diff options
Diffstat (limited to 'results/classifier/accel-gemma3:12b/kvm/734')
| -rw-r--r-- | results/classifier/accel-gemma3:12b/kvm/734 | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/results/classifier/accel-gemma3:12b/kvm/734 b/results/classifier/accel-gemma3:12b/kvm/734 new file mode 100644 index 00000000..28d5fa70 --- /dev/null +++ b/results/classifier/accel-gemma3:12b/kvm/734 @@ -0,0 +1,29 @@ + +aarch64 tlb range invalidate is not accurate +Description of problem: +In this (https://gitlab.com/qemu-project/qemu/-/commit/84940ed82552d3c7c7327c83076b02cee7978257) commit, tlb range invalidate support is added, and I think qemu's range calculation is wrong. + +In `tlbi_aa64_range_get_length` function, `num`, `scale`, `page_size_granule` is caculated as below. + + +``` + num = extract64(value, 39, 4); + scale = extract64(value, 44, 2); + page_size_granule = extract64(value, 46, 2); + + page_shift = page_size_granule * 2 + 12; +``` + +As [Arm documentation](https://developer.arm.com/documentation/ddi0595/2021-06/AArch64-Instructions/TLBI-RVALE1--TLBI-RVALE1NXS--TLB-Range-Invalidate-by-VA--Last-level--EL1), NUM bits's length is 5, but the code above only extract 4bits. + +And `page_shift` also should be calculated as `(page_size_granule-1) <<1) + 12` rather than `page_size_granule * 2 + 12`. +Steps to reproduce: +1. +2. +3. +Additional information: +I found this issue while debugging a phenomenon that kernel panic occurs randomly in my qemu fork. + +I'm pretty sure this is one of the causes, but even if I roughly correct it, my problem has not been solved. + +I think my problem is TLB invalidate related issue, so if I find any more problems, I'll comment here. |