Re: [PATCH -next] PCI: Use GFP_ATOMIC in resource_alignment_store()

[Logan Gunthorpe <logang@deltatee.com>]

On 2019-09-02 1:50 a.m., Christoph Hellwig wrote:
> On Sat, Aug 31, 2019 at 12:49:32PM +0000, YueHaibing wrote:
>> When allocating memory, the GFP_KERNEL cannot be used during the
>> spin_lock period. It may cause scheduling when holding spin_lock.
>>
>> Fixes: f13755318675 ("PCI: Move pci_[get|set]_resource_alignment_param() into their callers")
>> Signed-off-by: YueHaibing <yuehaibing@huawei.com>
>> ---
>>  drivers/pci/pci.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>> index 484e35349565..0b5fc6736f3f 100644
>> --- a/drivers/pci/pci.c
>> +++ b/drivers/pci/pci.c
>> @@ -6148,7 +6148,7 @@ static ssize_t resource_alignment_store(struct bus_type *bus,
>>  	spin_lock(&resource_alignment_lock);
>>  
>>  	kfree(resource_alignment_param);
>> -	resource_alignment_param = kstrndup(buf, count, GFP_KERNEL);
>> +	resource_alignment_param = kstrndup(buf, count, GFP_ATOMIC);
>>  
>>  	spin_unlock(&resource_alignment_lock);
> 
> Why not move the allocation outside the lock? Something like this
> seems much more sensible:

Yes, that seems like a good way to do it. Bjorn, can you squash
Christoph's patch or do you want me to resend a new one?

Thanks,

Logan

> 
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 484e35349565..fe205829f676 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -6145,14 +6145,16 @@ static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
>  static ssize_t resource_alignment_store(struct bus_type *bus,
>  					const char *buf, size_t count)
>  {
> -	spin_lock(&resource_alignment_lock);
> +	char *param = kstrndup(buf, count, GFP_KERNEL);
>  
> -	kfree(resource_alignment_param);
> -	resource_alignment_param = kstrndup(buf, count, GFP_KERNEL);
> +	if (!param)
> +		return -ENOMEM;
>  
> +	spin_lock(&resource_alignment_lock);
> +	kfree(resource_alignment_param);
> +	resource_alignment_param = param;
>  	spin_unlock(&resource_alignment_lock);
> -
> -	return resource_alignment_param ? count : -ENOMEM;
> +	return count;
>  }
>  
>  static BUS_ATTR_RW(resource_alignment);
>