Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 19 Jul 2023 09:28:50 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Sandipan Das <sandipan.das@amd.com>
Cc:     linux-kernel@vger.kernel.org, linux-perf-users@vger.kernel.org,
        x86@kernel.org, mingo@redhat.com, acme@kernel.org,
        mark.rutland@arm.com, alexander.shishkin@linux.intel.com,
        jolsa@kernel.org, namhyung@kernel.org, irogers@google.com,
        adrian.hunter@intel.com, tglx@linutronix.de, bp@alien8.de,
        dave.hansen@linux.intel.com, hpa@zytor.com, eranian@google.com,
        ananth.narayan@amd.com, ravi.bangoria@amd.com,
        santosh.shukla@amd.com
Subject: Re: [PATCH 3/6] x86/cpuid: Add smp helper
Message-ID: <20230719072850.GA3529734@hirez.programming.kicks-ass.net>
References: <cover.1689748843.git.sandipan.das@amd.com>
 <827723d8f506411700c68bccc5072ec8d918d2de.1689748843.git.sandipan.das@amd.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <827723d8f506411700c68bccc5072ec8d918d2de.1689748843.git.sandipan.das@amd.com>
Precedence: bulk

On Wed, Jul 19, 2023 at 12:25:38PM +0530, Sandipan Das wrote:
> Depending on which CPU the CPUID instruction is executed, some leaves
> can report different values. There are cases where it may be required
> to know all possible values.
> 
> E.g. for AMD Zen 4 processors, the ActiveUmcMask field from leaf
> 0x80000022 ECX, which provides a way to determine the active memory
> controllers, can have different masks on CPUs belonging to different
> sockets as each socket can follow a different DIMM population scheme.
> Each memory channel is assigned a memory controller (UMC) and if no
> DIMMs are attached to a channel, the corresponding memory controller
> is inactive. There are performance monitoring counters exclusive to
> each memory controller which need to be represented under separate
> PMUs. So, it will be necessary to know the active memory controllers
> on each socket during the initialization of the UMC PMUs irrespective
> of where the uncore driver's module init runs.
> 
> Add a new helper that executes CPUID on a particular CPU and returns
> the EAX, EBX, ECX and EDX values.
> 

So I hate all this for multiple reasons:

 - the wohle foo_on_cpu() model generally leads to atrocious code that
   does multiple IPIs, I've seen rdmsr_on_cpu() followed by
   wrmsr_on_cpu() and worse things, just don't do this.

 - The whole CPUID thing is insane; we should read CPUID -- all of it --
   *ONCE* at bringup and thereafter never touch the instruction ever
   again. It could be people are already working on patches to this
   effect.

 - Different CPUID values for different CPUs is a pain :/