Nouveau - Feb 2019 - About Porting Android to nouveau

Hi, guys:
	I’m a developer of FydeOS. We porting ChromiumOS to amd64 and arm platforms.
Now, I’m woking on porting android environment to Nvidia graphic cards. I have
experience to port android to Vmware(SVGA).
	I found two display formats were not supported to display in nouveau driver:
PIPE_FORMAT_R8G8B8A8_UNORM, PIPE_FORMAT_R8G8B8X8_UNORM. Which are defined at
nv50/nv50_formats.c line: 130,131.

C4(A, R8G8B8A8_UNORM, RGBA8_UNORM, R, G, B, A, UNORM, A8B8G8R8, IB),
F3(A, R8G8B8X8_UNORM, RGBX8_UNORM, R, G, B, xx, UNORM, A8B8G8R8, TB),

	I want the last params to be “TD” or “ID” (support PIPE_BIND_DISPLAY_TARGET) ,
to support android. I modified these code, but got display issues and kernel
issue:

[   93.313995] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002c0014
[   93.314082] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d0016
[   93.314112] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d0018
[   93.314215] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001a
[   93.314244] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001c
[   93.314272] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001e
[   93.314301] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002e0020
[   93.314325] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5 [007f6c6000
RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002e0022

	Any idea?

Best regards,

	Yang Tsao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<https://lists.freedesktop.org/archives/nouveau/attachments/20181121/60f8cef5/attachment-0001.html>

On Fri, Nov 23, 2018 at 11:00 PM Yang Tsao <yang at fydeos.io>
wrote:
>
> Hi, guys:
> I’m a developer of FydeOS. We porting ChromiumOS to amd64 and arm
platforms. Now, I’m woking on porting android environment to Nvidia graphic
cards. I have experience to port android to Vmware(SVGA).
> I found two display formats were not supported to display in nouveau
driver: PIPE_FORMAT_R8G8B8A8_UNORM, PIPE_FORMAT_R8G8B8X8_UNORM. Which are
defined at nv50/nv50_formats.c line: 130,131.
>
> C4(A, R8G8B8A8_UNORM, RGBA8_UNORM, R, G, B, A, UNORM, A8B8G8R8, IB),
> F3(A, R8G8B8X8_UNORM, RGBX8_UNORM, R, G, B, xx, UNORM, A8B8G8R8, TB),
>
> I want the last params to be “TD” or “ID” (support
PIPE_BIND_DISPLAY_TARGET) , to support android. I modified these code, but got
display issues and kernel issue:
That's generally fine (IB -> ID, TB -> TD), but you might want to flip
the BGRA formats to not have the display bits. Some winsys's get upset
if you have both RGBA and BGRA ordering though. (The reason that we
prefer BGRA8 on X is largely legacy.)
> [   93.313995] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002c0014
> [   93.314082] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d0016
> [   93.314112] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d0018
> [   93.314215] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001a
> [   93.314244] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001c
> [   93.314272] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001e
> [   93.314301] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002e0020
> [   93.314325] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch 5
[007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002e0022
I don't think this is related. Although I'm a bit puzzled as to what
this means. This is from index element submission (VB_ELEMENT_U16).
There's no real way to have a data error there... perhaps it think
it's out of bounds, or something else is not configured properly. This
could happen if there's multi-threaded rendering going on -- please
ensure that only a single thread ever calls into nouveau at any given
time in a process.

Cheers,

  -ilia

On Sat, Feb 16, 2019 at 2:27 AM Yang Tsao <yang at fydeos.io>
wrote:
>
> Thx for your reply.
> Chromium OS is running based DRM driver. There is no X server frame work. I
modified nv50/nv50_formats.c.
>
> And there also a few code related the display format:
> nv50_surface.c
> Function nv50_2d_format:
> I add:
>
> ———————————
>    switch (format) {
>    case PIPE_FORMAT_R8G8B8X8_UNORM:
>         return G80_SURFACE_FORMAT_RGBX8_UNORM;
>         break;
>    case PIPE_FORMAT_R8G8B8A8_UNORM:
>         return G80_SURFACE_FORMAT_RGBA8_UNORM;
>         break;
>    default:
>         break;
>    }
I don't understand why that was necessary. All 4 of RGBX8, RGBA8,
BGRX8, BGRA8 appear to be in NV50_ENG2D_SUPPORTED_FORMATS, so that
should fall under the

