| Message ID | 4CF7A5C4.6010303@ge.com |
|---|---|
| State | Superseded |
| Delegated to: | Sandeep Paulraj |
| Headers | show |
Hi Nick, On Thu, Dec 2, 2010 at 8:57 AM, Nick Thompson <nick.thompson@ge.com> wrote: > This change allows the davinci timer functions to be used before > relocation since it avoids using static variables prior to BSS being > made available. > > The code is based on that used in the at91 timers, modified to use > a davinci specific hardware timer. It also maintains reset_timer() > to allow deprecated timer usage to continue to work (for example, > in nand_base.c) > > Signed-off-by: Nick Thompson <nick.thompson@ge.com> > --- > This patch relies on http://patchwork.ozlabs.org/patch/73758/ which has not been > accepted yet. I have made a comment on it, suggesting that it should be changed: > http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/90040/focus=90162 > > arch/arm/cpu/arm926ejs/davinci/timer.c | 77 ++++++++++++-------------------- > 1 files changed, 28 insertions(+), 49 deletions(-) Applies cleanly to v2010.12-rc2 on top of http://patchwork.ozlabs.org/patch/73758/ Tested on da850evm w/o any config changes. This fixes the freeze on boot: 10/10 times. You're awesome, Nick! Tested-by: Ben Gardiner <bengardiner@nanometrics.ca> > diff --git a/arch/arm/cpu/arm926ejs/davinci/timer.c b/arch/arm/cpu/arm926ejs/davinci/timer.c > [...] > void __udelay(unsigned long usec) > { > - ulong tmo; > - ulong endtime; > - signed long diff; > - > - tmo = CONFIG_SYS_HZ_CLOCK / 1000; > - tmo *= usec; > - tmo /= (1000 * TIM_CLK_DIV); > - > - endtime = get_timer_raw() + tmo; > + unsigned long long endtime; > > - do { > - ulong now = get_timer_raw(); > - diff = endtime - now; > - } while (diff >= 0); > -} > + endtime = ((unsigned long long )usec * gd->timer_rate_hz) / 1000000UL; just one minor checkpatch.pl warning; should be (unsigned long long) -- no space before ')' Best Regards, Ben Gardiner --- Nanometrics Inc. http://www.nanometrics.ca
On 02/12/10 14:18, Ben Gardiner wrote: > Hi Nick, > > On Thu, Dec 2, 2010 at 8:57 AM, Nick Thompson <nick.thompson@ge.com> wrote: >> This change allows the davinci timer functions to be used before >> relocation since it avoids using static variables prior to BSS being >> made available. >> Signed-off-by: Nick Thompson <nick.thompson@ge.com> >> --- >> This patch relies on http://patchwork.ozlabs.org/patch/73758/ which has not been >> accepted yet. I have made a comment on it, suggesting that it should be changed: >> http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/90040/focus=90162 > > Applies cleanly to v2010.12-rc2 on top of > http://patchwork.ozlabs.org/patch/73758/ > > Tested on da850evm w/o any config changes. This fixes the freeze on > boot: 10/10 times. > > You're awesome, Nick! > > Tested-by: Ben Gardiner <bengardiner@nanometrics.ca> I thought it sounded like the problem you where having. Glad it helps. Thanks for the Tested-by. >> + endtime = ((unsigned long long )usec * gd->timer_rate_hz) / 1000000UL; > > just one minor checkpatch.pl warning; should be (unsigned long long) > -- no space before ')' Good spot. I've fixed that locally and will submit a v2 if the depending patch finds favour - Alternative approaches have been suggested, this one is *my* short-term favourite. Thanks, Nick.
