linux/drivers/video/fbdev/efifb.c

438 lines
12 KiB
C
Raw Normal View History

/*
* Framebuffer driver for EFI/UEFI based system
*
* (c) 2006 Edgar Hucek <gimli@dark-green.com>
* Original efi driver written by Gerd Knorr <kraxel@goldbach.in-berlin.de>
*
*/
#include <linux/kernel.h>
#include <linux/efi.h>
#include <linux/errno.h>
#include <linux/fb.h>
#include <linux/pci.h>
#include <linux/platform_device.h>
#include <linux/screen_info.h>
#include <video/vga.h>
#include <asm/efi.h>
static bool request_mem_succeeded = false;
efifb: allow user to disable write combined mapping. This patch allows the user to disable write combined mapping of the efifb framebuffer console using an nowc option. A customer noticed major slowdowns while logging to the console with write combining enabled, on other tasks running on the same CPU. (10x or greater slow down on all other cores on the same CPU as is doing the logging). I reproduced this on a machine with dual CPUs. Intel(R) Xeon(R) CPU E5-2609 v3 @ 1.90GHz (6 core) I wrote a test that just mmaps the pci bar and writes to it in a loop, while this was running in the background one a single core with (taskset -c 1), building a kernel up to init/version.o (taskset -c 8) went from 13s to 133s or so. I've yet to explain why this occurs or what is going wrong I haven't managed to find a perf command that in any way gives insight into this. 11,885,070,715 instructions # 1.39 insns per cycle vs 12,082,592,342 instructions # 0.13 insns per cycle is the only thing I've spotted of interest, I've tried at least: dTLB-stores,dTLB-store-misses,L1-dcache-stores,LLC-store,LLC-store-misses,LLC-load-misses,LLC-loads,\mem-loads,mem-stores,iTLB-loads,iTLB-load-misses,cache-references,cache-misses For now it seems at least a good idea to allow a user to disable write combining if they see this until we can figure it out. Note also most users get a real framebuffer driver loaded when kms kicks in, it just happens on these machines the kernel didn't support the gpu specific driver. Signed-off-by: Dave Airlie <airlied@redhat.com> Acked-by: Peter Jones <pjones@redhat.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
2017-08-01 00:45:41 +08:00
static bool nowc = false;
static struct fb_var_screeninfo efifb_defined = {
.activate = FB_ACTIVATE_NOW,
.height = -1,
.width = -1,
.right_margin = 32,
.upper_margin = 16,
.lower_margin = 4,
.vsync_len = 4,
.vmode = FB_VMODE_NONINTERLACED,
};
static struct fb_fix_screeninfo efifb_fix = {
.id = "EFI VGA",
.type = FB_TYPE_PACKED_PIXELS,
.accel = FB_ACCEL_NONE,
.visual = FB_VISUAL_TRUECOLOR,
};
static int efifb_setcolreg(unsigned regno, unsigned red, unsigned green,
unsigned blue, unsigned transp,
struct fb_info *info)
{
/*
* Set a single color register. The values supplied are
* already rounded down to the hardware's capabilities
* (according to the entries in the `var' structure). Return
* != 0 for invalid regno.
