Drivers fail to configure - device already claimed

Submitted by d3c..@..il.com

Assigned to Xorg Project Team

Link to original bug (#91748)

Description

I recently put together a new system, and have had issues getting X to come up and be useful. I'm using Arch linux.

X.Org X Server 1.17.2 Release Date: 2015-06-16 X Protocol Version 11, Revision 0 Build Operating System: Linux 4.1.5-1-ARCH x86_64 Current Operating System: Linux tower2 4.1.5-1-ARCH #1 (closed) SMP PREEMPT Tue Aug 11 15:41:14 CEST 2015 x86_64 Kernel command line: BOOT_IMAGE=/vmlinuz-linux root=UUID=e126c645-2cb7-421f-8e0a-0f42c36337a7 rw quiet Build Date: 24 August 2015 01:48:03AM

I was getting 'ErrorF("No devices to configure. Configuration failed.\n");' I thought it was xf86-video-intel reporting no devices, but after adding tracing logging to that found it was returning success. The issue comes up later in xf86pciBus.c

/*

• Check if the slot requested is free. If it is already in use, return FALSE. */ Bool xf86CheckPciSlot(const struct pci_device *d) ... for (i = 0; i < xf86NumEntities; i++) { const EntityPtr p = xf86Entities[i]; ... #ifdef XSERVER_PLATFORM_BUS if ((p->bus.type == BUS_PLATFORM) && (p->bus.id.plat->pdev)) { struct pci_device *ud = p->bus.id.plat->pdev; if (MATCH_PCI_DEVICES(ud, d))

the MATCH_PCI_DEVICES finds the same device already registered from intel driver's drv->platformProbe... ( xf86platformProbeDev(drv) ) this was called earlier when the driver was first loaded, which added the device to xf86Entities. So later, when it is in xf86PciAddMatchingDev it fails because it thinks the device was already claimed; however it is claimed by the driver itself.

I found that this patch... http://cgit.freedesktop.org/xorg/xserver/commit/hw/xfree86/common/xf86pciBus.c?id=cf66471353ac5899383b573a3cfca407e90d501e

is the one that added the MATCH_PCI_DEVICES code mentioend above, but that was in 2012-07-06; so I don't know how any system is managing to function, unless it's a combination of options passed to ./configure that's causing it

This is the configure command line it uses...

./configure --prefix=/usr
--enable-ipv6
--enable-dri
--enable-dmx
--enable-xvfb
--enable-xnest
--enable-composite
--enable-xcsecurity
--enable-libunwind
--enable-xorg
--enable-xephyr
--enable-glamor
--enable-xwayland
--enable-glx-tls
--enable-kdrive
--enable-kdrive-evdev
--enable-kdrive-kbd
--enable-kdrive-mouse
--enable-config-udev
--enable-systemd-logind
--enable-suid-wrapper
--disable-install-setuid
--enable-record
--disable-xfbdev
--disable-xfake
--disable-static
--libexecdir=/usr/lib/xorg-server
--sysconfdir=/etc
--localstatedir=/var
--with-xkb-path=/usr/share/X11/xkb
--with-xkb-output=/var/lib/xkb
--with-fontrootdir=/usr/share/fonts
--with-sha1=libgcrypt

it does --enable-config-udev which results in (I think) XSERVER_PLATFORM_BUS being defined. (it's actually because CONFIG_UDEV_KMS is defined in the script, but I didn't see an option specifically for udev_kms)

---

I managed to get it to work by commenting out the return FALSE in xf86CheckPciSlot. But later had an issue that xf86screens entry didn't have a PreInit function defined...

/* xf86Configure.c approx line 729 */

/* added check to see if preinit function was defined before calling it, otherwise it called 0x00000 address */

    if ((*xf86Screens[dev2screen[j]]->PreInit )&&
        (*xf86Screens[dev2screen[j]]->PreInit) (xf86Screens[dev2screen[j]],
                                                PROBE_DETECT) &&
        ConfiguredMonitor) {
        monitor_ptr = configureDDCMonitorSection(j);
    }
    else {
        monitor_ptr = configureMonitorSection(j);
    }

---

After the first (hack) and the second fix was able to get Xorg :0 to start from a ssh terminal, but not from a console terminal.

Version: git