| Message ID | 20170627161936.2301-1-ard.biesheuvel@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 27 June 2017 at 16:19, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote: > 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 the 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 > '#ifdef !CONFIG_X86 that surrounds it' helpfully removed by git-commit due to the leading # > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> > --- > drivers/video/fbdev/efifb.c | 24 ++++++++++++-------- > 1 file changed, 15 insertions(+), 9 deletions(-) > > diff --git a/drivers/video/fbdev/efifb.c b/drivers/video/fbdev/efifb.c > index b827a8113e26..6220de3e25d3 100644 > --- a/drivers/video/fbdev/efifb.c > +++ b/drivers/video/fbdev/efifb.c > @@ -146,6 +146,9 @@ ATTRIBUTE_GROUPS(efifb); > > static bool pci_dev_disabled; /* FB base matches BAR of a disabled device */ > > +static struct resource *bar_resource; > +static u64 bar_offset; > + > static int efifb_probe(struct platform_device *dev) > { > struct fb_info *info; > @@ -200,6 +203,13 @@ static int efifb_probe(struct platform_device *dev) > efifb_fix.smem_start |= ext_lfb_base; > } > > + if (bar_resource && > + bar_resource->start + bar_offset != efifb_fix.smem_start) { > + > + pr_warn("efifb: PCI BAR has moved, updating fb 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; > @@ -364,11 +374,11 @@ static struct platform_driver efifb_driver = { > > builtin_platform_driver(efifb_driver); > > -#if defined(CONFIG_PCI) && !defined(CONFIG_X86) > +#if defined(CONFIG_PCI) > > static bool pci_bar_found; /* did we find a BAR matching the efifb base? */ > > -static void claim_efifb_bar(struct pci_dev *dev, int idx) > +static void record_efifb_bar_resource(struct pci_dev *dev, int idx, u64 offset) > { > u16 word; > > @@ -383,12 +393,8 @@ static void claim_efifb_bar(struct pci_dev *dev, int idx) > return; > } > > - if (pci_claim_resource(dev, idx)) { > - pci_dev_disabled = true; > - dev_err(&dev->dev, > - "BAR %d: failed to claim resource for efifb!\n", idx); > - return; > - } > + bar_resource = &dev->resource[idx]; > + bar_offset = offset; > > dev_info(&dev->dev, "BAR %d: assigned to efifb\n", idx); > } > @@ -415,7 +421,7 @@ static void efifb_fixup_resources(struct pci_dev *dev) > continue; > > if (res->start <= base && res->end >= base + size - 1) { > - claim_efifb_bar(dev, i); > + record_efifb_bar_resource(dev, i, base - res->start); > break; > } > } > -- > 2.9.3 > -- To unsubscribe from this list: send the line "unsubscribe linux-fbdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Jun 27, 2017 at 04:19:36PM +0000, Ard Biesheuvel wrote: > 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 the 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 > #ifdef !CONFIG_X86 that surrounds it. > > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> > --- > drivers/video/fbdev/efifb.c | 24 ++++++++++++-------- > 1 file changed, 15 insertions(+), 9 deletions(-) > > diff --git a/drivers/video/fbdev/efifb.c b/drivers/video/fbdev/efifb.c > index b827a8113e26..6220de3e25d3 100644 > --- a/drivers/video/fbdev/efifb.c > +++ b/drivers/video/fbdev/efifb.c > @@ -146,6 +146,9 @@ ATTRIBUTE_GROUPS(efifb); > > static bool pci_dev_disabled; /* FB base matches BAR of a disabled device */ > > +static struct resource *bar_resource; > +static u64 bar_offset; > + > static int efifb_probe(struct platform_device *dev) > { > struct fb_info *info; > @@ -200,6 +203,13 @@ static int efifb_probe(struct platform_device *dev) > efifb_fix.smem_start |= ext_lfb_base; > } > > + if (bar_resource && > + bar_resource->start + bar_offset != efifb_fix.smem_start) { > + > + pr_warn("efifb: PCI BAR has moved, updating fb address\n"); > + efifb_fix.smem_start = bar_resource->start + bar_offset; Does this really need a pr_warn()? Seems like it should be dev_info(), the point of the patch is to make moving it not be a problem.
