Re: [PATCH v2 01/17] remoteproc: Add new operation and state machine for MCU synchronisation

[Suman Anna <s-anna@ti.com>]

On 4/1/20 4:53 PM, Mathieu Poirier wrote:
> On Mon, Mar 30, 2020 at 05:49:11PM -0500, Suman Anna wrote:
>> On 3/30/20 5:46 PM, Suman Anna wrote:
>>> Hi Mathieu,
>>>
>>> On 3/24/20 4:45 PM, Mathieu Poirier wrote:
>>>> Add a new rproc_ops sync_ops to support use cases where the remoteproc
>>>> core is synchronisting with the MCU.  When exactly to use the sync_ops is
>>>
>>> typo on syschronisting..
>>>
>>>> directed by the states in struct rproc_sync_states.
>>>>
>>>> Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
>>>> ---
>>>>  drivers/remoteproc/remoteproc_debugfs.c  | 31 ++++++++++++++++++++++++
>>>>  drivers/remoteproc/remoteproc_internal.h |  5 ++++
>>>>  include/linux/remoteproc.h               | 23 +++++++++++++++++-
>>>>  3 files changed, 58 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/remoteproc/remoteproc_debugfs.c b/drivers/remoteproc/remoteproc_debugfs.c
>>>> index dd93cf04e17f..187bcc67f997 100644
>>>> --- a/drivers/remoteproc/remoteproc_debugfs.c
>>>> +++ b/drivers/remoteproc/remoteproc_debugfs.c
>>>> @@ -311,6 +311,35 @@ static const struct file_operations rproc_carveouts_ops = {
>>>>  	.release	= single_release,
>>>>  };
>>>>  
>>>> +/* Expose synchronisation states via debugfs */
>>>> +static int rproc_sync_states_show(struct seq_file *seq, void *p)
>>>> +{
>>>> +	struct rproc *rproc = seq->private;
>>>> +
>>>> +	seq_printf(seq, "Sync with MCU: %s\n",
>>>> +		   rproc->sync_with_mcu ? "true" : "false");
>>>> +	seq_printf(seq, "On init: %s\n",
>>>> +		   rproc->sync_states->on_init ? "true" : "false");
>>>> +	seq_printf(seq, "After stop: %s\n",
>>>> +		   rproc->sync_states->after_stop ? "true" : "false");
>>>> +	seq_printf(seq, "After crash: %s\n",
>>>> +		   rproc->sync_states->after_crash ? "true" : "false");
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static int rproc_sync_states_open(struct inode *inode, struct file *file)
>>>> +{
>>>> +	return single_open(file, rproc_sync_states_show, inode->i_private);
>>>> +}
>>>> +
>>>> +static const struct file_operations rproc_sync_states_ops = {
>>>> +	.open		= rproc_sync_states_open,
>>>> +	.read		= seq_read,
>>>> +	.llseek		= seq_lseek,
>>>> +	.release	= single_release,
>>>> +};
>>>> +
>>>>  void rproc_remove_trace_file(struct dentry *tfile)
>>>>  {
>>>>  	debugfs_remove(tfile);
>>>> @@ -357,6 +386,8 @@ void rproc_create_debug_dir(struct rproc *rproc)
>>>>  			    rproc, &rproc_rsc_table_ops);
>>>>  	debugfs_create_file("carveout_memories", 0400, rproc->dbg_dir,
>>>>  			    rproc, &rproc_carveouts_ops);
>>>> +	debugfs_create_file("sync_states", 0400, rproc->dbg_dir,
>>>> +			    rproc, &rproc_sync_states_ops);
>>>>  }
>>>>  
>>>>  void __init rproc_init_debugfs(void)
>>>> diff --git a/drivers/remoteproc/remoteproc_internal.h b/drivers/remoteproc/remoteproc_internal.h
>>>> index 493ef9262411..5c93de5e00bb 100644
>>>> --- a/drivers/remoteproc/remoteproc_internal.h
>>>> +++ b/drivers/remoteproc/remoteproc_internal.h
>>>> @@ -63,6 +63,11 @@ struct resource_table *rproc_elf_find_loaded_rsc_table(struct rproc *rproc,
>>>>  struct rproc_mem_entry *
>>>>  rproc_find_carveout_by_name(struct rproc *rproc, const char *name, ...);
>>>>  
>>>> +static inline bool rproc_sync_with_mcu(struct rproc *rproc)
>>>> +{
>>>> +	return rproc->sync_with_mcu;
>>>> +}
>>>> +
>>>
>>> Since you are using this mostly for checking and as a boolean, I suggest
>>> you rename this appropriately, something like rproc_needs_sync,
>>> rproc_has_sync or rproc_uses_sync().
> 
> I will rename to rproc_syncing_with_rproc()

Hmm, I want this to reflect a boolean answer for better code
flow/readability.

