Re: [PATCH 1/2] x86/msr: add on cpu read/modify/write function

[Borislav Petkov <bp@alien8.de>]

On Sun, Dec 20, 2015 at 02:28:48PM +0100, Thomas Gleixner wrote:
> So this is a copy of the above !SMP inline. What's wrong with providing:
> 
>  int rmwmsrl_safe(msr_no, clear_mask, set_mask)
> 
> in x86/lib/msr.c and make the !SMP variant of rdmsrl_safe_on_cpu() and that
> variant for the SMP case a simple wrapper around it?
> 
> static void remote_rmwmsrl_safe(void *info)
> {
> 	struct msr_action *ma = info;
> 
> 	return rmwmsrl_safe(ma->msr, ma->clear_mask, ma->set_mask);
> }
> 
> No gotos, no pointless code duplication. Just simple.

TBH, I find this new "rmwmsrl" interface (the name is unreadable, btw)
silly:

It provides a plain read-modify-write on a MSR and nothing more but
patch 2 immediately shows that this interface is insufficient for the
other cases, i.e. package_power_limit_irq_save() for example, where you
need to do something more like check bits or error handling.

So there we do smp_call_function_single() with a function which does the
MSR accesses and whatever else is needed.

So why add the former interface in the first place?

Having driver-specific functions do whatever it is required and then
using a single IPI to run them is much cleaner than adding that
unfortunate function which doesn't really suffice.

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.