add zero-shot results

1 files changed, 30 insertions, 0 deletions

diff --git a/results/classifier/zero-shot-user-mode/output/instruction/1128 b/results/classifier/zero-shot-user-mode/output/instruction/1128
new file mode 100644
index 00000000..60212a6a
--- /dev/null
+++ b/results/classifier/zero-shot-user-mode/output/instruction/1128
@@ -0,0 +1,30 @@
+instruction: 0.608
+runtime: 0.223
+syscall: 0.169
+
+
+
+PPC: `spr_write_xer` doesn't set flag bits in `cpu_xer`
+Description of problem:
+`spr_write_xer()` does not set the `ca`, `ov`, `so`, `ca32`, `ov32` etc. flag bits in the `cpu_xer` variable.
+
+In fact it copies all bits from the source `GPR` and _excludes_ each flag bit.
+
+This is not a problem for execution since `spr_read_xer()` gets the flag bits from `cpu_ca/ov/so...` and not from `cpu_xer`.
+
+Nonetheless it is problem for tools which trace the execution in QEMU (e.g. https://github.com/BinaryAnalysisPlatform/qemu). 
+
+A fix would be to remove the `~` in https://gitlab.com/qemu-project/qemu/-/blob/master/target/ppc/translate.c#L481
+Steps to reproduce:
+Haven't found out yet how to debug QEMU so the TCGv values can be investigated. But in general one need to:
+
+- Execute a binary which executes something like:
+```
+r4 = 0xffffffffffffffff
+mtxer r4
+```
+and check the `cpu_xer` value after the `xer` write.
+
+Checking the debug logs (`in_asm,cpu`) doesn't work, since the `xer` value in the logs is not taken directly from `cpu_xer`.
+Additional information:
+Code ref: https://gitlab.com/qemu-project/qemu/-/blob/master/target/ppc/translate.c#L480-L483