linux-kernel - Re: [PATCH v9 4/9] clocksource/drivers/arm_arch

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]

Date:	Thu, 28 Jul 2016 14:53:21 +0100
From:	Will Deacon <will.deacon@....com>
To:	Timur Tabi <timur@...eaurora.org>
Cc:	Fu Wei <fu.wei@...aro.org>,
	"Rafael J. Wysocki" <rjw@...ysocki.net>,
	Len Brown <lenb@...nel.org>,
	Daniel Lezcano <daniel.lezcano@...aro.org>,
	Thomas Gleixner <tglx@...utronix.de>,
	Marc Zyngier <marc.zyngier@....com>,
	Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
	Sudeep Holla <sudeep.holla@....com>,
	Hanjun Guo <hanjun.guo@...aro.org>,
	linux-arm-kernel@...ts.infradead.org,
	Linaro ACPI Mailman List <linaro-acpi@...ts.linaro.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	ACPI Devel Maling List <linux-acpi@...r.kernel.org>,
	rruigrok@...eaurora.org, harba@...eaurora.org,
	Christopher Covington <cov@...eaurora.org>,
	G Gregory <graeme.gregory@...aro.org>,
	Al Stone <al.stone@...aro.org>, Jon Masters <jcm@...hat.com>,
	wei@...hat.com, Arnd Bergmann <arnd@...db.de>,
	Wim Van Sebroeck <wim@...ana.be>,
	Catalin Marinas <catalin.marinas@....com>,
	Suravee Suthikulpanit <Suravee.Suthikulpanit@....com>,
	Leo Duran <leo.duran@....com>,
	Guenter Roeck <linux@...ck-us.net>,
	linux-watchdog@...r.kernel.org
Subject: Re: [PATCH v9 4/9] clocksource/drivers/arm_arch_timer: use readq to
 get 64-bit CNTVCT

On Tue, Jul 26, 2016 at 09:11:49AM -0500, Timur Tabi wrote:
> Will Deacon wrote:
> >The kernel really needs to support both of those platforms :/
> >
> >For the memory-mapped counter registers, the architecture says:
> >
> >   `If the implementation supports 64-bit atomic accesses, then the
> >    CNTV_CVAL register must be accessible as an atomic 64-bit value.'
> >
> >which is borderline tautological. If we take the generous reading that
> >this means AArch64 CPUs can use readq (and I'm not completely
> >comfortable with that assertion, particularly as you say that it breaks
> >the model), then you still need to use readq_relaxed here to avoid a
> >DSB. Furthermore, what are you going to do for AArch32? readq doesn't
> >exist over there, and if you use the generic implementation then it's
> >not atomic. In which case, we end up with the current code, as well as a
> >readq_relaxed guarded by a questionable #ifdef that is known to break a
> >supported platform for an unknown performance improvement. Hardly a big
> >win.
> 
> I know Fu dropped this patch, and I don't want to kick a dead horse, but I
> was wondering if it would be okay to do this:
> 
> static u64 arch_counter_get_cntvct_mem(void)
> {
> #ifdef readq_relaxed
> 	return readq_relaxed(arch_counter_base + CNTVCT_LO);
> #else
> 	u32 vct_lo, vct_hi, tmp_hi;
> 
> 	do {
> 		vct_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
> 		vct_lo = readl_relaxed(arch_counter_base + CNTVCT_LO);
> 		tmp_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
> 	} while (vct_hi != tmp_hi);
> 
> 	return ((u64) vct_hi << 32) | vct_lo;
> #endif
> }
> 
> readq and readq_relaxed are defined in arch/arm64/include/asm/io.h.  Why
> would the function exist if AArch64 CPUs can't use it?
> 
> Do we need something like ARCH_HAS_64BIT_ATOMIC_READ in order to decide
> whether readq is safe?

No, I'm still not ok with this. If you want to use readq_relaxed we need
the following guarantees:

  1. readq_relaxed is provided by the architecture
  2. readq_relaxed is single-copy atomic from the CPU's perspective
  3. The memory-mapped timer has been integrated in such a way that it
     can be accessed using 64-bit transactions.

(1) is easy, and you have that above. For (2), we just need to avoid
include/linux/io-64-nonatomic-*.h. (3), however, is not something we
can safely probe. If this optimisation really is worthwhile, then we
need to extend the device-tree binding for the counter so that we can
tell the kernel that it's ok to use 64-bit accesses for the counter
without tearing.

I have confirmed this with the architects here at ARM.

Will