[dpdk-dev] [PATCH] net/szedata2: fix incorrect device memory access

Ferruh Yigit ferruh.yigit at intel.com
Tue Jan 24 17:03:16 CET 2017


On 1/24/2017 3:55 PM, Matej Vido wrote:
> On 24.01.2017 16:11, Ferruh Yigit wrote:
>> On 1/24/2017 2:02 PM, Matej Vido wrote:
>>> On 24.01.2017 12:58, Ferruh Yigit wrote:
>>>> On 1/24/2017 10:49 AM, Matej Vido wrote:
>>>>> Fixes: 8acba705b119 ("net/szedata2: localize handling of PCI resources")
>>>>>
>>>>> Signed-off-by: Matej Vido <vido at cesnet.cz>
>>>> Unrelated from this patch, in maintainers file, you have your other mail
>>>> address: "Matej Vido <matejvido at gmail.com>", do you want to update it?
>>> Hi Ferruh,
>>>
>>> yes, I will send the patch.
>>>>> ---
>>>>>    drivers/net/szedata2/rte_eth_szedata2.h | 2 +-
>>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/net/szedata2/rte_eth_szedata2.h b/drivers/net/szedata2/rte_eth_szedata2.h
>>>>> index b58adb6..afe8a38 100644
>>>>> --- a/drivers/net/szedata2/rte_eth_szedata2.h
>>>>> +++ b/drivers/net/szedata2/rte_eth_szedata2.h
>>>>> @@ -192,7 +192,7 @@ struct szedata {
>>>>>    }
>>>>>    
>>>>>    #define SZEDATA2_PCI_RESOURCE_PTR(rsc, offset, type) \
>>>>> -	((type)((uint8_t *)(rsc)->addr) + (offset))
>>>>> +	((type)(((uint8_t *)(rsc)->addr) + (offset)))
>>>> Although output will be same, (in all uses, type is a pointer),
>> Of course won't be same, please forget about it J, I am confused.
>>
>> So these two differs a lot, taking into account that offset numbers used
>> are big numbers (0x8000..), it should be accessing very unrelated addresses.
>>
>> So how this was working before?
> 
> The macro was fine before the patch [1]. It hasn't been working since 
> the acceptance of that patch, but I didn't manage to test the 
> functionality until now.

I see, thanks for clarification, seems broken relatively new, and thanks
for fixing.

> 
> [1] http://dpdk.org/ml/archives/dev/2016-December/053241.html
> 
> Regards,
> Matej
> 
>>
>>> this
>>>> seems the intention, so:
>>>>
>>>> Reviewed-by: Ferruh Yigit <ferruh.yigit at intel.com>
>>>>
>>>> btw, following will do same, right, not sure if it is better:
>>>> ((type)(rsc)->addr + (offset))
>>> This is also wrong. The intention of the macro is to add an offset to
>>> the base address and typecast the result.
>>>
>>> Regards,
>>> Matej
>>>>>    
>>>>>    enum szedata2_link_speed {
>>>>>    	SZEDATA2_LINK_SPEED_DEFAULT = 0,
>>>>>
> 



More information about the dev mailing list