Re: Xorg vs wifi scanning
On Mon, Feb 06, 2017 at 03:54:56PM +0100, Martin Pieuchot wrote:
> On 06/02/17(Mon) 12:27, Martin Pieuchot wrote:
> > Paul and Antoine reported that, since the NET_LOCK() went in, doing a
> > channel scan with iwm(4) and iwn(4) freezes X.
> >
> > This is a deadlock due to the fact that wireless drivers sleep during
> > a long time holding the NET_LOCK() while the X server tries to
> > communicate with unix sockets, that still need the NET_LOCK().
> >
> > The obvious solution to this problem is to not hold the NET_LOCK() for
> > unix socket related operations. We're working on that.
> >
> > However since hardware ioctl(2) can sleep there's no problem to release
> > the NET_LOCK() there. This is true for all ioctls that are not
> > modifying any stack state (address, multicast group, etc). Hence the
> > diff below that should fix the problem in a generic way.
>
> tb@ found that another code path needs to be unlocked. Diff below does
> that. These ioctl(2)s are only messing at the driver level, so it is
> safe to drop the NET_LOCK() there as well.
>
> ok?
OK pirofti@. Thanks!
>
> Index: netinet/in.c
> ===
> RCS file: /cvs/src/sys/netinet/in.c,v
> retrieving revision 1.133
> diff -u -p -r1.133 in.c
> --- netinet/in.c 20 Dec 2016 12:35:38 - 1.133
> +++ netinet/in.c 6 Feb 2017 13:43:06 -
> @@ -390,7 +390,11 @@ in_ioctl(u_long cmd, caddr_t data, struc
> default:
> if (ifp->if_ioctl == NULL)
> return (EOPNOTSUPP);
> - return ((*ifp->if_ioctl)(ifp, cmd, data));
> + /* XXXSMP breaks atomicity */
> + rw_exit_write();
> + error = ((*ifp->if_ioctl)(ifp, cmd, data));
> + rw_enter_write();
> + return (error);
> }
> return (0);
> }
> Index: netinet6/in6.c
> ===
> RCS file: /cvs/src/sys/netinet6/in6.c,v
> retrieving revision 1.197
> diff -u -p -r1.197 in6.c
> --- netinet6/in6.c5 Feb 2017 16:04:14 - 1.197
> +++ netinet6/in6.c6 Feb 2017 13:43:06 -
> @@ -190,6 +190,7 @@ in6_ioctl(u_long cmd, caddr_t data, stru
> struct in6_ifaddr *ia6 = NULL;
> struct in6_aliasreq *ifra = (struct in6_aliasreq *)data;
> struct sockaddr_in6 *sa6;
> + int error;
>
> if (ifp == NULL)
> return (EOPNOTSUPP);
> @@ -469,9 +470,13 @@ in6_ioctl(u_long cmd, caddr_t data, stru
> break;
>
> default:
> - if (ifp == NULL || ifp->if_ioctl == 0)
> + if (ifp->if_ioctl == NULL)
> return (EOPNOTSUPP);
> - return ((*ifp->if_ioctl)(ifp, cmd, data));
> + /* XXXSMP breaks atomicity */
> + rw_exit_write();
> + error = ((*ifp->if_ioctl)(ifp, cmd, data));
> + rw_enter_write();
> + return (error);
> }
>
> return (0);
> Index: dev/pci/if_iwm.c
> ===
> RCS file: /cvs/src/sys/dev/pci/if_iwm.c,v
> retrieving revision 1.162
> diff -u -p -r1.162 if_iwm.c
> --- dev/pci/if_iwm.c 4 Feb 2017 19:20:59 - 1.162
> +++ dev/pci/if_iwm.c 6 Feb 2017 14:52:24 -
> @@ -6133,18 +6133,13 @@ iwm_ioctl(struct ifnet *ifp, u_long cmd,
> struct ifreq *ifr;
> int s, err = 0;
>
> - /* XXXSMP breaks atomicity */
> - rw_exit_write();
> -
> /*
>* Prevent processes from entering this function while another
>* process is tsleep'ing in it.
