vermilion.c

Linux环境下视频显示卡设备的驱动程序源代码

/*
 * Copyright (c) Intel Corp. 2007.
 * All Rights Reserved.
 *
 * Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
 * develop this driver.
 *
 * This file is part of the Vermilion Range fb driver.
 * The Vermilion Range fb driver is free software;
 * you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * The Vermilion Range fb driver is distributed
 * in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this driver; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 *
 * Authors:
 *   Thomas Hellström <thomas-at-tungstengraphics-dot-com>
 *   Michel Dänzer <michel-at-tungstengraphics-dot-com>
 *   Alan Hourihane <alanh-at-tungstengraphics-dot-com>
 */

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/string.h>
#include <linux/delay.h>
#include <linux/mm.h>
#include <linux/fb.h>
#include <linux/pci.h>
#include <asm/cacheflush.h>
#include <asm/tlbflush.h>
#include <linux/mmzone.h>

/* #define VERMILION_DEBUG */

#include "vermilion.h"

#define MODULE_NAME "vmlfb"

#define VML_TOHW(_val, _width) ((((_val) << (_width)) + 0x7FFF - (_val)) >> 16)

static struct mutex vml_mutex;
static struct list_head global_no_mode;
static struct list_head global_has_mode;
static struct fb_ops vmlfb_ops;
static struct vml_sys *subsys = NULL;
static char *vml_default_mode = "1024x768@60";
static struct fb_videomode defaultmode = {
	NULL, 60, 1024, 768, 12896, 144, 24, 29, 3, 136, 6,
	0, FB_VMODE_NONINTERLACED
};

static u32 vml_mem_requested = (10 * 1024 * 1024);
static u32 vml_mem_contig = (4 * 1024 * 1024);
static u32 vml_mem_min = (4 * 1024 * 1024);

static u32 vml_clocks[] = {
	6750,
	13500,
	27000,
	29700,
	37125,
	54000,
	59400,
	74250,
	120000,
	148500
};

static u32 vml_num_clocks = ARRAY_SIZE(vml_clocks);

/*
 * Allocate a contiguous vram area and make its linear kernel map
 * uncached.
 */

static int vmlfb_alloc_vram_area(struct vram_area *va, unsigned max_order,
				 unsigned min_order)
{
	gfp_t flags;
	unsigned long i;

	max_order++;
	do {
		/*
		 * Really try hard to get the needed memory.
		 * We need memory below the first 32MB, so we
		 * add the __GFP_DMA flag that guarantees that we are
		 * below the first 16MB.
		 */

		flags = __GFP_DMA | __GFP_HIGH;
		va->logical =
			 __get_free_pages(flags, --max_order);
	} while (va->logical == 0 && max_order > min_order);

	if (!va->logical)
		return -ENOMEM;

	va->phys = virt_to_phys((void *)va->logical);
	va->size = PAGE_SIZE << max_order;
	va->order = max_order;

	/*
	 * It seems like __get_free_pages only ups the usage count
	 * of the first page. This doesn't work with fault mapping, so
	 * up the usage count once more (XXX: should use split_page or
	 * compound page).
	 */

	memset((void *)va->logical, 0x00, va->size);
	for (i = va->logical; i < va->logical + va->size; i += PAGE_SIZE) {
		get_page(virt_to_page(i));
	}

	/*
	 * Change caching policy of the linear kernel map to avoid
	 * mapping type conflicts with user-space mappings.
	 */
	set_pages_uc(virt_to_page(va->logical), va->size >> PAGE_SHIFT);

	printk(KERN_DEBUG MODULE_NAME
	       ": Allocated %ld bytes vram area at 0x%08lx\n",
	       va->size, va->phys);

	return 0;
}

/*
 * Free a contiguous vram area and reset its linear kernel map
 * mapping type.
 */

static void vmlfb_free_vram_area(struct vram_area *va)
{
	unsigned long j;

	if (va->logical) {

		/*
		 * Reset the linear kernel map caching policy.
		 */

		set_pages_wb(virt_to_page(va->logical),
				 va->size >> PAGE_SHIFT);

