On Mon, Apr 18, 2011 at 10:03:06PM +0100, Alan Cox wrote:
So has this been benchmarked - intuitively I'd agree and expect that a shadowfb driver ought to give best performance.
No, but it's noticably nicer to use under virt-manager. I'll try to come up with some numbers.
+/* Map the framebuffer from the card and configure the core */ +static int cirrus_vram_init(struct cirrus_device *cdev) +{
- int ret;
- /* BAR 0 is VRAM */
- cdev->mc.vram_base = pci_resource_start(cdev->ddev->pdev, 0);
- /* We have 4MB of VRAM */
- cdev->mc.vram_size = 4 * 1024 * 1024;
For real hardware at least you should check that the pci resource is at least 4Mb long before doing this, otherwise the resulting progression in the fail case is that you map some other device into user address space, which isn't good!
True. The PCI table is restrictive enough that it won't bind to real hardware, so I don't know if it's worth it to be paranoid.
+static void cirrus_encoder_mode_set(struct drm_encoder *encoder,
struct drm_display_mode *mode,struct drm_display_mode *adjusted_mode)+{
- struct drm_device *dev = encoder->dev;
- struct cirrus_device *cdev = dev->dev_private;
- unsigned char tmp;
- switch (encoder->crtc->fb->bits_per_pixel) {
- case 8:
tmp = 0x0;break;- case 16:
/* Enable 16 bit mode */WREG_HDR(0x01);If you switch back from 16 does this not need clearing ?
Nope. qemu just looks at this to distinguish between 15 and 16 bit, and I've no intention of supporting 15 bit...