Re: [PATCH V2 13/32] omap_hsmmc: context save/restore support

[Adrian Hunter <adrian.hunter@nokia.com>]

Madhusudhan wrote:
> 
>> -----Original Message-----
>> From: linux-omap-owner@vger.kernel.org [mailto:linux-omap-
>> owner@vger.kernel.org] On Behalf Of Adrian Hunter
>> Sent: Tuesday, July 28, 2009 5:40 AM
>> To: Andrew Morton
>> Cc: Jarkko Lavinen; Adrian Hunter; linux-omap Mailing List; Pierre Ossman;
>> Denis Karpov; Matt Fleming; lkml
>> Subject: [PATCH V2 13/32] omap_hsmmc: context save/restore support
>>
>> From 43e9fa346d7e386328876a8535dc8619bd1f47ae Mon Sep 17 00:00:00 2001
>> From: Denis Karpov <ext-denis.2.karpov@nokia.com>
>> Date: Wed, 22 Apr 2009 16:04:25 +0200
>> Subject: [PATCH] omap_hsmmc: context save/restore support
>>
>> Keep the context over PM dynamic OFF states.
>>
>> Signed-off-by: Denis Karpov <ext-denis.2.karpov@nokia.com>
>> Signed-off-by: Adrian Hunter <adrian.hunter@nokia.com>
>> ---
>>  drivers/mmc/host/omap_hsmmc.c |  194
>> ++++++++++++++++++++++++++++++++++++++--
>>  1 files changed, 184 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/mmc/host/omap_hsmmc.c b/drivers/mmc/host/omap_hsmmc.c
>> index c28d055..ac1a3bf 100644
>> --- a/drivers/mmc/host/omap_hsmmc.c
>> +++ b/drivers/mmc/host/omap_hsmmc.c
>> @@ -37,6 +37,7 @@
>>
>>  /* OMAP HSMMC Host Controller Registers */
>>  #define OMAP_HSMMC_SYSCONFIG	0x0010
>> +#define OMAP_HSMMC_SYSSTATUS	0x0014
>>  #define OMAP_HSMMC_CON		0x002C
>>  #define OMAP_HSMMC_BLK		0x0104
>>  #define OMAP_HSMMC_ARG		0x0108
>> @@ -94,6 +95,8 @@
>>  #define DUAL_VOLT_OCR_BIT	7
>>  #define SRC			(1 << 25)
>>  #define SRD			(1 << 26)
>> +#define SOFTRESET		(1 << 1)
>> +#define RESETDONE		(1 << 0)
>>
>>  /*
>>   * FIXME: Most likely all the data using these _DEVID defines should come
>> @@ -152,6 +155,8 @@ struct mmc_omap_host {
>>  	int			slot_id;
>>  	int			dbclk_enabled;
>>  	int			response_busy;
>> +	int			context_loss;
>> +
>>  	struct	omap_mmc_platform_data	*pdata;
>>  };
>>
>> @@ -166,6 +171,166 @@ static void omap_mmc_stop_clock(struct mmc_omap_host
>> *host)
>>  		dev_dbg(mmc_dev(host->mmc), "MMC Clock is not stoped\n");
>>  }
>>
>> +#ifdef CONFIG_PM
>> +
>> +/*
>> + * Restore the MMC host context, if it was lost as result of a
>> + * power state change.
>> + */
>> +static int omap_mmc_restore_ctx(struct mmc_omap_host *host)
>> +{
>> +	struct mmc_ios *ios = &host->mmc->ios;
>> +	struct omap_mmc_platform_data *pdata = host->pdata;
>> +	int context_loss = 0;
>> +	u32 hctl, capa, con;
>> +	u16 dsor = 0;
>> +	unsigned long timeout;
>> +
>> +	if (pdata->get_context_loss_count) {
>> +		context_loss = pdata->get_context_loss_count(host->dev);
>> +		if (context_loss < 0)
>> +			return 1;
>> +	}
> This seems to restore the context always. The context should be restored
> only if the card was powered OFF, Right?

Not exactly. The "context" is just the host controller registers, which get
lost if power management decides to power off the host controller, which it
will if it decides it is not doing anything.  For us, switching off the
functional clock lets PM power off the host controller.  The card can be in
any power state: on, off, card/regulator sleep.

The context is restored only when it has been lost - see the following
lines.  But this is done from the "enable" method, so this is only called
if someone is trying to access the card.

+	dev_dbg(mmc_dev(host->mmc), "context was %slost\n",
+		context_loss == host->context_loss ? "not " : "");
+	if (host->context_loss == context_loss)
+		return 1;

And then further down, we skip some things if the card is off:

+	/* Do not initialize card-specific things if the power is off */
+	if (host->power_mode == MMC_POWER_OFF)
+		goto out;


> The context could also be lost if the CORE transitions to OFF. I assume that
> case gets handled here without anything extra required if "power_saving" is
> set to true. Am I right?

Yes, "power_saving" is not related to "Power Management".  "power_saving"
is about controlling the power regulators, which the host controller does
itself without any support from PM.

To put it another way, "power_saving" and CONFIG_PM can be set independently
of one another.

> How about the case of eMMC? Since it is not a removable device the card is
> not powered OFF. But the CORE OFF would result in context loss. Do we hit
> the restore_ctx path in that case?

On the contrary, eMMC *is* powered off.  After 100ms of inactivity, eMMC is
put to card sleep, and the regulator is put to sleep also - in fact there
are two regulators and one is powered off here.  After 8 seconds
the other regulator is powered off and consequently after that the eMMC must
be reinitialised (rescanned) before it can be used.