linux-kernel - Re: [PATCH V2 rdma-next 3/4] RDMA/hns: Add reset process for RoCE in hip08

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 linux-cve-announce 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]

Message-ID: <5B08BCBF.9040001@huawei.com>
Date:   Sat, 26 May 2018 09:47:43 +0800
From:   "Wei Hu (Xavier)" <xavier.huwei@...wei.com>
To:     Jason Gunthorpe <jgg@...pe.ca>
CC:     <dledford@...hat.com>, <linux-rdma@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH V2 rdma-next 3/4] RDMA/hns: Add reset process for RoCE in
 hip08



On 2018/5/25 22:55, Jason Gunthorpe wrote:
> On Fri, May 25, 2018 at 01:54:31PM +0800, Wei Hu (Xavier) wrote:
>>
>> On 2018/5/25 5:31, Jason Gunthorpe wrote:
>>>>  static const struct hnae3_client_ops hns_roce_hw_v2_ops = {
>>>>  	.init_instance = hns_roce_hw_v2_init_instance,
>>>>  	.uninit_instance = hns_roce_hw_v2_uninit_instance,
>>>> +	.reset_notify = hns_roce_hw_v2_reset_notify,
>>>>  };
>>>>  
>>>>  static struct hnae3_client hns_roce_hw_v2_client = {
>>>> diff --git a/drivers/infiniband/hw/hns/hns_roce_main.c b/drivers/infiniband/hw/hns/hns_roce_main.c
>>>> index 1b79a38..ac51372 100644
>>>> +++ b/drivers/infiniband/hw/hns/hns_roce_main.c
>>>> @@ -332,6 +332,9 @@ static struct ib_ucontext *hns_roce_alloc_ucontext(struct ib_device *ib_dev,
>>>>  	struct hns_roce_ib_alloc_ucontext_resp resp = {};
>>>>  	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
>>>>  
>>>> +	if (!hr_dev->active)
>>>> +		return ERR_PTR(-EAGAIN);
>>> This still doesn't make sense, ib_unregister_device already makes sure
>>> that hns_roce_alloc_ucontext isn't running and won't be called before
>>> returning, don't need another flag to do that.
>>>
>>> Since this is the only place the active flag is tested it can just be deleted
>>> entirely.
>> Hi, Jason
>>    
>>     roce reset process:
>>     1. hr_dev->active = false;  //make sure no any process call
>> ibv_open_device.   
>>     2. call ib_dispatch_event() function to report IB_EVENT_DEVICE_FATAL
>> event.
>>     3. msleep(100);   // for some app to free resources
>>     4. call ib_unregister_device().   
>>     5. ...
>>     6. ...
>>
>>     There are 2 steps as above before calling ib_unregister_device(), we
>> evaluate
>> hr_dev->active with false to avoid no any process call
>> ibv_open_device.
> If you think this is the right flow then it is core issue to block new
> opens, not an individual driver issue, send a core patch - eg add a
> 'ib_driver_fatal_error()' call that does the dispatch and call it from
> all the drivers using this IB_EVENT_DEVICE_FATAL
Hi, Jason

It seem to be no difference between calling ib_driver_fatal_error and
calling ib_dispatch_event  directly in manufacturer driver.

void ib_driver_fatal_error(struct ib_device *ib_dev, u8 port_num)
 {
       struct ib_event event;
                
  event.event = IB_EVENT_DEVICE_FATAL;
    event.device = ib_dev;
    event.element.port_num = port_num;
    ib_dispatch_event(&event);
}

    Regards
Wei Hu
> I'm not completely sure this makes sense though, it might be better
> for the core code to force stuff a IB_EVENT_DEVICE_FATAL to contexts
> that open after the fatal event..
>
> Jason
>
> .
>