		/*
		 * Decrease the usage count on the pages we've used
		 * to compensate for upping when allocating.
		 */

		for (j = va->logical; j < va->logical + va->size;
		     j += PAGE_SIZE) {
			(void)put_page_testzero(virt_to_page(j));
		}

		printk(KERN_DEBUG MODULE_NAME
		       ": Freeing %ld bytes vram area at 0x%08lx\n",
		       va->size, va->phys);
		free_pages(va->logical, va->order);

		va->logical = 0;
	}
}

/*
 * Free allocated vram.
 */

static void vmlfb_free_vram(struct vml_info *vinfo)
{
	int i;

	for (i = 0; i < vinfo->num_areas; ++i) {
		vmlfb_free_vram_area(&vinfo->vram[i]);
	}
	vinfo->num_areas = 0;
}

/*
 * Allocate vram. Currently we try to allocate contiguous areas from the
 * __GFP_DMA zone and puzzle them together. A better approach would be to
 * allocate one contiguous area for scanout and use one-page allocations for
 * offscreen areas. This requires user-space and GPU virtual mappings.
 */

static int vmlfb_alloc_vram(struct vml_info *vinfo,
			    size_t requested,
			    size_t min_total, size_t min_contig)
{
	int i, j;
	int order;
	int contiguous;
	int err;
	struct vram_area *va;
	struct vram_area *va2;

	vinfo->num_areas = 0;
	for (i = 0; i < VML_VRAM_AREAS; ++i) {
		va = &vinfo->vram[i];
		order = 0;

		while (requested > (PAGE_SIZE << order) && order < MAX_ORDER)
			order++;

		err = vmlfb_alloc_vram_area(va, order, 0);

		if (err)
			break;

		if (i == 0) {
			vinfo->vram_start = va->phys;
			vinfo->vram_logical = (void __iomem *) va->logical;
			vinfo->vram_contig_size = va->size;
			vinfo->num_areas = 1;
		} else {
			contiguous = 0;

			for (j = 0; j < i; ++j) {
				va2 = &vinfo->vram[j];
				if (va->phys + va->size == va2->phys ||
				    va2->phys + va2->size == va->phys) {
					contiguous = 1;
					break;
				}
			}

			if (contiguous) {
				vinfo->num_areas++;
				if (va->phys < vinfo->vram_start) {
					vinfo->vram_start = va->phys;
					vinfo->vram_logical =
						(void __iomem *)va->logical;
				}
				vinfo->vram_contig_size += va->size;
			} else {
				vmlfb_free_vram_area(va);
				break;
			}
		}

		if (requested < va->size)
			break;
		else
			requested -= va->size;
	}

	if (vinfo->vram_contig_size > min_total &&
	    vinfo->vram_contig_size > min_contig) {

		printk(KERN_DEBUG MODULE_NAME
		       ": Contiguous vram: %ld bytes at physical 0x%08lx.\n",
		       (unsigned long)vinfo->vram_contig_size,
		       (unsigned long)vinfo->vram_start);

		return 0;
	}

	printk(KERN_ERR MODULE_NAME
	       ": Could not allocate requested minimal amount of vram.\n");

	vmlfb_free_vram(vinfo);

	return -ENOMEM;
}

/*
 * Find the GPU to use with our display controller.
 */

static int vmlfb_get_gpu(struct vml_par *par)
{
	mutex_lock(&vml_mutex);

	par->gpu = pci_get_device(PCI_VENDOR_ID_INTEL, VML_DEVICE_GPU, NULL);

	if (!par->gpu) {
		mutex_unlock(&vml_mutex);
		return -ENODEV;
	}

	mutex_unlock(&vml_mutex);

	if (pci_enable_device(par->gpu) < 0)
		return -ENODEV;

	return 0;
}

/*
 * Find a contiguous vram area that contains a given offset from vram start.
 */
static int vmlfb_vram_offset(struct vml_info *vinfo, unsigned long offset)
{
	unsigned long aoffset;
	unsigned i;

	for (i = 0; i < vinfo->num_areas; ++i) {
		aoffset = offset - (vinfo->vram[i].phys - vinfo->vram_start);

		if (aoffset < vinfo->vram[i].size) {
			return 0;
		}
	}

	return -EINVAL;
}

/*
 * Remap the MMIO register spaces of the VDC and the GPU.
 */

static int vmlfb_enable_mmio(struct vml_par *par)
{
	int err;

