On 2/21/23 12:18, Richard Fitzgerald wrote:
> On 21/02/2023 16:45, Pierre-Louis Bossart wrote:
>>
>>> +static int cs35l56_sdw_interrupt(struct sdw_slave *peripheral,
>>> + struct sdw_slave_intr_status *status)
>>> +{
>>> + struct cs35l56_private *cs35l56 =
>>> dev_get_drvdata(&peripheral->dev);
>>> +
>>> + /* SoundWire core holds our pm_runtime when calling this
>>> function. */
>>> +
>>> + dev_dbg(cs35l56->dev, "int control_port=%#x\n",
>>> status->control_port);
>>> +
>>> + if ((status->control_port & SDW_SCP_INT1_IMPL_DEF) == 0)
>>> + return 0;
>>> +
>>> + /* Prevent host controller suspending before we handle the
>>> interrupt */
>>> + pm_runtime_get_noresume(cs35l56->dev);
>>
>> can this happen that the manager suspends in this function?
>>
>> Or is this needed because of the queued work which the manager has no
>> knowledge of?
>>
>
> Because you issue a Bus-Reset when you suspend and clock-stop, if we
> didn't take our pm_runtime there is a small window of time where we
> could be reset before we've handled the interrupt. It's unlikely to
> happen but better to be safe than to rely on autosuspend delays.
What I meant is that the pm_runtime refcount should be increased prior
to the SoundWire core using the interrupt_callback.
So the only window I see is in the second part of in the interrupt
handling, before the workqueue is scheduled.
>
>>> +
>>> + /*
>>> + * Mask and clear until it has been handled. The read of
>>> GEN_INT_STAT_1
>>> + * is required as per the SoundWire spec for interrupt status bits
>>> + * to clear. GEN_INT_MASK_1 masks the _inputs_ to GEN_INT_STAT1.
>>> + * None of the interrupts are time-critical so use the
>>> + * power-efficient queue.
>>> + */
>>> + sdw_write_no_pm(peripheral, CS35L56_SDW_GEN_INT_MASK_1, 0);
>>> + sdw_read_no_pm(peripheral, CS35L56_SDW_GEN_INT_STAT_1);
>>> + sdw_write_no_pm(peripheral, CS35L56_SDW_GEN_INT_STAT_1, 0xFF);
>>> + queue_work(system_power_efficient_wq, &cs35l56->sdw_irq_work);
>>> +
>>> + return 0;
>>> +}
>>
>>> +static int __maybe_unused cs35l56_sdw_handle_unattach(struct
>>> cs35l56_private *cs35l56)
>>> +{
>>> + struct sdw_slave *peripheral = cs35l56->sdw_peripheral;
>>> +
>>> + if (peripheral->unattach_request) {
>>> + /* Cannot access registers until master re-attaches. */
>>
>> not sure what the comment means, the manager does not attach. did you
>> mean resume the bus?
>>
>
> If the manager has forced us to reset we can't access the registers
> until the manager has recovered its state.
"until the manager restarted the bus and re-enumerated devices" ?
>
>>> + dev_dbg(cs35l56->dev, "Wait for initialization_complete\n");
>>> + if
>>> (!wait_for_completion_timeout(&peripheral->initialization_complete,
>>> + msecs_to_jiffies(5000))) {
>>> + dev_err(cs35l56->dev, "initialization_complete timed
>>> out\n");
>>> + return -ETIMEDOUT;
>>> + }
>>> +
>>> + peripheral->unattach_request = 0;
>>> +
>>> + /*
>>> + * Don't call regcache_mark_dirty(), we can't be sure that the
>>> + * Manager really did issue a Bus Reset.
>>> + */
>>> + }
>>> +
>>> + return 0;
>>> +}
>> ...
>>
>>> +static void cs35l56_dsp_work(struct work_struct *work)
>>> +{
>>> + struct cs35l56_private *cs35l56 = container_of(work,
>>> + struct cs35l56_private,
>>> + dsp_work);
>>> + unsigned int reg;
>>> + unsigned int val;
>>> + int ret = 0;
>>> +
>>> + if (!wait_for_completion_timeout(&cs35l56->init_completion,
>>> + msecs_to_jiffies(5000))) {
>>> + dev_err(cs35l56->dev, "%s: init_completion timed out\n",
>>> __func__);
>>> + goto complete;
>>> + }
>>> +
>>> + if (!cs35l56->init_done || cs35l56->removing)
>>> + goto complete;
>>> +
>>> + cs35l56->dsp.part = devm_kasprintf(cs35l56->dev, GFP_KERNEL,
>>> "cs35l56%s-%02x",
>>> + cs35l56->secured ? "s" : "", cs35l56->rev);
>>> +
>>> + if (!cs35l56->dsp.part)
>>> + goto complete;
>>> +
>>> + pm_runtime_get_sync(cs35l56->dev);
>>
>> test that this is successful?
