Re: [PATCH -next] um: vector: use GFP_ATOMIC under spin lock

[Richard Weinberger <richard@nod.at>]

----- Ursprüngliche Mail -----
> Von: "anton ivanov" <anton.ivanov@cambridgegreys.com>
> An: "Dan Carpenter" <dan.carpenter@oracle.com>, "Wei Yongjun" <weiyongjun1@huawei.com>
> CC: "Song Liu" <songliubraving@fb.com>, "Daniel Borkmann" <daniel@iogearbox.net>, "kernel-janitors"
> <kernel-janitors@vger.kernel.org>, "richard" <richard@nod.at>, "Jeff Dike" <jdike@addtoit.com>, "linux-um"
> <linux-um@lists.infradead.org>, "Alexei Starovoitov" <ast@kernel.org>, "netdev" <netdev@vger.kernel.org>,
> bpf@vger.kernel.org, "Martin KaFai Lau" <kafai@fb.com>
> Gesendet: Donnerstag, 28. November 2019 09:18:30
> Betreff: Re: [PATCH -next] um: vector: use GFP_ATOMIC under spin lock

> On 28/11/2019 08:06, Dan Carpenter wrote:
>> On Thu, Nov 28, 2019 at 02:01:47AM +0000, Wei Yongjun wrote:
>>> A spin lock is taken here so we should use GFP_ATOMIC.
>>>
>>> Fixes: 9807019a62dc ("um: Loadable BPF "Firmware" for vector drivers")
>>> Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
>>> ---
>>>   arch/um/drivers/vector_kern.c | 4 ++--
>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/um/drivers/vector_kern.c b/arch/um/drivers/vector_kern.c
>>> index 92617e16829e..6ff0065a271d 100644
>>> --- a/arch/um/drivers/vector_kern.c
>>> +++ b/arch/um/drivers/vector_kern.c
>>> @@ -1402,7 +1402,7 @@ static int vector_net_load_bpf_flash(struct net_device
>>> *dev,
>>>   		kfree(vp->bpf->filter);
>>>   		vp->bpf->filter = NULL;
>>>   	} else {
>>> -		vp->bpf = kmalloc(sizeof(struct sock_fprog), GFP_KERNEL);
>>> +		vp->bpf = kmalloc(sizeof(struct sock_fprog), GFP_ATOMIC);
>>>   		if (vp->bpf == NULL) {
>>>   			netdev_err(dev, "failed to allocate memory for firmware\n");
>>>   			goto flash_fail;
>>> @@ -1414,7 +1414,7 @@ static int vector_net_load_bpf_flash(struct net_device
>>> *dev,
>>>   	if (request_firmware(&fw, efl->data, &vdevice->pdev.dev))
>>              ^^^^^^^^^^^^^^^^
>> 
>> Is it really possible to call request_firmware() while holding a
>> spin_lock?  I was so sure that read from the disk.
> 
> Works, I tested the patch quite a few times.

It works because of the nature of UML ->no  SMP or PREEMPT.
But better request the firmware before taking the spinlock.
request_firmware() can block.
Same for the kmalloc(), just allocate the buffer before and then assign
the pointer under the lock. That way you don't need GFP_ATOMIC.

Thanks,
//richard