On 28 February 2023 20:17:19 GMT, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>On Tue, Feb 28 2023 at 17:09, David Woodhouse wrote:
>> On Tue, 2023-02-28 at 17:13 +0100, Thomas Gleixner wrote:
>>> As this patch is now part of the parallel boot series and actually
>>> introduces smpboot_control, the above is neither accurate nor useful.
>>
>> Better commit message, add a comment where we abuse current->thread.sp
>> in the sleep path. Didn't remove the {} which would be added back in
>> the very next patch. Pushed to my tree for Usama's next round.
>
>Ok.
>
>> However, we start by introducing one more: smpboot_control. For now this
>
>s/we// :)
Yeah, actually spotted that one as I hit send and it's different in the git tree already.
>> merely holds the CPU# of the CPU which is coming up. That CPU can then
>> find its own per-cpu data, and everything else it needs can be found from
>> there, allowing the other global variables to be removed.
>>
>> First to be removed is initial_stack. Each CPU can load %rsp from its
>> current_task->thread.sp instead. That is already set up with the correct
>> idle thread for APs. Set up the .sp field in INIT_THREAD on x86 so that
>> the BSP also finds a suitable stack pointer in the static per-cpu data
>> when coming up on first boot.
>>
>> On resume from S3, the CPU needs a temporary stack because its idle task
>> is already active. Instead of setting initial_stack, the sleep code can
>> simply set its own current->thread.sp to point to the temporary stack.
>> The true stack pointer will get restored with the rest of the CPU
>> context in do_suspend_lowlevel().
>
>Thanks for writing this up!
>
>> + /*
>> + * As each CPU starts up, it will find its own stack pointer
>> + * from its current_task->thread.sp. Typically that will be
>> + * the idle thread for a newly-started AP, or even the boot
>> + * CPU which will find it set to &init_task in the static
>> + * per-cpu data.
>> + *
>> + * Make the resuming CPU use the temporary stack at startup
>> + * by setting current->thread.sp to point to that. The true
>> + * %rsp will be restored with the rest of the CPU context,
>> + * by do_suspend_lowlevel().
>
>Right, but what restores current->thread.sp? thread.sp is used by
>unwinders...
Unwinding a thread that is actually *on* the CPU? By the time it's taken off, won't ->thread.sp have been written out again? I figured it was just a dead variable while the actual %rsp was in use?