	par->vdc_mem_base = pci_resource_start(par->vdc, 0);
	par->vdc_mem_size = pci_resource_len(par->vdc, 0);
	if (!request_mem_region(par->vdc_mem_base, par->vdc_mem_size, "vmlfb")) {
		printk(KERN_ERR MODULE_NAME
		       ": Could not claim display controller MMIO.\n");
		return -EBUSY;
	}
	par->vdc_mem = ioremap_nocache(par->vdc_mem_base, par->vdc_mem_size);
	if (par->vdc_mem == NULL) {
		printk(KERN_ERR MODULE_NAME
		       ": Could not map display controller MMIO.\n");
		err = -ENOMEM;
		goto out_err_0;
	}

	par->gpu_mem_base = pci_resource_start(par->gpu, 0);
	par->gpu_mem_size = pci_resource_len(par->gpu, 0);
	if (!request_mem_region(par->gpu_mem_base, par->gpu_mem_size, "vmlfb")) {
		printk(KERN_ERR MODULE_NAME ": Could not claim GPU MMIO.\n");
		err = -EBUSY;
		goto out_err_1;
	}
	par->gpu_mem = ioremap_nocache(par->gpu_mem_base, par->gpu_mem_size);
	if (par->gpu_mem == NULL) {
		printk(KERN_ERR MODULE_NAME ": Could not map GPU MMIO.\n");
		err = -ENOMEM;
		goto out_err_2;
	}

	return 0;

out_err_2:
	release_mem_region(par->gpu_mem_base, par->gpu_mem_size);
out_err_1:
	iounmap(par->vdc_mem);
out_err_0:
	release_mem_region(par->vdc_mem_base, par->vdc_mem_size);
	return err;
}

/*
 * Unmap the VDC and GPU register spaces.
 */

static void vmlfb_disable_mmio(struct vml_par *par)
{
	iounmap(par->gpu_mem);
	release_mem_region(par->gpu_mem_base, par->gpu_mem_size);
	iounmap(par->vdc_mem);
	release_mem_region(par->vdc_mem_base, par->vdc_mem_size);
}

/*
 * Release and uninit the VDC and GPU.
 */

static void vmlfb_release_devices(struct vml_par *par)
{
	if (atomic_dec_and_test(&par->refcount)) {
		pci_set_drvdata(par->vdc, NULL);
		pci_disable_device(par->gpu);
		pci_disable_device(par->vdc);
	}
}

/*
 * Free up allocated resources for a device.
 */

static void __devexit vml_pci_remove(struct pci_dev *dev)
{
	struct fb_info *info;
	struct vml_info *vinfo;
	struct vml_par *par;

	info = pci_get_drvdata(dev);
	if (info) {
		vinfo = container_of(info, struct vml_info, info);
		par = vinfo->par;
		mutex_lock(&vml_mutex);
		unregister_framebuffer(info);
		fb_dealloc_cmap(&info->cmap);
		vmlfb_free_vram(vinfo);
		vmlfb_disable_mmio(par);
		vmlfb_release_devices(par);
		kfree(vinfo);
		kfree(par);
		mutex_unlock(&vml_mutex);
	}
}

static void vmlfb_set_pref_pixel_format(struct fb_var_screeninfo *var)
{
	switch (var->bits_per_pixel) {
	case 16:
		var->blue.offset = 0;
		var->blue.length = 5;
		var->green.offset = 5;
		var->green.length = 5;
		var->red.offset = 10;
		var->red.length = 5;
		var->transp.offset = 15;
		var->transp.length = 1;
		break;
	case 32:
		var->blue.offset = 0;
		var->blue.length = 8;
		var->green.offset = 8;
		var->green.length = 8;
		var->red.offset = 16;
		var->red.length = 8;
		var->transp.offset = 24;
		var->transp.length = 0;
		break;
	default:
		break;
	}

	var->blue.msb_right = var->green.msb_right =
	    var->red.msb_right = var->transp.msb_right = 0;
}

