Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/22 01:38, Bjorn Helgaas wrote: On Fri, Aug 18, 2023 at 12:09:29PM +0800, suijingfeng wrote: On 2023/8/18 06:08, Bjorn Helgaas wrote: On Wed, Aug 16, 2023 at 06:05:27AM +0800, Sui Jingfeng wrote: Currently, the vga_is_firmware_default() function only works on x86 and ia64, it is a no-op on ARM, ARM64, PPC, RISC-V, etc. This patch completes the implementation for the rest of the architectures. The added code tries to identify the PCI(e) VGA device that owns the firmware framebuffer before PCI resource reallocation happens. As far as I can tell, this is basically identical to the existing vga_is_firmware_default(), except that this patch funs that code as a header fixup, so it happens before any PCI BAR reallocations happen. Yes, what you said is right in overall. But I think I should mention a few tiny points that make a difference. 1) My version is *less arch-dependent* Of course. If we make the patch simple and the commit log simple by removing extraneous details, this will all be obvious. 2) My version focus on the address in ranges, weaken the size parameter. Which make the code easy to read and follow the canonical convention to express the address range. while the vga_is_firmware_default() is not. Whether it's start/size or start/end is a trivial question. We don't need to waste time on it now. 3) A tiny change make a big difference. The original vga_is_firmware_default() only works with the assumption that the PCI resource reallocation won't happens. While I see no clue that why this is true even on X86 and IA64. The original patch[1] not mention this assumption explicitly. [1] 86fd887b7fe3 ('vgaarb: Don't default exclusively to first video device with mem+io') That sounds like a good idea, because this is all based on the framebuffer in screen_info, and screen_info was initialized before PCI enumeration, and it certainly doesn't account for any BAR changes done by the PCI core. Yes. So why would we keep vga_is_firmware_default() at all? If the header fixup has already identified the firmware framebuffer, it seems pointless to look again later. It need another patch to do the cleanup work, while my patch just add code to solve the real problem. It focus on provide a solution for the architectures which have a decent way set up the screen_info. Other things except that is secondary. I don't want both mechanisms when only one of them is useful. PCI BAR reassignment is completely fine, and keeping the assumption in vga_is_firmware_default() that we can compare reassigned BAR values to the pre-reassignment screen_info range is a trap that we should remove. OK,it's clear now. I know what to do next. Thanks. Bjorn
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/22 01:33, Bjorn Helgaas wrote: On Fri, Aug 18, 2023 at 09:48:46AM +0800, suijingfeng wrote: On 2023/8/18 06:08, Bjorn Helgaas wrote: + if (resource_type(res) != IORESOURCE_MEM) + continue; + + if (!res->start || !res->end) + continue; + + if (res->start <= fb_start && fb_end <= res->end) { + pdev_boot_vga = pdev; + + vgaarb_info(&pdev->dev, + "BAR %d contains firmware FB\n", i); Print the BAR with %pR and include the framebuffer region from screen_info in the same format. I do remember that you already told me to do this in V3, sorry for not replying to you at V3. Most of the time, what you tell me is right.But here, I think I need to explain. Because doing it that way will make the code line too long,and it will exceed 80 characters in the column if we print too much. I believe that the vgaarb_info() at here is already the most compact and simplest form. Printing the BAR with %pR is not absolute necessary, because we can get the additional information by: $ lspci | grep VGA $ dmesg | grep 05:00.0 $ dmesg | grep :03:00.0 $ dmesg | grep PCI Fair enough. The BAR info is already there. But I don't think the screen_info framebuffer data is in the dmesg log anywhere, and I think that would be useful. It's fine if dmesg lines or kernel printk lines exceed 80 columns. Actually, I only add something that is absolute necessary. Printing BAR with %pR and/or Printing the framebuffer region is consider to only for *debugging* purpose. I think printing the framebuffer region is important because it makes it clear *why* we're selecting the device as the default VGA device. It's more than just debugging; it helps make the system more transparent and more understandable. OK, I'm clear what to do next. The printing will be added at the next version. Bjorn
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
On Fri, Aug 18, 2023 at 12:09:29PM +0800, suijingfeng wrote: > On 2023/8/18 06:08, Bjorn Helgaas wrote: > > On Wed, Aug 16, 2023 at 06:05:27AM +0800, Sui Jingfeng wrote: > > > Currently, the vga_is_firmware_default() function only works on x86 and > > > ia64, it is a no-op on ARM, ARM64, PPC, RISC-V, etc. This patch completes > > > the implementation for the rest of the architectures. The added code tries > > > to identify the PCI(e) VGA device that owns the firmware framebuffer > > > before PCI resource reallocation happens. > > > > As far as I can tell, this is basically identical to the existing > > vga_is_firmware_default(), except that this patch funs that code as a > > header fixup, so it happens before any PCI BAR reallocations happen. > > Yes, what you said is right in overall. > But I think I should mention a few tiny points that make a difference. > > 1) My version is *less arch-dependent* Of course. If we make the patch simple and the commit log simple by removing extraneous details, this will all be obvious. > 2) My version focus on the address in ranges, weaken the size parameter. > > Which make the code easy to read and follow the canonical convention to > express the address range. while the vga_is_firmware_default() is not. Whether it's start/size or start/end is a trivial question. We don't need to waste time on it now. > 3) A tiny change make a big difference. > > The original vga_is_firmware_default() only works with the assumption > that the PCI resource reallocation won't happens. While I see no clue > that why this is true even on X86 and IA64. The original patch[1] not > mention this assumption explicitly. > [1] 86fd887b7fe3 ('vgaarb: Don't default exclusively to first video device > with mem+io') > > > That sounds like a good idea, because this is all based on the > > framebuffer in screen_info, and screen_info was initialized before PCI > > enumeration, and it certainly doesn't account for any BAR changes done > > by the PCI core. > > Yes. > > > So why would we keep vga_is_firmware_default() at all? If the header > > fixup has already identified the firmware framebuffer, it seems > > pointless to look again later. > > It need another patch to do the cleanup work, while my patch just > add code to solve the real problem. It focus on provide a solution > for the architectures which have a decent way set up the > screen_info. Other things except that is secondary. I don't want both mechanisms when only one of them is useful. PCI BAR reassignment is completely fine, and keeping the assumption in vga_is_firmware_default() that we can compare reassigned BAR values to the pre-reassignment screen_info range is a trap that we should remove. Bjorn
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
On Fri, Aug 18, 2023 at 09:48:46AM +0800, suijingfeng wrote: > On 2023/8/18 06:08, Bjorn Helgaas wrote: > > > + if (resource_type(res) != IORESOURCE_MEM) > > > + continue; > > > + > > > + if (!res->start || !res->end) > > > + continue; > > > + > > > + if (res->start <= fb_start && fb_end <= res->end) { > > > + pdev_boot_vga = pdev; > > > + > > > + vgaarb_info(&pdev->dev, > > > + "BAR %d contains firmware FB\n", i); > > Print the BAR with %pR and include the framebuffer region from > > screen_info in the same format. > > I do remember that you already told me to do this in V3, sorry for not > replying to you at V3. Most of the time, what you tell me is right.But here, > I think I need to explain. Because doing it that way will make the code line > too long,and it will exceed 80 characters in the column if we print too > much. > I believe that the vgaarb_info() at here is already the most compact and > simplest form. Printing the BAR with %pR is not absolute necessary, because > we can get the additional information by: $ lspci | grep VGA > > $ dmesg | grep 05:00.0 > $ dmesg | grep :03:00.0 > $ dmesg | grep PCI Fair enough. The BAR info is already there. But I don't think the screen_info framebuffer data is in the dmesg log anywhere, and I think that would be useful. It's fine if dmesg lines or kernel printk lines exceed 80 columns. > Actually, I only add something that is absolute necessary. > Printing BAR with %pR and/or Printing the framebuffer region > is consider to only for *debugging* purpose. I think printing the framebuffer region is important because it makes it clear *why* we're selecting the device as the default VGA device. It's more than just debugging; it helps make the system more transparent and more understandable. Bjorn
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 18:20, suijingfeng wrote: 1) The "weird" logic completely overrides whatever decision VGAARB ever made. It seems to say that the decision ever made by VGAARB is useless. Well, I think VGAARB shouldn't endure this; VGAARB has to be small. VGAARB have to be smart! The "weird logic" here refer to the weird pci_fixup_vgadev() function.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: I guess the point here is that: - 03:00.0 BAR 0 is [mem 0xe005000-0xe005fff] - screen_info says the framebuffer is [mem 0xe005000-0xe005fff] (or part of it) - Therefore, we want 03:00.0 to be the default VGA - PCI core reassigns 03:00.0 BAR 0 to [mem 0xe003000-0xe003fff] - PCI core assigns a 00:06.1 BAR to contain [mem 0xe005000-0xe005fff] - vga_is_firmware_default() incorrectly decides 00:06.1 should be the default VGA because it has a BAR that contains the screen_info address range Is that right? Yes, The 00:06.1 is loongson integrated display controller, integrated in LS7A1000 North bridge. On loongarch, before apply apply any patch, VGAARB always select 00:06.1 as the default boot device. because it is enumerated first. It is always the first VGA compatible device found on our system. Because its PCI domain, bus, function number is smallest. And it own IO and MEM, so the 00:06.1 will always be selected as the default boot device. Even you plug a discrete GPU on the mother board. Therefore we need help the vga_is_firmware_default() to overriding previous. On a multiple GPU co-exist case (on loongson platform), if no "overriding previous" being printed. then there something wrong. On normal case, we need the discrete GPU overriding the integrated one. Because the discrete GPU is more powerful than the platform integrated. But what we want is let the VGAARB determine the primary GPU by referencing the firmware. If firmware put the firmware framebuffer in the VRAM of the integrated display card(00:06.1). then the integrated display card should be the primary GPU. If firmware put the firmware framebuffer in the VRAM of the discrete display card, then the discrete display card should be the primary GPU.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: + +/* + * Identify the PCI VGA device that contains the firmware framebuffer + */ +static void pci_boot_vga_finder(struct pci_dev *pdev) +{ + resource_size_t fb_start; + resource_size_t fb_end; + unsigned int i; + + /* Already found the pdev which has firmware framebuffer ownership */ + if (pdev_boot_vga) + return; + + if (!vga_arb_get_firmware_fb_range(&fb_start, &fb_end)) + return; + + for (i = 0; i < PCI_STD_NUM_BARS; i++) { + struct resource *res = &pdev->resource[i]; This is essentially identical to vga_is_firmware_default() so it should look the same, i.e., it should use pci_dev_for_each_resource(). OK, will be fixed at the next version. Thanks.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: Please note that before apply this patch, vgaarb can not select the right boot vga due to weird logic introduced with the commit 57fc7323a8e7c ("LoongArch: Add PCI controller support") If we need this reference to 57fc7323a8e7c, we need more specifics about what the "weird logic" is. pci_fixup_vgadev() is the only obvious VGA connection, so I suppose it's related to that. Sorry, re-reading your reviews, I realized that I may miss the point, here, the sentence is not well phased, Correct expression should be: Please note that before apply this patch, there don't have a reasonable, system level solution on loongarch. VGAARB select a wrong device as primary GPU due to the pci_fixup_vgadev() function introduced with the commit 57fc7323a8e7c ("LoongArch: Add PCI controller support").
