disco is dead (tomorrow), marking wontfix for disco.
** Description changed:
[Impact]
QEMU crashes when passing through 8 GPU devices on an AMD Rome-based system
which is configured (via BIOS) as a single NUMA domain.
[Test Case]
[Fix]
+ the index into the bitmap array of batch_notify_vqs incremented itself
+ by BITS_PER_LONG but then incorrectly indexed the unsigned long bitmap
+ array by the full index value; when the number of vqs was under
+ BITS_PER_LONG, the index was always 0, but once the number of vqs
+ increased over BITS_PER_LONG, the bitmap array was indexed with (e.g.
+ using BITS_PER_LONG == 64) bitmap[64] instead of bitmap[1]. Fix it to
+ use the proper index by dividing the index counter by BITS_PER_LONG to
+ get the bitmap array index.
+
[Regression Risk]
+
+ as this changes/fixes the index into an array, regressions would likely
+ occur around notifications to the guest about virtio-blk device updates,
+ or out-of-bounds errors in qemu itself, causing crash.
+
+ [Scope]
+
+ This is needed in b/e/f.
+
+ This bug was introduced by commit
+ e21737ab150c2742dd94089017db96c472dd4b87 which was introduced in version
+ 2.7.0, so this bug does not exist in Xenial or earlier.
+
+ This is fixed by commit 725fe5d10dbd4259b1853b7d253cef83a3c0d22a which
+ is not yet in focal.
** Description changed:
[Impact]
QEMU crashes when passing through 8 GPU devices on an AMD Rome-based system
which is configured (via BIOS) as a single NUMA domain.
[Test Case]
[Fix]
the index into the bitmap array of batch_notify_vqs incremented itself
by BITS_PER_LONG but then incorrectly indexed the unsigned long bitmap
array by the full index value; when the number of vqs was under
BITS_PER_LONG, the index was always 0, but once the number of vqs
increased over BITS_PER_LONG, the bitmap array was indexed with (e.g.
using BITS_PER_LONG == 64) bitmap[64] instead of bitmap[1]. Fix it to
use the proper index by dividing the index counter by BITS_PER_LONG to
get the bitmap array index.
[Regression Risk]
as this changes/fixes the index into an array, regressions would likely
occur around notifications to the guest about virtio-blk device updates,
or out-of-bounds errors in qemu itself, causing crash.
[Scope]
This is needed in b/e/f.
This bug was introduced by commit
e21737ab150c2742dd94089017db96c472dd4b87 which was introduced in version
2.7.0, so this bug does not exist in Xenial or earlier.
This is fixed by commit 725fe5d10dbd4259b1853b7d253cef83a3c0d22a which
- is not yet in focal.
+ is not yet in focal, but per comment 3 is included in the pending MR for
+ focal.
** Changed in: qemu (Ubuntu Disco)
Status: New => Won't Fix
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1859527
Title:
vring_get_region_caches: Assertion `caches != NULL' failed.
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/qemu/+bug/1859527/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
