[dpdk-dev] [PATCH v4 06/24] ethdev: enable hotplug on multi-process

[Burakov, Anatoly]

On 26-Jun-18 1:19 PM, Zhang, Qi Z wrote:
> 
> 
>> -----Original Message-----
>> From: Burakov, Anatoly
>> Sent: Tuesday, June 26, 2018 8:09 PM
>> To: Zhang, Qi Z <qi.z.zhang at intel.com>; thomas at monjalon.net
>> Cc: Ananyev, Konstantin <konstantin.ananyev at intel.com>; dev at dpdk.org;
>> Richardson, Bruce <bruce.richardson at intel.com>; Yigit, Ferruh
>> <ferruh.yigit at intel.com>; Shelton, Benjamin H
>> <benjamin.h.shelton at intel.com>; Vangati, Narender
>> <narender.vangati at intel.com>
>> Subject: Re: [PATCH v4 06/24] ethdev: enable hotplug on multi-process
>>
>> On 26-Jun-18 8:08 AM, Qi Zhang wrote:
>>> We are going to introduce the solution to handle different hotplug
>>> cases in multi-process situation, it include below scenario:
>>>
>>> 1. Attach a share device from primary
>>> 2. Detach a share device from primary
>>> 3. Attach a share device from secondary 4. Detach a share device from
>>> secondary 5. Attach a private device from secondary 6. Detach a
>>> private device from secondary 7. Detach a share device from secondary
>>> privately 8. Attach a share device from secondary privately
>>>
>>> In primary-secondary process model, we assume device is shared by default.
>>> that means attach or detach a device on any process will broadcast to
>>> all other processes through mp channel then device information will be
>>> synchronized on all processes.
>>>
>>> Any failure during attaching process will cause inconsistent status
>>> between processes, so proper rollback action should be considered.
>>> Also it is not safe to detach a share device when other process still
>>> use it, so a handshake mechanism is introduced.
>>>
>>> This patch covers the implementation of case 1,2,5,6,7,8.
>>> Case 3,4 will be implemented on separate patch as well as handshake
>>> mechanism.
>>>
>>> Scenario for Case 1, 2:
>>>
>>> attach device
>>> a) primary attach the new device if failed goto h).
>>> b) primary send attach sync request to all secondary.
>>> c) secondary receive request and attach device and send reply.
>>> d) primary check the reply if all success go to i).
>>> e) primary send attach rollback sync request to all secondary.
>>> f) secondary receive the request and detach device and send reply.
>>> g) primary receive the reply and detach device as rollback action.
>>> h) attach fail
>>> i) attach success
>>>
>>> detach device
>>> a) primary perform pre-detach check, if device is locked, goto i).
>>> b) primary send pre-detach sync request to all secondary.
>>> c) secondary perform pre-detach check and send reply.
>>> d) primary check the reply if any fail goto i).
>>> e) primary send detach sync request to all secondary
>>> f) secondary detach the device and send reply (assume no fail)
>>> g) primary detach the device.
>>> h) detach success
>>> i) detach failed
>>>
>>> Case 5, 6:
>>> Secondary process can attach private device which only visible to
>>> itself, in this case no IPC is involved, primary process is not
>>> allowed to have private device so far.
>>>
>>> Case 7, 8:
>>> Secondary process can also temporally to detach a share device "privately"
>>> then attach it back later, this action also not impact other processes.
>>>
>>> APIs changes:
>>>
>>> rte_eth_dev_attach and rte_eth_dev_attach are extended to support
>>> share device attach/detach in primary-secondary process model, it will
>>> be called in case 1,2,3,4.
>>>
>>> New API rte_eth_dev_attach_private and rte_eth_dev_detach_private are
>>> introduced to cover case 5,6,7,8, this API can only be invoked in
>>> secondary process.
>>>
>>> Signed-off-by: Qi Zhang <qi.z.zhang at intel.com>
>>> ---
>>
>> <snip>
>>
>>> +static int
>>> +handle_primary_request(const struct rte_mp_msg *msg, const void
>>> +*peer) {
>>> +
>>> +	struct rte_mp_msg mp_resp;
>>> +	const struct eth_dev_mp_req *req =
>>> +		(const struct eth_dev_mp_req *)msg->param;
>>> +	struct eth_dev_mp_req *resp =
>>> +		(struct eth_dev_mp_req *)mp_resp.param;
>>> +	struct mp_reply_bundle *bundle;
>>> +	int ret = 0;
>>> +
>>> +	memset(&mp_resp, 0, sizeof(mp_resp));
>>> +	strlcpy(mp_resp.name, ETH_DEV_MP_ACTION_REQUEST,
>> sizeof(mp_resp.name));
>>> +	mp_resp.len_param = sizeof(*req);
>>> +	memcpy(resp, req, sizeof(*resp));
>>> +
>>> +	bundle = calloc(1, sizeof(*bundle));
>>> +	if (bundle == NULL) {
>>> +		resp->result = -ENOMEM;
>>> +		ret = rte_mp_reply(&mp_resp, peer);
>>> +		if (ret) {
>>> +			ethdev_log(ERR, "failed to send reply to primary request\n");
>>> +			return ret;
>>> +		}
>>> +	}
>>> +
>>> +	bundle->msg = *msg;
>>> +	bundle->peer = peer;
>>> +
>>> +	ret = rte_eal_mp_task_add(__handle_primary_request, bundle);
>>> +	if (ret) {
>>> +		resp->result = ret;
>>> +		ret = rte_mp_reply(&mp_resp, peer);
>>> +		if (ret) {
>>> +			ethdev_log(ERR, "failed to send reply to primary request\n");
>>> +			return ret;
>>> +		}
>>> +	}
>>
>> What you're doing here is quite dangerous. The parameter "const void *peer"
>> is only guaranteed to be valid at the time of the callback - not necessarily
>> afterwards. So, if you're handing off sending replies to a separate thread,
>> things might blow up because the pointer may no longer be valid.
> 
> OK, so what about clone the content a buffer, I think the content should be valid before reply is sent, right?

Yes, but even if you clone the content of the buffer, where would you 
send it *to*? You'll need the peer parameter to know where to send your 
response.

> 
> Thanks
> Qi
>>
>> --
>> Thanks,
>> Anatoly


-- 
Thanks,
Anatoly