Date: Fri, 14 Sep 2018 11:06:53 -0600 From: Khalid Aziz <khalid.aziz@...cle.com> To: Julian Stecklina <jsteckli@...zon.de>, Linus Torvalds <torvalds@...ux-foundation.org> Cc: David Woodhouse <dwmw@...zon.co.uk>, Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>, juerg.haefliger@....com, deepa.srinivasan@...cle.com, Jim Mattson <jmattson@...gle.com>, Andrew Cooper <andrew.cooper3@...rix.com>, Linux Kernel Mailing List <linux-kernel@...r.kernel.org>, Boris Ostrovsky <boris.ostrovsky@...cle.com>, linux-mm <linux-mm@...ck.org>, Thomas Gleixner <tglx@...utronix.de>, joao.m.martins@...cle.com, pradeep.vincent@...cle.com, Andi Kleen <ak@...ux.intel.com>, kanth.ghatraju@...cle.com, Liran Alon <liran.alon@...cle.com>, Kees Cook <keescook@...gle.com>, Kernel Hardening <kernel-hardening@...ts.openwall.com>, chris.hyser@...cle.com, Tyler Hicks <tyhicks@...onical.com>, John Haxby <john.haxby@...cle.com>, Jon Masters <jcm@...hat.com> Subject: Re: Redoing eXclusive Page Frame Ownership (XPFO) with isolated CPUs in mind (for KVM to isolate its guests per CPU) On 09/12/2018 09:37 AM, Julian Stecklina wrote: > Julian Stecklina <jsteckli@...zon.de> writes: > >> Linus Torvalds <torvalds@...ux-foundation.org> writes: >> >>> On Fri, Aug 31, 2018 at 12:45 AM Julian Stecklina <jsteckli@...zon.de> wrote: >>>> >>>> I've been spending some cycles on the XPFO patch set this week. For the >>>> patch set as it was posted for v4.13, the performance overhead of >>>> compiling a Linux kernel is ~40% on x86_64. The overhead comes almost >>>> completely from TLB flushing. If we can live with stale TLB entries >>>> allowing temporary access (which I think is reasonable), we can remove >>>> all TLB flushing (on x86). This reduces the overhead to 2-3% for >>>> kernel compile. >>> >>> I have to say, even 2-3% for a kernel compile sounds absolutely horrendous. >> >> Well, it's at least in a range where it doesn't look hopeless. >> >>> Kernel bullds are 90% user space at least for me, so a 2-3% slowdown >>> from a kernel is not some small unnoticeable thing. >> >> The overhead seems to come from the hooks that XPFO adds to >> alloc/free_pages. These hooks add a couple of atomic operations per >> allocated (4K) page for book keeping. Some of these atomic ops are only >> for debugging and could be removed. There is also some opportunity to >> streamline the per-page space overhead of XPFO. > > I've updated my XPFO branch to make some of the debugging optional > and also integrated the XPFO bookkeeping with struct page, instead of > requiring CONFIG_PAGE_EXTENSION, which removes some checks in the hot > path. These changes push the overhead down to somewhere between 1.5 and > 2% for my quad core box in kernel compile. This is close to the > measurement noise, so I take suggestions for a better benchmark here. > > Of course, if you hit contention on the xpfo spinlock then performance > will suffer. I guess this is what happened on Khalid's large box. > > I'll try to remove the spinlocks and add fixup code to the pagefault > handler to see whether this improves the situation on large boxes. This > might turn out to be ugly, though. > Hi Julian, I ran tests with your updated code and gathered lock statistics. Change in system time for "make -j60" was in the noise margin (It actually went up by about 2%). There is some contention on xpfo_lock. Average wait time does not look high compared to other locks. Max hold time looks a little long. From /proc/lock_stat: &(&page->xpfo_lock)->rlock: 29698 29897 0.06 134.39 15345.58 0.51 422474670 960222532 0.05 30362.05 195807002.62 0.20 Nevertheless even a smaller average wait time can add up. -- Khalid
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.