>*/
> err = rw_enter(>ioctl_rwl, RW_WRITE | RW_INTR);
> - if (err) {
> - rw_enter_write();
> + if (err)
> return err;
> - }
>
> s = splnet();
>
> @@ -6190,7 +6185,6 @@ iwm_ioctl(struct ifnet *ifp, u_long cmd,
>
> splx(s);
> rw_exit(>ioctl_rwl);
> - rw_enter_write();
>
> return err;
> }
> Index: net/if.c
> ===
> RCS file: /cvs/src/sys/net/if.c,v
> retrieving revision 1.485
> diff -u -p -r1.485 if.c
> --- net/if.c 1 Feb 2017 02:02:01 - 1.485
> +++ net/if.c 6 Feb 2017 14:27:47 -
> @@ -2029,7 +2029,10 @@ ifioctl(struct socket *so, u_long cmd, c
> case SIOCGIFPARENT:
> if (ifp->if_ioctl == 0)
> return (EOPNOTSUPP);
> + /* XXXSMP breaks atomicity */
> + rw_exit_write();
> error = (*ifp->if_ioctl)(ifp, cmd, data);
> + rw_enter_write();
> break;
>
> case SIOCGIFDESCR:
Re: Xorg vs wifi scanning
On Mon, Feb 06, 2017 at 03:54:56PM +0100, Martin Pieuchot wrote: > tb@ found that another code path needs to be unlocked. Diff below does > that. These ioctl(2)s are only messing at the driver level, so it is > safe to drop the NET_LOCK() there as well. > > ok? I am happy with this. OK. And thank you for taking care of this problem!
Re: Xorg vs wifi scanning
On 06/02/17(Mon) 12:27, Martin Pieuchot wrote:
> Paul and Antoine reported that, since the NET_LOCK() went in, doing a
> channel scan with iwm(4) and iwn(4) freezes X.
>
> This is a deadlock due to the fact that wireless drivers sleep during
> a long time holding the NET_LOCK() while the X server tries to
> communicate with unix sockets, that still need the NET_LOCK().
>
> The obvious solution to this problem is to not hold the NET_LOCK() for
> unix socket related operations. We're working on that.
>
> However since hardware ioctl(2) can sleep there's no problem to release
> the NET_LOCK() there. This is true for all ioctls that are not
> modifying any stack state (address, multicast group, etc). Hence the
> diff below that should fix the problem in a generic way.
tb@ found that another code path needs to be unlocked. Diff below does
that. These ioctl(2)s are only messing at the driver level, so it is
safe to drop the NET_LOCK() there as well.
ok?
Index: netinet/in.c
===
RCS file: /cvs/src/sys/netinet/in.c,v
retrieving revision 1.133
diff -u -p -r1.133 in.c
--- netinet/in.c20 Dec 2016 12:35:38 - 1.133
+++ netinet/in.c6 Feb 2017 13:43:06 -
@@ -390,7 +390,11 @@ in_ioctl(u_long cmd, caddr_t data, struc
default:
if (ifp->if_ioctl == NULL)
return (EOPNOTSUPP);
- return ((*ifp->if_ioctl)(ifp, cmd, data));
+ /* XXXSMP breaks atomicity */
+ rw_exit_write();
+ error = ((*ifp->if_ioctl)(ifp, cmd, data));
+ rw_enter_write();
+ return (error);
}
return (0);
}
Index: netinet6/in6.c
===
RCS file: /cvs/src/sys/netinet6/in6.c,v
retrieving revision 1.197
diff -u -p -r1.197 in6.c
--- netinet6/in6.c 5 Feb 2017 16:04:14 - 1.197
+++ netinet6/in6.c 6 Feb 2017 13:43:06 -
@@ -190,6 +190,7 @@ in6_ioctl(u_long cmd, caddr_t data, stru
struct in6_ifaddr *ia6 = NULL;
struct in6_aliasreq *ifra = (struct in6_aliasreq *)data;
struct sockaddr_in6 *sa6;
+ int error;
if (ifp == NULL)
return (EOPNOTSUPP);
@@ -469,9 +470,13 @@ in6_ioctl(u_long cmd, caddr_t data, stru
break;
default:
- if (ifp == NULL || ifp->if_ioctl == 0)
+ if (ifp->if_ioctl == NULL)
return (EOPNOTSUPP);
- return ((*ifp->if_ioctl)(ifp, cmd, data));
+ /* XXXSMP breaks atomicity */
+ rw_exit_write();
+ error = ((*ifp->if_ioctl)(ifp, cmd, data));
+ rw_enter_write();
+ return (error);
}
return (0);
Index: dev/pci/if_iwm.c
===
RCS file: /cvs/src/sys/dev/pci/if_iwm.c,v
retrieving revision 1.162
diff -u -p -r1.162 if_iwm.c
--- dev/pci/if_iwm.c4 Feb 2017 19:20:59 - 1.162
+++ dev/pci/if_iwm.c6 Feb 2017 14:52:24 -
@@ -6133,18 +6133,13 @@ iwm_ioctl(struct ifnet *ifp, u_long cmd,
struct ifreq *ifr;
int s, err = 0;
- /* XXXSMP breaks atomicity */
- rw_exit_write();
-
/*
* Prevent processes from entering this function while another
* process is tsleep'ing in it.
