Re: [PATCH] vboxguest: add missing devm_free_irq

[Hans de Goede]

Hi,

On 6/10/22 17:00, Pascal Terjan wrote:
> On Fri, 10 Jun 2022 at 13:50, Greg Kroah-Hartman
> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>>
>> On Thu, Jun 09, 2022 at 02:40:57PM +0100, Pascal Terjan wrote:
>>> This fixes the following warning when unloading the module:
>>>
>>> [249348.837181] remove_proc_entry: removing non-empty directory 'irq/20', leaking at least 'vboxguest'
>>> [249348.837219] WARNING: CPU: 0 PID: 6708 at fs/proc/generic.c:715 remove_proc_entry+0x119/0x140
>>>
>>> [249348.837379] Call Trace:
>>> [249348.837385]  unregister_irq_proc+0xbd/0xe0
>>> [249348.837392]  free_desc+0x23/0x60
>>> [249348.837396]  irq_free_descs+0x4a/0x70
>>> [249348.837401]  irq_domain_free_irqs+0x160/0x1a0
>>> [249348.837452]  mp_unmap_irq+0x5c/0x60
>>> [249348.837458]  acpi_unregister_gsi_ioapic+0x29/0x40
>>> [249348.837463]  acpi_unregister_gsi+0x17/0x30
>>> [249348.837467]  acpi_pci_irq_disable+0xbf/0xe0
>>> [249348.837473]  pcibios_disable_device+0x20/0x30
>>> [249348.837478]  pci_disable_device+0xef/0x120
>>> [249348.837482]  vbg_pci_remove+0x6c/0x70 [vboxguest]
>>>
>>> Signed-off-by: Pascal Terjan <pterjan@xxxxxxxxxx>
>>> ---
>>>  drivers/virt/vboxguest/vboxguest_linux.c | 1 +
>>>  1 file changed, 1 insertion(+)
>>>
>>> diff --git a/drivers/virt/vboxguest/vboxguest_linux.c b/drivers/virt/vboxguest/vboxguest_linux.c
>>> index 6e8c0f1c1056..faa4bc9f625c 100644
>>> --- a/drivers/virt/vboxguest/vboxguest_linux.c
>>> +++ b/drivers/virt/vboxguest/vboxguest_linux.c
>>> @@ -423,6 +423,7 @@ static void vbg_pci_remove(struct pci_dev *pci)
>>>       vbg_gdev = NULL;
>>>       mutex_unlock(&vbg_gdev_mutex);
>>>
>>> +     devm_free_irq(gdev->dev, pci->irq, gdev);
>>
>> The whope point of using devm_* calls is so you don't have to do stuff
>> like this.  Perhaps this should not be using devm_() for the irq at all?
> 
> My initial assumption was that some sort of dependency was missing
> somewhere to ensure this gets freed first, but I failed to find any
> documentation on how this is supposed to work, so I went with a fix
> that would work.
> 
> But you are obviously right, if we manually free it in the normal path
> then using devm_{request,free}_irq sounds like just overhead without
> benefits.

Right, please also move the irq-request over to be non devm and
also don't forgot that the error-exit path from the probe()
function also needs to free the irq before disabling the pci-dev.

Regards,

Hans