*/
if (regno >= info->cmap.len)
return 1;
if (regno < 16) {
red >>= 16 - info->var.red.length;
green >>= 16 - info->var.green.length;
blue >>= 16 - info->var.blue.length;
((u32 *)(info->pseudo_palette))[regno] =
(red << info->var.red.offset) |
(green << info->var.green.offset) |
(blue << info->var.blue.offset);
}
return 0;
}
static void efifb_destroy(struct fb_info *info)
{
if (info->screen_base)
iounmap(info->screen_base);
if (request_mem_succeeded)
release_mem_region(info->apertures->ranges[0].base,
info->apertures->ranges[0].size);
fb_dealloc_cmap(&info->cmap);
}
static struct fb_ops efifb_ops = {
.owner = THIS_MODULE,
.fb_destroy = efifb_destroy,
.fb_setcolreg = efifb_setcolreg,
.fb_fillrect = cfb_fillrect,
.fb_copyarea = cfb_copyarea,
.fb_imageblit = cfb_imageblit,
};
static int efifb_setup(char *options)
{
char *this_opt;
if (options && *options) {
while ((this_opt = strsep(&options, ",")) != NULL) {
if (!*this_opt) continue;
efifb_setup_from_dmi(&screen_info, this_opt);
if (!strncmp(this_opt, "base:", 5))
screen_info.lfb_base = simple_strtoul(this_opt+5, NULL, 0);
else if (!strncmp(this_opt, "stride:", 7))
screen_info.lfb_linelength = simple_strtoul(this_opt+7, NULL, 0) * 4;
else if (!strncmp(this_opt, "height:", 7))
screen_info.lfb_height = simple_strtoul(this_opt+7, NULL, 0);
else if (!strncmp(this_opt, "width:", 6))
screen_info.lfb_width = simple_strtoul(this_opt+6, NULL, 0);
efifb: allow user to disable write combined mapping. This patch allows the user to disable write combined mapping of the efifb framebuffer console using an nowc option. A customer noticed major slowdowns while logging to the console with write combining enabled, on other tasks running on the same CPU. (10x or greater slow down on all other cores on the same CPU as is doing the logging). I reproduced this on a machine with dual CPUs. Intel(R) Xeon(R) CPU E5-2609 v3 @ 1.90GHz (6 core) I wrote a test that just mmaps the pci bar and writes to it in a loop, while this was running in the background one a single core with (taskset -c 1), building a kernel up to init/version.o (taskset -c 8) went from 13s to 133s or so. I've yet to explain why this occurs or what is going wrong I haven't managed to find a perf command that in any way gives insight into this. 11,885,070,715 instructions # 1.39 insns per cycle vs 12,082,592,342 instructions # 0.13 insns per cycle is the only thing I've spotted of interest, I've tried at least: dTLB-stores,dTLB-store-misses,L1-dcache-stores,LLC-store,LLC-store-misses,LLC-load-misses,LLC-loads,\mem-loads,mem-stores,iTLB-loads,iTLB-load-misses,cache-references,cache-misses For now it seems at least a good idea to allow a user to disable write combining if they see this until we can figure it out. Note also most users get a real framebuffer driver loaded when kms kicks in, it just happens on these machines the kernel didn't support the gpu specific driver. Signed-off-by: Dave Airlie <airlied@redhat.com> Acked-by: Peter Jones <pjones@redhat.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
2017-08-01 00:45:41 +08:00
else if (!strcmp(this_opt, "nowc"))
nowc = true;
}
}
return 0;
}
static inline bool fb_base_is_valid(void)
{
if (screen_info.lfb_base)
return true;
if (!(screen_info.capabilities & VIDEO_CAPABILITY_64BIT_BASE))
return false;
if (screen_info.ext_lfb_base)
return true;
return false;
}
#define efifb_attr_decl(name, fmt) \
static ssize_t name##_show(struct device *dev, \
struct device_attribute *attr, \
char *buf) \
{ \
return sprintf(buf, fmt "\n", (screen_info.lfb_##name)); \
} \
static DEVICE_ATTR_RO(name)
efifb_attr_decl(base, "0x%x");
efifb_attr_decl(linelength, "%u");
efifb_attr_decl(height, "%u");
efifb_attr_decl(width, "%u");
efifb_attr_decl(depth, "%u");
static struct attribute *efifb_attrs[] = {
&dev_attr_base.attr,
&dev_attr_linelength.attr,
&dev_attr_width.attr,
&dev_attr_height.attr,
&dev_attr_depth.attr,
NULL
};
ATTRIBUTE_GROUPS(efifb);
static bool pci_dev_disabled; /* FB base matches BAR of a disabled device */
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
static struct pci_dev *efifb_pci_dev; /* dev with BAR covering the efifb */
static struct resource *bar_resource;
static u64 bar_offset;
static int efifb_probe(struct platform_device *dev)
{
struct fb_info *info;
int err;
unsigned int size_vmode;
unsigned int size_remap;
unsigned int size_total;
char *option = NULL;
if (screen_info.orig_video_isVGA != VIDEO_TYPE_EFI || pci_dev_disabled)
return -ENODEV;
if (fb_get_options("efifb", &option))
return -ENODEV;
efifb_setup(option);
/* We don't get linelength from UGA Draw Protocol, only from
* EFI Graphics Protocol. So if it's not in DMI, and it's not
* passed in from the user, we really can't use the framebuffer.
