Re: [Qemu-devel] Re: [PATCH] Reducing X communication bandwidth, take 2

[Anthony Liguori <anthony@codemonkey.ws>]

Brian Johnson wrote:
> Paul Brook wrote:
>>>> Will this work also for the CL542x adaptor?  (Does that fall in the
>>>> category of vga?)  My current hack works for with/without -std-vga 
>>>> and I
>>>> think that's because it lives "underneath" both, in the connection to
>>>> SDL.
>>> Each adapter will have to do it's own minimization but that's sort of
>>> the write thing anyway IMHO.  How granular each update is really only
>>> depends on the adapter.  For instance, the VMware adapter really
>>> shouldn't need to do any minimization at all.
>>
>> It would be nice if we could share the framebuffer blitting routines. 
>> We've currently got 3 different implementations (vga/cirrus, tcx and 
>> pl110) of basically the same framebuffer rendering routines.
>
> Take a look at the video code in BasiliskII / SheepShaver, a 68k/PPC 
> classic MacOS emulator written by Gwenolé Beauchesne:
>
> http://gwenole.beauchesne.info/projects/sheepshaver/
>
> It contains optimized code (source level) for blitting between various 
> bit depths and endiannesses.  See 
> SheepShaver-2.3/src/Unix/video_blit.{h,cpp} in the sources.
>
> It also uses a technique called "video on segfault" (VOSF) to improve 
> performance on platforms which support it:  

This is what QEMU already does.  In fact, it's an exceedingly common 
technique.

Regards,

Anthony Liguori

> rather than testing each store to see if it modifies the framebuffer, 
> it keeps the framebuffer write-protected (via mprotect(), or the 
> equivalent on non-POSIX systems) and uses a SIGSEGV handler to catch 
> stores to the buffer.  When a page receives a store, the handler 
> unprotects the page and updates a bitmap of modified pages.  Every so 
> often a display update thread wakes up, consults the bitmap, 
> calculates the updated region, blits it to the screen (using the 
> optimized blitters), and clears the bitmap.  See 
> SheepShaver-2.3/src/Unix/video_vosf.h, 
> SheepShaver-2.3/src/Unix/video_x.cpp, and other files.
>
> The emulators also have alternative techniques for tracking update 
> regions on systems for which VOSF is not supported.  But VOSF is 
> almost always a big win.  And most modern OSes can support it without 
> trouble.
>
> The code is in C++ so it can't be dropped in directly, but Some of the 
> techniques may be useful in qemu.
>
>
> Brian J. Johnson
>
>
>