linux-kernel - Re: [PATCH v4 03/11] mm/gup: Applies counting method to monitor gup_pgd

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>] [day] [month] [year] [list]

Message-ID: <1f5f2f689385ceeb1240b4cc66ef3f4b66638ab0.camel@linux.ibm.com>
Date:   Tue, 01 Oct 2019 16:40:59 -0300
From:   Leonardo Bras <leonardo@...ux.ibm.com>
To:     John Hubbard <jhubbard@...dia.com>, linuxppc-dev@...ts.ozlabs.org,
        linux-kernel@...r.kernel.org, kvm-ppc@...r.kernel.org,
        linux-arch@...r.kernel.org, linux-mm@...ck.org
Cc:     Arnd Bergmann <arnd@...db.de>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        YueHaibing <yuehaibing@...wei.com>,
        Nicholas Piggin <npiggin@...il.com>,
        Mike Rapoport <rppt@...ux.ibm.com>,
        Keith Busch <keith.busch@...el.com>,
        Jason Gunthorpe <jgg@...pe.ca>,
        Paul Mackerras <paulus@...ba.org>,
        "Aneesh Kumar K.V" <aneesh.kumar@...ux.ibm.com>,
        Allison Randal <allison@...utok.net>,
        Mahesh Salgaonkar <mahesh@...ux.vnet.ibm.com>,
        Ganesh Goudar <ganeshgr@...ux.ibm.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ira Weiny <ira.weiny@...el.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Dan Williams <dan.j.williams@...el.com>
Subject: Re: [PATCH v4 03/11] mm/gup: Applies counting method to monitor
 gup_pgd_range

On Tue, 2019-10-01 at 12:04 -0700, John Hubbard wrote:
> On 10/1/19 10:56 AM, Leonardo Bras wrote:
> > On Mon, 2019-09-30 at 14:51 -0700, John Hubbard wrote:
> > > On 9/27/19 4:40 PM, Leonardo Bras wrote:
> ...
> > > > diff --git a/mm/gup.c b/mm/gup.c
> > > > index 98f13ab37bac..7105c829cf44 100644
> > > > --- a/mm/gup.c
> > > > +++ b/mm/gup.c
> > > > @@ -2325,6 +2325,7 @@ static bool gup_fast_permitted(unsigned long start, unsigned long end)
> > > >  int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
> > > >  			  struct page **pages)
> > > >  {
> > > > +	struct mm_struct *mm;
> > > 
> > > I don't think that this local variable adds any value, so let's not use it.
> > > Similar point in a few other patches too.
> > 
> > It avoids 1 deference of current->mm, it's a little performance gain.
> > 
> 
> No, it isn't. :) 
> 
> Longer answer: at this level (by which I mean, "wrote the C code, haven't looked
> at the generated asm yet, and haven't done a direct perf test yet"), none of us
> C programmers are entitled to imagine that we can second guess both the compiler 
> and the CPU well enough to claim that  declaring a local pointer variable on the
> stack will even *affect* performance, much less know which way it will go!
> 

I did this based on how costly can be 'current', and I could notice
reduction in assembly size most of the time. (powerpc)
But I get what you mean, maybe the (possible) performance gain don't
worth the extra work.

> The compiler at -O2 will *absolutely* optimize away any local variables that
> it doesn't need.
> 
> And that leads to how kernel programmers routinely decide about that kind of 
> variable: "does the variable's added clarity compensate for the extra visual 
> noise and for the need to manage the variable?"

That's a good way to decide it. :)

> 
> Here, and in most (all?) other points in the patchset where you've added an
> mm local variable, the answer is no.
> 

Well, IMHO it's cleaner that way. But I get that other people may
disagree. 

> 
> ...	start_lockless_pgtbl_walk(mm);
> > > Minor: I'd like to rename this register_lockless_pgtable_walker().
> > > 
> > > >  		local_irq_disable();
> > > >  		gup_pgd_range(addr, end, gup_flags, pages, &nr);
> > > >  		local_irq_enable();
> > > > +		end_lockless_pgtbl_walk(mm);
> > > 
> > > ...and deregister_lockless_pgtable_walker().
> > > 
> > 
> > I have no problem changing the name, but I don't register/deregister
> > are good terms for this. 
> > 
> > I would rather use start/finish, begin/end, and so on. Register sounds
> > like something more complicated than what we are trying to achieve
> > here. 
> > 
> 
> OK, well, I don't want to bikeshed on naming more than I usually do, and 
> what you have is reasonable, so I'll leave that alone. :)
> 
> thanks,

Thank for the feedback,


Download attachment "signature.asc" of type "application/pgp-signature" (834 bytes)