vermilion.c

Go to the documentation of this file.

/*
 * 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/slab.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;
        }
    }
    return vml_clocks[cur_index];
}

static int vmlfb_check_var_locked(struct fb_var_screeninfo *var,
                  struct vml_info *vinfo)
{
    u32 pitch;
    u64 mem;
    int nearest_clock;
    int clock;
    int clock_diff;
    struct fb_var_screeninfo v;

    v = *var;
    clock = PICOS2KHZ(var->pixclock);

    if (subsys && subsys->nearest_clock) {
        nearest_clock = subsys->nearest_clock(subsys, clock);
    } else {
        nearest_clock = vml_nearest_clock(clock);
    }

    /*
     * Accept a 20% diff.
     */

    clock_diff = nearest_clock - clock;
    clock_diff = (clock_diff < 0) ? -clock_diff : clock_diff;
    if (clock_diff > clock / 5) {
#if 0
        printk(KERN_DEBUG MODULE_NAME ": Diff failure. %d %d\n",clock_diff,clock);
#endif
        return -EINVAL;
    }

    v.pixclock = KHZ2PICOS(nearest_clock);

    if (var->xres > VML_MAX_XRES || var->yres > VML_MAX_YRES) {
        printk(KERN_DEBUG MODULE_NAME ": Resolution failure.\n");
        return -EINVAL;
    }
    if (var->xres_virtual > VML_MAX_XRES_VIRTUAL) {
        printk(KERN_DEBUG MODULE_NAME
               ": Virtual resolution failure.\n");
        return -EINVAL;
    }
    switch (v.bits_per_pixel) {
    case 0 ... 16:
        v.bits_per_pixel = 16;
        break;
    case 17 ... 32:
        v.bits_per_pixel = 32;
        break;
    default:
        printk(KERN_DEBUG MODULE_NAME ": Invalid bpp: %d.\n",
               var->bits_per_pixel);
        return -EINVAL;
    }

    pitch = ALIGN((var->xres * var->bits_per_pixel) >> 3, 0x40);
    mem = pitch * var->yres_virtual;
    if (mem > vinfo->vram_contig_size) {
        return -ENOMEM;
    }

    switch (v.bits_per_pixel) {
    case 16:
        if (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) {
            vmlfb_set_pref_pixel_format(&v);
        }
        break;
    case 32:
        if (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.length != 0 && var->transp.length != 8) ||
            (var->transp.length == 8 && var->transp.offset != 24)) {
            vmlfb_set_pref_pixel_format(&v);
        }
        break;
    default:
        return -EINVAL;
    }

    *var = v;

    return 0;
}

static int vmlfb_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
{
    struct vml_info *vinfo = container_of(info, struct vml_info, info);
    int ret;

    mutex_lock(&vml_mutex);
    ret = vmlfb_check_var_locked(var, vinfo);
    mutex_unlock(&vml_mutex);

    return ret;
}

static void vml_wait_vblank(struct vml_info *vinfo)
{
    /* Wait for vblank. For now, just wait for a 50Hz cycle (20ms)) */
    mdelay(20);
}

static void vmlfb_disable_pipe(struct vml_info *vinfo)
{
    struct vml_par *par = vinfo->par;

    /* Disable the MDVO pad */
    VML_WRITE32(par, VML_RCOMPSTAT, 0);
    while (!(VML_READ32(par, VML_RCOMPSTAT) & VML_MDVO_VDC_I_RCOMP)) ;

    /* Disable display planes */
    VML_WRITE32(par, VML_DSPCCNTR,
            VML_READ32(par, VML_DSPCCNTR) & ~VML_GFX_ENABLE);
    (void)VML_READ32(par, VML_DSPCCNTR);
    /* Wait for vblank for the disable to take effect */
    vml_wait_vblank(vinfo);

