Re: [PATCH RFC] x86/smpboot: Set safer __max_logical_packages limit

From: Boris Ostrovsky <boris.ostrovsky@oracle.com>
To: Peter Zijlstra <peterz@infradead.org>,
	Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Prarit Bhargava <prarit@redhat.com>,
	Juergen Gross <jgross@suse.com>,
	x86@kernel.org, linux-kernel@vger.kernel.org,
	Ingo Molnar <mingo@redhat.com>, "H. Peter Anvin" <hpa@zytor.com>,
	xen-devel@lists.xenproject.org,
	Thomas Gleixner <tglx@linutronix.de>,
	Borislav Petkov <bp@suse.de>
Subject: Re: [PATCH RFC] x86/smpboot: Set safer __max_logical_packages limit
Date: Thu, 20 Apr 2017 13:09:43 -0400	[thread overview]
Message-ID: <a7b88190-8250-df76-70fc-f64beebb33eb__38570.2980951921$1492708290$gmane$org@oracle.com> (raw)
In-Reply-To: <20170420161503.vr7nc2czjht7g2v3@hirez.programming.kicks-ass.net>

On 04/20/2017 12:15 PM, Peter Zijlstra wrote:
> On Thu, Apr 20, 2017 at 05:40:37PM +0200, Vitaly Kuznetsov wrote:
>>> This is getting ludicrous. Xen is plain broken, and instead of fixing
>>> it, you propose to somehow deal with its obviously crack induced
>>> behaviour :-(
>> Totally agree and I don't like the solution I propose (and that's why
>> this is RFC)... The problem is that there are such Xen setups in the
>> wild and with the recent changes some guests will BUG() :-(
>>
>> Alternatively, we can just remove the BUG() and do something with CPUs
>> which have their pkg >= __max_logical_packages, e.g. assign them to the
>> last package. Far from ideal but will help to avoid the regression.
> So currently none of the stuff that uses this should appear in Xen. Its
> all drivers for hardware that isn't virtualized (afaik). So assigning to
> the last package 'works'.
>
> But the moment this ends up getting used that explodes, because we'll
> need different object instances for each piece of hardware.

This already gets used. I don't remember details but we had to fix
something due to RAPL code referencing topology info (and yes, there is
no reason for a guest to use RAPL, but we shouldn't crash neither)

>
> There just isn't a good solution; on the one hand the BIOS is prone to
> providing crap numbers, on the other hand virt (esp. Xen as it turns
> out) provides absolutely bonkers/inconsistent topology information.
>
> Very frustrating :-/
>

So we might need a way to bypass topology discovery and present some
sort of default topology (single package, or 1 CPU per package, for
example) and ignore APICID and all that.

-boris

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel