From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-6.4 required=3.0 tests=BAYES_00,DKIMWL_WL_HIGH, DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,MAILING_LIST_MULTI,SPF_HELO_NONE, SPF_PASS autolearn=no autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 280B9C433ED for ; Tue, 18 May 2021 19:35:24 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id EDD77611BF for ; Tue, 18 May 2021 19:35:23 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1351844AbhERTgl (ORCPT ); Tue, 18 May 2021 15:36:41 -0400 Received: from mail.kernel.org ([198.145.29.99]:33864 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S229813AbhERTgl (ORCPT ); Tue, 18 May 2021 15:36:41 -0400 Received: by mail.kernel.org (Postfix) with ESMTPSA id 9795A60FE7; Tue, 18 May 2021 19:35:22 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1621366522; bh=8KOBawNhLkuA1r+CMebCDDsP1hDph3sC3ZIg/eaUa2M=; h=Date:From:To:Cc:Subject:In-Reply-To:From; b=WeP6xyDB2OQV473TN2IhFNKm6OqBkRvfHHGwgUdSGsiYARc9l6J/nhW4OOECv/8UK 1NCBImNXg8vQQ/T04XGhP/jdwhqdpyoc5R914hs20c2QhQrMXQUKUlonxaK9Fghcoa +8QdaNat30wCQK9heffry04dX6Fr/OLNsH3zlNvXuscGdVNTyXEqtK2R5+pbahALTp mJ4R0mrYV4qDYRSDTEW45kimGM45rSm2LqEYFstvCU+9AM4JkHscxWjPkEOHEMVlJ6 /zIio2U2kbWnuMtmFV1PBri1fXy9Krpv/9Z0BIJOUSsq0I8gPZGUloS8suhLtamblY 9j0rLHAE8mgRA== Date: Tue, 18 May 2021 14:35:21 -0500 From: Bjorn Helgaas To: Huacai Chen Cc: Huacai Chen , Bjorn Helgaas , linux-pci , Jiaxun Yang , Jingfeng Sui Subject: Re: [PATCH 5/5] PCI: Support ASpeed VGA cards behind a misbehaving bridge Message-ID: <20210518193521.GA145968@bjorn-Precision-5520> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: Precedence: bulk List-ID: X-Mailing-List: linux-pci@vger.kernel.org On Tue, May 18, 2021 at 03:13:43PM +0800, Huacai Chen wrote: > On Tue, May 18, 2021 at 2:28 AM Bjorn Helgaas wrote: > > On Mon, May 17, 2021 at 08:53:43PM +0800, Huacai Chen wrote: > > > On Sat, May 15, 2021 at 5:09 PM Huacai Chen wrote: > > > > On Fri, May 14, 2021 at 11:10 PM Bjorn Helgaas wrote: > > > > > On Fri, May 14, 2021 at 04:00:25PM +0800, Huacai Chen wrote: > > > > > > According to PCI-to-PCI bridge spec, bit 3 of Bridge Control Register is > > > > > > VGA Enable bit which modifies the response to VGA compatible addresses. > > > > > > > > > > The bridge spec is pretty old, and most of the content has been > > > > > incorporated into the PCIe spec. I think you can cite "PCIe r5.0, sec > > > > > 7.5.1.3.13" here instead. > > > > > > > > > > > If the VGA Enable bit is set, the bridge will decode and forward the > > > > > > following accesses on the primary interface to the secondary interface. > > > > > > > > > > *Which* following accesses? The structure of English requires that if > > > > > you say "the following accesses," you must continue by *listing* the > > > > > accesses. > > > > > > > > > > > The ASpeed AST2500 hardward does not set the VGA Enable bit on its > > > > > > bridge control register, which causes vgaarb subsystem don't think the > > > > > > VGA card behind the bridge as a valid boot vga device. > > > > > > > > > > s/hardward/bridge/ > > > > > s/vga/VGA/ (also in code comments and dmesg strings below) > > > > > > > > > > From the code, it looks like AST2500 ([1a03:2000]) is a VGA device, > > > > > since it apparently has a VGA class code. But here you say the > > > > > AST2500 has a Bridge Control register, which suggests that it's a > > > > > bridge. If AST2500 is some sort of combination that includes both a > > > > > bridge and a VGA device, please outline that topology. > > > > > > > > > > But the hardware defect is that some bridges forward VGA accesses even > > > > > though their VGA Enable bit is not set? The quirk should be attached > > > > > to broken *bridges*, not to VGA devices. > > > > > > > > > > If a bridge forwards VGA accesses