*/
if (!screen_info.lfb_linelength)
return -ENODEV;
if (!screen_info.lfb_depth)
screen_info.lfb_depth = 32;
if (!screen_info.pages)
screen_info.pages = 1;
if (!fb_base_is_valid()) {
printk(KERN_DEBUG "efifb: invalid framebuffer address\n");
return -ENODEV;
}
printk(KERN_INFO "efifb: probing for efifb\n");
/* just assume they're all unset if any are */
if (!screen_info.blue_size) {
screen_info.blue_size = 8;
screen_info.blue_pos = 0;
screen_info.green_size = 8;
screen_info.green_pos = 8;
screen_info.red_size = 8;
screen_info.red_pos = 16;
screen_info.rsvd_size = 8;
screen_info.rsvd_pos = 24;
}
efifb_fix.smem_start = screen_info.lfb_base;
if (screen_info.capabilities & VIDEO_CAPABILITY_64BIT_BASE) {
u64 ext_lfb_base;
ext_lfb_base = (u64)(unsigned long)screen_info.ext_lfb_base << 32;
efifb_fix.smem_start |= ext_lfb_base;
}
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
if (bar_resource &&
bar_resource->start + bar_offset != efifb_fix.smem_start) {
dev_info(&efifb_pci_dev->dev,
"BAR has moved, updating efifb address\n");
efifb_fix.smem_start = bar_resource->start + bar_offset;
}
efifb_defined.bits_per_pixel = screen_info.lfb_depth;
efifb_defined.xres = screen_info.lfb_width;
efifb_defined.yres = screen_info.lfb_height;
efifb_fix.line_length = screen_info.lfb_linelength;
/* size_vmode -- that is the amount of memory needed for the
* used video mode, i.e. the minimum amount of
* memory we need. */
size_vmode = efifb_defined.yres * efifb_fix.line_length;
/* size_total -- all video memory we have. Used for
* entries, ressource allocation and bounds
* checking. */
size_total = screen_info.lfb_size;
if (size_total < size_vmode)
size_total = size_vmode;
/* size_remap -- the amount of video memory we are going to
* use for efifb. With modern cards it is no
* option to simply use size_total as that
* wastes plenty of kernel address space. */
size_remap = size_vmode * 2;
if (size_remap > size_total)
size_remap = size_total;
if (size_remap % PAGE_SIZE)
size_remap += PAGE_SIZE - (size_remap % PAGE_SIZE);
efifb_fix.smem_len = size_remap;
if (request_mem_region(efifb_fix.smem_start, size_remap, "efifb")) {
request_mem_succeeded = true;
} else {
/* We cannot make this fatal. Sometimes this comes from magic
spaces our resource handlers simply don't know about */
pr_warn("efifb: cannot reserve video memory at 0x%lx\n",
efifb_fix.smem_start);
}
info = framebuffer_alloc(sizeof(u32) * 16, &dev->dev);
if (!info) {
pr_err("efifb: cannot allocate framebuffer\n");
err = -ENOMEM;
goto err_release_mem;
}
platform_set_drvdata(dev, info);
info->pseudo_palette = info->par;
info->par = NULL;
info->apertures = alloc_apertures(1);
if (!info->apertures) {
err = -ENOMEM;
goto err_release_fb;
}
info->apertures->ranges[0].base = efifb_fix.smem_start;
info->apertures->ranges[0].size = size_remap;
efifb: allow user to disable write combined mapping. This patch allows the user to disable write combined mapping of the efifb framebuffer console using an nowc option. A customer noticed major slowdowns while logging to the console with write combining enabled, on other tasks running on the same CPU. (10x or greater slow down on all other cores on the same CPU as is doing the logging). I reproduced this on a machine with dual CPUs. Intel(R) Xeon(R) CPU E5-2609 v3 @ 1.90GHz (6 core) I wrote a test that just mmaps the pci bar and writes to it in a loop, while this was running in the background one a single core with (taskset -c 1), building a kernel up to init/version.o (taskset -c 8) went from 13s to 133s or so. I've yet to explain why this occurs or what is going wrong I haven't managed to find a perf command that in any way gives insight into this. 11,885,070,715 instructions # 1.39 