[PATCH] D110869: [X86] Implement -fzero-call-used-regs option
Bill Wendling via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Tue Feb 1 16:43:55 PST 2022
void added a comment.
In D110869#3288485 <https://reviews.llvm.org/D110869#3288485>, @nickdesaulniers wrote:
> Hey! Looks like Diff 404763 for an x86 defconfig plus CONFIG_ZERO_CALL_USED_REGS=y starts booting! Looks like it panics though trying to launch `init`(pid 1) though.
>
> [ 0.702163] Run /bin/sh as init process
> [ 0.702913] Failed to execute /bin/sh (error -22)
> [ 0.703721] Run /sbin/init as init process
> [ 0.704454] Starting init: /sbin/init exists but couldn't execute it (error -22)
> [ 0.705702] Run /etc/init as init process
> [ 0.706390] Run /bin/init as init process
> [ 0.707037] Run /bin/sh as init process
> [ 0.707736] Starting init: /bin/sh exists but couldn't execute it (error -22)
> [ 0.708895] Kernel panic - not syncing: No working init found. Try passing init= option to kernel. See Linux Documentation/admin-guide/init.rst for guidance.
> [ 0.711246] CPU: 1 PID: 1 Comm: sh Not tainted 5.16.0-12116-g74e25defe135 #6
> [ 0.712426] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
> [ 0.713778] Call Trace:
> [ 0.714188] <TASK>
> [ 0.714578] dump_stack_lvl+0x65/0x9a
> [ 0.715190] panic+0x101/0x295
> [ 0.715743] ? _printk+0x54/0x80
> [ 0.716297] ? rest_init+0xd0/0xd0
> [ 0.716882] kernel_init+0x18b/0x190
> [ 0.717525] ret_from_fork+0x22/0x30
> [ 0.718138] </TASK>
> [ 0.721159] Kernel Offset: 0xe800000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
> [ 0.722977] ---[ end Kernel panic - not syncing: No working init found. Try passing init= option to kernel. See Linux Documentation/admin-guide/init.rst for guidance. ]---
>
> If I disable `CONFIG_ZERO_CALL_USED_REGS=y`, I can launch userspace just fine. (`Error -22` corresponds to `EINVAL` FWIW). Perhaps some kernel code related to launching init (or processes, generally) isn't happy with a zero'd register somewhere, or we have another codegen bug. Either way, that needs to be investigated+fixed before merging. Probably could sprinkle `subdir-ccflags-y += -fzero_call_used_regs=skip` (or whatever) in various Makefiles to pinpoint which object file is affected, then take a look at the disassembly from there. (or add function attributes to get more fine grained).
Strange. It boots to a login prompt for me under QEMU....
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D110869/new/
https://reviews.llvm.org/D110869
More information about the cfe-commits
mailing list