    /* Next, disable display pipes */
    VML_WRITE32(par, VML_PIPEACONF, 0);
    (void)VML_READ32(par, VML_PIPEACONF);

    vinfo->pipe_disabled = 1;
}

#ifdef VERMILION_DEBUG
static void vml_dump_regs(struct vml_info *vinfo)
{
    struct vml_par *par = vinfo->par;

    printk(KERN_DEBUG MODULE_NAME ": Modesetting register dump:\n");
    printk(KERN_DEBUG MODULE_NAME ": \tHTOTAL_A         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_HTOTAL_A));
    printk(KERN_DEBUG MODULE_NAME ": \tHBLANK_A         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_HBLANK_A));
    printk(KERN_DEBUG MODULE_NAME ": \tHSYNC_A          : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_HSYNC_A));
    printk(KERN_DEBUG MODULE_NAME ": \tVTOTAL_A         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_VTOTAL_A));
    printk(KERN_DEBUG MODULE_NAME ": \tVBLANK_A         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_VBLANK_A));
    printk(KERN_DEBUG MODULE_NAME ": \tVSYNC_A          : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_VSYNC_A));
    printk(KERN_DEBUG MODULE_NAME ": \tDSPCSTRIDE       : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_DSPCSTRIDE));
    printk(KERN_DEBUG MODULE_NAME ": \tDSPCSIZE         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_DSPCSIZE));
    printk(KERN_DEBUG MODULE_NAME ": \tDSPCPOS          : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_DSPCPOS));
    printk(KERN_DEBUG MODULE_NAME ": \tDSPARB           : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_DSPARB));
    printk(KERN_DEBUG MODULE_NAME ": \tDSPCADDR         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_DSPCADDR));
    printk(KERN_DEBUG MODULE_NAME ": \tBCLRPAT_A        : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_BCLRPAT_A));
    printk(KERN_DEBUG MODULE_NAME ": \tCANVSCLR_A       : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_CANVSCLR_A));
    printk(KERN_DEBUG MODULE_NAME ": \tPIPEASRC         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_PIPEASRC));
    printk(KERN_DEBUG MODULE_NAME ": \tPIPEACONF        : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_PIPEACONF));
    printk(KERN_DEBUG MODULE_NAME ": \tDSPCCNTR         : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_DSPCCNTR));
    printk(KERN_DEBUG MODULE_NAME ": \tRCOMPSTAT        : 0x%08x\n",
           (unsigned)VML_READ32(par, VML_RCOMPSTAT));
    printk(KERN_DEBUG MODULE_NAME ": End of modesetting register dump.\n");
}
#endif

static int vmlfb_set_par_locked(struct vml_info *vinfo)
{
    struct vml_par *par = vinfo->par;
    struct fb_info *info = &vinfo->info;
    struct fb_var_screeninfo *var = &info->var;
    u32 htotal, hactive, hblank_start, hblank_end, hsync_start, hsync_end;
    u32 vtotal, vactive, vblank_start, vblank_end, vsync_start, vsync_end;
    u32 dspcntr;
    int clock;

    vinfo->bytes_per_pixel = var->bits_per_pixel >> 3;
    vinfo->stride = ALIGN(var->xres_virtual * vinfo->bytes_per_pixel, 0x40);
    info->fix.line_length = vinfo->stride;

    if (!subsys)
        return 0;

    htotal =
        var->xres + var->right_margin + var->hsync_len + var->left_margin;
    hactive = var->xres;
    hblank_start = var->xres;
    hblank_end = htotal;
    hsync_start = hactive + var->right_margin;
    hsync_end = hsync_start + var->hsync_len;

    vtotal =
        var->yres + var->lower_margin + var->vsync_len + var->upper_margin;
    vactive = var->yres;
    vblank_start = var->yres;
    vblank_end = vtotal;
    vsync_start = vactive + var->lower_margin;
    vsync_end = vsync_start + var->vsync_len;

    dspcntr = VML_GFX_ENABLE | VML_GFX_GAMMABYPASS;
    clock = PICOS2KHZ(var->pixclock);

    if (subsys->nearest_clock) {
        clock = subsys->nearest_clock(subsys, clock);
    } else {
        clock = vml_nearest_clock(clock);
    }
    printk(KERN_DEBUG MODULE_NAME
           ": Set mode Hfreq : %d kHz, Vfreq : %d Hz.\n", clock / htotal,
           ((clock / htotal) * 1000) / vtotal);

    switch (var->bits_per_pixel) {
    case 16:
        dspcntr |= VML_GFX_ARGB1555;
        break;
    case 32:
        if (var->transp.length == 8)
            dspcntr |= VML_GFX_ARGB8888 | VML_GFX_ALPHAMULT;
        else
            dspcntr |= VML_GFX_RGB0888;
        break;
    default:
        return -EINVAL;
    }

    vmlfb_disable_pipe(vinfo);
    mb();

    if (subsys->set_clock)
        subsys->set_clock(subsys, clock);
    else
        return -EINVAL;

