Enumerate all allocated evtchns in lsevtchn
v2 cover Enumerate all allocated evtchns in lsevtchn
Currently, lsevtchn aborts its event channel enumeration when it hits its first hypercall error, namely:
* When an event channel doesn't exist at the specified port
* When the event channel is owned by Xen
This results in lsevtchn missing potential relevant event channels with higher port numbers, since lsevtchn cannot determine the cause of hypercall errors.
This patch adds error status indicators for the evtchn_status hypercall for when no further event channels will be yielded for higher port numbers, allowing lsevtchn to terminate when all event channels have been enumerated over.
Matthew Barnes (2):
evtchn: Add error status indicators for evtchn_status hypercall tools/lsevtchn: Use new status identifiers in loop
tools/xcutils/lsevtchn.c | 11 ++++++++++- xen/common/event_channel.c | 12 +++++++++++- xen/include/public/event_channel.h | 2 ++ 3 files changed, 23 insertions(+), 2 deletions(-)
Signature · 1 line
2.34.1
PATCH v2 1/2 evtchn: Add error status indicators for evtchn_status hypercall
When the evtchn_status hypercall fails, it is not possible to determine the cause of the error, since the library call returns -1 to the tool and not the errno.
Because of this, lsevtchn is unable to determine whether to continue event channel enumeration upon an evtchn_status hypercall error.
Add error status indicators for the eventchn_status hypercall for lsevtchn to terminate its loop on, and keep other errors as failed hypercalls.
--- xen/common/event_channel.c | 12 +++++++++++- xen/include/public/event_channel.h | 2 ++ 2 files changed, 13 insertions(+), 1 deletion(-)
xen/common/event_channel.c +11 -1
@@ -1030,7 +1030,17 @@ int evtchn_status(evtchn_status_t *status) d =rcu_lock_domain_by_any_id (dom);if ( d ==NULL ) -return -ESRCH ; + { + status->status = EVTCHNSTAT_dom_invalid; +return 0 ; + } + +if ( !port_is_valid (d, port) ) + { + status->status = EVTCHNSTAT_port_invalid; +rcu_unlock_domain (d); +return 0 ; + } chn =_evtchn_from_port (d, port);if ( !chn )
xen/include/public/event_channel.h +2 -0
@@ -200,6 +200,8 @@ struct evtchn_status {#define EVTCHNSTAT_pirq 3/* Channel is bound to a phys IRQ line. */ #define EVTCHNSTAT_virq 4/* Channel is bound to a virtual IRQ line */ #define EVTCHNSTAT_ipi 5/* Channel is bound to a virtual IPI line */ +#define EVTCHNSTAT_dom_invalid 6/* Given domain ID is not a valid domain */ +#define EVTCHNSTAT_port_invalid 7/* Given port is not within valid range */ uint32_t status;uint32_t vcpu;/* VCPU to which this channel is bound. */ union {
Signature · 1 line
2.34.1
Re: [XEN PATCH v2 1/2] evtchn: Add error status indicators for evtchn_status hypercall
Quoted 3 lines · Matthew Barnes
When the
evtchn_statushypercall fails, it is not possible to determine the cause of the error, since the library call returns -1 to the tool and not the errno.
That's normal behavior for such library functions. If you want to know the specific error number, you ought to look at the errno variable. Or are you saying that errno isn't set correctly in this case (I can't spot such an issue when looking at do_evtchn_op(), backing xc_evtchn_status())?
xen/common/event_channel.c int evtchn_status(evtchn_status_t *status) · line 1035
Quoted 10 lines
xen/common/event_channel.c +3
-1@@ -1030,7 +1030,17 @@ int evtchn_status(evtchn_status_t *status) d =rcu_lock_domain_by_any_id (dom);if ( d ==NULL ) -return -ESRCH ; + { + status->status = EVTCHNSTAT_dom_invalid; +return 0 ;
This surely ought to remain -ESRCH. You may not break existing callers.
xen/common/event_channel.c int evtchn_status(evtchn_status_t *status) · line 1043
Quoted 8 lines
xen/common/event_channel.c +8
-0+ } + +if ( !port_is_valid (d, port) ) + { + status->status = EVTCHNSTAT_port_invalid; +rcu_unlock_domain (d); +return 0 ; + }
I can see that for the purpose of patch 2 this wants distinguishing from
xen/common/event_channel.c int evtchn_status(evtchn_status_t *status) · line 1046
Quoted 2 lines
xen/common/event_channel.c +0
-0chn =_evtchn_from_port (d, port);if ( !chn )
... the -EINVAL returned here. Yet "success" doesn't look correct there either. -ENOENT, -EBADF, -ENFILE, or -EDOM maybe?
xen/include/public/event_channel.h struct evtchn_status { · line 207
Quoted 11 lines
xen/include/public/event_channel.h +2
-0@@ -200,6 +200,8 @@ struct evtchn_status {#define EVTCHNSTAT_pirq 3/* Channel is bound to a phys IRQ line. */ #define EVTCHNSTAT_virq 4/* Channel is bound to a virtual IRQ line */ #define EVTCHNSTAT_ipi 5/* Channel is bound to a virtual IPI line */ +#define EVTCHNSTAT_dom_invalid 6/* Given domain ID is not a valid domain */ +#define EVTCHNSTAT_port_invalid 7/* Given port is not within valid range */ uint32_t status;uint32_t vcpu;/* VCPU to which this channel is bound. */ union {
If such indicators are to be added, I'm pretty sure they want to be discontiguous from the presently used range. Sadly, with status having unsigned type, using negative values wouldn't feel quite right.
Jan
PATCH v2 2/2 tools/lsevtchn: Use new status identifiers in loop
In reply to Matthew Barneslsevtchn terminates the loop when the hypercall returns an error, even if there are still event channels with higher port numbers to be enumerated over.
Continue the loop even on hypercall errors, and use the new status identifiers for the evtchn_status hypercall, namely "port invalid" and "domain invalid", to determine when to break the loop.
--- tools/xcutils/lsevtchn.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
tools/xcutils/lsevtchn.c +10 -1
@@ -24,11 +24,20 @@ int main(int argc, char **argv) status.port = port; rc =xc_evtchn_status (xch, &status);if ( rc <0 ) -break ; +continue ;if ( status.status == EVTCHNSTAT_closed )continue ; +if ( status.status == EVTCHNSTAT_dom_invalid ) + { +printf ("Domain ID '%d' does not correspond to valid domain.\n" , domid); +break ; + } + +if ( status.status == EVTCHNSTAT_port_invalid ) +break ; +printf ("%4d: VCPU %u: " , port, status.vcpu); switch ( status.status )
Signature · 1 line
2.34.1