| Message ID | 3135179.iXEx1Rk87m@vostro.rjw.lan (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
On Sun, Dec 29, 2013 at 10:36:56PM +0100, Rafael J. Wysocki wrote: > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > The changes in the ACPI-based PCI hotplug (ACPIPHP) subsystem made > during the 3.12 development cycle uncovered a problem with VGA > switcheroo that on some systems, when the device-specific method > (ATPX in the radeon case, _DSM in the nouveau case) is used to turn > off the discrete graphics, the BIOS generates ACPI hotplug events for > that device and those events cause ACPIPHP to attempt to remove the > device from the system (they are events for a device that was present > previously and is not present any more, so that's what should be done > according to the spec). Then, the system stops functioning correctly. > > Since the hotplug events in question were simply silently ignored > previously, the least intrusive way to address that problem is to > make ACPIPHP ignore them again. For this purpose, introduce a new > ACPI device flag, no_hotplug, and modify ACPIPHP to ignore hotplug > events for PCI devices whose ACPI companions have that flag set. > Next, make the radeon and nouveau switcheroo detection code set the > no_hotplug flag for the discrete graphics' ACPI companion. > > Fixes: bbd34fcdd1b2 (ACPI / hotplug / PCI: Register all devices under the given bridge) > References: https://bugzilla.kernel.org/show_bug.cgi?id=61891 > References: https://bugzilla.kernel.org/show_bug.cgi?id=64891 > Reported-and-tested-by: Mike Lothian <mike@fireburn.co.uk> > Reported-and-tested-by: <madcatx@atlas.cz> > Reported-by: Joaquín Aramendía <samsagax@gmail.com> > Cc: Alex Deucher <alexdeucher@gmail.com> > Cc: Dave Airlie <airlied@linux.ie> > Cc: Takashi Iwai <tiwai@suse.de> > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > Cc: 3.12+ <stable@vger.kernel.org> # 3.12+ FWIW, Thunderbolt hotplug still works fine after this patch is applied :) -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Monday, December 30, 2013 11:20:43 AM Mika Westerberg wrote: > On Sun, Dec 29, 2013 at 10:36:56PM +0100, Rafael J. Wysocki wrote: > > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > > > The changes in the ACPI-based PCI hotplug (ACPIPHP) subsystem made > > during the 3.12 development cycle uncovered a problem with VGA > > switcheroo that on some systems, when the device-specific method > > (ATPX in the radeon case, _DSM in the nouveau case) is used to turn > > off the discrete graphics, the BIOS generates ACPI hotplug events for > > that device and those events cause ACPIPHP to attempt to remove the > > device from the system (they are events for a device that was present > > previously and is not present any more, so that's what should be done > > according to the spec). Then, the system stops functioning correctly. > > > > Since the hotplug events in question were simply silently ignored > > previously, the least intrusive way to address that problem is to > > make ACPIPHP ignore them again. For this purpose, introduce a new > > ACPI device flag, no_hotplug, and modify ACPIPHP to ignore hotplug > > events for PCI devices whose ACPI companions have that flag set. > > Next, make the radeon and nouveau switcheroo detection code set the > > no_hotplug flag for the discrete graphics' ACPI companion. > > > > Fixes: bbd34fcdd1b2 (ACPI / hotplug / PCI: Register all devices under the given bridge) > > References: https://bugzilla.kernel.org/show_bug.cgi?id=61891 > > References: https://bugzilla.kernel.org/show_bug.cgi?id=64891 > > Reported-and-tested-by: Mike Lothian <mike@fireburn.co.uk> > > Reported-and-tested-by: <madcatx@atlas.cz> > > Reported-by: Joaquín Aramendía <samsagax@gmail.com> > > Cc: Alex Deucher <alexdeucher@gmail.com> > > Cc: Dave Airlie <airlied@linux.ie> > > Cc: Takashi Iwai <tiwai@suse.de> > > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > Cc: 3.12+ <stable@vger.kernel.org> # 3.12+ > > FWIW, Thunderbolt hotplug still works fine after this patch is applied :) I've checked that, but thanks for the confirmation! Rafael -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Index: linux-pm/drivers/gpu/drm/radeon/radeon_atpx_handler.c =================================================================== --- linux-pm.orig/drivers/gpu/drm/radeon/radeon_atpx_handler.c +++ linux-pm/drivers/gpu/drm/radeon/radeon_atpx_handler.c @@ -33,6 +33,7 @@ static struct radeon_atpx_priv { bool atpx_detected; /* handle for device - and atpx */ acpi_handle dhandle; + acpi_handle other_handle; struct radeon_atpx atpx; } radeon_atpx_priv; @@ -451,9 +452,10 @@ static bool radeon_atpx_pci_probe_handle return false; status = acpi_get_handle(dhandle, "ATPX", &atpx_handle); - if (ACPI_FAILURE(status)) + if (ACPI_FAILURE(status)) { + radeon_atpx_priv.other_handle = dhandle; return false; - + } radeon_atpx_priv.dhandle = dhandle; radeon_atpx_priv.atpx.handle = atpx_handle; return true; @@ -526,10 +528,24 @@ static bool radeon_atpx_detect(void) } if (has_atpx && vga_count == 2) { + struct acpi_device *adev = NULL; + acpi_get_name(radeon_atpx_priv.atpx.handle, ACPI_FULL_PATHNAME, &buffer); printk(KERN_INFO "VGA switcheroo: detected switching method %s handle\n", acpi_method_name); radeon_atpx_priv.atpx_detected = true; + /* + * On some systems hotplug events are generated for the device + * being switched off when ATPX is executed. They cause ACPI + * hotplug to trigger and attempt to remove the device from + * the system, which causes it to break down. Prevent that from + * happening by setting the no_hotplug flag for the ACPI device + * object in question. + */ + acpi_bus_get_device(radeon_atpx_priv.other_handle, &adev); + if (adev) + adev->flags.no_hotplug = true; + return true; } return false; Index: linux-pm/drivers/gpu/drm/nouveau/nouveau_acpi.c =================================================================== --- linux-pm.orig/drivers/gpu/drm/nouveau/nouveau_acpi.c +++ linux-pm/drivers/gpu/drm/nouveau/nouveau_acpi.c @@ -51,6 +51,7 @@ static struct nouveau_dsm_priv { bool dsm_detected; bool optimus_detected; acpi_handle dhandle; + acpi_handle other_handle; acpi_handle rom_handle; } nouveau_dsm_priv; @@ -260,9 +261,10 @@ static int nouveau_dsm_pci_probe(struct if (!dhandle) return false; - if (!acpi_has_method(dhandle, "_DSM")) + if (!acpi_has_method(dhandle, "_DSM")) { + nouveau_dsm_priv.other_handle = dhandle; return false; - + } if (nouveau_test_dsm(dhandle, nouveau_dsm, NOUVEAU_DSM_POWER)) retval |= NOUVEAU_DSM_HAS_MUX; @@ -333,11 +335,25 @@ static bool nouveau_dsm_detect(void) nouveau_dsm_priv.optimus_detected = true; ret = true; } else if (vga_count == 2 && has_dsm && guid_valid) { + struct acpi_device *adev = NULL; + acpi_get_name(nouveau_dsm_priv.dhandle, ACPI_FULL_PATHNAME, &buffer); printk(KERN_INFO "VGA switcheroo: detected DSM switching method %s handle\n", acpi_method_name); nouveau_dsm_priv.dsm_detected = true; + /* + * On some systems hotplug events are generated for the device + * being switched off when _DSM is executed. They cause ACPI + * hotplug to trigger and attempt to remove the device from + * the system, which causes it to break down. Prevent that from + * happening by setting the no_hotplug flag for the ACPI device + * object in question. + */ + acpi_bus_get_device(nouveau_dsm_priv.other_handle, &adev); + if (adev) + adev->flags.no_hotplug = true; + ret = true; } Index: linux-pm/include/acpi/acpi_bus.h =================================================================== --- linux-pm.orig/include/acpi/acpi_bus.h +++ linux-pm/include/acpi/acpi_bus.h @@ -169,7 +169,8 @@ struct acpi_device_flags { u32 ejectable:1; u32 power_manageable:1; u32 match_driver:1; - u32 reserved:27; + u32 no_hotplug:1; + u32 reserved:26; }; /* File System */ Index: linux-pm/drivers/pci/hotplug/acpiphp_glue.c =================================================================== --- linux-pm.orig/drivers/pci/hotplug/acpiphp_glue.c +++ linux-pm/drivers/pci/hotplug/acpiphp_glue.c @@ -643,6 +643,24 @@ static void disable_slot(struct acpiphp_ slot->flags &= (~SLOT_ENABLED); } +static bool acpiphp_no_hotplug(acpi_handle handle) +{ + struct acpi_device *adev = NULL; + + acpi_bus_get_device(handle, &adev); + return adev && adev->flags.no_hotplug; +} + +static bool slot_no_hotplug(struct acpiphp_slot *slot) +{ + struct acpiphp_func *func; + + list_for_each_entry(func, &slot->funcs, sibling) + if (acpiphp_no_hotplug(func_to_handle(func))) + return true; + + return false; +} /** * get_slot_status - get ACPI slot status @@ -701,7 +719,8 @@ static void trim_stale_devices(struct pc unsigned long long sta; status = acpi_evaluate_integer(handle, "_STA", NULL, &sta); - alive = ACPI_SUCCESS(status) && sta == ACPI_STA_ALL; + alive = (ACPI_SUCCESS(status) && sta == ACPI_STA_ALL) + || acpiphp_no_hotplug(handle); } if (!alive) { u32 v; @@ -741,8 +760,9 @@ static void acpiphp_check_bridge(struct struct pci_dev *dev, *tmp; mutex_lock(&slot->crit_sect); - /* wake up all functions */ - if (get_slot_status(slot) == ACPI_STA_ALL) { + if (slot_no_hotplug(slot)) { + ; /* do nothing */ + } else if (get_slot_status(slot) == ACPI_STA_ALL) { /* remove stale devices if any */ list_for_each_entry_safe(dev, tmp, &bus->devices, bus_list)