vmware-svga/examples/screen-simple/main.c
Micah Dowty 68478eab4b There are many new tests and examples that we wrote internally at VMware,
but couldn't release immediately since they depended on virtual GPU features
which were not yet publicly released in any products.  This checkin moves those
features from our internal repository to the open source repository. Future   
development on these tests and examples will take place directly in the open
source repository.

The primary feature added by this patch is 'Screen Object', a new dynamic
display management extension supported by Workstation 7.0 and Fusion 3.0.
See the README for a quick explanation.
2009-10-21 20:20:49 +00:00

196 lines
6.3 KiB
C

/*
* Simple example for Screen Objects in the SVGA device.
*
* This is an optional feature of the VMware SVGA device which allows
* multiple screens to be managed dynamically. It replaces legacy
* multi-monitor mode. It also allows the video driver to manage its
* own framebuffer memory. The host will never DMA to or from this
* memory unless the guest requests a DMA operation, and the guest is
* free to allocate or reallocate framebuffer memory at any time.
*
* This is a bare-bones example which uses the Screen Object extension
* to create a single screen and draw a test pattern to it.
*
* This example does not use the legacy Guest Framebuffer (GFB) memory
* at BAR1 at all. A system memory GMR is used as a data source for
* blits to the screen.
*/
#include "svga.h"
#include "gmr.h"
#include "screen.h"
#include "intr.h"
#include "datafile.h"
DECLARE_DATAFILE(testPatternData, testpattern_z);
/*
* main --
*
* Main loop and initialization.
*/
int
main(void)
{
Intr_Init();
Intr_SetFaultHandlers(SVGA_DefaultFaultHandler);
SVGA_Init();
GMR_Init();
Heap_Reset();
/*
* When you use Screen Objects, you no longer need the legacy SVGA
* framebuffer or display mode. You still need to call SetMode to
* enable the device and the command FIFO, but it doesn't have to
* have a valid video mode. Setting a 0x0 video mode explicitly
* tells the device that we don't need the legacy framebuffer.
*/
SVGA_SetMode(0, 0, 32);
/*
* Screen_Init requires the SVGA FIFO, which is set up in SVGA_SetMode.
*/
Screen_Init();
/*
* Define a screen.
*
* This screen has ID zero, which is both the first valid ID and
* the default ID used by the legacy SVGA device interfaces. This
* will replace any legacy screen with our new screen.
*
* We mark this screen as 'primary', and root it in the virtual
* coordinate space. We'll just pick a totally arbitrary root
* position.
*/
SVGAScreenObject myScreen = {
.structSize = sizeof(SVGAScreenObject),
.id = 0,
.flags = SVGA_SCREEN_HAS_ROOT | SVGA_SCREEN_IS_PRIMARY,
.size = { 640, 480 },
.root = { -500, 10000 },
};
Screen_Define(&myScreen);
/*
* Create a system memory framebuffer.
*
* This step is optional. Most drivers will probably want a
* framebuffer, but the SVGA device does not require that you have
* one. For example, a character-cell driver could point the GMRFB
* at a font table, and draw characters using individual
* GMRFB-to-screen blits from the font table. A more complex
* driver which wants to avoid tearing could double-buffer the
* whole screen, or even create a queue of smaller update-sized
* buffers which would be used as GMRFB memory.
*
* There are three steps for creating a framebuffer:
*
* 1. Allocate system memory. This is done using a trivial
* heap allocator, which uses a chunk of contiguous system
* memory that begins immediately after our binary's last
* segment ends.
*
* This memory doesn't need to be contiguous, but contiguous
* memory may give better performance, so it's preferred.
*
* 2. Define a Guest Memory Region (GMR), which is basically a
* page table which lets the SVGA device access this memory.
* GMRs are how we support discontiguous memory. Any pages
* which are part of a GMR should be locked down, so your
* OS's virtual memory subsystem can't move them or page
* them out.
*
* Defining a GMR is relatively costly, so it's best if
* drivers use GMRs for very coarse-grained memory allocation.
* It would make sense to use a GMR for a general-purpose DMA
* heap, or for a 2D framebuffer- but not for a single DMA
* buffer.
*
* 3. Define a GMRFB. This is a tiny piece of state which tells
* future blit operations where to pull their source data
* from. It's perfectly fine to redefine the GMRFB as often
* as you like, but in this simple example we'll set it once
* and leave it alone.
*/
/*
* Steps 1 and 2 are handled by GMR_DefineContiguous, in our
* small 'gmr.c' utility library.
*
* Use GMR ID 0, the first user-defined GMR. We'll allocate
* enough memory for a 32 bit per pixel framebuffer.
*/
const uint32 gmrId = 0;
const uint32 bitsPerPixel = 32;
const uint32 colorDepth = 24;
const uint32 bytesPerPixel = bitsPerPixel >> 3;
const uint32 fbBytesPerLine = myScreen.size.width * bytesPerPixel;
const uint32 fbSizeInBytes = fbBytesPerLine * myScreen.size.height;
const uint32 fbSizeInPages = (fbSizeInBytes + PAGE_MASK) / PAGE_SIZE;
PPN fbFirstPage = GMR_DefineContiguous(gmrId, fbSizeInPages);
uint32 *fbPointer = PPN_POINTER(fbFirstPage);
/*
* Step 3: Use the DEFINE_GMRFB command to tell the device about
* our framebuffer and its format. It is in our user-defined GMR,
* at offset zero.
*
* the SVGAGMRImageFormat type is a packed 32-bit word which
* encodes our pixel size and color depth. (The upper 16 bits are
* reserved for future formats.)
*/
SVGAGuestPtr fbGuestPtr = {
.gmrId = gmrId,
.offset = 0,
};
SVGAGMRImageFormat fbFormat = {{{
.bitsPerPixel = bitsPerPixel,
.colorDepth = colorDepth,
}}};
Screen_DefineGMRFB(fbGuestPtr, fbBytesPerLine, fbFormat);
/*
* Now we have a framebuffer! Fill it with a test pattern,
* and blit it to the screen.
*/
SVGASignedPoint blitOrigin = { 0, 0 };
SVGASignedRect blitDest = { 0, 0, myScreen.size.width, myScreen.size.height };
DataFile_Decompress(testPatternData, (void*) fbPointer, fbSizeInBytes);
Screen_BlitFromGMRFB(&blitOrigin, &blitDest, myScreen.id);
uint32 dmaFence = SVGA_InsertFence();
/*
* Wait for the blit's DMA to complete, then clobber our
* framebuffer with a different value, to keep the SVGA device
* honest. It shouldn't access the GMRFB memory any more now that
* the blit is done. If we were doing animation, we could use this
* guarantee to perform double-buffering, for example.
*/
SVGA_SyncToFence(dmaFence);
memset(fbPointer, 0x42, fbSizeInBytes);
/*
* Our work here is done. Sleep indefinitely.
*/
while (1) {
Intr_Halt();
}
return 0;
}