There are sleep in atomic context bugs when dm_fsync_timer_callback is
executing. The root cause is that the memory allocation functions with
GFP_KERNEL parameter are called in dm_fsync_timer_callback which is a
timer handler. The call paths that could trigger bugs are shown below:
(interrupt context)
dm_fsync_timer_callback
write_nic_byte
kzalloc(sizeof(data), GFP_KERNEL); //may sleep
write_nic_dword
kzalloc(sizeof(data), GFP_KERNEL); //may sleep
This patch changes allocation mode from GFP_KERNEL to GFP_ATOMIC
in order to prevent atomic context sleeping. The GFP_ATOMIC flag
makes memory allocation operation could be used in atomic context.
Signed-off-by: Duoming Zhou <[email protected]>
---
drivers/staging/rtl8192u/r8192U_core.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/rtl8192u/r8192U_core.c b/drivers/staging/rtl8192u/r8192U_core.c
index ce807c9d421..679c362baad 100644
--- a/drivers/staging/rtl8192u/r8192U_core.c
+++ b/drivers/staging/rtl8192u/r8192U_core.c
@@ -267,7 +267,7 @@ int write_nic_byte(struct net_device *dev, int indx, u8 data)
struct r8192_priv *priv = (struct r8192_priv *)ieee80211_priv(dev);
struct usb_device *udev = priv->udev;
- u8 *usbdata = kzalloc(sizeof(data), GFP_KERNEL);
+ u8 *usbdata = kzalloc(sizeof(data), GFP_ATOMIC);
if (!usbdata)
return -ENOMEM;
@@ -319,7 +319,7 @@ int write_nic_dword(struct net_device *dev, int indx, u32 data)
struct r8192_priv *priv = (struct r8192_priv *)ieee80211_priv(dev);
struct usb_device *udev = priv->udev;
- u32 *usbdata = kzalloc(sizeof(data), GFP_KERNEL);
+ u32 *usbdata = kzalloc(sizeof(data), GFP_ATOMIC);
if (!usbdata)
return -ENOMEM;
--
2.17.1
On Fri, May 20, 2022 at 02:15:41PM +0800, Duoming Zhou wrote:
> There are sleep in atomic context bugs when dm_fsync_timer_callback is
> executing. The root cause is that the memory allocation functions with
> GFP_KERNEL parameter are called in dm_fsync_timer_callback which is a
> timer handler. The call paths that could trigger bugs are shown below:
>
> (interrupt context)
> dm_fsync_timer_callback
> write_nic_byte
> kzalloc(sizeof(data), GFP_KERNEL); //may sleep
> write_nic_dword
> kzalloc(sizeof(data), GFP_KERNEL); //may sleep
>
> This patch changes allocation mode from GFP_KERNEL to GFP_ATOMIC
> in order to prevent atomic context sleeping. The GFP_ATOMIC flag
> makes memory allocation operation could be used in atomic context.
>
> Signed-off-by: Duoming Zhou <[email protected]>
What commit id does this fix?
And how did you find this issue? Did you run the code to verify it
still works properly?
thanks,
greg k-h
Hello,
On Mon, 6 Jun 2022 07:43:11 +0200 greg k-h wrote:
> On Fri, May 20, 2022 at 02:15:41PM +0800, Duoming Zhou wrote:
> > There are sleep in atomic context bugs when dm_fsync_timer_callback is
> > executing. The root cause is that the memory allocation functions with
> > GFP_KERNEL parameter are called in dm_fsync_timer_callback which is a
> > timer handler. The call paths that could trigger bugs are shown below:
> >
> > (interrupt context)
> > dm_fsync_timer_callback
> > write_nic_byte
> > kzalloc(sizeof(data), GFP_KERNEL); //may sleep
> > write_nic_dword
> > kzalloc(sizeof(data), GFP_KERNEL); //may sleep
> >
> > This patch changes allocation mode from GFP_KERNEL to GFP_ATOMIC
> > in order to prevent atomic context sleeping. The GFP_ATOMIC flag
> > makes memory allocation operation could be used in atomic context.
> >
> > Signed-off-by: Duoming Zhou <[email protected]>
>
> What commit id does this fix?
Thanks for your time and reply!
The commit id this patch fix is 8fc8598e61f6 ("Staging: Added Realtek rtl8192u driver to staging").
> And how did you find this issue? Did you run the code to verify it
> still works properly?
I find this issue by writing codeql query. I am trying to use usb raw-gadget to simulate
rtl8192u card in order to test this code.
What`s more, I found the usb_control_msg() with GFP_NOIO parameter in write_nic_byte()
and write_nic_dword() may also sleep. So I think use the delayed queue to replace timer
is better.
Best regards,
Duoming Zhou
On Mon, Jun 06, 2022 at 03:11:39PM +0800, [email protected] wrote:
> Hello,
>
> On Mon, 6 Jun 2022 07:43:11 +0200 greg k-h wrote:
>
> > On Fri, May 20, 2022 at 02:15:41PM +0800, Duoming Zhou wrote:
> > > There are sleep in atomic context bugs when dm_fsync_timer_callback is
> > > executing. The root cause is that the memory allocation functions with
> > > GFP_KERNEL parameter are called in dm_fsync_timer_callback which is a
> > > timer handler. The call paths that could trigger bugs are shown below:
> > >
> > > (interrupt context)
> > > dm_fsync_timer_callback
> > > write_nic_byte
> > > kzalloc(sizeof(data), GFP_KERNEL); //may sleep
> > > write_nic_dword
> > > kzalloc(sizeof(data), GFP_KERNEL); //may sleep
> > >
> > > This patch changes allocation mode from GFP_KERNEL to GFP_ATOMIC
> > > in order to prevent atomic context sleeping. The GFP_ATOMIC flag
> > > makes memory allocation operation could be used in atomic context.
> > >
> > > Signed-off-by: Duoming Zhou <[email protected]>
> >
> > What commit id does this fix?
>
> Thanks for your time and reply!
> The commit id this patch fix is 8fc8598e61f6 ("Staging: Added Realtek rtl8192u driver to staging").
>
> > And how did you find this issue? Did you run the code to verify it
> > still works properly?
>
> I find this issue by writing codeql query. I am trying to use usb raw-gadget to simulate
> rtl8192u card in order to test this code.
>
> What`s more, I found the usb_control_msg() with GFP_NOIO parameter in write_nic_byte()
> and write_nic_dword() may also sleep. So I think use the delayed queue to replace timer
> is better.
Please explain all of this in the changelog text when you resend this.
thanks,
greg k-h