On Wednesday, February 20, 2013 01:23:34 PM Toshi Kani wrote:
> On Wed, 2013-02-20 at 14:23 +0100, Rafael J. Wysocki wrote:
> > On Tuesday, February 19, 2013 03:43:08 PM Yasuaki Ishimatsu wrote:
> :
> > >
> > > > - status = acpi_os_hotplug_execute(acpi_bus_hot_remove_device, ej_event);
> > > > - if (ACPI_FAILURE(status)) {
> > > > - put_device(&acpi_device->dev);
> > > > - kfree(ej_event);
> > > > + acpi_evaluate_hotplug_ost(acpi_device->handle, ost_source,
> > > > + ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
> > > > + get_device(&acpi_device->dev);
> > > > + ret = acpi_scan_hot_remove(acpi_device);
> > >
> > > Why don't you use acpi_os_hotplug_execute()? Do you have some reason?
> >
> > Yes, I do. acpi_eject_store() is run in a separate thread anyway (started by
> > user space), so there's no need to use the workqueue for delayed execution here
> > and we are under acpi_scan_lock anyway, so there shouldn't be any concurrency
> > issues.
>
> Well, there is an issue... I just tested your patchset and hit the
> following hang when I tried to delete a container through its sysfs
> eject. This thread got stuck in trying to delete the sysfs eject file
> of the container. I believe this is because the shell is still opening
> this sysfs eject file.
You're right.
> PID: 1518 TASK: ffff88005f09c950 CPU: 1 COMMAND: "bash"
> #0 [ffff88003392baf8] __schedule at ffffffff8151ba75
> #1 [ffff88003392bb70] schedule at ffffffff8151bdc7
> #2 [ffff88003392bb80] schedule_timeout at ffffffff8151aa55
> #3 [ffff88003392bc00] wait_for_common at ffffffff8151bc43
> #4 [ffff88003392bc70] wait_for_completion at ffffffff8151bd60
> #5 [ffff88003392bc80] sysfs_addrm_finish at ffffffff811984ad
> #6 [ffff88003392bcd0] sysfs_hash_and_remove at ffffffff81196deb
> #7 [ffff88003392bd10] sysfs_remove_file at ffffffff81197051
> #8 [ffff88003392bd40] device_remove_file at ffffffff81332950
> #9 [ffff88003392bd50] acpi_device_unregister at ffffffff812a0556
> #10 [ffff88003392bd80] acpi_bus_remove at ffffffff812a0658
> #11 [ffff88003392bda0] acpi_bus_trim at ffffffff812a090e
> #12 [ffff88003392bdd0] acpi_scan_hot_remove at ffffffff812a09c9
> #13 [ffff88003392be30] acpi_eject_store at ffffffff812a0b45
> #14 [ffff88003392be70] dev_attr_store at ffffffff81332038
> #15 [ffff88003392be80] sysfs_write_file at ffffffff81197212
> #16 [ffff88003392bee0] vfs_write at ffffffff8113a3cb
> #17 [ffff88003392bf20] sys_write at ffffffff8113a5fd
> #18 [ffff88003392bf80] system_call_fastpath at ffffffff81523759
> RIP: 00000033a16e4950 RSP: 00007fff4a5f5368 RFLAGS: 00000206
> RAX: 0000000000000001 RBX: ffffffff81523759 RCX: ffffffffffffffff
> RDX: 0000000000000002 RSI: 00007f2f8a3d8000 RDI: 0000000000000001
> RBP: 00007f2f8a3d8000 R8: 000000000000000a R9: 00007f2f8a3c4740
> R10: 00007f2f8a3c4740 R11: 0000000000000246 R12: 00000033a19b1260
> R13: 0000000000000002 R14: ffff880000000000 R15: ffff88003395d680
> ORIG_RAX: 0000000000000001 CS: 0033 SS: 002b
Well, admittedly, I didn't think about this situation.
Since the eject attribute is under the device we're going to remove, the
removal has to be done from a different thread (e.g. workqueue).
OK, I'll fix up the patch.
Thanks,
Rafael
--
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
