linux-kernel - Re: [PATCH 5/6] perf_counter: kerneltop: mmap

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: <20090325123514.GB28639@elte.hu>
Date:	Wed, 25 Mar 2009 13:35:14 +0100
From:	Ingo Molnar <mingo@...e.hu>
To:	Peter Zijlstra <a.p.zijlstra@...llo.nl>
Cc:	linux-kernel@...r.kernel.org, Paul Mackerras <paulus@...ba.org>,
	Mike Galbraith <efault@....de>,
	Arjan van de Ven <arjan@...radead.org>,
	Wu Fengguang <fengguang.wu@...el.com>
Subject: Re: [PATCH 5/6] perf_counter: kerneltop: mmap_pages argument

* Peter Zijlstra <a.p.zijlstra@...llo.nl> wrote:

> On Wed, 2009-03-25 at 13:18 +0100, Ingo Molnar wrote:
> > * Peter Zijlstra <a.p.zijlstra@...llo.nl> wrote:
> > 
> > > provide a knob to set the number of mmap data pages.
> > 
> > > +	" -m pages  --mmap_pages=<pages> # number of mmap data pages\n"
> > 
> > Btw., we really want this to be auto-tuning to a large degree. If 
> > the kernel observes missed events, it should create a 
> > PERF_EVENT_OVERFLOW==0x3 record, with the number of missed events - 
> > or something like that.
> 
> Well, who's to say we ever see that overflow record if we're 
> having trouble tracking the output as is?

it would overwrite the last (new) record - so it's deterministic and 
the tail does not consume the head - just bites itself a bit.

But it would still be somewhat racy, if user-space _just_ managed to 
process those records ...

> How important is it for people to have accurate overflow 
> information other than the current -- we can't keep up -- kind?

it's somewhat important and could pave the way for the kernel to 
react to overflow more intelligently (via iterim buffering or 
whatever future mechanism).

It's also a general quality of implementation principle for kernel 
code: if we want to hide information we want to hide it from 
_user-space_, not the kernel. Hiding information from the kernel 
almost always causes trouble down the line.

> One possible solution is making the control page writable and 
> writing the userspace read position to it, then the kernel can, on 
> perf_output_begin() detect the overflow and count the number of 
> overwritten events.
> 
> This overflow count could then be published back into the control 
> page.

Ok, that's a nice idea - it keeps the amount of dirty cachelines 
minimal.

> TBH I'm not much of a fan, making all these pages writable just 
> opens a whole can of worms, and that accurate overflow tracking 
> will put more code in the output path.

What can of worms can you see there? (It would not be COW-ed - if 
you share those pages without knowing that they are shared then 
confused user-space will have to keep broken pieces of iteself.)

> Also, when mixing streams (events,mmap) is a single: you missed 
> 'n' events still good?

How would such mixing work? Multiple counters streaming into the 
same mmap area?

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/