Message-ID: <20161011202850.GK19318@brightrain.aerifal.cx>
Date:   Tue, 11 Oct 2016 16:28:50 -0400
From:   Rich Felker <dalias@...c.org>
To:     Daniel Lezcano <daniel.lezcano@...aro.org>
Cc:     devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-sh@...r.kernel.org, Rob Herring <robh+dt@...nel.org>,
        Mark Rutland <mark.rutland@....com>,
        Thomas Gleixner <tglx@...utronix.de>,
        "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
Subject: Re: [PATCH v8 2/2] clocksource: add J-Core timer/clocksource driver

On Tue, Oct 11, 2016 at 08:18:12PM +0200, Daniel Lezcano wrote:
> 
> Hi Rich,
> 
> On Sun, Oct 09, 2016 at 05:34:22AM +0000, Rich Felker wrote:
> > At the hardware level, the J-Core PIT is integrated with the interrupt
> > controller, but it is represented as its own device and has an
> > independent programming interface. It provides a 12-bit countdown
> > timer, which is not presently used, and a periodic timer. The interval
> > length for the latter is programmable via a 32-bit throttle register
> > whose units are determined by a bus-period register. The periodic
> > timer is used to implement both periodic and oneshot clock event
> > modes; in oneshot mode the interrupt handler simply disables the timer
> > as soon as it fires.
> > 
> > Despite its device tree node representing an interrupt for the PIT,
> > the actual irq generated is programmable, not hard-wired. The driver
> > is responsible for programming the PIT to generate the hardware irq
> > number that the DT assigns to it.
> > 
> > On SMP configurations, J-Core provides cpu-local instances of the PIT;
> > no broadcast timer is needed. This driver supports the creation of the
> > necessary per-cpu clock_event_device instances.
> 
> For my personnal information, why no broadcast timer is needed ?

Broadcast timer is only needed if you don't have percpu local timers.
Early on in SMP development I actually tested with an ipi broadcast
timer and performance was noticably worse.

> Are the CPUs on always-on power down ?

For now they are always on and don't even have the sleep instruction
(i.e. stop cpu clock until interrupt) implemented. Adding sleep will
be the first power-saving step, and perhaps the only one for now,
since there doesn't seem to be any indication (according to the ppl
working on the hardware) that a deeper sleep would provide significant
additional savings.

> > A nanosecond-resolution clocksource is provided using the J-Core "RTC"
> > registers, which give a 64-bit seconds count and 32-bit nanoseconds
> > that wrap every second. The driver converts these to a full-range
> > 32-bit nanoseconds count.
> > 
> > Signed-off-by: Rich Felker <dalias@...c.org>
> > ---
> >  drivers/clocksource/Kconfig     |  10 ++
> >  drivers/clocksource/Makefile    |   1 +
> >  drivers/clocksource/jcore-pit.c | 231 ++++++++++++++++++++++++++++++++++++++++
> >  include/linux/cpuhotplug.h      |   1 +
> >  4 files changed, 243 insertions(+)
> >  create mode 100644 drivers/clocksource/jcore-pit.c
> > 
> > diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig
> > index 5677886..95dd78b 100644
> > --- a/drivers/clocksource/Kconfig
> > +++ b/drivers/clocksource/Kconfig
> > @@ -407,6 +407,16 @@ config SYS_SUPPORTS_SH_TMU
> >  config SYS_SUPPORTS_EM_STI
> >          bool
> >  
> > +config CLKSRC_JCORE_PIT
> > +	bool "J-Core PIT timer driver"
> > +	depends on OF && (SUPERH || COMPILE_TEST)
> 
> Actually the idea is to have the SUPERH to select this timer, not create
> a dependency on SUPERH from here.
> 
> We don't want to prompt in the configuration menu the drivers because it
> would be impossible to anyone to know which timer comes with which
> hardware, so we let the platform to select the timer it needs.

I thought we discussed this before. For users building a kernel for
legacy SH systems, especially in the current state where they're only
supported with hard-coded board files rather than device tree, it
makes no sense to build drivers for J-core hardware. It would make
sense to be on by default for CONFIG_SH_DEVICE_TREE with a compatible
CPU selection, but at least at this time, not for SUPERH in general.

Anyway I'd really like to do this non-invasively as long as we have a
mix of legacy and new stuff and the legacy stuff is not readily
testable. Once all of arch/sh is moved over to device tree, could we
revisit this and make all the drivers follow a common policy (on by
default if they're associated with boards/SoCs using a matching or
compatible CPU model, or something like that, but still able to be
disabled manually, since the user might be trying to get a tiny-ish
embedded kernel)?