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: Please note that before apply this patch, vgaarb can not select the right boot vga due to weird logic introduced with the commit 57fc7323a8e7c ("LoongArch: Add PCI controller support") If we need this reference to 57fc7323a8e7c, we need more specifics about what the "weird logic" is. pci_fixup_vgadev() is the only obvious VGA connection, so I suppose it's related to that. Yes, you are right. The pci_fixup_vgadev() function will set the last VGA device enumerated as the default boot device. By "the last" VGA device, I mean that this device has the largest PCI bus, domain, and function triple. Thus, it is added to vgaarb in the end of all VGA device. So that logic expresses that the last one added will be the default. This probably is not what we want. On the LS3A5000+LS7A1000 platform, the last VGA device is a S3 graphics (08:00.0). This GPU has two cores. Say the log below: $ lspci | grep VGA 00:06.1 VGA compatible controller: Loongson Technology LLC DC (Display Controller) (rev 01) 03:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Caicos XT [Radeon HD 7470/8470 / R5 235/310 OEM] 07:00.0 VGA compatible controller: S3 Graphics Ltd. Device 9070 (rev 01) 08:00.0 VGA compatible controller: S3 Graphics Ltd. Device 9070 (rev 01) [0.361781] vgaarb: loaded [0.367838] pci :00:06.1: Overriding boot device as 1002:6778 [0.367841] pci :00:06.1: Overriding boot device as 5333:9070 [0.367843] pci :00:06.1: Overriding boot device as 5333:9070 1) The "weird" logic completely overrides whatever decision VGAARB ever made. It seems to say that the decision ever made by VGAARB is useless. Well, I think VGAARB shouldn't endure this; VGAARB has to be small. 2) The results it gives are not correct either. In the first testing example in my commit message, it overrides the S3 graphics as the default boot VGA instead of the AMD/ATI GPU. Actually, the firmware chooses the AMD/ATI GPU as the "frimware default". 3) It tries to make the decision for the end user instead of the firmware. Therefore, that function is always wrong. Again, it's a policy, not a mechanism. Since that already have been merge, I'm fine. Maybe Huacai is busy, he might don't has the time to carry on a deep thinking. But I think we should correct the mistake ever made, let's merge this patch to make vgaarb great again ? Well, that commit is not a dependency, I don't mind delete the referencing to that commit. After all, I think my patch will be effective on other architectures. Is additional testing on ARM64 and X86 is needed, if so I have to find the machine to carry on the testing.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: This patch makes the vga_is_firmware_default() function works on whatever arch that has UEFI GOP support. But we make it available only on platforms where PCI resource relocation happens. if the provided method proves to be effective and reliable, it can be expanded to other arch easily. v2: * Fix test robot warnnings and fix typos v3: * Fix linkage problems if the global screen_info is not exported v4: * Handle linkage problems by hiding behind of CONFIG_SYSFB, * Drop side-effects and simplify. The v2, v3, v4 changelog is nice, but we don't need it in the commit log itself, where it will become part of the git history. It should go in a cover letter or after the "---" marker: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/submitting-patches.rst?id=v6.0#n678 Thanks for point it out, now I know. Will be fixed at the next version.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: On Wed, Aug 16, 2023 at 06:05:27AM +0800, Sui Jingfeng wrote: Currently, the vga_is_firmware_default() function only works on x86 and ia64, it is a no-op on ARM, ARM64, PPC, RISC-V, etc. This patch completes the implementation for the rest of the architectures. The added code tries to identify the PCI(e) VGA device that owns the firmware framebuffer before PCI resource reallocation happens. As far as I can tell, this is basically identical to the existing vga_is_firmware_default(), except that this patch funs that code as a header fixup, so it happens before any PCI BAR reallocations happen. Yes, what you said is right in overall. But I think I should mention a few tiny points that make a difference. 