[PATCH 03/11] ARM: imx: Add reset routine for i.MX28
Uwe Kleine-König
u.kleine-koenig at pengutronix.de
Wed Nov 17 08:44:35 EST 2010
On Wed, Nov 17, 2010 at 07:17:39PM +0800, Shawn Guo wrote:
> 2010/11/16 Uwe Kleine-König <u.kleine-koenig at pengutronix.de>:
> > On Mon, Nov 15, 2010 at 10:36:27PM +0800, Shawn Guo wrote:
> >> +#ifdef CONFIG_ARCH_MX28
> >> +int mxc_reset_block(void __iomem *reg_addr)
> >> +{
> >> + u32 reg;
> >> + int timeout;
> >> +
> >> + /*
> >> + * The process of software reset of IP block is done
> >> + * in several steps:
> >> + *
> >> + * 1) clear SFTRST and wait it cleared;
> >> + * 2) clear CLKGATE, set SFTRST, wait CLKGATE set;
> >> + * 3) clear SFTRST and wait it cleared;
> >> + * 4) clear CLKGATE and wait it cleared.
> >> + */
> >> +
> >> + /* Clear SFTRST */
> >> + reg = __raw_readl(reg_addr);
> >> + reg &= ~(1 << 31);
> >> + __raw_writel(reg, reg_addr);
> >> + /* Wait SFTRST cleared */
> >> + timeout = 1000;
> >> + do {
> >> + mdelay(1);
> >> + if ((__raw_readl(reg_addr) & (1 << 31)) == 0)
> >> + break;
> >> + } while (--timeout > 0);
> > Does the mdelay make sence here?
> >
> i.MX28 Application Processor Reference Manual, Chapter 39.5.10
> "Correct Way to Software Reset a Block".
Quoting the chapter you pointed out:
// Prepare for soft-reset by making sure that SFTRST is not currently
// asserted. Also clear CLKGATE so we can wait for its assertion below.
HW_GPMI_CTRL0_CLR(BM_GPMI_CTRL0_SFTRST);
// Wait at least a microsecond for SFTRST to deassert. In actuality, we
// need to wait 3 GPMI clocks, but this is much easier to implement.
musecs = hw_profile_GetMicroseconds();
while (HW_GPMI_CTRL0.B.SFTRST || (hw_profile_GetMicroseconds() - musecs <
DDI_NAND_HAL_GPMI_SOFT_RESET_LATENCY));
[...]
This talks about a microsecond, not a millisecond.
Expecting that most of the time not a whole millisecond is needed, I'd
do something like that:
val = __raw_readl(reset_addr);
val &= ~MX28_MODULERESET_SFTRST;
__raw_writel(val, reset_addr);
/*
* SFTRST needs 3 GPMI clocks to settle, the reference manual
* recommends to wait 1us.
*/
udelay(1);
timeout = 0x400;
while ((__raw_readl(reset_addr) & MX28_MODULERESET_SFTRST) && --timeout)
/* nothing /*;
if (!timeout)
goto error;
That is do a busy read and don't busy wait a millisecond between two
consecutive reads just twiddling thumbs.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | http://www.pengutronix.de/ |
More information about the linux-arm-kernel
mailing list