insns per cycle vs 12,082,592,342 instructions # 0.13 insns per cycle is the only thing I've spotted of interest, I've tried at least: dTLB-stores,dTLB-store-misses,L1-dcache-stores,LLC-store,LLC-store-misses,LLC-load-misses,LLC-loads,\mem-loads,mem-stores,iTLB-loads,iTLB-load-misses,cache-references,cache-misses For now it seems at least a good idea to allow a user to disable write combining if they see this until we can figure it out. Note also most users get a real framebuffer driver loaded when kms kicks in, it just happens on these machines the kernel didn't support the gpu specific driver. Signed-off-by: Dave Airlie <airlied@redhat.com> Acked-by: Peter Jones <pjones@redhat.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
2017-08-01 00:45:41 +08:00
if (nowc)
info->screen_base = ioremap(efifb_fix.smem_start, efifb_fix.smem_len);
else
info->screen_base = ioremap_wc(efifb_fix.smem_start, efifb_fix.smem_len);
if (!info->screen_base) {
pr_err("efifb: abort, cannot ioremap video memory 0x%x @ 0x%lx\n",
efifb_fix.smem_len, efifb_fix.smem_start);
err = -EIO;
goto err_release_fb;
}
pr_info("efifb: framebuffer at 0x%lx, using %dk, total %dk\n",
efifb_fix.smem_start, size_remap/1024, size_total/1024);
pr_info("efifb: mode is %dx%dx%d, linelength=%d, pages=%d\n",
efifb_defined.xres, efifb_defined.yres,
efifb_defined.bits_per_pixel, efifb_fix.line_length,
screen_info.pages);
efifb_defined.xres_virtual = efifb_defined.xres;
efifb_defined.yres_virtual = efifb_fix.smem_len /
efifb_fix.line_length;
pr_info("efifb: scrolling: redraw\n");
efifb_defined.yres_virtual = efifb_defined.yres;
/* some dummy values for timing to make fbset happy */
efifb_defined.pixclock = 10000000 / efifb_defined.xres *
1000 / efifb_defined.yres;
efifb_defined.left_margin = (efifb_defined.xres / 8) & 0xf8;
efifb_defined.hsync_len = (efifb_defined.xres / 8) & 0xf8;
efifb_defined.red.offset = screen_info.red_pos;
efifb_defined.red.length = screen_info.red_size;
efifb_defined.green.offset = screen_info.green_pos;
efifb_defined.green.length = screen_info.green_size;
efifb_defined.blue.offset = screen_info.blue_pos;
efifb_defined.blue.length = screen_info.blue_size;
efifb_defined.transp.offset = screen_info.rsvd_pos;
efifb_defined.transp.length = screen_info.rsvd_size;
pr_info("efifb: %s: "
"size=%d:%d:%d:%d, shift=%d:%d:%d:%d\n",
"Truecolor",
screen_info.rsvd_size,
screen_info.red_size,
screen_info.green_size,
screen_info.blue_size,
screen_info.rsvd_pos,
screen_info.red_pos,
screen_info.green_pos,
screen_info.blue_pos);
efifb_fix.ypanstep = 0;
efifb_fix.ywrapstep = 0;
info->fbops = &efifb_ops;
info->var = efifb_defined;
info->fix = efifb_fix;
info->flags = FBINFO_FLAG_DEFAULT | FBINFO_MISC_FIRMWARE;
err = sysfs_create_groups(&dev->dev.kobj, efifb_groups);
if (err) {
pr_err("efifb: cannot add sysfs attrs\n");
goto err_unmap;
}
err = fb_alloc_cmap(&info->cmap, 256, 0);
if (err < 0) {
pr_err("efifb: cannot allocate colormap\n");
goto err_groups;
}
err = register_framebuffer(info);
if (err < 0) {
pr_err("efifb: cannot register framebuffer\n");
goto err_fb_dealoc;
}
fb_info(info, "%s frame buffer device\n", info->fix.id);
return 0;
err_fb_dealoc:
fb_dealloc_cmap(&info->cmap);
err_groups:
sysfs_remove_groups(&dev->dev.kobj, efifb_groups);
err_unmap:
iounmap(info->screen_base);
err_release_fb:
framebuffer_release(info);
err_release_mem:
if (request_mem_succeeded)
release_mem_region(efifb_fix.smem_start, size_total);
return err;
}
static int efifb_remove(struct platform_device *pdev)
{
struct fb_info *info = platform_get_drvdata(pdev);
unregister_framebuffer(info);
sysfs_remove_groups(&pdev->dev.kobj, efifb_groups);
framebuffer_release(info);
return 0;
}
static struct platform_driver efifb_driver = {