*/
err = rw_enter(>ioctl_rwl, RW_WRITE | RW_INTR);
- if (err) {
- rw_enter_write();
+ if (err)
return err;
- }
s = splnet();
@@ -6190,7 +6185,6 @@ iwm_ioctl(struct ifnet *ifp, u_long cmd,
splx(s);
rw_exit(>ioctl_rwl);
- rw_enter_write();
return err;
}
Index: net/if.c
===
RCS file: /cvs/src/sys/net/if.c,v
retrieving revision 1.485
diff -u -p -r1.485 if.c
--- net/if.c1 Feb 2017 02:02:01 - 1.485
+++ net/if.c6 Feb 2017 14:27:47 -
@@ -2029,7 +2029,10 @@ ifioctl(struct socket *so, u_long cmd, c
case SIOCGIFPARENT:
if (ifp->if_ioctl == 0)
return (EOPNOTSUPP);
+ /* XXXSMP breaks atomicity */
+ rw_exit_write();
error = (*ifp->if_ioctl)(ifp, cmd, data);
+ rw_enter_write();
break;
case SIOCGIFDESCR:
Re: Xorg vs wifi scanning
On 06/02/17(Mon) 12:59, Stefan Sperling wrote: > On Mon, Feb 06, 2017 at 12:44:52PM +0100, Martin Pieuchot wrote: > > On 06/02/17(Mon) 12:39, Theo Buehler wrote: > > > This fixes the issue on my iwn0, but re-introduces the problem on my iwm0. > > > > That means another ioctl(2) triggers the problem with iwm(4). It would > > help if you could figure out which one. > > Probably SIOCS80211SCAN. Your diff doesn't touch ieee80211_ioctl(). My diff does touch ieee80211_ioctl(). Like that: ifioctl() -> pr_usrreq(PRU_CONTROL -> in_control( -> iwm_ioctl( -> ieee80211_ioctl(). Or do you know another path to reach this ioctl handler?
Re: Xorg vs wifi scanning
On Mon, Feb 06, 2017 at 12:44:52PM +0100, Martin Pieuchot wrote: > On 06/02/17(Mon) 12:39, Theo Buehler wrote: > > This fixes the issue on my iwn0, but re-introduces the problem on my iwm0. > > That means another ioctl(2) triggers the problem with iwm(4). It would > help if you could figure out which one. Probably SIOCS80211SCAN. Your diff doesn't touch ieee80211_ioctl().
Re: Xorg vs wifi scanning
On 06/02/17(Mon) 12:39, Theo Buehler wrote: > On Mon, Feb 06, 2017 at 12:27:15PM +0100, Martin Pieuchot wrote: > > Paul and Antoine reported that, since the NET_LOCK() went in, doing a > > channel scan with iwm(4) and iwn(4) freezes X. > > > > This is a deadlock due to the fact that wireless drivers sleep during > > a long time holding the NET_LOCK() while the X server tries to > > communicate with unix sockets, that still need the NET_LOCK(). > > > > The obvious solution to this problem is to not hold the NET_LOCK() for > > unix socket related operations. We're working on that. > > > > However since hardware ioctl(2) can sleep there's no problem to release > > the NET_LOCK() there. This is true for all ioctls that are not > > modifying any stack state (address, multicast group, etc). Hence the > > diff below that should fix the problem in a generic way. > > > > ok? > > This fixes the issue on my iwn0, but re-introduces the problem on my iwm0. That means another ioctl(2) triggers the problem with iwm(4). It would help if you could figure out which one. > Tested with > > iwn0 at pci2 dev 0 function 0 "Intel Centrino Ultimate-N 6300" rev 0x35: msi, > MIMO 3T3R, MoW > > and > > iwm0 at pci2 dev 0 function 0 "Intel Dual Band Wireless AC 7265" rev 0x59, msi > iwm0: hw rev 0x210, fw ver 16.242414.0 >
Re: Xorg vs wifi scanning
On Mon, Feb 06, 2017 at 12:27:15PM +0100, Martin Pieuchot wrote: > Paul and Antoine reported that, since the NET_LOCK() went in, doing a > channel scan with iwm(4) and iwn(4) freezes X. > > This is a deadlock due to the fact that wireless drivers sleep during > a long time holding the NET_LOCK() while the X server tries to > communicate with unix sockets, that still need the NET_LOCK(). > > The obvious solution to this problem is to not hold the NET_LOCK() for > unix socket related operations. We're working on that. > > However since hardware ioctl(2) can sleep there's no problem to release > the NET_LOCK() there. This is true for all ioctls that are not > modifying any stack state (address, multicast group, etc). Hence the > diff below that should fix the problem in a generic way. > > ok? This fixes the issue on my iwn0, but re-introduces the problem on my iwm0. Tested with iwn0 at pci2 dev 0 function 0 "Intel Centrino Ultimate-N 6300" rev 0x35: msi, MIMO 3T3R, MoW and iwm0 at pci2 dev 0 function 0 "Intel Dual Band Wireless AC 7265" rev 0x59, msi iwm0: hw rev 0x210, fw ver 16.242414.0
Xorg vs wifi scanning
Paul and Antoine reported that, since the NET_LOCK() went in, doing a
channel scan with iwm(4) and iwn(4) freezes X.
