On Sat, Apr 15 2023 at 23:06, Thomas Gleixner wrote: > On Sat, Apr 15 2023 at 09:22, Brian Gerst wrote: >> On Fri, Apr 14, 2023 at 7:45 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote: >>> @@ -248,10 +311,20 @@ SYM_INNER_LABEL(secondary_startup_64_no_ >>> * >>> * RDX contains the per-cpu offset >>> */ >>> - movq pcpu_hot + X86_current_task(%rdx), %rax >>> - movq TASK_threadsp(%rax), %rsp >>> + movq pcpu_hot + X86_top_of_stack(%rdx), %rsp >> >> Switching to using pcpu_hot.top_of_stack is ok, but it's not >> completely equivalent. top_of_stack points to the end of the pt_regs >> structure, while the kernel stack starts below pt_regs even for kernel >> threads. So you need to subtract PTREGS_SIZE from the stack pointer >> after this. >> >> This change should also be a separate patch. > > You're right on both counts. Actually no. We can't do that as this breaks suspend/resume (again). /me drops it.
