graphic: 0.903
device: 0.862
performance: 0.683
socket: 0.638
architecture: 0.608
network: 0.597
arm: 0.570
risc-v: 0.519
vnc: 0.475
kernel: 0.472
user-level: 0.471
semantic: 0.469
i386: 0.442
VMM: 0.403
boot: 0.399
PID: 0.390
register: 0.383
KVM: 0.368
x86: 0.365
debug: 0.355
peripherals: 0.334
assembly: 0.319
ppc: 0.318
TCG: 0.297
hypervisor: 0.255
files: 0.251
virtual: 0.249
mistranslation: 0.230
permissions: 0.112

Heap-use-after-free through double-fetch in ehci

Hello,
I don't have a qtest reproducer for this crash because it involves a DMA double-fetch, and I don't think we can reproduce those with qtest.

Instead, I attached the pseudo-qtest trace produced by the fuzzer, along with some trace events.
The lines annotated with [DMA] are write commands that were triggered by a callback from a DMA read by the device. The lines annotated with [DOUBLE-FETCH] are DMA accesses that hit the same address more than once (possible double-fetches).

I am still thinking of nicer ways of presenting this trace and providing a reproducer.
-Alex



Hi Alexander! Have you ever been able to create a reproducer for this problem?

No. If we figure out some way to consistently reproduce double-fetches in a non-fuzzer build, I'll report the issue again, but this can probably be closed

Ok, let's close this one since it was not reproducible. If you find a reproducer, please open a new ticket in the gitlab tracker instead.