Hi Nick, On Thu, Dec 02, 2010 at 19:27:24, Nick Thompson wrote: > This change allows the davinci timer functions to be used before > relocation since it avoids using static variables prior to BSS being > made available. > > The code is based on that used in the at91 timers, modified to use > a davinci specific hardware timer. It also maintains reset_timer() > to allow deprecated timer usage to continue to work (for example, > in nand_base.c) > > Signed-off-by: Nick Thompson <nick.thompson@ge.com> > --- > This patch relies on http://patchwork.ozlabs.org/patch/73758/ which has not been > accepted yet. I have made a comment on it, suggesting that it should be changed: > http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/90040/focus=90162 > > arch/arm/cpu/arm926ejs/davinci/timer.c | 77 ++++++++++++-------------------- > 1 files changed, 28 insertions(+), 49 deletions(-) > I applied this patch on top of http://patchwork.ozlabs.org/patch/73758/ and tested on DM365 without any config changes. Now the board boots fine without any hang. Tested-by: Sudhakar Rajashekhara <sudhakar.raj@ti.com> Thanks, Sudhakar
diff --git a/arch/arm/cpu/arm926ejs/davinci/timer.c b/arch/arm/cpu/arm926ejs/davinci/timer.c index 9da7443..bfc6b40 100644 --- a/arch/arm/cpu/arm926ejs/davinci/timer.c +++ b/arch/arm/cpu/arm926ejs/davinci/timer.c @@ -40,6 +40,8 @@ #include <common.h> #include <asm/io.h> +DECLARE_GLOBAL_DATA_PTR; + struct davinci_timer { u_int32_t pid12; u_int32_t emumgt; @@ -57,11 +59,9 @@ struct davinci_timer { static struct davinci_timer * const timer = (struct davinci_timer *)CONFIG_SYS_TIMERBASE; -#define TIMER_LOAD_VAL (CONFIG_SYS_HZ_CLOCK / CONFIG_SYS_HZ) -#define TIM_CLK_DIV 16 +#define TIMER_LOAD_VAL 0xffffffff -static ulong timestamp; -static ulong lastinc; +#define TIM_CLK_DIV 16 int timer_init(void) { @@ -71,72 +71,51 @@ int timer_init(void) writel(0x06 | ((TIM_CLK_DIV - 1) << 8), &timer->tgcr); writel(0x0, &timer->tim34); writel(TIMER_LOAD_VAL, &timer->prd34); - lastinc = 0; - timestamp = 0; writel(2 << 22, &timer->tcr); + gd->timer_rate_hz = CONFIG_SYS_HZ_CLOCK / TIM_CLK_DIV; + gd->timer_reset_value = 0; return(0); } void reset_timer(void) { - writel(0x0, &timer->tcr); - writel(0x0, &timer->tim34); - lastinc = 0; - timestamp = 0; - writel(2 << 22, &timer->tcr); + gd->timer_reset_value = get_ticks(); } -static ulong get_timer_raw(void) +/* + * Get the current 64 bit timer tick count + */ +unsigned long long get_ticks(void) { - ulong now = readl(&timer->tim34); - - if (now >= lastinc) { - /* normal mode */ - timestamp += now - lastinc; - } else { - /* overflow ... */ - timestamp += now + TIMER_LOAD_VAL - lastinc; - } - lastinc = now; - return timestamp; + unsigned long now = readl(&timer->tim34); + + /* increment tbu if tbl has rolled over */ + if (now < gd->tbl) + gd->tbu++; + gd->tbl = now; + + return (((unsigned long long)gd->tbu) << 32) | gd->tbl; } ulong get_timer(ulong base) { - return((get_timer_raw() / (TIMER_LOAD_VAL / TIM_CLK_DIV)) - base); -} + unsigned long long timer_diff; -void set_timer(ulong t) -{ - timestamp = t; + timer_diff = get_ticks() - gd->timer_reset_value; + + return (timer_diff / (gd->timer_rate_hz / CONFIG_SYS_HZ)) - base; } void __udelay(unsigned long usec) { - ulong tmo; - ulong endtime; - signed long diff; - - tmo = CONFIG_SYS_HZ_CLOCK / 1000; - tmo *= usec; - tmo /= (1000 * TIM_CLK_DIV); - - endtime = get_timer_raw() + tmo; + unsigned long long endtime; - do { - ulong now = get_timer_raw(); - diff = endtime - now; - } while (diff >= 0); -} + endtime = ((unsigned long long )usec * gd->timer_rate_hz) / 1000000UL; + endtime += get_ticks(); -/* - * This function is derived from PowerPC code (read timebase as long long). - * On ARM it just returns the timer value. - */ -unsigned long long get_ticks(void) -{ - return(get_timer(0)); + while (get_ticks() < endtime) + ; } /*
This change allows the davinci timer functions to be used before relocation since it avoids using static variables prior to BSS being made available. The code is based on that used in the at91 timers, modified to use a davinci specific hardware timer. It also maintains reset_timer() to allow deprecated timer usage to continue to work (for example, in nand_base.c) Signed-off-by: Nick Thompson <nick.thompson@ge.com> --- This patch relies on http://patchwork.ozlabs.org/patch/73758/ which has not been accepted yet. I have made a comment on it, suggesting that it should be changed: http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/90040/focus=90162 arch/arm/cpu/arm926ejs/davinci/timer.c | 77 ++++++++++++-------------------- 1 files changed, 28 insertions(+), 49 deletions(-)