[dpdk-stable] [PATCH v2] malloc: notify primary process about hotplug in secondary

[Thomas Monjalon]

07/12/2018 21:30, Stojaczyk, Dariusz:
> Hi Kevin, is the merge window for DPDK 18.08.1 and 18.05.2 still open? This fix is critical for multi-process memory hotplug and you might want to pull it in.

First thing to do is to make sure stable at dpdk.org is Cc'ed
in the patch, below Fixes:.


> > -----Original Message-----
> > From: Howell, Seth
> > Sent: Friday, December 7, 2018 9:11 PM
> > To: Burakov, Anatoly <anatoly.burakov at intel.com>
> > Cc: dev at dpdk.org; stable at dpdk.org; Howell, Seth
> > <seth.howell at intel.com>; Stojaczyk, Dariusz <dariusz.stojaczyk at intel.com>
> > Subject: [PATCH v2] malloc: notify primary process about hotplug in
> > secondary
> > 
> > When secondary process hotplugs memory, it sends a request
> > to primary, which then performs the real mmap() and sends
> > sync requests to all secondary processes. Upon receiving
> > such sync request, each secondary process will notify the
> > upper layers of hotplugged memory (and will call all
> > locally registered event callbacks).
> > 
> > In the end we'll end up with memory event callbacks fired
> > in all the processes except the primary, which is a bug.
> > 
> > This gets critical if memory is hotplugged while a VFIO
> > device is attached, as the VFIO memory registration -
> > which is done from a memory event callback present in the
> > primary process only - is never called.
> > 
> > After this patch, a primary process fires memory event
> > callbacks before secondary processes start their
> > synchronizations - both for hotplug and hotremove.
> > 
> > Fixes: 07dcbfe0101f ("malloc: support multiprocess memory hotplug")
> > 
> > Signed-off-by: Seth Howell <seth.howell at intel.com>
> > Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk at intel.com>
> > ---
> >  lib/librte_eal/common/malloc_mp.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> > 
> > diff --git a/lib/librte_eal/common/malloc_mp.c
> > b/lib/librte_eal/common/malloc_mp.c
> > index 5f2d4e0be..f3a13353b 100644
> > --- a/lib/librte_eal/common/malloc_mp.c
> > +++ b/lib/librte_eal/common/malloc_mp.c
> > @@ -209,6 +209,8 @@ handle_alloc_request(const struct malloc_mp_req
> > *m,
> > 
> >  	map_addr = ms[0]->addr;
> > 
> > +	eal_memalloc_mem_event_notify(RTE_MEM_EVENT_ALLOC,
> > map_addr, alloc_sz);
> > +
> >  	/* we have succeeded in allocating memory, but we still need to sync
> >  	 * with other processes. however, since DPDK IPC is single-threaded,
> > we
> >  	 * send an asynchronous request and exit this callback.
> > @@ -258,6 +260,9 @@ handle_request(const struct rte_mp_msg *msg,
> > const void *peer __rte_unused)
> >  	if (m->t == REQ_TYPE_ALLOC) {
> >  		ret = handle_alloc_request(m, entry);
> >  	} else if (m->t == REQ_TYPE_FREE) {
> > +		eal_memalloc_mem_event_notify(RTE_MEM_EVENT_FREE,
> > +				m->free_req.addr, m->free_req.len);
> > +
> >  		ret = malloc_heap_free_pages(m->free_req.addr,
> >  				m->free_req.len);
> >  	} else {
> > @@ -436,6 +441,9 @@ handle_sync_response(const struct rte_mp_msg
> > *request,
> >  		memset(&rb_msg, 0, sizeof(rb_msg));
> > 
> >  		/* we've failed to sync, so do a rollback */
> > +		eal_memalloc_mem_event_notify(RTE_MEM_EVENT_FREE,
> > +				state->map_addr, state->map_len);
> > +
> >  		rollback_expand_heap(state->ms, state->ms_len, state-
> > >elem,
> >  				state->map_addr, state->map_len);
> > 
> > --
> > 2.17.2
>