[dpdk-dev] [PATCH v10 05/19] eal: enable hotplug on multi-process

[Zhang, Qi Z]

> -----Original Message-----
> From: Burakov, Anatoly
> Sent: Tuesday, July 10, 2018 10:01 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 v10 05/19] eal: enable hotplug on multi-process
> 
> On 09-Jul-18 4:36 AM, Qi Zhang wrote:
> > We are going to introduce the solution to handle hotplug in
> > multi-process, it includes the below scenario:
> >
> > 1. Attach a device from the primary
> > 2. Detach a device from the primary
> > 3. Attach a device from a secondary
> > 4. Detach a device from a secondary
> >
> > In the primary-secondary process model, we assume devices are shared
> > by default. that means attaches or detaches 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/detaching process will cause inconsistent
> > status between processes, so proper rollback action should be considered.
> >
> > This patch covers the implementation of case 1,2.
> > Case 3,4 will be implemented on a separate patch.
> >
> > IPC scenario for Case 1, 2:
> >
> > attach a 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 the device and send a reply.
> > d) primary check the reply if all success goes to i).
> > e) primary send attach rollback sync request to all secondary.
> > f) secondary receive the request and detach the device and send a reply.
> > g) primary receive the reply and detach device as rollback action.
> > h) attach fail
> > i) attach success
> >
> > detach a device
> > a) primary send detach sync request to all secondary
> > b) secondary detach the device and send reply
> > c) primary check the reply if all success goes to f).
> > d) primary send detach rollback sync request to all secondary.
> > e) secondary receive the request and attach back device. goto g)
> > f) primary detach the device if success goto g), else goto d)
> > g) detach fail.
> > h) detach success.
> >
> > Signed-off-by: Qi Zhang <qi.z.zhang at intel.com>
> > ---
> 
> <snip>
> 
> > +	req.t = EAL_DEV_REQ_TYPE_ATTACH;
> > +	strlcpy(req.busname, busname, RTE_BUS_NAME_MAX_LEN);
> > +	strlcpy(req.devname, devname, RTE_DEV_NAME_MAX_LEN);
> > +	strlcpy(req.devargs, devargs, RTE_DEV_ARGS_MAX_LEN);
> > +
> > +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> > +		return -ENOTSUP;
> 
> Nitpick, but maybe do this before strlcpy?

On the next patch, these strlcpy can be reused when implemented secondary process case

> 
> > +
> > +	/**
> > +	 * attach a device from primary start from here:
> > +	 *
> > +	 * 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 the device and send a reply.
> > +	 * d) primary check the reply if all success goes to i).
> 
> <snip>
> 
> > +
> > +	memset(&req, 0, sizeof(req));
> > +	req.t = EAL_DEV_REQ_TYPE_DETACH;
> > +	strlcpy(req.busname, busname, RTE_BUS_NAME_MAX_LEN);
> > +	strlcpy(req.devname, devname, RTE_DEV_NAME_MAX_LEN);
> > +
> > +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> > +		return -ENOTSUP;
> 
> Same nitpick, probably move this above.
> 
> > +
> > +	/**
> > +	 * detach a device from primary start from here:
> > +	 *
> > +	 * a) primary send detach sync request to all secondary
> > +	 * b) secondary detach the device and send reply
> 
> <snip>
> 
> > +	struct mp_reply_bundle *bundle = param;
> > +	struct rte_mp_msg *msg = &bundle->msg;
> > +	const struct eal_dev_mp_req *req =
> > +		(const struct eal_dev_mp_req *)msg->param;
> > +	struct rte_mp_msg mp_resp;
> > +	struct eal_dev_mp_req *resp =
> > +		(struct eal_dev_mp_req *)mp_resp.param;
> > +	int ret = 0;
> > +
> > +	memset(&mp_resp, 0, sizeof(mp_resp));
> > +
> > +	switch (req->t) {
> > +	case EAL_DEV_REQ_TYPE_ATTACH:
> > +	case EAL_DEV_REQ_TYPE_DETACH_ROLLBACK:
> > +		ret = do_dev_hotplug_add(req->busname, req->devname, "");
> 
> I'm not too familiar with devargs and hotplug, but why are we passing empty
> devargs string here? Is it possible for it to be not empty?

For secondary process, devargs is ignored, so we just need the device unique identity <busname, devname>

> 
> > +		break;
> > +	case EAL_DEV_REQ_TYPE_DETACH:
> > +	case EAL_DEV_REQ_TYPE_ATTACH_ROLLBACK:
> > +		ret = do_dev_hotplug_remove(req->busname, req->devname);
> > +		break;
> > +	default:
> > +		ret = -EINVAL;
> > +	}
> 
> <snip>
> 
> > +	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> > +		ret = rte_mp_action_register(EAL_DEV_MP_ACTION_REQUEST,
> > +					handle_secondary_request);
> > +		if (ret) {
> > +			RTE_LOG(ERR, EAL, "Couldn't register '%s' action\n",
> > +				EAL_DEV_MP_ACTION_REQUEST);
> > +			return ret;
> > +		}
> > +	} else {
> > +		ret = rte_mp_action_register(EAL_DEV_MP_ACTION_REQUEST,
> > +		handle_primary_request);
> 
> ^^ wrong indentation.

Will fix.
> 
> > +		if (ret) {
> > +			RTE_LOG(ERR, EAL, "Couldn't register '%s' action\n",
> > +				EAL_DEV_MP_ACTION_REQUEST);
> > +			return ret;
> > +		}
> > +	}
> 
> <snip>
> 
> > +
> > +#endif /* _HOTPLUG_MP_H_ */
> > diff --git a/lib/librte_eal/common/include/rte_bus.h
> > b/lib/librte_eal/common/include/rte_bus.h
> > index eb9eded4e..720f7c3c8 100644
> > --- a/lib/librte_eal/common/include/rte_bus.h
> > +++ b/lib/librte_eal/common/include/rte_bus.h
> > @@ -197,6 +197,9 @@ struct rte_bus_conf {
> >   typedef enum rte_iova_mode (*rte_bus_get_iommu_class_t)(void);
> >
> >
> > +/* Max length for a bus name */
> > +#define RTE_BUS_NAME_MAX_LEN 32
> 
> Is this enforced anywhere in the bus codebase? Can we guarantee that bus
> name will never be bigger than this?

I think 32 should be enough for a bus name even in future.

> 
> > +
> >   /**
> >    * A structure describing a generic bus.
> >    */
> > diff --git a/lib/librte_eal/common/include/rte_dev.h
> > b/lib/librte_eal/common/include/rte_dev.h
> > index 3879ff3ca..667df20f0 100644
> > --- a/lib/librte_eal/common/include/rte_dev.h
> > +++ b/lib/librte_eal/common/include/rte_dev.h
> > @@ -152,6 +152,9 @@ struct rte_driver {
> >    */
> >   #define RTE_DEV_NAME_MAX_LEN 64
> >
> > +/* Max devargs length be allowed */
> > +#define RTE_DEV_ARGS_MAX_LEN 128
> 
> Same - is this enforced anywhere in the codebase related to devargs?

I'm not sure, but I guess it is big enough for all exist driver :)

Thanks
Qi

> 
> --
> Thanks,
> Anatoly