RE: [PATCH 1/6] ACPI/EC: Introduce STARTED/STOPPED flags to replace BLOCKED flag.

[Zheng, Lv]

Hi, Rafael

There is one thing I should let you know.

Originally this patchset is dependent on the GPE "dead lock" fix.
Because this patch will invoke acpi_enable_gpe()/acpi_disable_gpe() with EC lock held.

I saw system hang during suspending using only this patchset, so we have to find a solution.

> From: Zheng, Lv
> Sent: Monday, November 03, 2014 1:16 PM
> 
> By using the 2 flags, we can indicate an inter-mediate state where the
> current transactions should be completed while the new transactions should
> be dropped.
> 
> The comparison of the old flag and the new flags:
>   Old			New
>   about to set BLOCKED	STOPPED set / STARTED set
>   BLOCKED set		STOPPED clear / STARTED clear
>   BLOCKED clear		STOPPED clear / STARTED set
> The new period is between the point where we are about to set BLOCKED and
> the point when the BLOCKED is set. The GPE is disabled during this period.
> The new flags allow us to add acpi_ec_stopped() check to only check with
> STOPPED flag to implement transaction flushing. This is not done in this
> patch.
> 
> No functional changes except that after applying this patch, the GPE
> enabling/disabling is protected by the EC specific lock. We can do this
> because of recent ACPICA GPE API enhancement. This is reasonable as the GPE
> disabling/enabling state should only be determined by the EC driver's state
> machine which is protected by the EC spinlock.

This paragraph is talking about the dependency.

> 
> Signed-off-by: Lv Zheng <lv.zheng@xxxxxxxxx>
> Tested-by: Ortwin GlÃck <odi@xxxxxx>
> ---
>  drivers/acpi/ec.c |   56 +++++++++++++++++++++++++++++++++++++++++++++--------
>  1 file changed, 48 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/acpi/ec.c b/drivers/acpi/ec.c
> index 5f9b74b..192cd11 100644
> --- a/drivers/acpi/ec.c
> +++ b/drivers/acpi/ec.c
> @@ -79,7 +79,8 @@ enum {
>  	EC_FLAGS_GPE_STORM,		/* GPE storm detected */
>  	EC_FLAGS_HANDLERS_INSTALLED,	/* Handlers for GPE and
>  					 * OpReg are installed */
> -	EC_FLAGS_BLOCKED,		/* Transactions are blocked */
> +	EC_FLAGS_STARTED,		/* Driver is started */
> +	EC_FLAGS_STOPPED,		/* Driver is stopped */
>  };
> 
>  #define ACPI_EC_COMMAND_POLL		0x01 /* Available for command byte */
> @@ -129,6 +130,16 @@ static int EC_FLAGS_CLEAR_ON_RESUME; /* Needs acpi_ec_clear() on boot/resume */
>  static int EC_FLAGS_QUERY_HANDSHAKE; /* Needs QR_EC issued when SCI_EVT set */
> 
>  /* --------------------------------------------------------------------------
> + *                           Device Flags
> + * -------------------------------------------------------------------------- */
> +
> +static bool acpi_ec_started(struct acpi_ec *ec)
> +{
> +	return test_bit(EC_FLAGS_STARTED, &ec->flags) &&
> +	       !test_bit(EC_FLAGS_STOPPED, &ec->flags);
> +}
> +
> +/* --------------------------------------------------------------------------
>   *                           Transaction Management
>   * -------------------------------------------------------------------------- */
> 
> @@ -354,7 +365,7 @@ static int acpi_ec_transaction(struct acpi_ec *ec, struct transaction *t)
>  	if (t->rdata)
>  		memset(t->rdata, 0, t->rlen);
>  	mutex_lock(&ec->mutex);
> -	if (test_bit(EC_FLAGS_BLOCKED, &ec->flags)) {
> +	if (!acpi_ec_started(ec)) {
>  		status = -EINVAL;
>  		goto unlock;
>  	}
> @@ -511,6 +522,35 @@ static void acpi_ec_clear(struct acpi_ec *ec)
>  		pr_info("%d stale EC events cleared\n", i);
>  }
> 
> +static void acpi_ec_start(struct acpi_ec *ec)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ec->lock, flags);
> +	if (!test_and_set_bit(EC_FLAGS_STARTED, &ec->flags)) {
> +		pr_debug("+++++ Starting EC +++++\n");
> +		acpi_enable_gpe(NULL, ec->gpe);

This can work without "GPE dead lock" fix applied because:
1. During boot, this API is called when the EC GPE is disabled.
2. During resume, this API is called when the EC GPE is disabled (because EC GPE is always not wake capable).

