    /*
> +     * We do maths with the frame and element fields which exceeds
> +     * a signed 32-bit integer, so convert all these to 64 bit for future 
> use.
> +     */
> +    frames64 = a->frames;
> +    frame64 = a->frame;
> +    elements64 = a->elements;
> +    element64 = a->element;

The real hardware is surely not doing 64-bit arithmetic though.
It's presumably working with elements * frames as unsigned 32 bit
multiplication.

-- PMM

Reply via email to