[dpdk-dev] [PATCH V4 6/9] eal: add failure handle mechanism for hot plug

[Ananyev, Konstantin]

> -----Original Message-----
> From: Guo, Jia
> Sent: Friday, June 29, 2018 12:15 PM
> To: Ananyev, Konstantin <konstantin.ananyev at intel.com>; stephen at networkplumber.org; Richardson, Bruce
> <bruce.richardson at intel.com>; Yigit, Ferruh <ferruh.yigit at intel.com>; gaetan.rivet at 6wind.com; Wu, Jingjing
> <jingjing.wu at intel.com>; thomas at monjalon.net; motih at mellanox.com; matan at mellanox.com; Van Haaren, Harry
> <harry.van.haaren at intel.com>; Zhang, Qi Z <qi.z.zhang at intel.com>; He, Shaopeng <shaopeng.he at intel.com>; Iremonger, Bernard
> <bernard.iremonger at intel.com>
> Cc: jblunck at infradead.org; shreyansh.jain at nxp.com; dev at dpdk.org; Zhang, Helin <helin.zhang at intel.com>
> Subject: Re: [PATCH V4 6/9] eal: add failure handle mechanism for hot plug
> 
> hi,konstantin
> 
> 
> On 6/29/2018 6:49 PM, Ananyev, Konstantin wrote:
> > Hi Jeff,
> >
> >> This patch introduces a failure handler mechanism to handle device
> >> hot plug removal event.
> >>
> >> First register sigbus handler, once sigbus error be captured, will
> >> check the failure address and accordingly remap the invalid memory
> >> for the corresponding device. Bese on this mechanism, it could
> >> guaranty the application not to be crash when hot unplug devices.
> >>
> >> Signed-off-by: Jeff Guo <jia.guo at intel.com>
> >> ---
> >> v4->v3:
> >> split patches to be small and clear.
> >> ---
> >>   lib/librte_eal/linuxapp/eal/eal_dev.c | 88 ++++++++++++++++++++++++++++++++++-
> >>   1 file changed, 87 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/lib/librte_eal/linuxapp/eal/eal_dev.c b/lib/librte_eal/linuxapp/eal/eal_dev.c
> >> index 1cf6aeb..c9dddab 100644
> >> --- a/lib/librte_eal/linuxapp/eal/eal_dev.c
> >> +++ b/lib/librte_eal/linuxapp/eal/eal_dev.c
> >> @@ -4,6 +4,8 @@
> >>
> >>   #include <string.h>
> >>   #include <unistd.h>
> >> +#include <fcntl.h>
> >> +#include <signal.h>
> >>   #include <sys/socket.h>
> >>   #include <linux/netlink.h>
> >>
> >> @@ -14,15 +16,24 @@
> >>   #include <rte_malloc.h>
> >>   #include <rte_interrupts.h>
> >>   #include <rte_alarm.h>
> >> +#include <rte_bus.h>
> >> +#include <rte_eal.h>
> >> +#include <rte_spinlock.h>
> >> +#include <rte_errno.h>
> >>
> >>   #include "eal_private.h"
> >>
> >>   static struct rte_intr_handle intr_handle = {.fd = -1 };
> >>   static bool monitor_started;
> >>
> >> +extern struct rte_bus_list rte_bus_list;
> >> +
> >>   #define EAL_UEV_MSG_LEN 4096
> >>   #define EAL_UEV_MSG_ELEM_LEN 128
> >>
> >> +/* spinlock for device failure process */
> >> +static rte_spinlock_t dev_failure_lock = RTE_SPINLOCK_INITIALIZER;
> >> +
> >>   static void dev_uev_handler(__rte_unused void *param);
> >>
> >>   /* identify the system layer which reports this event. */
> >> @@ -33,6 +44,34 @@ enum eal_dev_event_subsystem {
> >>   	EAL_DEV_EVENT_SUBSYSTEM_MAX
> >>   };
> >>
> >> +static void sigbus_handler(int signum __rte_unused, siginfo_t *info,
> >> +				void *ctx __rte_unused)
> >> +{
> >> +	int ret;
> >> +
> >> +	RTE_LOG(DEBUG, EAL, "Thread[%d] catch SIGBUS, fault address:%p\n",
> >> +		(int)pthread_self(), info->si_addr);
> >> +
> >> +	rte_spinlock_lock(&dev_failure_lock);
> >> +	ret = rte_bus_sigbus_handler(info->si_addr);
> >> +	rte_spinlock_unlock(&dev_failure_lock);
> >> +	if (!ret)
> >> +		RTE_LOG(INFO, EAL,
> >> +			"Success to handle SIGBUS error for hotplug!\n");
> >> +	else
> >> +		rte_exit(EXIT_FAILURE,
> >> +			 "A generic SIGBUS error, (rte_errno: %s)!",
> >> +			 strerror(rte_errno));
> >> +}
> > As I said in comments for previous versions:
> > I think we need to distinguish why do we fail -
> >   1) address doesn't belong to any device,
> >   2) we failed to remap
> > For 1) we probably need to call previous sigbus handler.
> 
> i know your point, but i think what ever 1) or 2), we should also need
> to call previous sigbus handler to show exception of the memory error,
> to cut down any other after try to run.

