linux-kernel - Re: [PATCH 3/3] KVM: x86/vPMU: Add lazy mechanism to release perf

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <795f5e36-0211-154f-fcf0-f2f1771bf724@redhat.com>
Date:   Wed, 9 Oct 2019 11:21:30 +0200
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Like Xu <like.xu@...ux.intel.com>, kvm@...r.kernel.org,
        rkrcmar@...hat.com, sean.j.christopherson@...el.com,
        vkuznets@...hat.com, Jim Mattson <jmattson@...gle.com>,
        Ingo Molnar <mingo@...hat.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        ak@...ux.intel.com, wei.w.wang@...el.com, kan.liang@...el.com,
        like.xu@...el.com, ehankland@...gle.com, arbel.moshe@...cle.com,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/3] KVM: x86/vPMU: Add lazy mechanism to release
 perf_event per vPMC

On 09/10/19 10:16, Peter Zijlstra wrote:
> On Wed, Oct 09, 2019 at 09:15:03AM +0200, Paolo Bonzini wrote:
>> For stuff like hardware registers, bitfields are probably a bad idea
>> anyway, so let's only consider the case of space optimization.
> 
> Except for hardware registers? I actually like bitfields to describe
> hardware registers.

In theory yes, in practice for MMIO it's a problem that you're not able
to see the exact compiler reads or writes.  Of course you can do:

	union {
		struct {
			/* some bitfields here
		} u;
		u32 val;
	}

and only use the bitfields after reading/writing from the register.

> But worse, as used in the parent thread:
> 
> 	u8	count:7;
> 	bool	flag:1;
> 
> Who says the @flag thing will even be the msb of the initial u8 and not
> a whole new variable due to change in base type?

Good point.

>> bool bitfields preserve the magic behavior where something like this:
>>
>>   foo->x = y;
>>
>> (x is a bool bitfield) would be compiled as
>>
>>   foo->x = (y != 0);
> 
> This is confusion; if y is a single bit bitfield, then there is
> absolutely _NO_ difference between these two expressions.

y is not in a struct so it cannot be a single bit bitfield. :) If y is
an int and foo->x is a bool bitfield, you get the following:

	foo->x = 6;	/* foo->x is 1, it would be 0 for int:1 */
	foo->x = 7;	/* foo->x is 1, it would be 1 for int:1 */

Anyway it's good that we agree on the important thing about the patch!

Paolo

> The _only_ thing about _Bool is that it magically casts values to 0,1.
> Single bit bitfield variables have no choice but to already be in that
> range.