    VML_WRITE32(par, VML_HTOTAL_A, ((htotal - 1) << 16) | (hactive - 1));
    VML_WRITE32(par, VML_HBLANK_A,
            ((hblank_end - 1) << 16) | (hblank_start - 1));
    VML_WRITE32(par, VML_HSYNC_A,
            ((hsync_end - 1) << 16) | (hsync_start - 1));
    VML_WRITE32(par, VML_VTOTAL_A, ((vtotal - 1) << 16) | (vactive - 1));
    VML_WRITE32(par, VML_VBLANK_A,
            ((vblank_end - 1) << 16) | (vblank_start - 1));
    VML_WRITE32(par, VML_VSYNC_A,
            ((vsync_end - 1) << 16) | (vsync_start - 1));
    VML_WRITE32(par, VML_DSPCSTRIDE, vinfo->stride);
    VML_WRITE32(par, VML_DSPCSIZE,
            ((var->yres - 1) << 16) | (var->xres - 1));
    VML_WRITE32(par, VML_DSPCPOS, 0x00000000);
    VML_WRITE32(par, VML_DSPARB, VML_FIFO_DEFAULT);
    VML_WRITE32(par, VML_BCLRPAT_A, 0x00000000);
    VML_WRITE32(par, VML_CANVSCLR_A, 0x00000000);
    VML_WRITE32(par, VML_PIPEASRC,
            ((var->xres - 1) << 16) | (var->yres - 1));

    wmb();
    VML_WRITE32(par, VML_PIPEACONF, VML_PIPE_ENABLE);
    wmb();
    VML_WRITE32(par, VML_DSPCCNTR, dspcntr);
    wmb();
    VML_WRITE32(par, VML_DSPCADDR, (u32) vinfo->vram_start +
            var->yoffset * vinfo->stride +
            var->xoffset * vinfo->bytes_per_pixel);

    VML_WRITE32(par, VML_RCOMPSTAT, VML_MDVO_PAD_ENABLE);

    while (!(VML_READ32(par, VML_RCOMPSTAT) &
         (VML_MDVO_VDC_I_RCOMP | VML_MDVO_PAD_ENABLE))) ;

    vinfo->pipe_disabled = 0;
#ifdef VERMILION_DEBUG
    vml_dump_regs(vinfo);
#endif

    return 0;
}

static int vmlfb_set_par(struct fb_info *info)
{
    struct vml_info *vinfo = container_of(info, struct vml_info, info);
    int ret;

    mutex_lock(&vml_mutex);
    list_move(&vinfo->head, (subsys) ? &global_has_mode : &global_no_mode);
    ret = vmlfb_set_par_locked(vinfo);

    mutex_unlock(&vml_mutex);
    return ret;
}

static int vmlfb_blank_locked(struct vml_info *vinfo)
{
    struct vml_par *par = vinfo->par;
    u32 cur = VML_READ32(par, VML_PIPEACONF);

    switch (vinfo->cur_blank_mode) {
    case FB_BLANK_UNBLANK:
        if (vinfo->pipe_disabled) {
            vmlfb_set_par_locked(vinfo);
        }
        VML_WRITE32(par, VML_PIPEACONF, cur & ~VML_PIPE_FORCE_BORDER);
        (void)VML_READ32(par, VML_PIPEACONF);
        break;
    case FB_BLANK_NORMAL:
        if (vinfo->pipe_disabled) {
            vmlfb_set_par_locked(vinfo);
        }
        VML_WRITE32(par, VML_PIPEACONF, cur | VML_PIPE_FORCE_BORDER);
        (void)VML_READ32(par, VML_PIPEACONF);
        break;
    case FB_BLANK_VSYNC_SUSPEND:
    case FB_BLANK_HSYNC_SUSPEND:
        if (!vinfo->pipe_disabled) {
            vmlfb_disable_pipe(vinfo);
        }
        break;
    case FB_BLANK_POWERDOWN:
        if (!vinfo->pipe_disabled) {
            vmlfb_disable_pipe(vinfo);
        }
        break;
    default:
        return -EINVAL;
    }