On 27 June 2017 at 20:02, Peter Jones <pjones@redhat.com> wrote: > On Tue, Jun 27, 2017 at 04:19:36PM +0000, Ard Biesheuvel wrote: >> 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 the 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 >> #ifdef !CONFIG_X86 that surrounds it. >> >> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> >> --- >> drivers/video/fbdev/efifb.c | 24 ++++++++++++-------- >> 1 file changed, 15 insertions(+), 9 deletions(-) >> >> diff --git a/drivers/video/fbdev/efifb.c b/drivers/video/fbdev/efifb.c >> index b827a8113e26..6220de3e25d3 100644 >> --- a/drivers/video/fbdev/efifb.c >> +++ b/drivers/video/fbdev/efifb.c >> @@ -146,6 +146,9 @@ ATTRIBUTE_GROUPS(efifb); >> >> static bool pci_dev_disabled; /* FB base matches BAR of a disabled device */ >> >> +static struct resource *bar_resource; >> +static u64 bar_offset; >> + >> static int efifb_probe(struct platform_device *dev) >> { >> struct fb_info *info; >> @@ -200,6 +203,13 @@ static int efifb_probe(struct platform_device *dev) >> efifb_fix.smem_start |= ext_lfb_base; >> } >> >> + if (bar_resource && >> + bar_resource->start + bar_offset != efifb_fix.smem_start) { >> + >> + pr_warn("efifb: PCI BAR has moved, updating fb address\n"); >> + efifb_fix.smem_start = bar_resource->start + bar_offset; > > Does this really need a pr_warn()? Seems like it should be dev_info(), > the point of the patch is to make moving it not be a problem. > Yes, good point. -- To unsubscribe from this list: send the line "unsubscribe linux-fbdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/video/fbdev/efifb.c b/drivers/video/fbdev/efifb.c index b827a8113e26..6220de3e25d3 100644 --- a/drivers/video/fbdev/efifb.c +++ b/drivers/video/fbdev/efifb.c @@ -146,6 +146,9 @@ ATTRIBUTE_GROUPS(efifb); static bool pci_dev_disabled; /* FB base matches BAR of a disabled device */ +static struct resource *bar_resource; +static u64 bar_offset; + static int efifb_probe(struct platform_device *dev) { struct fb_info *info; @@ -200,6 +203,13 @@ static int efifb_probe(struct platform_device *dev) efifb_fix.smem_start |= ext_lfb_base; } + if (bar_resource && + bar_resource->start + bar_offset != efifb_fix.smem_start) { + + pr_warn("efifb: PCI BAR has moved, updating fb 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; @@ -364,11 +374,11 @@ static struct platform_driver efifb_driver = { builtin_platform_driver(efifb_driver); -#if defined(CONFIG_PCI) && !defined(CONFIG_X86) +#if defined(CONFIG_PCI) static bool pci_bar_found; /* did we find a BAR matching the efifb base? */ -static void claim_efifb_bar(struct pci_dev *dev, int idx) +static void record_efifb_bar_resource(struct pci_dev *dev, int idx, u64 offset) { u16 word; @@ -383,12 +393,8 @@ static void claim_efifb_bar(struct pci_dev *dev, int idx) return; } - if (pci_claim_resource(dev, idx)) { - pci_dev_disabled = true; - dev_err(&dev->dev, - "BAR %d: failed to claim resource for efifb!\n", idx); - return; - } + bar_resource = &dev->resource[idx]; + bar_offset = offset; dev_info(&dev->dev, "BAR %d: assigned to efifb\n", idx); } @@ -415,7 +421,7 @@ static void efifb_fixup_resources(struct pci_dev *dev) continue; if (res->start <= base && res->end >= base + size - 1) { - claim_efifb_bar(dev, i); + record_efifb_bar_resource(dev, i, base - res->start); break; } }
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 the 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 Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> --- drivers/video/fbdev/efifb.c | 24 ++++++++++++-------- 1 file changed, 15 insertions(+), 9 deletions(-)