Re: [PATCH 1/2] pci: host: pci-hyperv: Replace GFP_ATOMIC with GFP_KERNEL in hv_pci_onchannelcallback

[Jia-Ju Bai]

On 2018/3/19 16:38, Dan Carpenter wrote:
> On Sun, Mar 18, 2018 at 10:53:02PM +0800, Jia-Ju Bai wrote:
> > hv_pci_onchannelcallback() is not called in atomic context.
> >
> > The call chain ending up at hv_pci_onchannelcallback() is:
> > [1] hv_pci_onchannelcallback() <- hv_pci_probe()
> > hv_pci_probe() is only set as ".probe" in hv_driver
> > structure "hv_pci_drv".
> Your static analysis tool is faulty and apparently so is Smatch.
>
> $ smdb function_ptrs hv_pci_onchannelcallback
>
> Says it can't find a caller.  When I look for function pointers I get:
>
> $ smdb function_ptr hv_pci_onchannelcallback
> hv_pci_onchannelcallback =  'hv_pci_onchannelcallback' , 'vmbus_open param 5' , '(struct vmbus_channel)->onchannel_callback' , '__read_once_size param 0'
>
> Anyway the call tree is:
>
> vmbus_chan_sched() <-- takes rcu_read_lock();
> -> vmbus_channel_isr()
>     -> channel->onchannel_callback() -> which is hv_pci_onchannelcallback(

Thanks for your reply :)
I admit my tool produces a false positive for this code...
Sorry for my incorrect patch.

Anyway, I find that function pointers are quite hard to analyze in the 
Linux kernel code, because their usages are often flexible.
Have you found a good way to handle function pointers? Or can you 
recommend some good tools to handle them?


Best wishes,
Jia-Ju Bai