    return 0;
}

static int vmlfb_blank(int blank_mode, struct fb_info *info)
{
    struct vml_info *vinfo = container_of(info, struct vml_info, info);
    int ret;

    mutex_lock(&vml_mutex);
    vinfo->cur_blank_mode = blank_mode;
    ret = vmlfb_blank_locked(vinfo);
    mutex_unlock(&vml_mutex);
    return ret;
}

static int vmlfb_pan_display(struct fb_var_screeninfo *var,
                 struct fb_info *info)
{
    struct vml_info *vinfo = container_of(info, struct vml_info, info);
    struct vml_par *par = vinfo->par;

    mutex_lock(&vml_mutex);
    VML_WRITE32(par, VML_DSPCADDR, (u32) vinfo->vram_start +
            var->yoffset * vinfo->stride +
            var->xoffset * vinfo->bytes_per_pixel);
    (void)VML_READ32(par, VML_DSPCADDR);
    mutex_unlock(&vml_mutex);

    return 0;
}

static int vmlfb_setcolreg(u_int regno, u_int red, u_int green, u_int blue,
               u_int transp, struct fb_info *info)
{
    u32 v;

    if (regno >= 16)
        return -EINVAL;

    if (info->var.grayscale) {
        red = green = blue = (red * 77 + green * 151 + blue * 28) >> 8;
    }

    if (info->fix.visual != FB_VISUAL_TRUECOLOR)
        return -EINVAL;

    red = VML_TOHW(red, info->var.red.length);
    blue = VML_TOHW(blue, info->var.blue.length);
    green = VML_TOHW(green, info->var.green.length);
    transp = VML_TOHW(transp, info->var.transp.length);

    v = (red << info->var.red.offset) |
        (green << info->var.green.offset) |
        (blue << info->var.blue.offset) |
        (transp << info->var.transp.offset);

    switch (info->var.bits_per_pixel) {
    case 16:
        ((u32 *) info->pseudo_palette)[regno] = v;
        break;
    case 24:
    case 32:
        ((u32 *) info->pseudo_palette)[regno] = v;
        break;
    }
    return 0;
}

static int vmlfb_mmap(struct fb_info *info, struct vm_area_struct *vma)
{
    struct vml_info *vinfo = container_of(info, struct vml_info, info);
    unsigned long size = vma->vm_end - vma->vm_start;
    unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
    int ret;

    if (vma->vm_pgoff > (~0UL >> PAGE_SHIFT))
        return -EINVAL;
    if (offset + size > vinfo->vram_contig_size)
        return -EINVAL;
    ret = vmlfb_vram_offset(vinfo, offset);
    if (ret)
        return -EINVAL;
    offset += vinfo->vram_start;
    pgprot_val(vma->vm_page_prot) |= _PAGE_PCD;
    pgprot_val(vma->vm_page_prot) &= ~_PAGE_PWT;
    if (remap_pfn_range(vma, vma->vm_start, offset >> PAGE_SHIFT,
                        size, vma->vm_page_prot))
        return -EAGAIN;
    return 0;
}

static int vmlfb_sync(struct fb_info *info)
{
    return 0;
}

static int vmlfb_cursor(struct fb_info *info, struct fb_cursor *cursor)
{
    return -EINVAL; /* just to force soft_cursor() call */
}

static struct fb_ops vmlfb_ops = {
    .owner = THIS_MODULE,
    .fb_open = vmlfb_open,
    .fb_release = vmlfb_release,
    .fb_check_var = vmlfb_check_var,
    .fb_set_par = vmlfb_set_par,
    .fb_blank = vmlfb_blank,
    .fb_pan_display = vmlfb_pan_display,
    .fb_fillrect = cfb_fillrect,
    .fb_copyarea = cfb_copyarea,
    .fb_imageblit = cfb_imageblit,
    .fb_cursor = vmlfb_cursor,
    .fb_sync = vmlfb_sync,
    .fb_mmap = vmlfb_mmap,
    .fb_setcolreg = vmlfb_setcolreg
};

