printk: queue wake_up_klogd irq_work only if per-CPU areas are ready

printk: queue wake_up_klogd irq_work only if per-CPU areas are ready

[Sergey Senozhatsky]

Hello,

Commit ab6f762f0f53162d41 Linus' HEAD.

printk_deferred() does not make sure that it's safe to write to
per-CPU data, which causes problems when printk_deferred() is
invoked "too early", before per-CPU areas are initialized. There
are multiple bug reports, e.g.
https://bugzilla.kernel.org/show_bug.cgi?id=206847

	-ss

Re: printk: queue wake_up_klogd irq_work only if per-CPU areas are ready

[Greg KH]

On Tue, Apr 14, 2020 at 09:06:13PM +0900, Sergey Senozhatsky wrote:
> Hello,
> 
> Commit ab6f762f0f53162d41 Linus' HEAD.
> 
> printk_deferred() does not make sure that it's safe to write to
> per-CPU data, which causes problems when printk_deferred() is
> invoked "too early", before per-CPU areas are initialized. There
> are multiple bug reports, e.g.
> https://bugzilla.kernel.org/show_bug.cgi?id=206847
> 
> 	-ss

So where do you want this commit backported to?

thanks,
greg k-h

Re: printk: queue wake_up_klogd irq_work only if per-CPU areas are ready

[Sergey Senozhatsky]

On (20/04/14 14:14), Greg KH wrote:
> On Tue, Apr 14, 2020 at 09:06:13PM +0900, Sergey Senozhatsky wrote:
> > Hello,
> > 
> > Commit ab6f762f0f53162d41 Linus' HEAD.
> > 
> > printk_deferred() does not make sure that it's safe to write to
> > per-CPU data, which causes problems when printk_deferred() is
> > invoked "too early", before per-CPU areas are initialized. There
> > are multiple bug reports, e.g.
> > https://bugzilla.kernel.org/show_bug.cgi?id=206847
> > 
> > 	-ss
> 
> So where do you want this commit backported to?

Well,  printk() is affected in all the kernels where
printk_deferred() relies on per-CPU data. Which may
translate to "pretty much all current stable kernels?"
This patch, however, uses printk_safe() bits, so it
won't apply on pre-printk_safe() kernels (not sure if
we have such -stable kernels though).

	-ss

Re: printk: queue wake_up_klogd irq_work only if per-CPU areas are ready

[Greg KH]

On Tue, Apr 14, 2020 at 10:13:09PM +0900, Sergey Senozhatsky wrote:
> On (20/04/14 14:14), Greg KH wrote:
> > On Tue, Apr 14, 2020 at 09:06:13PM +0900, Sergey Senozhatsky wrote:
> > > Hello,
> > > 
> > > Commit ab6f762f0f53162d41 Linus' HEAD.
> > > 
> > > printk_deferred() does not make sure that it's safe to write to
> > > per-CPU data, which causes problems when printk_deferred() is
> > > invoked "too early", before per-CPU areas are initialized. There
> > > are multiple bug reports, e.g.
> > > https://bugzilla.kernel.org/show_bug.cgi?id=206847
> > > 
> > > 	-ss
> > 
> > So where do you want this commit backported to?
> 
> Well,  printk() is affected in all the kernels where
> printk_deferred() relies on per-CPU data. Which may
> translate to "pretty much all current stable kernels?"
> This patch, however, uses printk_safe() bits, so it
> won't apply on pre-printk_safe() kernels (not sure if
> we have such -stable kernels though).

This is already in the 5.4.y tree, so I've added it to 4.19.y, but I
could not apply it to anything older (4.14.y, 4.9.y, 4.4.y) as it did
not apply there.

If you want it in those older kernels, and you think it is really
needed, please feel free to send a backported version for them.

thanks,

greg k-h