1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
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.
|