lkml.org 
[lkml]   [2021]   [Feb]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2] staging: gdm724x: Fix DMA from stack
On 21/02/10 12:04PM, Dan Carpenter wrote:
> On Wed, Feb 10, 2021 at 02:28:11PM +0530, Amey Narkhede wrote:
> > On 21/02/10 09:06AM, Greg KH wrote:
> > > On Wed, Feb 10, 2021 at 01:31:34PM +0530, Amey Narkhede wrote:
> > > > Stack allocated buffers cannot be used for DMA
> > > > on all architectures so allocate hci_packet buffer
> > > > using kzalloc().
> > > >
> > > > Signed-off-by: Amey Narkhede <ameynarkhede03@gmail.com>
> > > > ---
> > > > Changes in v2:
> > > > - Fixed build warning
> > > > - Fixed memory leak using kfree
> > > >
> > > > drivers/staging/gdm724x/gdm_usb.c | 9 +++++++--
> > > > 1 file changed, 7 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/drivers/staging/gdm724x/gdm_usb.c b/drivers/staging/gdm724x/gdm_usb.c
> > > > index dc4da66c3..c4a9b90c5 100644
> > > > --- a/drivers/staging/gdm724x/gdm_usb.c
> > > > +++ b/drivers/staging/gdm724x/gdm_usb.c
> > > > @@ -56,11 +56,15 @@ static int gdm_usb_recv(void *priv_dev,
> > > >
> > > > static int request_mac_address(struct lte_udev *udev)
> > > > {
> > > > - u8 buf[16] = {0,};
> > > > - struct hci_packet *hci = (struct hci_packet *)buf;
> > > > + u8 *buf;
> > > > + struct hci_packet *hci;
> > > > struct usb_device *usbdev = udev->usbdev;
> > > > int actual;
> > > > int ret = -1;
> > > > + buf = kzalloc(16, GFP_KERNEL);
> > >
> > > checkpatch did not complain about this?
> > No. checkpatch shows no errors and warnings.
> > >
> > > And why do you need 'buf' anymore now? Why not just allocate hci and
> > > pass that to the request instead? Saves you an extra cast and an extra
> > > pointer.
> > >
> > > thanks,
> > >
> > > greg k-h
> > Thanks. I'll send v3. I assume now we don't need kzalloc anymore as we initialize
> > the hci_packet so kmalloc(sizeof(struct hci_packet),..) will do.
>
> We only initialize the first five bytes, but it also seems as if we only
> use the first five bytes which raises the question of why we are
> allocating 16 bytes.
>
> regards,
> dan carpenter
>
That makes sense. I kept 16 bytes as original implementation allocated
16 bytes on stack.

Thanks,
Amey
[unhandled content-type:application/pgp-signature]
\
 
 \ /
  Last update: 2021-02-10 11:54    [W:0.115 / U:2.520 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site