ACPI boot memory leaks in 2.6.32-rc1/2

ACPI boot memory leaks in 2.6.32-rc1/2

[Zdenek Kabelac]

Hi

My qemu guest seems to be reporting now 46 memleaks  ACPI related.

My kernel   x86_64 - commit: 17d857be649a21ca90008c6dc425d849fa83db5c
(2.6.32-rc1)


unreferenced object 0xffff88001f8095d0 (size 8):
  comm "swapper", pid 1, jiffies 4294877484
  hex dump (first 8 bytes):
    5c 00 80 1f 00 88 ff ff                          \.......
  backtrace:
    [<ffffffff8140a0c6>] kmemleak_alloc+0x26/0x60
    [<ffffffff81128253>] __kmalloc+0x153/0x250
    [<ffffffff812bfff5>] acpi_os_allocate+0x25/0x27
    [<ffffffff812c0037>] acpi_ut_initialize_buffer+0x40/0x71
    [<ffffffff812b739b>] acpi_ns_handle_to_pathname+0x3f/0xa7
    [<ffffffff812b84c4>] acpi_get_name+0x45/0xb8
    [<ffffffff8129b854>] acpi_bus_scan+0x3e/0xa0
    [<ffffffff81878acd>] acpi_scan_init+0x45/0xba
    [<ffffffff81878906>] acpi_init+0x21f/0x26a
    [<ffffffff8100904c>] do_one_initcall+0x3c/0x1d0
    [<ffffffff818574e6>] kernel_init+0x150/0x1a6
    [<ffffffff8100d21a>] child_rip+0xa/0x20
    [<ffffffffffffffff>] 0xffffffffffffffff
unreferenced object 0xffff88001fa13380 (size 128):
  comm "swapper", pid 1, jiffies 4294877495
  hex dump (first 32 bytes):
    7d 00 00 00 50 43 49 30 06 00 00 00 00 0f 01 ff  }...PCI0........
    ff ff ff ff ff ff ff ff ff ff ff ff 00 00 00 00  ................
  backtrace:
    [<ffffffff8140a0c6>] kmemleak_alloc+0x26/0x60
    [<ffffffff81128253>] __kmalloc+0x153/0x250
    [<ffffffff812b829f>] acpi_get_object_info+0x161/0x341
    [<ffffffff8129afcc>] acpi_add_single_object+0x2b8/0x9ca
    [<ffffffff8129b7be>] acpi_bus_check_add+0xe0/0x138
    [<ffffffff812b98b4>] acpi_ns_walk_namespace+0xb2/0x127
    [<ffffffff812b6524>] acpi_walk_namespace+0x78/0xb2
    [<ffffffff8129b89d>] acpi_bus_scan+0x87/0xa0
    [<ffffffff81878acd>] acpi_scan_init+0x45/0xba
    [<ffffffff81878906>] acpi_init+0x21f/0x26a
    [<ffffffff8100904c>] do_one_initcall+0x3c/0x1d0
    [<ffffffff818574e6>] kernel_init+0x150/0x1a6
    [<ffffffff8100d21a>] child_rip+0xa/0x20
    [<ffffffffffffffff>] 0xffffffffffffffff
....

All seems to be acpi_bus_scan() related.

Zdenek

Re: ACPI boot memory leaks in 2.6.32-rc1/2

[Bjorn Helgaas]

On Tuesday 29 September 2009 03:18:53 am Zdenek Kabelac wrote:
> My qemu guest seems to be reporting now 46 memleaks  ACPI related.

Thanks a lot for the report!

Can you verify that the patch below fixes these leaks?


commit fff7774a805e392ce258f0342e48cbe682ef2b9b
Author: Bjorn Helgaas <bjorn.helgaas@hp.com>
Date:   Tue Sep 29 10:32:19 2009 -0600

    ACPI: fix bus scanning memory leaks
    
    Free an acpi_get_object_info() buffer when we're finished.  Skip the
    acpi_get_name() altogether -- it was only used for a printk that was
    really just for debug anyway.
    
    Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
    Reported-by: Zdenek Kabelac <zdenek.kabelac@gmail.com>

diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
index 468921b..14a7481 100644
--- a/drivers/acpi/scan.c
+++ b/drivers/acpi/scan.c
@@ -1052,6 +1052,8 @@ static void acpi_device_set_id(struct acpi_device *device)
 			device->flags.bus_address = 1;
 		}
 
+		kfree(info);
+
 		/*
 		 * Some devices don't reliably have _HIDs & _CIDs, so add
 		 * synthetic HIDs to make sure drivers can find them.
@@ -1325,13 +1327,8 @@ static int acpi_bus_scan(acpi_handle handle, struct acpi_bus_ops *ops,
 			 struct acpi_device **child)
 {
 	acpi_status status;
-	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 	void *device = NULL;
 
-	acpi_get_name(handle, ACPI_FULL_PATHNAME, &buffer);
-	printk(KERN_INFO PREFIX "Enumerating devices from [%s]\n",
-	       (char *) buffer.pointer);
-
 	status = acpi_bus_check_add(handle, 0, ops, &device);
 	if (ACPI_SUCCESS(status))
 		acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,

Re: ACPI boot memory leaks in 2.6.32-rc1/2

[Zdenek Kabelac]

2009/9/29 Bjorn Helgaas <bjorn.helgaas@hp.com>:
> On Tuesday 29 September 2009 03:18:53 am Zdenek Kabelac wrote:
>> My qemu guest seems to be reporting now 46 memleaks  ACPI related.
>
> Thanks a lot for the report!
>
> Can you verify that the patch below fixes these leaks?
>

Yes - leaks are fixed with this patch

>
> commit fff7774a805e392ce258f0342e48cbe682ef2b9b
> Author: Bjorn Helgaas <bjorn.helgaas@hp.com>
> Date:   Tue Sep 29 10:32:19 2009 -0600
>
>    ACPI: fix bus scanning memory leaks
>
>    Free an acpi_get_object_info() buffer when we're finished.  Skip the
>    acpi_get_name() altogether -- it was only used for a printk that was
>    really just for debug anyway.
>
>    Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
>    Reported-by: Zdenek Kabelac <zdenek.kabelac@gmail.com>


Reported-and-tested-by: Zdenek Kabelac <zdenek.kabelac@gmail.com>

>
> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
> index 468921b..14a7481 100644
> --- a/drivers/acpi/scan.c
> +++ b/drivers/acpi/scan.c
> @@ -1052,6 +1052,8 @@ static void acpi_device_set_id(struct acpi_device *device)
>                        device->flags.bus_address = 1;
>                }
>
> +               kfree(info);
> +
>                /*
>                 * Some devices don't reliably have _HIDs & _CIDs, so add
>                 * synthetic HIDs to make sure drivers can find them.
> @@ -1325,13 +1327,8 @@ static int acpi_bus_scan(acpi_handle handle, struct acpi_bus_ops *ops,
>                         struct acpi_device **child)
>  {
>        acpi_status status;
> -       struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
>        void *device = NULL;
>
> -       acpi_get_name(handle, ACPI_FULL_PATHNAME, &buffer);
> -       printk(KERN_INFO PREFIX "Enumerating devices from [%s]\n",
> -              (char *) buffer.pointer);
> -
>        status = acpi_bus_check_add(handle, 0, ops, &device);
>        if (ACPI_SUCCESS(status))
>                acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
>

Re: ACPI boot memory leaks in 2.6.32-rc1/2

[Len Brown]

applied.

Zdenek, Bjorn,
Thanks for the quick testing & fix!

Len Brown, Intel Open Source Technology Center