static struct pci_device_id vml_ids[] = {
    {PCI_DEVICE(PCI_VENDOR_ID_INTEL, VML_DEVICE_VDC)},
    {0}
};

static struct pci_driver vmlfb_pci_driver = {
    .name = "vmlfb",
    .id_table = vml_ids,
    .probe = vml_pci_probe,
    .remove = __devexit_p(vml_pci_remove)
};

static void __exit vmlfb_cleanup(void)
{
    pci_unregister_driver(&vmlfb_pci_driver);
}

static int __init vmlfb_init(void)
{

#ifndef MODULE
    char *option = NULL;

    if (fb_get_options(MODULE_NAME, &option))
        return -ENODEV;
#endif

    printk(KERN_DEBUG MODULE_NAME ": initializing\n");
    mutex_init(&vml_mutex);
    INIT_LIST_HEAD(&global_no_mode);
    INIT_LIST_HEAD(&global_has_mode);

    return pci_register_driver(&vmlfb_pci_driver);
}

int vmlfb_register_subsys(struct vml_sys *sys)
{
    struct vml_info *entry;
    struct list_head *list;
    u32 save_activate;

    mutex_lock(&vml_mutex);
    if (subsys != NULL) {
        subsys->restore(subsys);
    }
    subsys = sys;
    subsys->save(subsys);

    /*
     * We need to restart list traversal for each item, since we
     * release the list mutex in the loop.
     */

    list = global_no_mode.next;
    while (list != &global_no_mode) {
        list_del_init(list);
        entry = list_entry(list, struct vml_info, head);

        /*
         * First, try the current mode which might not be
         * completely validated with respect to the pixel clock.
         */

        if (!vmlfb_check_var_locked(&entry->info.var, entry)) {
            vmlfb_set_par_locked(entry);
            list_add_tail(list, &global_has_mode);
        } else {

            /*
             * Didn't work. Try to find another mode,
             * that matches this subsys.
             */

            mutex_unlock(&vml_mutex);
            save_activate = entry->info.var.activate;
            entry->info.var.bits_per_pixel = 16;
            vmlfb_set_pref_pixel_format(&entry->info.var);
            if (fb_find_mode(&entry->info.var,
                     &entry->info,
                     vml_default_mode, NULL, 0, NULL, 16)) {
                entry->info.var.activate |=
                    FB_ACTIVATE_FORCE | FB_ACTIVATE_NOW;
                fb_set_var(&entry->info, &entry->info.var);
            } else {
                printk(KERN_ERR MODULE_NAME
                       ": Sorry. no mode found for this subsys.\n");
            }
            entry->info.var.activate = save_activate;
            mutex_lock(&vml_mutex);
        }
        vmlfb_blank_locked(entry);
        list = global_no_mode.next;
    }
    mutex_unlock(&vml_mutex);

    printk(KERN_DEBUG MODULE_NAME ": Registered %s subsystem.\n",
                subsys->name ? subsys->name : "unknown");
    return 0;
}

EXPORT_SYMBOL_GPL(vmlfb_register_subsys);

void vmlfb_unregister_subsys(struct vml_sys *sys)
{
    struct vml_info *entry, *next;

    mutex_lock(&vml_mutex);
    if (subsys != sys) {
        mutex_unlock(&vml_mutex);
        return;
    }
    subsys->restore(subsys);
    subsys = NULL;
    list_for_each_entry_safe(entry, next, &global_has_mode, head) {
        printk(KERN_DEBUG MODULE_NAME ": subsys disable pipe\n");
        vmlfb_disable_pipe(entry);
        list_move_tail(&entry->head, &global_no_mode);
    }
    mutex_unlock(&vml_mutex);
}

EXPORT_SYMBOL_GPL(vmlfb_unregister_subsys);

module_init(vmlfb_init);
module_exit(vmlfb_cleanup);

MODULE_AUTHOR("Tungsten Graphics");
MODULE_DESCRIPTION("Initialization of the Vermilion display devices");
MODULE_VERSION("1.0.0");
MODULE_LICENSE("GPL");