1) My version is *less arch-dependent* Again, since the global screen_info is arch-dependent. The vga_is_firmware_default() mess up the arch-dependent part and arch-independent part. It's a mess and it's a bit harder to make the cleanup on the top of it. While my version is my version split the arch-dependent part and arch-independent part clearly. Since we decide to make it less arch-dependent, we have to bear the pain. Despite all other arches should always export the screen_info like the X86 and IA64 arch does, or at least a arch should give a Kconfig token (for example, CONFIG_ARCH_HAS_SCREEN_INFO) to demonstrate that an arch has the support for it. While currently, the fact is that the dependence just populated to everywhere. I think this is the hard part, you have to investigate how various arches defines and set up the screen_info. And then process dependency and the linkage problem across arch properly. 2) My version focus on the address in ranges, weaken the size parameter. Which make the code easy to read and follow the canonical convention to express the address range. while the vga_is_firmware_default() is not. 3) A tiny change make a big difference. The original vga_is_firmware_default() only works with the assumption that the PCI resource reallocation won't happens. While I see no clue that why this is true even on X86 and IA64. The original patch[1] not mention this assumption explicitly. [1] 86fd887b7fe3 ('vgaarb: Don't default exclusively to first video device with mem+io') That sounds like a good idea, because this is all based on the framebuffer in screen_info, and screen_info was initialized before PCI enumeration, and it certainly doesn't account for any BAR changes done by the PCI core. Yes. So why would we keep vga_is_firmware_default() at all? If the header fixup has already identified the firmware framebuffer, it seems pointless to look again later. It need another patch to do the cleanup work, while my patch just add code to solve the real problem. It focus on provide a solution for the architectures which have a decent way set up the screen_info. Other things except that is secondary.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Hi, On 2023/8/18 06:08, Bjorn Helgaas wrote: + if (resource_type(res) != IORESOURCE_MEM) + continue; + + if (!res->start || !res->end) + continue; + + if (res->start <= fb_start && fb_end <= res->end) { + pdev_boot_vga = pdev; + + vgaarb_info(&pdev->dev, + "BAR %d contains firmware FB\n", i); Print the BAR with %pR and include the framebuffer region from screen_info in the same format. I do remember that you already told me to do this in V3, sorry for not replying to you at V3. Most of the time, what you tell me is right.But here, I think I need to explain. Because doing it that way will make the code line too long,and it will exceed 80 characters in the column if we print too much. I believe that the vgaarb_info() at here is already the most compact and simplest form. Printing the BAR with %pR is not absolute necessary, because we can get the additional information by: $ lspci | grep VGA $ dmesg | grep 05:00.0 $ dmesg | grep :03:00.0 $ dmesg | grep PCI Actually, I only add something that is absolute necessary. Printing BAR with %pR and/or Printing the framebuffer region is consider to only for *debugging* purpose.
Re: [PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
On Wed, Aug 16, 2023 at 06:05:27AM +0800, Sui Jingfeng wrote: > Currently, the vga_is_firmware_default() function only works on x86 and > ia64, it is a no-op on ARM, ARM64, PPC, RISC-V, etc. This patch completes > the implementation for the rest of the architectures. The added code tries > to identify the PCI(e) VGA device that owns the firmware framebuffer > before PCI resource reallocation happens. As far as I can tell, this is basically identical to the existing vga_is_firmware_default(), except that this patch funs that code as a header fixup, so it happens before any PCI BAR reallocations happen. That sounds like a good idea, because this is all based on the framebuffer in screen_info, and screen_info was initialized before PCI enumeration, and it certainly doesn't account for any BAR changes done by the PCI core. So why would we keep vga_is_firmware_default() at all? If