[Letux-kernel] jz4730-i2c - clocksource

[Paul Boddie]

On Saturday, 6 March 2021 18:12:48 CET H. Nikolaus Schaller wrote:
> 
> Now there is also letux-current which is the latest complete tree that I did
> build and test.

So, I built the letux-current kernel and hopefully installed all the modules 
in a fresh Debian Stretch root ext2-formatted filesystem, with the uImage in 
its normal place on the primary FAT-formatted partition. Unfortunately, I 
still don't get anything shown on the screen whose backlight does come on (as 
usual).

[...]

> But what would happen if cpu_has_llsc is wrongly defined as "1"? Shouldn't
> this trigger some ILLEGAL INSTRUCTION fault in kernel and perhaps panic
> the kernel?

I imagine that it should be trapped and even emulated at some level. That 
might not seem to make much sense, given that the aim of using LL/SC is to 
avoid expensive transitions to kernel space, but if a program uses it then 
there has to be a way of supporting it.

The futex implementation seems to return an error code if it is invoked 
without LL/SC support, so the kernel presumably does not use the instruction 
if it is missing. However, if it does not know it is missing, maybe some 
emulation occurs that ends up being rather expensive within the kernel itself.

> > So it is possible that LL/SC support is missing and that this flag might
> > be unset for better performance. But that is just an idea.
> 
> We should give it a try...

If I can get the kernel working, then I might see what effect it has.

Paul