Hi, Bjorn, Why this series still missing in 5.17-rc1? :( Huacai On Fri, Jan 7, 2022 at 12:21 AM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote: > > On Thu, Jan 06, 2022 at 02:44:42PM +0800, Huacai Chen wrote: > > On Thu, Jan 6, 2022 at 8:07 AM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote: > > > Previously we selected a device that owns the boot framebuffer as the > > > default device in vga_arb_select_default_device(). This was only done in > > > the vga_arb_device_init() subsys_initcall, so devices enumerated later, > > > e.g., by pcibios_init(), were not eligible. > > > > > > Fix this by moving the framebuffer device selection from > > > vga_arb_select_default_device() to vga_arbiter_add_pci_device(), which is > > > called after every PCI device is enumerated, either by the > > > vga_arb_device_init() subsys_initcall or as an ADD_DEVICE notifier. > > > > > > Note that if vga_arb_select_default_device() found a device owning the boot > > > framebuffer, it unconditionally set it to be the default VGA device, and no > > > subsequent device could replace it. > > > > > > Link: https://lore.kernel.org/r/20211015061512.2941859-7-chenhuacai@xxxxxxxxxxx > > > Based-on-patch-by: Huacai Chen <chenhuacai@xxxxxxxxxxx> > > > Signed-off-by: Bjorn Helgaas <bhelgaas@xxxxxxxxxx> > > > Cc: Bruno Prémont <bonbons@xxxxxxxxxxxxxxxxx> > > > --- > > > drivers/gpu/vga/vgaarb.c | 37 +++++++++++++++++-------------------- > > > 1 file changed, 17 insertions(+), 20 deletions(-) > > > > > > diff --git a/drivers/gpu/vga/vgaarb.c b/drivers/gpu/vga/vgaarb.c > > > index b0ae0f177c6f..aefa4f406f7d 100644 > > > --- a/drivers/gpu/vga/vgaarb.c > > > +++ b/drivers/gpu/vga/vgaarb.c > > > @@ -72,6 +72,7 @@ struct vga_device { > > > unsigned int io_norm_cnt; /* normal IO count */ > > > unsigned int mem_norm_cnt; /* normal MEM count */ > > > bool bridge_has_one_vga; > > > + bool is_framebuffer; /* BAR covers firmware framebuffer */ > > > unsigned int (*set_decode)(struct pci_dev *pdev, bool decode); > > > }; > > > > > > @@ -565,10 +566,9 @@ void vga_put(struct pci_dev *pdev, unsigned int rsrc) > > > } > > > EXPORT_SYMBOL(vga_put); > > > > > > -static void __init vga_select_framebuffer_device(struct pci_dev *pdev) > > > +static bool vga_is_framebuffer_device(struct pci_dev *pdev) > > > { > > > #if defined(CONFIG_X86) || defined(CONFIG_IA64) > > > - struct device *dev = &pdev->dev; > > > u64 base = screen_info.lfb_base; > > > u64 size = screen_info.lfb_size; > > > u64 limit; > > > @@ -583,15 +583,6 @@ static void __init vga_select_framebuffer_device(struct pci_dev *pdev) > > > > > > limit = base + size; > > > > > > - /* > > > - * Override vga_arbiter_add_pci_device()'s I/O based detection > > > - * as it may take the wrong device (e.g. on Apple system under > > > - * EFI). > > > - * > > > - * Select the device owning the boot framebuffer if there is > > > - * one. > > > - */ > > > - > > > /* Does firmware framebuffer belong to us? */ > > > for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { > > > flags = pci_resource_flags(pdev, i); > > > @@ -608,13 +599,10 @@ static void __init vga_select_framebuffer_device(struct pci_dev *pdev) > > > if (base < start || limit >= end) > > > continue; > > > > > > - if (!vga_default_device()) > > > - vgaarb_info(dev, "setting as boot device\n"); > > > - else if (pdev != vga_default_device()) > > > - vgaarb_info(dev, "overriding boot device\n"); > > > - vga_set_default_device(pdev); > > > + return true; > > > } > > > #endif > > > + return false; > > > } > > > > > > static bool vga_arb_integrated_gpu(struct device *dev) > > > @@ -635,6 +623,7 @@ static bool vga_arb_integrated_gpu(struct device *dev) > > > static bool vga_is_boot_device(struct vga_device *vgadev) > > > { > > > struct vga_device *boot_vga = vgadev_find(vga_default_device()); > > > + struct pci_dev *pdev = vgadev->pdev; > > > > > > /* > > > * We select the default VGA device in this order: > > > @@ -645,6 +634,18 @@ static bool vga_is_boot_device(struct vga_device *vgadev) > > > * Other device (see vga_arb_select_default_device()) > > > */ > > > > > > + /* > > > + * We always prefer a firmware framebuffer, so if we've already > > > + * found one, there's no need to consider vgadev. > > > + */ > > > + if (boot_vga && boot_vga->is_framebuffer) > > > + return false; > > > + > > > + if (vga_is_framebuffer_device(pdev)) { > > > + vgadev->is_framebuffer = true; > > > + return true; > > > + } > > Maybe it is better to rename vga_is_framebuffer_device() to > > vga_is_firmware_device() and rename is_framebuffer to > > is_fw_framebuffer? > > That's a great point, thanks! > > The "framebuffer" term is way too generic. *All* VGA devices have a > framebuffer, so it adds no information. This is really about finding > the device that was used by firmware. > > I renamed: > > vga_is_framebuffer_device() -> vga_is_firmware_default() > vga_device.is_framebuffer -> vga_device.is_firmware_default > > I updated my local branch and pushed it to: > https://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci.git/log/?h=pci/vga > with head 0f4caffa1297 ("vgaarb: Replace full MIT license text with > SPDX identifier"). > > I don't maintain drivers/gpu/vga/vgaarb.c, so this branch is just for > reference. It'll ultimately be up to the DRM folks to handle this. > > I'll wait for any other comments or testing reports before reposting. > > > > /* > > > * A legacy VGA device has MEM and IO enabled and any bridges > > > * leading to it have PCI_BRIDGE_CTL_VGA enabled so the legacy > > > @@ -1531,10 +1532,6 @@ static void __init vga_arb_select_default_device(void) > > > struct pci_dev *pdev, *found = NULL; > > > struct vga_device *vgadev; > > > > > > - list_for_each_entry(vgadev, &vga_list, list) { > > > - vga_select_framebuffer_device(vgadev->pdev); > > > - } > > > - > > > if (!vga_default_device()) { > > > list_for_each_entry_reverse(vgadev, &vga_list, list) { > > > struct device *dev = &vgadev->pdev->dev; > > > -- > > > 2.25.1 > > >
