Misaligned patch-site bcm2836_arm_irqchip_handle_irq+0x0/0x88

[Mark Rutland]

Hi Florian,

On Sat, Feb 25, 2023 at 07:26:50AM -0800, Florian Fainelli wrote:
> On 2/25/2023 2:37 AM, Marc Zyngier wrote:
> > On Fri, 24 Feb 2023 23:39:24 +0000,
> > Florian Fainelli <f.fainelli at gmail.com> wrote:
> > > 
> > > Hello,
> > > 
> > > You will find a complete boot log [1] below that shows the failure as
> > > well as the configuration file [2], however I am hitting what you see
> > > below as of v6.2-10217-ga93e884edf61. Have not tried linux-next yet,
> > > so if this is a known issue, please ignore :)
> > > 
> > > This is seen with on a Raspberry pi 4B running in AArch64 mode.
> > > 
> > > [1]: https://gist.github.com/b1a977b5fea837d180ccb800f091218a
> > > [2]: https://gist.github.com/64cf18b6ee770d58f64371e1db01afb9
> > 
> >  From the config file:
> > 
> > CONFIG_ARM=y
> > 
> > Can you please provide the config file for the build you're seeing
> > this on?
> 
> Yes, sorry about that, here is the 64-bit configuration:
> 
> https://gist.github.com/4eb83740c25b10f75b54560f8c8febb1

I think the problem here is:

  CONFIG_CC_OPTIMIZE_FOR_SIZE=y

... since GCC is known to ignore function alignment when -Os is passed.

We can have DYNANMIC_FTRACE_WITH_CALL_OPS depend on
!CONFIG_CC_OPTIMIZE_FOR_SIZE to prevent that.

That said, you have FTRACE and CC_OPTIMIZE_FOR_SIZE selected; are you actually
seeing any beenfit from CC_OPTIMIZE_FOR_SIZE?

> In case this is relevant the toolchain is using:
> 
> GNU ld (GNU Binutils) 2.38
> gcc version 11.3.0 (Buildroot 2022.08-rc2-21-g4cbf7336914f)

Thanks for this info; I'll go check the above with the kernel.org GCC 11.3.0
cross toolchain.

Thanks,
Mark.