the header fixup has already identified the firmware framebuffer, it seems pointless to look again later. > This patch makes the vga_is_firmware_default() function works on whatever > arch that has UEFI GOP support. But we make it available only on platforms > where PCI resource relocation happens. if the provided method proves to be > effective and reliable, it can be expanded to other arch easily. > > v2: > * Fix test robot warnnings and fix typos > > v3: > * Fix linkage problems if the global screen_info is not exported > > v4: > * Handle linkage problems by hiding behind of CONFIG_SYSFB, > * Drop side-effects and simplify. The v2, v3, v4 changelog is nice, but we don't need it in the commit log itself, where it will become part of the git history. It should go in a cover letter or after the "---" marker: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/submitting-patches.rst?id=v6.0#n678 > Since only one GPU could owns the firmware fb in normal case, things > are almost done once we determine the boot VGA selected by firmware. > By using the DECLARE_PCI_FIXUP_CLASS_HEADER(), we also ensure that we > could identify the primary display (if there do have one) before PCI > resource reallocation happen. > > The method provided in this patch will be effective as long as the target > platform has a way set up the kernel's screen_info. For the machines with > UEFI firmware, the screen_info is typically set up by the UEFI stub with > respect to the UEFI GOP protocol. Since the UEFI stub executes in the > context of the decompressor, and it cannot access the kernel's screen_info > directly. Hence, the efi stub copies the screen_info provided by firmware > to kernel's counterpart. Therefore, we handle linkage problems by using > the CONFIG_EFI guard. Since when CONFIG_EFI is set, the kernel's > screen_info is guaranteed to be static linkable for arm, arm64, riscv. > V4 of this patch handle linkage problems by hiding behind of CONFIG_SYSFB, > since sysfb reference the global screen_info as well. > > This patch is tested on: > > 1) LS3A5000+LS7A1000 platform with three video cards > > $ lspci | grep VGA > > 00:06.1 VGA compatible controller: Loongson Technology LLC DC (Display > Controller) (rev 01) > 03:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] > Caicos XT [Radeon HD 7470/8470 / R5 235/310 OEM] > 07:00.0 VGA compatible controller: S3 Graphics Ltd. Device 9070 (rev 01) > 08:00.0 VGA compatible controller: S3 Graphics Ltd. Device 9070 (rev 01) > > Before apply this patch: > > [0.361748] pci :00:06.1: vgaarb: setting as boot VGA device > [0.361751] pci :00:06.1: vgaarb: bridge control possible > [0.361753] pci :00:06.1: vgaarb: VGA device added: > decodes=io+mem,owns=io+mem,locks=none > [0.361763] pci :03:00.0: vgaarb: bridge control possible > [0.361765] pci :03:00.0: vgaarb: VGA device added: > decodes=io+mem,owns=none,locks=none > [0.361771] pci :07:00.0: vgaarb: bridge control possible > [0.361773] pci :07:00.0: vgaarb: VGA device added: > decodes=io+mem,owns=none,locks=none > [0.361777] pci :08:00.0: vgaarb: bridge control possible > [0.361779] pci :08:00.0: vgaarb: VGA device added: > decodes=io+mem,owns=none,locks=none > [0.361781] vgaarb: loaded > [0.367838] pci :00:06.1: Overriding boot device as 1002:6778 > [0.367841] pci :00:06.1: Overriding boot device as 5333:9070 > [0.367843] pci :00:06.1: Overriding boot device as 5333:9070 > > After apply this patch: > > [0.357780] pci :03:00.0: vgaarb: BAR 0 contains firmware FB > [0.361726] pci :00:06.1: vgaarb: setting as boot VGA device > [0.361729] pci :00:06.1: vgaarb: bridge control possible > [0.361731] pci :00:06.1: vgaarb: VGA device added: > decodes=io+mem,owns=io+mem,locks=none > [0.361741] pci :03:00.0: vgaarb: setting as boot VGA device > (overriding previous) > [0.361743] pci :03:00.0: vgaarb: bridge control p
[PATCH v4] PCI/VGA: Make the vga_is_firmware_default() less arch-dependent