.driver = {
.name = "efi-framebuffer",
},
.probe = efifb_probe,
.remove = efifb_remove,
};
builtin_platform_driver(efifb_driver);
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
#if defined(CONFIG_PCI)
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
static void record_efifb_bar_resource(struct pci_dev *dev, int idx, u64 offset)
{
u16 word;
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
efifb_pci_dev = dev;
pci_read_config_word(dev, PCI_COMMAND, &word);
if (!(word & PCI_COMMAND_MEMORY)) {
pci_dev_disabled = true;
dev_err(&dev->dev,
"BAR %d: assigned to efifb but device is disabled!\n",
idx);
return;
}
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
bar_resource = &dev->resource[idx];
bar_offset = offset;
dev_info(&dev->dev, "BAR %d: assigned to efifb\n", idx);
}
static void efifb_fixup_resources(struct pci_dev *dev)
{
u64 base = screen_info.lfb_base;
u64 size = screen_info.lfb_size;
int i;
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
if (efifb_pci_dev || screen_info.orig_video_isVGA != VIDEO_TYPE_EFI)
return;
if (screen_info.capabilities & VIDEO_CAPABILITY_64BIT_BASE)
base |= (u64)screen_info.ext_lfb_base << 32;
if (!base)
return;
for (i = 0; i <= PCI_STD_RESOURCE_END; i++) {
struct resource *res = &dev->resource[i];
if (!(res->flags & IORESOURCE_MEM))
continue;
if (res->start <= base && res->end >= base + size - 1) {
drivers/fbdev/efifb: Allow BAR to be moved instead of claiming it On UEFI systems, the firmware may expose a Graphics Output Protocol (GOP) instance to which the efifb driver attempts to attach in order to provide a minimal, unaccelerated framebuffer. The GOP protocol itself is not very sophisticated, and only describes the offset and size of the framebuffer in memory, and the pixel format. If the GOP framebuffer is provided by a PCI device, it will have been configured and enabled by the UEFI firmware, and the GOP protocol will simply point into a live BAR region. However, the GOP protocol itself does not describe this relation, and so we have to take care not to reconfigure the BAR without taking efifb's dependency on it into account. Commit: 55d728a40d36 ("efi/fb: Avoid reconfiguration of BAR that covers the framebuffer") attempted to do so by claiming the BAR resource early on, which prevents the PCI resource allocation routines from changing it. However, it turns out that this only works if the PCI device is not behind any bridges, since the bridge resources need to be claimed first. So instead, allow the BAR to be moved, but make the efifb driver deal with that gracefully. So record the resource that covers the BAR early on, and if it turns out to have moved by the time we probe the efifb driver, update the framebuffer address accordingly. While this is less likely to occur on x86, given that the firmware's PCI resource allocation is more likely to be preserved, this is a worthwhile sanity check to have in place, and so let's remove the preprocessor conditional that makes it !X86 only. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> Reviewed-by: Peter Jones <pjones@redhat.com> Acked-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com> Acked-by: Bjorn Helgaas <bhelgaas@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Matt Fleming <matt@codeblueprint.co.uk> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: linux-efi@vger.kernel.org Link: http://lkml.kernel.org/r/20170818194947.19347-8-ard.biesheuvel@linaro.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-08-19 03:49:40 +08:00
record_efifb_bar_resource(dev, i, base - res->start);
break;
}
}
}
DECLARE_PCI_FIXUP_CLASS_HEADER(PCI_ANY_ID, PCI_ANY_ID, PCI_BASE_CLASS_DISPLAY,
16, efifb_fixup_resources);
#endif