/*
 * Device initialization.
 * We initialize one vml_par struct per device and one vml_info
 * struct per pipe. Currently we have only one pipe.
 */

static int __devinit vml_pci_probe(struct pci_dev *dev,
				   const struct pci_device_id *id)
{
	struct vml_info *vinfo;
	struct fb_info *info;
	struct vml_par *par;
	int err = 0;

	par = kzalloc(sizeof(*par), GFP_KERNEL);
	if (par == NULL)
		return -ENOMEM;

	vinfo = kzalloc(sizeof(*vinfo), GFP_KERNEL);
	if (vinfo == NULL) {
		err = -ENOMEM;
		goto out_err_0;
	}

	vinfo->par = par;
	par->vdc = dev;
	atomic_set(&par->refcount, 1);

	switch (id->device) {
	case VML_DEVICE_VDC:
		if ((err = vmlfb_get_gpu(par)))
			goto out_err_1;
		pci_set_drvdata(dev, &vinfo->info);
		break;
	default:
		err = -ENODEV;
		goto out_err_1;
		break;
	}

	info = &vinfo->info;
	info->flags = FBINFO_DEFAULT | FBINFO_PARTIAL_PAN_OK;

	err = vmlfb_enable_mmio(par);
	if (err)
		goto out_err_2;

	err = vmlfb_alloc_vram(vinfo, vml_mem_requested,
			       vml_mem_contig, vml_mem_min);
	if (err)
		goto out_err_3;

	strcpy(info->fix.id, "Vermilion Range");
	info->fix.mmio_start = 0;
	info->fix.mmio_len = 0;
	info->fix.smem_start = vinfo->vram_start;
	info->fix.smem_len = vinfo->vram_contig_size;
	info->fix.type = FB_TYPE_PACKED_PIXELS;
	info->fix.visual = FB_VISUAL_TRUECOLOR;
	info->fix.ypanstep = 1;
	info->fix.xpanstep = 1;
	info->fix.ywrapstep = 0;
	info->fix.accel = FB_ACCEL_NONE;
	info->screen_base = vinfo->vram_logical;
	info->pseudo_palette = vinfo->pseudo_palette;
	info->par = par;
	info->fbops = &vmlfb_ops;
	info->device = &dev->dev;

	INIT_LIST_HEAD(&vinfo->head);
	vinfo->pipe_disabled = 1;
	vinfo->cur_blank_mode = FB_BLANK_UNBLANK;

	info->var.grayscale = 0;
	info->var.bits_per_pixel = 16;
	vmlfb_set_pref_pixel_format(&info->var);

	if (!fb_find_mode
	    (&info->var, info, vml_default_mode, NULL, 0, &defaultmode, 16)) {
		printk(KERN_ERR MODULE_NAME ": Could not find initial mode\n");
	}

	if (fb_alloc_cmap(&info->cmap, 256, 1) < 0) {
		err = -ENOMEM;
		goto out_err_4;
	}

	err = register_framebuffer(info);
	if (err) {
		printk(KERN_ERR MODULE_NAME ": Register framebuffer error.\n");
		goto out_err_5;
	}

	printk("Initialized vmlfb\n");

	return 0;

out_err_5:
	fb_dealloc_cmap(&info->cmap);
out_err_4:
	vmlfb_free_vram(vinfo);
out_err_3:
	vmlfb_disable_mmio(par);
out_err_2:
	vmlfb_release_devices(par);
out_err_1:
	kfree(vinfo);
out_err_0:
	kfree(par);
	return err;
}

static int vmlfb_open(struct fb_info *info, int user)
{
	/*
	 * Save registers here?
	 */
	return 0;
}

static int vmlfb_release(struct fb_info *info, int user)
{
	/*
	 * Restore registers here.
	 */

	return 0;
}

static int vml_nearest_clock(int clock)
{

	int i;
	int cur_index;
	int cur_diff;
	int diff;

	cur_index = 0;
	cur_diff = clock - vml_clocks[0];
	cur_diff = (cur_diff < 0) ? -cur_diff : cur_diff;
	for (i = 1; i < vml_num_clocks; ++i) {
		diff = clock - vml_clocks[i];
		diff = (diff < 0) ? -diff : diff;
		if (diff < cur_diff) {
			cur_index = i;
			cur_diff = diff;