Re: [GIT PULL] printk for 5.10 (includes lockless ringbuffer)

[Petr Mladek <mladek.petr@gmail.com>]

On Wed 2020-10-14 16:58:27, Rasmus Villemoes wrote:
> On 14/10/2020 16.16, Geert Uytterhoeven wrote:
> > Hi Petr,
> > 
> > On Mon, Oct 12, 2020 at 4:50 PM Petr Mladek <pmladek@suse.com> wrote:
> >> - Fully lockless ringbuffer implementation, including the support for
> >>   continuous lines. It will allow to store and read messages in any
> >>   situation wihtout the risk of deadlocks and without the need
> >>   of temporary per-CPU buffers.
> > 
> >     linux-m68k-atari_defconfig$ bloat-o-meter vmlinux.old
> > vmlinux.lockless_ringbuffer
> >     add/remove: 39/16 grow/shrink: 9/15 up/down: 214075/-4362 (209713)
> >     Function                                     old     new   delta
> >     _printk_rb_static_infos                        -  180224 +180224
> >     _printk_rb_static_descs                        -   24576  +24576
> >     [...]
> > 
> > Seriously?!? Or am I being misled by the tools?
> > 
> >     linux-m68k-atari_defconfig$ size vmlinux.old vmlinux.lockless_ringbuffer
> >        text    data     bss     dec     hex filename
> >     3559108 941716 177772 4678596 4763c4 vmlinux.old
> >     3563922 1152496 175276 4891694 4aa42e vmlinux.lockless_ringbuffer
> > 
> > Apparently not...
> 
> Hm, that's quite a lot. And the only reason the buffers don't live
> entirely in .bss is because a few of their entries have non-zero
> initializers.
> 
> Perhaps one could add a .init.text.initialize_static_data section of
> function pointers, with the _DEFINE_PRINTKRB macro growing something like
> 
> static void __init __initialize_printkrb_##name(void) { \
>   _##name##_descs[_DESCS_COUNT(descbits) - 1] = ...; \
>   _##name##_infos[0] = ...; \
>   _##name##_infos[_DESCS_COUNT(descbits) - 1] = ...; \
> } \
> static_data_initializer(__initialize_printkrb_##name);
> 
> with static_data_initalizer being the obvious yoga for putting a
> function pointer in the .init.text.initialize_static_data section. Then
> very early in start_kernel(), probably first thing, iterate that section
> and call all the functions. But maybe that's not even early enough?

A solution might be to initialize the buffer during the first
printk() call. We could make sure that it is done in
setup_log_buf() at latest. It is called when only one CPU is
running so it should be safe. The only problem might be NMI.

Best Regards,
Petr