Re: [PATCH] HID: usbhid: Use GFP_KERNEL instead of GFP_ATOMIC when applicable

[Christophe JAILLET]

Hi, (and sorry if you receive this email twice. I've used a web mail 
which sends HTML by default and it was rejected by ML)

Le 01/08/2019 Ã 12:06, walter harms a ÃcritÂ:
> Am 01.08.2019 09:47, schrieb Christophe JAILLET:
> > There is no need to use GFP_ATOMIC when calling 'usb_alloc_coherent()'
> > here. These calls are done from probe functions and using GFP_KERNEL should
> > be safe.
> > The memory itself is used within some interrupts, but it is not a
> > problem, once it has been allocated.
> >
> > Signed-off-by: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
> > ---
> >   drivers/hid/usbhid/usbkbd.c   | 4 ++--
> >   drivers/hid/usbhid/usbmouse.c | 2 +-
> >   2 files changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/hid/usbhid/usbkbd.c b/drivers/hid/usbhid/usbkbd.c
> > index d5b7a696a68c..63e8ef8beb45 100644
> > --- a/drivers/hid/usbhid/usbkbd.c
> > +++ b/drivers/hid/usbhid/usbkbd.c
> > @@ -239,11 +239,11 @@ static int usb_kbd_alloc_mem(struct usb_device *dev, struct usb_kbd *kbd)
> >   		return -1;
> >   	if (!(kbd->led = usb_alloc_urb(0, GFP_KERNEL)))
> >   		return -1;
> > -	if (!(kbd->new = usb_alloc_coherent(dev, 8, GFP_ATOMIC, &kbd->new_dma)))
> > +	if (!(kbd->new = usb_alloc_coherent(dev, 8, GFP_KERNEL, &kbd->new_dma)))
> >   		return -1;
> >   	if (!(kbd->cr = kmalloc(sizeof(struct usb_ctrlrequest), GFP_KERNEL)))
> >   		return -1;
> > -	if (!(kbd->leds = usb_alloc_coherent(dev, 1, GFP_ATOMIC, &kbd->leds_dma)))
> > +	if (!(kbd->leds = usb_alloc_coherent(dev, 1, GFP_KERNEL, &kbd->leds_dma)))
> >   		return -1;
> >   
>
> the kernel style is usually:
>   kbd->new = usb_alloc_coherent(dev, 8, GFP_ATOMIC, &kbd->new_dma);
>   if (!kbd->new)
> 	return -1;

Searching with coccinelle with:
*   x = usb_alloc_coherent(..., <+... GFP_KERNEL ...+>, ...);
finds 67 files,

whereas:
*   x = usb_alloc_coherent(..., <+... GFP_ATOMIC ...+>, ...);
only finds 11 files.

> in usbmouse.c this is done, any reason for the change here ?

No real reason in fact, just to be consistent with surrounding code.

Unless some allocations are done within a spin_lock/spin_unlock, using 
both GFP_KERNEL and GFP_ATOMIC in the same function looks spurious to me.
Either there is a bug (GFP_KERNEL should be GFP_ATOMIC), or a useless 
constraint is given to the memory allocator.

CJ

> re,
>   wh
>
> >   	return 0;
> > diff --git a/drivers/hid/usbhid/usbmouse.c b/drivers/hid/usbhid/usbmouse.c
> > index 073127e65ac1..c89332017d5d 100644
> > --- a/drivers/hid/usbhid/usbmouse.c
> > +++ b/drivers/hid/usbhid/usbmouse.c
> > @@ -130,7 +130,7 @@ static int usb_mouse_probe(struct usb_interface *intf, const struct usb_device_i
> >   	if (!mouse || !input_dev)
> >   		goto fail1;
> >   
> > -	mouse->data = usb_alloc_coherent(dev, 8, GFP_ATOMIC, &mouse->data_dma);
> > +	mouse->data = usb_alloc_coherent(dev, 8, GFP_KERNEL, &mouse->data_dma);
> >   	if (!mouse->data)
> >   		goto fail1;
> >