On Tue, Jun 13, 2023 at 12:39 AM Philippe Mathieu-Daudé <phi...@linaro.org>
wrote:
> On 12/6/23 21:28, Bernhard Beschow wrote:
> >
> >
> > Am 9. Juni 2023 18:51:19 UTC schrieb Mark Cave-Ayland <
> mark.cave-ayl...@ilande.co.uk>:
> >> The aim here is to eliminate any device-specific registers from the
> main BMDMA
> >> bar memory region so it can potentially be moved into the shared PCI
> IDE device.
> >>
> >> For each BMDMA bus create a new cmd646-bmdma-specific memory region
> representing
> >> the device-specific BMDMA registers and then map them using aliases
> onto the
> >> existing BMDMAState memory region.
> >>
> >> Signed-off-by: Mark Cave-Ayland <mark.cave-ayl...@ilande.co.uk>
> >> ---
> >> hw/ide/cmd646.c | 111 +++++++++++++++++++++++++++++++---------
> >> include/hw/ide/cmd646.h | 4 ++
> >> 2 files changed, 90 insertions(+), 25 deletions(-)
>
>
> >> struct CMD646IDEState {
> >> PCIIDEState parent_obj;
> >> +
> >> + MemoryRegion bmdma_mem[2];
> >> + MemoryRegion bmdma_mem_alias[2][2];
> >
> > The added complexity of a two-dimensional alias array seems like a tough
> call for me. I'm not totally against it but I'm reluctant.
>
> Alternative:
>
> struct {
> MemoryRegion mem;
> MemoryRegion mem_alias[2];
>
If `mem_alias` became an anonymous struct as well we could avoid fiddling
with two indices in a matrix, lowering the complexity.
Best regards,
Bernhard
> } bmdma[2];
>
> >> };
> >>
> >> #endif
> >
>
>