I don't agree.
If 1) - that error doesn't belong to us (DPDK), but user app might know how to handle it.
So we just invoke previously saved previous (if any) sigbus handler.
for 2) - there is not much we can do but rte_exit(). 

> and for the previous sigbus handler, i still not find a explicit call to
> use it, i think it the sigbus handler could be restore but only will use
> when next error occur, right?
> if so, do you think i just use a rte_exit to replace this origin handler
> is make sense?
> 
> >> +
> >> +static int cmp_dev_name(const struct rte_device *dev,
> >> +	const void *_name)
> >> +{
> >> +	const char *name = _name;
> >> +
> >> +	return strcmp(dev->name, name);
> >> +}
> >> +
> >>   static int
> >>   dev_uev_socket_fd_create(void)
> >>   {
> >> @@ -147,6 +186,9 @@ dev_uev_handler(__rte_unused void *param)
> >>   	struct rte_dev_event uevent;
> >>   	int ret;
> >>   	char buf[EAL_UEV_MSG_LEN];
> >> +	struct rte_bus *bus;
> >> +	struct rte_device *dev;
> >> +	const char *busname;
> >>
> >>   	memset(&uevent, 0, sizeof(struct rte_dev_event));
> >>   	memset(buf, 0, EAL_UEV_MSG_LEN);
> >> @@ -171,13 +213,48 @@ dev_uev_handler(__rte_unused void *param)
> >>   	RTE_LOG(DEBUG, EAL, "receive uevent(name:%s, type:%d, subsystem:%d)\n",
> >>   		uevent.devname, uevent.type, uevent.subsystem);
> >>
> >> -	if (uevent.devname)
> >> +	switch (uevent.subsystem) {
> >> +	case EAL_DEV_EVENT_SUBSYSTEM_PCI:
> >> +	case EAL_DEV_EVENT_SUBSYSTEM_UIO:
> >> +		busname = "pci";
> >> +		break;
> >> +	default:
> >> +		break;
> >> +	}
> >> +
> >> +	if (uevent.devname) {
> >> +		if (uevent.type == RTE_DEV_EVENT_REMOVE) {
> >> +			bus = rte_bus_find_by_name(busname);
> >> +			if (bus == NULL) {
> >> +				RTE_LOG(ERR, EAL, "Cannot find bus (%s)\n",
> >> +					busname);
> >> +				return;
> >> +			}
> >> +			dev = bus->find_device(NULL, cmp_dev_name,
> >> +					       uevent.devname);
> >> +			if (dev == NULL) {
> >> +				RTE_LOG(ERR, EAL, "Cannot find device (%s) on "
> >> +					"bus (%s)\n", uevent.devname, busname);
> >> +				return;
> >> +			}
> >> +			rte_spinlock_lock(&dev_failure_lock);
> >> +			ret = bus->hotplug_handler(dev);
> >> +			rte_spinlock_unlock(&dev_failure_lock);
> > Ok, but this function is executed from interrupt thread, correct?
> 
> yes.
> 
> > What would happen if user would do dev-detach() at the same time and dev would not be valid anymore?
> > Shouldn't we have a lock (per bus?) that we would grab before find_device() and release after hotplug_handler?
> > Though in that case we probably need to revisit other bus ops too.
> 
> make sense, i think should be the case and need lock any bus ops here to
> sync.
> 
> >> +			if (ret) {
> >> +				RTE_LOG(ERR, EAL, "Can not handle hotplug for "
> >> +					"device (%s)\n", dev->name);
> >> +				return;
> >> +			}
> >> +		}
> >>   		dev_callback_process(uevent.devname, uevent.type);
> >> +	}
> >>   }
> >>
> >>   int __rte_experimental
> >>   rte_dev_event_monitor_start(void)
> >>   {
> >> +	sigset_t mask;
> >> +	struct sigaction action;
> >>   	int ret;
> >>
> >>   	if (monitor_started)
> >> @@ -197,6 +274,14 @@ rte_dev_event_monitor_start(void)
> >>   		return -1;
> >>   	}
> >>
> >> +	/* register sigbus handler */
> >> +	sigemptyset(&mask);
> >> +	sigaddset(&mask, SIGBUS);
> >> +	action.sa_flags = SA_SIGINFO;
> >> +	action.sa_mask = mask;
> >> +	action.sa_sigaction = sigbus_handler;
> >> +	sigaction(SIGBUS, &action, NULL);
> >> +
> > I still think we have to save (and restore at monitor_stop) previous sigbus handler.
> 
> ok, i think i missing here, if monitor_stop, definitely should restore
> the previous sigbus handler.
> 
> >>   	monitor_started = true;
> >>
> >>   	return 0;
> >> @@ -220,5 +305,6 @@ rte_dev_event_monitor_stop(void)
> >>   	close(intr_handle.fd);
> >>   	intr_handle.fd = -1;
> >>   	monitor_started = false;
> >> +
> >>   	return 0;
> >>   }
> >> --
> >> 2.7.4