On Monday, February 03, 2014 12:44:08 PM Mika Westerberg wrote: > On Sun, Feb 02, 2014 at 01:19:33AM +0100, Rafael J. Wysocki wrote: > > On Monday, January 27, 2014 01:37:17 AM Rafael J. Wysocki wrote: > > > Hi All, > > > > > > ACPIPHP can be simplified a bit on top of some PCI and ACPI changes merged > > > recently and the following series of patches implements those simplifications: > > > > > > [1/11] Fix up two kerneldoc comments in acpiphp_glue.c. > > > [2/11] Get rid of an unnecessary label in register_slot(). > > > [3/11] Drop acpiphp_bus_trim() and use acpi_bus_trim() instead of it directly. > > > [4/11] Move the acpi_bus_get_device() call out of acpiphp_no_hotplug(). > > > [5/11] Store struct acpi_device pointers instead of ACPI handles in struct acpiphp_context. > > > [6/11] Drop acpiphp_bus_add() (which has only one user). > > > [7/11] Drop crit_sect mutexes (that are redundant). > > > [8/11] Clean up the usage of the slot variable in hotplug_event(). > > > [9/11] Drop dev_in_slot() and rework disable_slot() to walk bus->devices directly. > > > [10/11] Use acpi_handle_debug() in hotplug_event() instead of open-coded stuff. > > > [11/11] Drop handle argument from the member functions of struct acpi_dock_ops. > > > > > > All of that is relateively straightforward, but I have some more intrusive changes > > > on top of it in the works. They will be posted separately later this week. > > > > I've learned a couple of things since I sent this patchset. First, all > > bus->devices list walks that may remove PCI devices should be done in reverse > > order or they can crash if virtual functions are involved. Second, hotplug_event() > > (in acpiphp_glue.c) has to acquire pci_rescan_remove_lock by itself, because it > > may be called from multiple places and all of them need that lock to be held. > > That is done by patches [1-2/13] which I'm planning to push as fixes for 3.14-rc2. > > > > The rest is pretty much the same as last time except that the old patch [9/11] > > became [3/13] in this series and it has been changed so that the list is walked in > > reverse order. > > On Intel NUC and DZ77RE-75K, > > Tested-by: Mika Westerberg <mika.westerberg@xxxxxxxxxxxxxxx> Thanks! As I said previously, I have found two concurrency-related bugs more in ACPIPHP and I needed to put fixes for those bugs at the top of the series (after previous [1-2/13] and the patch at https://patchwork.kernel.org/patch/3567701/). For this reason, I also had to rebase the ACPI/PCI hotplug consolidation patchset (https://lkml.org/lkml/2014/2/2/87) which got a few additional cosmetic changes too. The following is a combination of the two series with one patch dropped (it would conflict with [4-5/24]) and a few patches added. It is on top of 3.14-rc1. Patches [1-5/24], that are regarded as 3.14-rc2 material, are on the bleeding-edge branch of linux-pm.git. The remaining patches will show up in bleeding-edge shortly. [1/24] Remove entries from bus->devices in reverse order (in ACPIPHP). [2/24] Move PCI rescan-remove locking to hotplug_event(). [3/24] Scan root bus under the PCI rescan-remove lock [4/24] Fix race in handle_hotplug_event() related to concurrent bridge removal. [5/24] Fix race in hotplug_event() related to dock events and concurrent bridge removal. [6/24] Drop dev_in_slot() and rework disable_slot() to walk bus->devices directly. [7/24] Fix up two kerneldoc comments in acpiphp_glue.c. [8/24] Get rid of an unnecessary label in register_slot(). [9/24] Drop acpiphp_bus_trim() and use acpi_bus_trim() instead of it directly. [10/24] Move the acpi_bus_get_device() call out of acpiphp_no_hotplug(). [11/24] Store struct acpi_device pointers instead of ACPI handles in struct acpiphp_context. [12/24] Drop acpiphp_bus_add() (which has only one user). [13/24] Drop crit_sect mutexes (that are redundant). [14/24] Clean up the usage of the slot variable in hotplug_event(). [15/24] Use acpi_handle_debug() in hotplug_event() instead of open-coded stuff. [16/24] Do not pass ACPI handle to hotplug_event(). [17/24] Add a new function to ACPICA allowing a callback to be executed under the namespace mutex after calling acpi_ns_get_attached_data(). [18/24] Use the new ACPICA's function to fix a couple of potential races related to ACPI notifies. [19/24] Move the hotplug context lock definition to the ACPI core (from ACPIPHP). [20/24] Consolidate ACPI hotplug signaling for PCI and ACPI core. [21/24] Rework the handling of eject requests in the ACPI core. [22/24] Simplify a routine for installing hotplug notify handlers. [23/24] Dispatch ACPI hotplug notifications for "core" devices and PCI from acpi_bus_notify(). [24/24] Pass struct acpi_device pointer to acpiphp_check_host_bridge(). Thanks! -- I speak only for myself. Rafael J. Wysocki, Intel Open Source Technology Center. -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