   if ((id >= 0xc0) && (NV50_ENG2D_SUPPORTED_FORMATS & (1ULL
<< (id - 0xc0))))
      return id;

logic. And if they weren't, the 2d engine wouldn't be happy if you
handed it unknown formats.
> ———————————
> Nv50_blit.h
>
> ———————————
> Function nv50_2d_format_supported
>
> Function nv50_2d_src_format_faithful
>
> ———————————————
>
> My question is : why only support
> The format  which id bigger than 0xc0 ?
>
> (id >= 0xc0) && (mask & (1ULL << (id - 0xc0)))
Well, the masks are just like that... we define, for nv50:

#define NV50_ENG2D_SUPPORTED_FORMATS 0xff0843e080608409ULL
#define NV50_ENG2D_NOCONVERT_FORMATS 0x0008402000000000ULL
#define NV50_ENG2D_LUMINANCE_FORMATS 0x0008402000000000ULL
#define NV50_ENG2D_INTENSITY_FORMATS 0x0000000000000000ULL
#define NV50_ENG2D_OPERATION_FORMATS 0x060001c000608000ULL

But the actual enum values are like

#define G80_SURFACE_FORMAT_RGBA32_FLOAT 0x000000c0
#define G80_SURFACE_FORMAT_RGBA32_SINT 0x000000c1
#define G80_SURFACE_FORMAT_RGBA32_UINT 0x000000c2

etc

So ... we just subtract 0xc0 to have the capability masks fit into
64-bit integers. IMO it's a bit of a premature optimization, but it's
way more like I'll screw something up when undoing it than just
leaving it alone. (Why did NVIDIA choose to start their hardware
format enums at 0xc0? No clue. But it's what the HW expects...)

Hope this helps,

  -ilia
>
>
> > On 24 Nov 2018, at 12:20 PM, Ilia Mirkin <imirkin at
alum.mit.edu> wrote:
> >
> > On Fri, Nov 23, 2018 at 11:00 PM Yang Tsao <yang at fydeos.io>
wrote:
> >>
> >> Hi, guys:
> >> I’m a developer of FydeOS. We porting ChromiumOS to amd64 and arm
platforms. Now, I’m woking on porting android environment to Nvidia graphic
cards. I have experience to port android to Vmware(SVGA).
> >> I found two display formats were not supported to display in
nouveau driver: PIPE_FORMAT_R8G8B8A8_UNORM, PIPE_FORMAT_R8G8B8X8_UNORM. Which
are defined at nv50/nv50_formats.c line: 130,131.
> >>
> >> C4(A, R8G8B8A8_UNORM, RGBA8_UNORM, R, G, B, A, UNORM, A8B8G8R8,
IB),
> >> F3(A, R8G8B8X8_UNORM, RGBX8_UNORM, R, G, B, xx, UNORM, A8B8G8R8,
TB),
> >>
> >> I want the last params to be “TD” or “ID” (support
PIPE_BIND_DISPLAY_TARGET) , to support android. I modified these code, but got
display issues and kernel issue:
> >
> > That's generally fine (IB -> ID, TB -> TD), but you might
want to flip
> > the BGRA formats to not have the display bits. Some winsys's get
upset
> > if you have both RGBA and BGRA ordering though. (The reason that we
> > prefer BGRA8 on X is largely legacy.)
> >
> >> [   93.313995] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002c0014
> >> [   93.314082] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d0016
> >> [   93.314112] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d0018
> >> [   93.314215] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001a
> >> [   93.314244] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001c
> >> [   93.314272] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002d001e
> >> [   93.314301] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002e0020
> >> [   93.314325] nouveau 0000:01:00.0: gr: DATA_ERROR 0000009c [] ch
5 [007f6c6000 RenderThread[5395]] subc 0 class a097 mthd 15f0 data 002e0022
> >
> > I don't think this is related. Although I'm a bit puzzled as
to what
> > this means. This is from index element submission (VB_ELEMENT_U16).
> > There's no real way to have a data error there... perhaps it think
> > it's out of bounds, or something else is not configured properly.
This
> > could happen if there's multi-threaded rendering going on --
please
> > ensure that only a single thread ever calls into nouveau at any given
> > time in a process.
> >
> > Cheers,
> >
> >  -ilia
>