memory: remove assert to avoid unnecessary coredump
PATCH 1/1 memory: remove assert to avoid unnecessary coredump
It is too strict to use assert to make qemu coredump if the notification does not overlap with registered range. Skip it is fine enough.
During test, we found such a case for vhost net device:
memory_region_notify_one: entry->iova=0xfee00000, entry_end=0xfeffffff, notifier->start=0xfef00000, notifier->end=0xffffffffffffffff
Skip this notification but not coredump makes everything work well.
--- memory.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)
memory.c +2 -3
@@ -1921,12 +1921,11 @@ void memory_region_notify_one(IOMMUNotifier *notifier,* Skip the notification if the notification does not overlap * with registered range. */ -if (notifier->start > entry_end || notifier->end < entry->iova) { +if (notifier->start > entry_end || notifier->end < entry->iova || + entry->iova < notifier->start || entry_end > notifier->end) {return ; } -assert (entry->iova >= notifier->start && entry_end <= notifier->end); -if (entry->perm &IOMMU_RW ) { request_flags =IOMMU_NOTIFIER_MAP ; }else {
Signature · 1 line
2.15.1
Quoted 7 lines · Yi Sun
It is too strict to use assert to make qemu coredump if the notification does not overlap with registered range. Skip it is fine enough.
During test, we found such a case for vhost net device:
memory_region_notify_one: entry->iova=0xfee00000, entry_end=0xfeffffff, notifier->start=0xfef00000, notifier->end=0xffffffffffffffff
so for range from 0xfef00000 to 0xfefffff, would notification for this range get lost?
Thanks
Yan
Quoted 30 lines
Skip this notification but not coredump makes everything work well.
Signed-off-by Yi Sun <[email protected]>--- memory.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)memory.c +2
-3@@ -1921,12 +1921,11 @@ void memory_region_notify_one(IOMMUNotifier *notifier,* Skip the notification if the notification does not overlap * with registered range. */ -if (notifier->start > entry_end || notifier->end < entry->iova) { +if (notifier->start > entry_end || notifier->end < entry->iova || + entry->iova < notifier->start || entry_end > notifier->end) {return ; } -assert (entry->iova >= notifier->start && entry_end <= notifier->end); -if (entry->perm &IOMMU_RW ) { request_flags =IOMMU_NOTIFIER_MAP ; }else {Signature · 1 line
2.15.1
Quoted 11 lines · Yan Zhao
Quoted 7 lines · Yi Sun
It is too strict to use assert to make qemu coredump if the notification does not overlap with registered range. Skip it is fine enough.
During test, we found such a case for vhost net device:
memory_region_notify_one: entry->iova=0xfee00000, entry_end=0xfeffffff, notifier->start=0xfef00000, notifier->end=0xffffffffffffffffso for range from
0xfef00000to0xfefffff, would notification for this range get lost?
Yes, that is an issue although there is no any problem found during test with this fix.
I think we should notify the intersection between entry and notifier. How do you think?
Quoted 33 lines
Thanks
YanQuoted 30 lines
Skip this notification but not coredump makes everything work well.
Signed-off-by Yi Sun <[email protected]>--- memory.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)memory.c +2
-3@@ -1921,12 +1921,11 @@ void memory_region_notify_one(IOMMUNotifier *notifier,* Skip the notification if the notification does not overlap * with registered range. */ -if (notifier->start > entry_end || notifier->end < entry->iova) { +if (notifier->start > entry_end || notifier->end < entry->iova || + entry->iova < notifier->start || entry_end > notifier->end) {return ; } -assert (entry->iova >= notifier->start && entry_end <= notifier->end); -if (entry->perm &IOMMU_RW ) { request_flags =IOMMU_NOTIFIER_MAP ; }else {Signature · 1 line
2.15.1
Quoted 9 lines · Yi Sun
Quoted 8 lines · Yan Zhao
Quoted 7 lines · Yi Sun
It is too strict to use assert to make qemu coredump if the notification does not overlap with registered range. Skip it is fine enough.
During test, we found such a case for vhost net device:
memory_region_notify_one: entry->iova=0xfee00000, entry_end=0xfeffffff, notifier->start=0xfef00000, notifier->end=0xffffffffffffffff
Quoted 9 lines
Quoted 3 lines
so for range from
0xfef00000to0xfefffff, would notification for this range get lost?Yes, that is an issue although there is no any problem found during test with this fix.
I think we should notify the intersection between entry and notifier. How do you think?
no. please refer to the link below.
https://lists.gnu.org/archive/html/qemu-devel/2019-06/msg04218.html
Thanks
Yan
Quoted 33 lines
Quoted 33 lines
Thanks
YanQuoted 30 lines
Skip this notification but not coredump makes everything work well.
Signed-off-by Yi Sun <[email protected]>--- memory.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)memory.c +2
-3@@ -1921,12 +1921,11 @@ void memory_region_notify_one(IOMMUNotifier *notifier,* Skip the notification if the notification does not overlap * with registered range. */ -if (notifier->start > entry_end || notifier->end < entry->iova) { +if (notifier->start > entry_end || notifier->end < entry->iova || + entry->iova < notifier->start || entry_end > notifier->end) {return ; } -assert (entry->iova >= notifier->start && entry_end <= notifier->end); -if (entry->perm &IOMMU_RW ) { request_flags =IOMMU_NOTIFIER_MAP ; }else {Signature · 1 line
2.15.1