Thread (170 messages) 170 messages, 19 authors, 2012-09-16

Re: [PATCH v2 28/31] arm64: Generic timers support

From: Linus Walleij <hidden>
Date: 2012-08-16 12:40:09
Also in: linux-arm-kernel, lkml

On Tue, Aug 14, 2012 at 7:52 PM, Catalin Marinas
[off-list ref] wrote:
From: Marc Zyngier <redacted>
(...)
+static void __init arch_timer_calibrate(void)
I think you wrot in the last review thread that this should be renamed
"arch_timer_get_freq()".
+{
+       if (arch_timer_rate == 0) {
+               arch_timer_reg_write(ARCH_TIMER_REG_CTRL, 0);
+               arch_timer_rate = arch_timer_reg_read(ARCH_TIMER_REG_FREQ);
+
+               /* Check the timer frequency. */
+               if (arch_timer_rate == 0)
+                       panic("Architected timer frequency is set to zero.\n"
+                             "You must set this in your .dts file\n");
This comment about the .dts file is completely out-of-place.
The error message should be about the register containing 0, right?

Also, at this very point in the code, I think you should convert that panic()
to pr_err() and insert:

#include <linux/clk.h>

struct clk *clk;

clk = clk_get_sys("arm_generic_timer", NULL);
if (IS_ERR(clk))
               panic("Not even a clk to get freq off! Giving up.\n");
err = clk_prepare_enable(clk);
if (err) {
           pr_err("smp_twd: clock failed to prepare: %d\n", err);
           clk_put(clk);
           panic("Not even a clk to get freq off! Giving up.\n");
}
arch_timer_rate = clk_get_rate(clk);

Possibly the clk should even be able to override the register value,
then it's the other way around.
+       }
+
+       /* Cache the sched_clock multiplier to save a divide in the hot path. */
+
+       sched_clock_mult = NSEC_PER_SEC / arch_timer_rate;
+
+       pr_info("Architected local timer running at %u.%02uMHz.\n",
+                arch_timer_rate / 1000000, (arch_timer_rate / 10000) % 100);
This multiplier and print should be moved below the call site since
the DT frequency
overrides it and then you get no print which is sad, and incorrect
sched_clock_mult
which is a bug.
+}
(...)
+       /* Try to determine the frequency from the device tree or CNTFRQ */
+       if (!of_property_read_u32(np, "clock-frequency", &freq))
+               arch_timer_rate = freq;
+       arch_timer_calibrate();
rename arch_timer_get_freq()

And move that print here.

Yours,
Linus Walleij
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help