> The exception is to enable in order to compile on non-native platform to
> compile-test a bunch of drivers (eg. compile most of the clocksource / 
> clockevents drivers on a x86 big server).
> 
> That's why we should have:
> 
> config CLKSRC_JCORE_PIT
> 	bool "J-Core PIT timer driver" if COMPILE_TEST
> 
> So the jcore pit driver option appears only if compile test is enabled
> otherwise it is a silent option and the user doesn't have to deal with
> it. Having consistent dependency like the other drivers will help future
> changes to consolidate the Kconfig.

I don't think this matches the user expectation for the arch yet. For
now we have a j2_defconfig that enables the drivers and other kernel
settings expected to be useful for J-core socs. I want to modernize
this all but that's a separate project.

> [ ... ]              
> 
> > +#define REG_PITEN		0x00
> > +#define REG_THROT		0x10
> > +#define REG_COUNT		0x14
> > +#define REG_BUSPD		0x18
> > +#define REG_SECHI		0x20
> > +#define REG_SECLO		0x24
> > +#define REG_NSEC		0x28
> > +
> > +struct jcore_pit {
> > +	struct clock_event_device	ced;
> > +	__iomem void			*base;
> 
> It is not '__iomem void *' but 'void __iomem *'. This appears multiple
> times in this code.

OK, I'll change that.

> > +	unsigned long			periodic_delta;
> > +	unsigned			cpu;
> 
> This field is pointless, 'cpu' is only used for a trace in the init
> function which has already the cpu has parameter.

OK, will remove. It was needed for the old notify framework but not
for cpu hotplug framework, I think.

> > +	u32				enable_val;
> > +};
> > +
> > +static __iomem void		*jcore_pit_base;
> 
> static void __iomem *jcore_pit_base;

OK.

> > +struct jcore_pit __percpu	*jcore_pit_percpu;
> 
> static.

OK.

> [ ... ]
> 
> > +static int __init jcore_pit_init(struct device_node *node)
> > +{
> 
> [ ... ]

Was there something analogous you wanted me to change here, or just
leftover quoting?

> > +	/*
> > +	 * The J-Core PIT is not hard-wired to a particular IRQ, but
> > +	 * integrated with the interrupt controller such that the IRQ it
> > +	 * generates is programmable. The programming interface has a
> > +	 * legacy field which was an interrupt priority for AIC1, but
> > +	 * which is OR'd onto bits 2-5 of the generated IRQ number when
> > +	 * used with J-Core AIC2, so set it to match these bits.
> > +	 */
> > +	hwirq = irq_get_irq_data(pit_irq)->hwirq;
> > +	irqprio = (hwirq >> 2) & PIT_PRIO_MASK;
> > +	enable_val = (1U << PIT_ENABLE_SHIFT)
> > +		   | (hwirq << PIT_IRQ_SHIFT)
> > +		   | (irqprio << PIT_PRIO_SHIFT);
> > +
> 
> Why mention AIC1 if there is not test to check if AIC1 || AIC2 ?
> 
> Will be the same information available if the irqchip is AIC1 ?

The bit layout of the PIT enable register is:

	.....e..ppppiiiiiiii............

where the .'s indicate unrelated/unused bits, e is enable, p is
priority, and i is hard irq number.

For the PIT included in AIC1 (obsolete but still in use), any hard irq
(trap number) can be programmed via the 8 iiiiiiii bits, and a
priority (0-15) is programmable separately in the pppp bits.

For the PIT included in AIC2 (current), the programming interface is
equivalent modulo interrupt mapping. This is why a different
compatible tag was not used. However only traps 64-127 (the ones
actually intended to be used for interrupts, rather than
syscalls/exceptions/etc.) can be programmed (the high 2 bits of i are
ignored) and the priority pppp is <<2'd and or'd onto the irq number.
This was a poor decision made on the hardware engineering side based
on a wrong assumption that preserving old priority mapping of outdated
software was important, whereas priorities weren't/aren't even being
used.

When we do the next round of interrupt controller improvements (AIC3)
the PIT programming interface should remain compatible with the
driver; likely the priority bits will just be ignored.

If we do want to change the programming interface beyond this at some
point (that maay be a good idea, since we have identified several
things that are less than ideal for Linux, like the sechi/seclo/ns
clocksource), a new compatible tag will be added instead.

> I have to admin I find strange this driver has to invoke irq_get_irq_data(),
> it is the only one and it sounds even strange it has to be stored per cpu below.

The timer will not actually generate the irq it's assigned to (or any
interrupt at all) unless/until it's programmed for the (hard) irq
number. The need to use irq_get_irq_data comes from the fact that the
hardware needs the hard irq number, not a virq number, which could in
principle be different.

There's probably some argument that could have been made that the
irqchip and clocksource/timer driver should have been unified since
they're unified in the hardware and have this awkward programming
relationship, but that didn't fit the Linux model very well and I
think having them factored like this encourages future versions of the
hardware to adapt to the model the software wants.

Rich

Powered by blists - more mailing lists