Re: [PATCH] x86/msr: do not warn on writes to OC_MAILBOX

From: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
To: Borislav Petkov <bp@alien8.de>
Cc: Sultan Alsawaf <sultan@kerneltoast.com>,
	"Jason A. Donenfeld" <Jason@zx2c4.com>,
	kitsunyan <kitsunyan@airmail.cc>,
	"Brown, Len" <len.brown@intel.com>, X86 ML <x86@kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [PATCH] x86/msr: do not warn on writes to OC_MAILBOX
Date: Tue, 20 Oct 2020 11:40:28 -0700	[thread overview]
Message-ID: <2147d5853a5a3e0d753fcc0a4ed330f61b29aa83.camel@linux.intel.com> (raw)
In-Reply-To: <20201020174741.GJ11583@zn.tnic>

On Tue, 2020-10-20 at 19:47 +0200, Borislav Petkov wrote:
> On Tue, Oct 20, 2020 at 10:21:48AM -0700, Srinivas Pandruvada wrote:
> > These command id are model specific. There is no guarantee that
> > even
> > meaning changes. So I don't think we should write any code in
> > kernel
> > which can't stick.
> 
> Ok, is there a common *set* of values present on all models
Sorry, don't know.

> 
> A common set which we can abstract out from the MSR and have
> userspace
> write them into sysfs and the kernel does the model-specific write?
> 
> The sysfs interface should simply provide the functionality, like,
> for
> example say: "we have X valid undervolt indices, choose one".
> 
> Userspace doesn't have to deal with *how* that write happens and
> which
> bits need to be set in the MSR and depend on the model - that's all
> abstracted away by the kernel. All userspace needs to care about is
> *what* it wants done to the hw. The *how exactly* is done by the
> kernel.
> 
> And then the differences are done with x86 model tests.
> 
> Does that make more sense?
> 
> > May be something like this:
> > - Separate mailbox stuff from intel_turbo_max_3.c
> 
> Yah, that makes sense.
> 
> > - Create a standalone module which creates a debugfs interface
> > - This debugs interface takes one 64 bit value from user space and
> > use
> > protocol to avoid contention
> 
> We can't make debugfs an API - debugfs can change at any point in
> time.
> If you want an API, you put it in sysfs or in a separate fs.
Ok we can create a sysfs entry.

> 
> > - Warns users on writes via new interfaces you suggested above
> > > #define MSR_ADDR_TEMPERATURE 0x1a2
> > Need to check use case for undervolt.
> 
> throttled uses it too. I asked them today to talk to us to design a
> proper interface which satisfies their needs:
> 
> https://github.com/erpalma/throttled/issues/215
> 
> > > #define MSR_ADDR_UNITS 0x606
> > Why not reuse powercap rapl interface. That interface will take
> > care of
> > units.
> 
> Sure.
> 
> Btw, you should have a look at those tools - they all poke at all
> kinds
> of MSRs and correcting that is like a whack-a-mole game... ;-\
> 
> Oh, and the kernel pokes at them too so imagine the surprise one
> would have when
> some kernel driver like
> 
> drivers/thermal/intel/int340x_thermal/processor_thermal_device.c
> 
> went and read some MSRs and then all of a sudden they changed because
> some userspace daemon wrote them underneath it. Not good.
Agree, that poking MSR from user space is not a right thing to do.

Thanks,
Srinivas

> 
> Thx.
>