On Fri, May 26, 2023 at 09:38:25PM +0800, Zhangjin Wu wrote:
Hi, Conor.
On Fri, May 26, 2023 at 06:25:18PM +0800, Zhangjin Wu wrote:
On 2023-05-25 01:52:29+0800, Zhangjin Wu wrote:
riscv qemu has a builtin bios (opensbi), but it may not match the latest kernel and some old versions may hang during boot, let's allow user pass a newer version to qemu via the -bios option.
Nitpick:
This seems very specific and hopefully only necessary temporarily.
RISC-V is such a new ISA and the Spec (especially the SBI) changes very frequently ;-)
Huh. Could you please expand on which versions of QEMU will hang while booting an upstream or stable kernel? Which kernels would be good to know too.
As the cover letter listed (in the Environment section), the softwares we used are:
Not super interested in those ones since they work ;)
The kernel version is the one this patchset based on (Willy's nolibc repo), it is v6.4-rc1.
qemu v4.2.1 is the one systematically installed (/usr/bin) from the qemu-system-misc package and used to test this patchset in my Ubuntu 20.04 based test docker image.
Okay, in the context of RISC-V, that is pretty ancient ;)
Just installed a v7.0.0 qemu from ppa:canonical-server/server-backports, there is no default opensbi, and re-checked, there is one prebuilt opensbi for rv64, but still no prebuilt opensbi for rv32.
Ah, I see.
The hang issue I mentioned may be using one of my older prebuilt version of opensbi, I can not find which one it exactly is, so, please ignore that info, will update that description too.
Okay. If you do manage to reproduce it, LMK! I was/am just worried we have some regressions because you should be able to keep booting with those older opensbi versions, modulo some Kconfig changes - although if it is something like qemu 4.2.1 specific I don't think I care all that much about dinosaurs ;)
Btw, something not about this patch: qemu v8.0.0 seems not boot non-mmu v6.3, both sides have issues, not dig into it carefully, so, not report it yet.
Cool. Feel free to CC me on whatever you discover. nommu gets little enough testing in mainline, and even less in stable kernels. That reminds me, I do need to add 32-bit nommu to the patchwork automation for linux-riscv.
Thanks, Conor.