regardless of how its VGA Enable bit > > > > > is set, that means VGA arbitration (in vgaarb.c) cannot work > > > > > correctly, so merely setting the default VGA device once in a quirk is > > > > > not sufficient. You would have to somehow disable any future attempts > > > > > to use other VGA devices. Only the VGA device below this defective > > > > > bridge is usable. Any other VGA devices in the system would be > > > > > useless. > > > > > > > > > > > So we provide a quirk to fix Xorg auto-detection. > > > > > > > > > > > > See similar bug: > > > > > > > > > > > > https://patchwork.kernel.org/project/linux-pci/patch/20170619023528.11532-1-dja@axtens.net/ > > > > > > > > > > This patch was never merged. If we merged a revised version, please > > > > > cite the SHA1 instead. > > > > > > > > This patch has never merged, and I found that it is unnecessary after > > > > commit a37c0f48950b56f6ef2ee637 ("vgaarb: Select a default VGA device > > > > even if there's no legacy VGA"). Maybe this ASpeed patch is also > > > > unnecessary. If it is still needed, I'll investigate the root cause. > > > > > > I found that vga_arb_device_init() and pcibios_init() are both wrapped > > > by subsys_initcall(), which means their sequence is unpredictable. And > > > unfortunately, in our platform vga_arb_device_init() is called before > > > pcibios_init(), which makes vga_arb_device_init() fail to set a > > > default vga device. This is the root cause why we thought that we > > > still need a quirk for AST2500. > > > > Does this mean there is no hardware defect here? The VGA Enable bit > > works correctly? > > > No, VGA Enable bit still doesn't set, but with commit > a37c0f48950b56f6ef2ee637 ("vgaarb: Select a default VGA device even if > there's no legacy VGA") we no longer depend on VGA Enable. Correct me if I'm wrong: - On the AST2500 bridge [1a03:1150], the VGA Enable bit is read-only 0. - The AST2500 bridge never forwards VGA accesses ([mem 0xa0000-0xbffff], [io 0x3b0-0x3bb], [io 0x3c0-0x3df]) to its secondary bus. The VGA Enable bit is optional, and if both the above are true, the bridge is working correctly per spec, and the quirk below is not the right solution, and whatever solution we come up with should not claim that the bridge is misbehaving. > > > I think the best solution is make vga_arb_device_init() be wrapped by > > > subsys_initcall_sync(), do you think so? > > > > Hmm. Unfortunately the semantics of subsys_initcall_sync() are not > > documented, so I'm not sure exactly *why* such a change would work and > > whether we could rely on it to continue working. > > > > pcibios_init() isn't very consistent across arches. On some, > > including alpha, microblaze, some MIPS platforms, powerpc, and sh, it > > enumerates PCI devices. On others (ia64, parisc, sparc, x86), it does > > basically nothing. That makes life a little difficult. > > subsys_initcall_sync() is ensured after all subsys_initcall() > functions, so at least it can solve the problem on platforms which use > pcibios_init() to enumerate PCI devices (x86 and other ACPI-based > platforms are also OK, because they use acpi_init() > -->acpi_scan_init() -->pci_acpi_scan_root() to enumerate devices). More details in my response to suijingfeng: https://lore.kernel.org/r/20210518193100.GA148462@bjorn-Precision-5520 I'd rather not fiddle with the initcall ordering. That mechanism is fragile and I'd prefer something more robust. I'm wondering whether it's practical to do something in the normal PCI enumeration path, e.g., in pci_init_capabilities(). Maybe we can detect the default VGA device as we enumerate it. Then we wouldn't have this weird process of "find all PCI devices first, then scan for the default VGA device, and oh, by the way, also check for VGA devices hot-added later." Bjorn