Currently, the vga_is_firmware_default() function only works on x86 and ia64, it is a no-op on ARM, ARM64, PPC, RISC-V, etc. This patch completes the implementation for the rest of the architectures. The added code tries to identify the PCI(e) VGA device that owns the firmware framebuffer before PCI resource reallocation happens. This patch makes the vga_is_firmware_default() function works on whatever arch that has UEFI GOP support. But we make it available only on platforms where PCI resource relocation happens. if the provided method proves to be effective and reliable, it can be expanded to other arch easily. v2: * Fix test robot warnnings and fix typos v3: * Fix linkage problems if the global screen_info is not exported v4: * Handle linkage problems by hiding behind of CONFIG_SYSFB, * Drop side-effects and simplify. Since only one GPU could owns the firmware fb in normal case, things are almost done once we determine the boot VGA selected by firmware. By using the DECLARE_PCI_FIXUP_CLASS_HEADER(), we also ensure that we could identify the primary display (if there do have one) before PCI resource reallocation happen. The method provided in this patch will be effective as long as the target platform has a way set up the kernel's screen_info. For the machines with UEFI firmware, the screen_info is typically set up by the UEFI stub with respect to the UEFI GOP protocol. Since the UEFI stub executes in the context of the decompressor, and it cannot access the kernel's screen_info directly. Hence, the efi stub copies the screen_info provided by firmware to kernel's counterpart. Therefore, we handle linkage problems by using the CONFIG_EFI guard. Since when CONFIG_EFI is set, the kernel's screen_info is guaranteed to be static linkable for arm, arm64, riscv. V4 of this patch handle linkage problems by hiding behind of CONFIG_SYSFB, since sysfb reference the global screen_info as well. This patch is tested on: 1) LS3A5000+LS7A1000 platform with three video cards $ lspci | grep VGA 00:06.1 VGA compatible controller: Loongson Technology LLC DC (Display Controller) (rev 01) 03:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Caicos XT [Radeon HD 7470/8470 / R5 235/310 OEM] 07:00.0 VGA compatible controller: S3 Graphics Ltd. Device 9070 (rev 01) 08:00.0 VGA compatible controller: S3 Graphics Ltd. Device 9070 (rev 01) Before apply this patch: [0.361748] pci :00:06.1: vgaarb: setting as boot VGA device [0.361751] pci :00:06.1: vgaarb: bridge control possible [0.361753] pci :00:06.1: vgaarb: VGA device added: decodes=io+mem,owns=io+mem,locks=none [0.361763] pci :03:00.0: vgaarb: bridge control possible [0.361765] pci :03:00.0: vgaarb: VGA device added: decodes=io+mem,owns=none,locks=none [0.361771] pci :07:00.0: vgaarb: bridge control possible [0.361773] pci :07:00.0: vgaarb: VGA device added: decodes=io+mem,owns=none,locks=none [0.361777] pci :08:00.0: vgaarb: bridge control possible [0.361779] pci :08:00.0: vgaarb: VGA device added: decodes=io+mem,owns=none,locks=none [0.361781] vgaarb: loaded [0.367838] pci :00:06.1: Overriding boot device as 1002:6778 [0.367841] pci :00:06.1: Overriding boot device as 5333:9070 [0.367843] pci :00:06.1: Overriding boot device as 5333:9070 After apply this patch: [0.357780] pci :03:00.0: vgaarb: BAR 0 contains firmware FB [0.361726] pci :00:06.1: vgaarb: setting as boot VGA device [0.361729] pci :00:06.1: vgaarb: bridge control possible [0.361731] pci :00:06.1: vgaarb: VGA device added: decodes=io+mem,owns=io+mem,locks=none [0.361741] pci :03:00.0: vgaarb: setting as boot VGA device (overriding previous) [0.361743] pci :03:00.0: vgaarb: bridge control possible [0.361745] pci :03:00.0: vgaarb: VGA device added: decodes=io+mem,owns=none,locks=none [0.361751] pci :07:00.0: vgaarb: bridge control possible [0.361753] pci :07:00.0: vgaarb: VGA device added: decodes=io+mem,owns=none,locks=none [0.361757] pci :08:00.0: vgaarb: bridge control possible [0.361759] pci :08:00.0: vgaarb: VGA device added: decodes=io+mem,owns=none,locks=none [0.361761] vgaarb: loaded [ 14.730255] radeon :03:00.0: vgaarb: changed VGA decodes: olddecodes=io+mem,decodes=none:owns=none Please note that before apply this patch, vgaarb can not select the right boot vga due to weird logic introduced with the commit 57fc7323a8e7c ("LoongArch: Add PCI controller support") Please also note that VARM Bar do moves on LoongArch: $ dmesg | grep :03:00.0 [0.357688] pci :03:00.0: [1002:6778] type 00 class 0x03 [0.357713] pci :03:00.0: reg 0x10: [mem 0xe005000-0xe005fff 64bit pref] [0.357730] pci :03:00.0: reg 0x18: [mem 0xe006530-0xe006531 64bit] [0.357741] pci :03:00.0: reg 0x20: [io 0