> +		pr_info("+++++ EC started +++++\n");
> +	}
> +	spin_unlock_irqrestore(&ec->lock, flags);
> +}
> +
> +static void acpi_ec_stop(struct acpi_ec *ec)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ec->lock, flags);
> +	if (acpi_ec_started(ec)) {
> +		pr_debug("+++++ Stopping EC +++++\n");
> +		set_bit(EC_FLAGS_STOPPED, &ec->flags);
> +		acpi_disable_gpe(NULL, ec->gpe);

But this cannot work without "GPE dead lock" fix applied because:

In acpi_pm_freeze(), the call graph would be:
	acpi_pm_freeze()
		acpi_disable_all_gpes()
		acpi_os_wait_events_complete()
		acpi_ec_block_transactions()
			acpi_ec_stop()
				hold EC lock
				acpi_disable_gpe()
				hold GPE lock

And in the GPE handler acpi_irq(), the call graph would be:
	acpi_irq()
		acpi_ev_sci_xrupt_handler()
			acpi_ev_gpe_detect()
				hold GPE lock
					acpi_ev_gpe_dispatch()
						acpi_ec_gpe_handler()
							hold EC lock

Since acpi_os_wait_events_complete() cannot flush GPE but can only flush _Lxx/_Exx evaluation work queue currently.
The reversed ordered dead lock can happen.
We need to fix the acpi_os_wait_events_complete() prior than this series.
I have a fix to invoke synchronize_irq() in acpi_os_wait_events_complete().
Let me send it to you.
This cleanup should be applied after that fix.

Thanks and best regards
-Lv

> +		clear_bit(EC_FLAGS_STARTED, &ec->flags);
> +		clear_bit(EC_FLAGS_STOPPED, &ec->flags);
> +		pr_info("+++++ EC stopped +++++\n");
> +	}
> +	spin_unlock_irqrestore(&ec->lock, flags);
> +}
> +
>  void acpi_ec_block_transactions(void)
>  {
>  	struct acpi_ec *ec = first_ec;
> @@ -520,7 +560,7 @@ void acpi_ec_block_transactions(void)
> 
>  	mutex_lock(&ec->mutex);
>  	/* Prevent transactions from being carried out */
> -	set_bit(EC_FLAGS_BLOCKED, &ec->flags);
> +	acpi_ec_stop(ec);
>  	mutex_unlock(&ec->mutex);
>  }
> 
> @@ -533,7 +573,7 @@ void acpi_ec_unblock_transactions(void)
> 
>  	mutex_lock(&ec->mutex);
>  	/* Allow transactions to be carried out again */
> -	clear_bit(EC_FLAGS_BLOCKED, &ec->flags);
> +	acpi_ec_start(ec);
> 
>  	if (EC_FLAGS_CLEAR_ON_RESUME)
>  		acpi_ec_clear(ec);
> @@ -548,7 +588,7 @@ void acpi_ec_unblock_transactions_early(void)
>  	 * atomic context during wakeup, so we don't need to acquire the mutex).
>  	 */
>  	if (first_ec)
> -		clear_bit(EC_FLAGS_BLOCKED, &first_ec->flags);
> +		acpi_ec_start(first_ec);
>  }
> 
>  static int acpi_ec_query_unlocked(struct acpi_ec *ec, u8 *data)
> @@ -816,7 +856,7 @@ static int ec_install_handlers(struct acpi_ec *ec)
>  	if (ACPI_FAILURE(status))
>  		return -ENODEV;
> 
> -	acpi_enable_gpe(NULL, ec->gpe);
> +	acpi_ec_start(ec);
>  	status = acpi_install_address_space_handler(ec->handle,
>  						    ACPI_ADR_SPACE_EC,
>  						    &acpi_ec_space_handler,
> @@ -831,7 +871,7 @@ static int ec_install_handlers(struct acpi_ec *ec)
>  			pr_err("Fail in evaluating the _REG object"
>  				" of EC device. Broken bios is suspected.\n");
>  		} else {
> -			acpi_disable_gpe(NULL, ec->gpe);
> +			acpi_ec_stop(ec);
>  			acpi_remove_gpe_handler(NULL, ec->gpe,
>  				&acpi_ec_gpe_handler);
>  			return -ENODEV;
> @@ -844,7 +884,7 @@ static int ec_install_handlers(struct acpi_ec *ec)
> 
>  static void ec_remove_handlers(struct acpi_ec *ec)
>  {
> -	acpi_disable_gpe(NULL, ec->gpe);
> +	acpi_ec_stop(ec);
>  	if (ACPI_FAILURE(acpi_remove_address_space_handler(ec->handle,
>  				ACPI_ADR_SPACE_EC, &acpi_ec_space_handler)))
>  		pr_err("failed to remove space handler\n");
> --
> 1.7.10