This is a deadlock due to the fact that wireless drivers sleep during
a long time holding the NET_LOCK() while the X server tries to
communicate with unix sockets, that still need the NET_LOCK().
The obvious solution to this problem is to not hold the NET_LOCK() for
unix socket related operations. We're working on that.
However since hardware ioctl(2) can sleep there's no problem to release
the NET_LOCK() there. This is true for all ioctls that are not
modifying any stack state (address, multicast group, etc). Hence the
diff below that should fix the problem in a generic way.
ok?
Index: netinet/in.c
===
RCS file: /cvs/src/sys/netinet/in.c,v
retrieving revision 1.133
diff -u -p -r1.133 in.c
--- netinet/in.c20 Dec 2016 12:35:38 - 1.133
+++ netinet/in.c6 Feb 2017 11:15:45 -
@@ -390,7 +390,11 @@ in_ioctl(u_long cmd, caddr_t data, struc
default:
if (ifp->if_ioctl == NULL)
return (EOPNOTSUPP);
- return ((*ifp->if_ioctl)(ifp, cmd, data));
+ /* XXXSMP breaks atomicity */
+ rw_exit_write();
+ error = ((*ifp->if_ioctl)(ifp, cmd, data));
+ rw_enter_write();
+ return (error);
}
return (0);
}
Index: netinet6/in6.c
===
RCS file: /cvs/src/sys/netinet6/in6.c,v
retrieving revision 1.197
diff -u -p -r1.197 in6.c
--- netinet6/in6.c 5 Feb 2017 16:04:14 - 1.197
+++ netinet6/in6.c 6 Feb 2017 11:16:09 -
@@ -190,6 +190,7 @@ in6_ioctl(u_long cmd, caddr_t data, stru
struct in6_ifaddr *ia6 = NULL;
struct in6_aliasreq *ifra = (struct in6_aliasreq *)data;
struct sockaddr_in6 *sa6;
+ int error;
if (ifp == NULL)
return (EOPNOTSUPP);
@@ -469,9 +470,13 @@ in6_ioctl(u_long cmd, caddr_t data, stru
break;
default:
- if (ifp == NULL || ifp->if_ioctl == 0)
+ if (ifp->if_ioctl == NULL)
return (EOPNOTSUPP);
- return ((*ifp->if_ioctl)(ifp, cmd, data));
+ /* XXXSMP breaks atomicity */
+ rw_exit_write();
+ error = ((*ifp->if_ioctl)(ifp, cmd, data));
+ rw_enter_write();
+ return (error);
}
return (0);
Index: dev/pci/if_iwm.c
===
RCS file: /cvs/src/sys/dev/pci/if_iwm.c,v
retrieving revision 1.162
diff -u -p -r1.162 if_iwm.c
--- dev/pci/if_iwm.c4 Feb 2017 19:20:59 - 1.162
+++ dev/pci/if_iwm.c6 Feb 2017 11:15:56 -
@@ -6133,18 +6133,13 @@ iwm_ioctl(struct ifnet *ifp, u_long cmd,
struct ifreq *ifr;
int s, err = 0;
- /* XXXSMP breaks atomicity */
- rw_exit_write();
-
/*
* Prevent processes from entering this function while another
* process is tsleep'ing in it.
*/
err = rw_enter(>ioctl_rwl, RW_WRITE | RW_INTR);
- if (err) {
- rw_enter_write();
+ if (err)
return err;
- }
s = splnet();
@@ -6190,7 +6185,6 @@ iwm_ioctl(struct ifnet *ifp, u_long cmd,
splx(s);
rw_exit(>ioctl_rwl);
- rw_enter_write();
return err;
}