>>
>
> Could do. Wasn't really expecting it to fail unless the hardware is
> already broken.
it's not supposed to happen indeed, but our CI caught a couple of issues
over the last two years. Better add a check.
>
>>> +
>>> + /*
>>> + * Disable SoundWire interrupts to prevent race with IRQ work.
>>> + * Setting sdw_irq_no_unmask prevents the handler re-enabling
>>> + * the SoundWire interrupt.
>>> + */
>>> + if (cs35l56->sdw_peripheral) {
>>> + cs35l56->sdw_irq_no_unmask = true;
>>> + cancel_work_sync(&cs35l56->sdw_irq_work);
>>> + sdw_write_no_pm(cs35l56->sdw_peripheral,
>>> CS35L56_SDW_GEN_INT_MASK_1, 0);
>>> + sdw_read_no_pm(cs35l56->sdw_peripheral,
>>> CS35L56_SDW_GEN_INT_STAT_1);
>>> + sdw_write_no_pm(cs35l56->sdw_peripheral,
>>> CS35L56_SDW_GEN_INT_STAT_1, 0xFF);
>>> + }
>>> +
>>> + ret = cs35l56_mbox_send(cs35l56, CS35L56_MBOX_CMD_SHUTDOWN);
>>> + if (ret) {
>>> + dev_dbg(cs35l56->dev, "%s: CS35L56_MBOX_CMD_SHUTDOWN ret
>>> %d\n", __func__, ret);
>>> + goto err;
>>> + }
>>> +
>>> + if (cs35l56->rev < CS35L56_REVID_B0)
>>> + reg = CS35L56_DSP1_PM_CUR_STATE_A1;
>>> + else
>>> + reg = CS35L56_DSP1_PM_CUR_STATE;
>>> +
>>> + ret = regmap_read_poll_timeout(cs35l56->regmap, reg,
>>> + val, (val == CS35L56_HALO_STATE_SHUTDOWN),
>>> + CS35L56_HALO_STATE_POLL_US,
>>> + CS35L56_HALO_STATE_TIMEOUT_US);
>>> + if (ret < 0)
>>> + dev_err(cs35l56->dev, "Failed to poll PM_CUR_STATE to 1 is
>>> %d (ret %d)\n",
>>> + val, ret);
>>> +
>>> + /* Use wm_adsp to load and apply the firmware patch and
>>> coefficient files */
>>> + ret = wm_adsp_power_up(&cs35l56->dsp);
>>> + if (ret) {
>>> + dev_dbg(cs35l56->dev, "%s: wm_adsp_power_up ret %d\n",
>>> __func__, ret);
>>> + goto err;
>>> + }
>>> +
>>> + if (cs35l56->removing)
>>> + goto err;
>>> +
>>> + mutex_lock(&cs35l56->irq_lock);
>>> +
>>> + init_completion(&cs35l56->init_completion);
>>> +
>>> + cs35l56_system_reset(cs35l56);
>>> +
>>> + if (cs35l56->sdw_peripheral) {
>>> + if (!wait_for_completion_timeout(&cs35l56->init_completion,
>>> + msecs_to_jiffies(5000))) {
>>> + dev_err(cs35l56->dev, "%s: init_completion timed out
>>> (SDW)\n", __func__);
>>
>> shouldn't do the same routine as for a regular pm_runtime resume,
>> including re-synching regmaps?
>>
>
> Not sure it would help. It's not the same as runtime_resume because
> we've changed the firmware and rebooted it (the firmware is retained
> in a runtime_suspend). We need to do some of the first-time init()
> code again, which we don't need to do in runtime_resume.
>
> Also would create a circular dependency between this driver and the
> cs35l56-sdw driver. (We _could_ call our dev->pm->runtime_resume pointer
> but that's a bit ugly)
I wasn't suggesting using a pm_runtime suspend/resume cycle but rather
use a common helper called from here and from the pm_runtime_resume.
It's a suggestion only.
>
>>
>>> + goto err_unlock;
>>> + }
>>> + } else {
>>> + if (cs35l56_init(cs35l56))
>>> + goto err_unlock;
>>> + }
>>> +
>>> + cs35l56->fw_patched = true;
>>> +
>>> +err_unlock:
>>> + mutex_unlock(&cs35l56->irq_lock);
>>> +err:
>>> + pm_runtime_mark_last_busy(cs35l56->dev);
>>> + pm_runtime_put_autosuspend(cs35l56->dev);
>>> +
>>> + /* Re-enable SoundWire interrupts */
>>> + if (cs35l56->sdw_peripheral) {
>>> + cs35l56->sdw_irq_no_unmask = false;
>>> + sdw_write_no_pm(cs35l56->sdw_peripheral,
>>> CS35L56_SDW_GEN_INT_MASK_1,
>>> + CS35L56_SDW_INT_MASK_CODEC_IRQ);
>>> + }
>>> +
>>> +complete:
>>> + complete_all(&cs35l56->dsp_ready_completion);
>>> +}
