summary refs log tree commit diff stats
path: root/results/classifier/zero-shot/108/other/2578
diff options
context:
space:
mode:
Diffstat (limited to 'results/classifier/zero-shot/108/other/2578')
-rw-r--r--results/classifier/zero-shot/108/other/257829
1 files changed, 29 insertions, 0 deletions
diff --git a/results/classifier/zero-shot/108/other/2578 b/results/classifier/zero-shot/108/other/2578
new file mode 100644
index 000000000..8d58638cf
--- /dev/null
+++ b/results/classifier/zero-shot/108/other/2578
@@ -0,0 +1,29 @@
+graphic: 0.890
+device: 0.814
+KVM: 0.776
+network: 0.664
+socket: 0.644
+vnc: 0.579
+semantic: 0.512
+performance: 0.483
+PID: 0.444
+other: 0.398
+boot: 0.365
+permissions: 0.323
+debug: 0.279
+files: 0.200
+
+x86: exception during hardware interrupt pushes wrong error code
+Description of problem:
+Exceptions during IDT traversal push the wrong error code when triggered by a hardware interrupt.
+The EXT bit in TCG mode is never set.  However, it works fine in KVM mode as hardware is generating the number.
+Steps to reproduce:
+1. load a short IDT e.g. with 64 entries
+2. trigger a self IPI through the LAPIC with a vector 100
+3. the pushed error code is 802 instead of 803.
+Additional information:
+It can be fixed in the lines `raise_exception_err(env, EXCP0D_GPF, intno * 8 + 2);` in `seg_helper.c` 
+which must include the `is_hw` field when calculating the error number. Something like `intno * 8 + 2 + (is_hw != 0)` 
+works here.
+
+Nevertheless, all the other exception cases in the `do_interrupt_*` functions have to set the same bit as well.