netdev - Re: [PATCH] vDPA/ifcvf: assign nr_vring to the MSI vector of config

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 13 Apr 2022 17:16:06 +0800
From:   "Zhu, Lingshan" <lingshan.zhu@...el.com>
To:     Jason Wang <jasowang@...hat.com>, mst@...hat.com
Cc:     virtualization@...ts.linux-foundation.org, netdev@...r.kernel.org,
        Dan Carpenter <dan.carpenter@...cle.com>
Subject: Re: [PATCH] vDPA/ifcvf: assign nr_vring to the MSI vector of
 config_intr by default



On 4/13/2022 4:14 PM, Jason Wang wrote:
>
> 在 2022/4/8 下午8:10, Zhu Lingshan 写道:
>> This commit assign struct ifcvf_hw.nr_vring to the MSIX vector of the
>> config interrupt by default in ifcvf_request_config_irq().
>> ifcvf_hw.nr_vring is the most likely and the ideal case for
>> the device config interrupt handling, means every virtqueue has
>> an individual MSIX vector(0 ~ nr_vring - 1), and the config interrupt 
>> has
>> its own MSIX vector(number nr_vring).
>>
>> This change can also make GCC W = 2 happy, silence the
>> "uninitialized" warning.
>>
>> Signed-off-by: Zhu Lingshan <lingshan.zhu@...el.com>
>> Reviewed-by: Dan Carpenter <dan.carpenter@...cle.com>
>> ---
>>   drivers/vdpa/ifcvf/ifcvf_main.c | 8 ++++----
>>   1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/vdpa/ifcvf/ifcvf_main.c 
>> b/drivers/vdpa/ifcvf/ifcvf_main.c
>> index 4366320fb68d..b500fb941dab 100644
>> --- a/drivers/vdpa/ifcvf/ifcvf_main.c
>> +++ b/drivers/vdpa/ifcvf/ifcvf_main.c
>> @@ -290,13 +290,13 @@ static int ifcvf_request_config_irq(struct 
>> ifcvf_adapter *adapter)
>>       struct ifcvf_hw *vf = &adapter->vf;
>>       int config_vector, ret;
>>   +    /* vector 0 ~ vf->nr_vring for vqs, num vf->nr_vring vector 
>> for config interrupt */
>
>
> The comment is right before this patch, but probably wrong for 
> MSIX_VECTOR_DEV_SHARED.
This comment is for the case when every vq and config interrupt has its 
own vector, how
about a better comment "The ideal the default case, vector 0 ~ 
vf->nr_vring for vqs, num vf->nr_vring vector for config interrupt"
>
>
>> +    config_vector = vf->nr_vring;
>> +
>> +    /* re-use the vqs vector */
>>       if (vf->msix_vector_status == MSIX_VECTOR_DEV_SHARED)
>>           return 0;
>>   -    if (vf->msix_vector_status == MSIX_VECTOR_PER_VQ_AND_CONFIG)
>> -        /* vector 0 ~ vf->nr_vring for vqs, num vf->nr_vring vector 
>> for config interrupt */
>> -        config_vector = vf->nr_vring;
>> -
>>       if (vf->msix_vector_status == MSIX_VECTOR_SHARED_VQ_AND_CONFIG)
>>           /* vector 0 for vqs and 1 for config interrupt */
>>           config_vector = 1;
>
>
> Actually, I prefer to use if ... else ... here.
IMHO, if else may lead to mistakes.

The code:
         /* The ideal the default case, vector 0 ~ vf->nr_vring for vqs, 
num vf->nr_vring vector for config interrupt */
         config_vector = vf->nr_vring;

         /* re-use the vqs vector */
         if (vf->msix_vector_status == MSIX_VECTOR_DEV_SHARED)
                 return 0;

         if (vf->msix_vector_status == MSIX_VECTOR_SHARED_VQ_AND_CONFIG)
                 /* vector 0 for vqs and 1 for config interrupt */
                 config_vector = 1;


here by default config_vector = vf->nr_vring;
If msix_vector_status == MSIX_VECTOR_DEV_SHARED, it will reuse the dev 
shared vector, means using the vector(value 0) for data-vqs.
If msix_vector_status == MSIX_VECTOR_SHARED_VQ_AND_CONFIG, it will use 
vector=1(vector 0 for data-vqs).

If we use if...else, it will be:

         /* re-use the vqs vector */
         if (vf->msix_vector_status == MSIX_VECTOR_DEV_SHARED)
                 return 0;
         else
                 config_vector = 1;

This looks like config_vector can only be 0(re-used vector for the 
data-vqs, which is 0) or 1. It shadows the ideal and default case
config_vector = vf->nr_vring

Thanks,
Zhu Lingshan

>
> Thanks
>
>