> 
> 
>>>
>>> And I am wondering if it is actually better to introduce the sync state
>>> to check against here, rather than using the stored sync state and
>>> return. The current way makes it confusing to read the state machine.
> 
> I decided to proceed this way because there may not be a direct correlation
> between the current synchronisation state and the location of the check itself.
> for instance, in firmware_show(), what sync state should be key on?

Yeah OK. Its the combinations of ops (11 callbacks) plus sync states (3)
that kinda makes it hard to read the state-machine.

> 
>>>
>>>>  static inline
>>>>  int rproc_fw_sanity_check(struct rproc *rproc, const struct firmware *fw)
>>>>  {
>>>> diff --git a/include/linux/remoteproc.h b/include/linux/remoteproc.h
>>>> index 16ad66683ad0..d115e47d702d 100644
>>>> --- a/include/linux/remoteproc.h
>>>> +++ b/include/linux/remoteproc.h
>>>> @@ -353,6 +353,21 @@ enum rsc_handling_status {
>>>>  	RSC_IGNORED	= 1,
>>>>  };
>>>>  
>>>> +/**
>>>> + * struct rproc_sync_states - platform specific states indicating which
>>>> + *			      rproc_ops to use at specific times during
>>>> + *			      the MCU lifecycle.
>>>> + * @on_init: true if synchronising with MCU at system initialisation time
>>>> + * @after_stop: true if synchronising with MCU after stopped from the
>>>> + *		command line
>>>> + * @after_crash: true if synchonising with MCU after the MCU has crashed
>>>> + */
>>>> +struct rproc_sync_states {
>>>> +	bool on_init;
>>>> +	bool after_stop;
>>>> +	bool after_crash;
>>>> +};
>>>> +
>>>
>>> Overall, this patch can move down the order, and better to add it in
>>> the patches where you actually introduce these code. And the debugfs
>>> pieces can be added as a separate patch by itself.
>>
>> Also, actually sounds more like flags than states..
> 
> I thought about this in terms of "states" in which a decision should be made.
> I'm not sure those are flags...

I see them as just decision flags for sync, it is not reflecting a state
like rproc->state. The rproc structure variable holds the current sync
flag state though.

> 
>>
>> regards
>> Suman
>>
>>>
>>>>  /**
>>>>   * struct rproc_ops - platform-specific device handlers
>>>>   * @start:	power on the device and boot it
>>>> @@ -456,6 +471,9 @@ struct rproc_dump_segment {
>>>>   * @firmware: name of firmware file to be loaded
>>>>   * @priv: private data which belongs to the platform-specific rproc module
>>>>   * @ops: platform-specific start/stop rproc handlers
>>>> + * @sync_ops: paltform-specific start/stop rproc handlers when
>>>
>>> typo on platform
> 
> No matter how many times you read your own code, there's always something like
> this that escapes...

he he, indeed :)

> 
>>>
>>>> + *	      synchronising with a remote processor.
>>>> + * @sync_states: Determine the rproc_ops to choose in specific states.
>>>>   * @dev: virtual device for refcounting and common remoteproc behavior
>>>>   * @power: refcount of users who need this rproc powered up
>>>>   * @state: state of the device
>>>> @@ -479,6 +497,7 @@ struct rproc_dump_segment {
>>>>   * @table_sz: size of @cached_table
>>>>   * @has_iommu: flag to indicate if remote processor is behind an MMU
>>>>   * @auto_boot: flag to indicate if remote processor should be auto-started
>>>> + * @sync_with_mcu: true if currently synchronising with MCU
>>>>   * @dump_segments: list of segments in the firmware
>>>>   * @nb_vdev: number of vdev currently handled by rproc
>>>>   */
>>>> @@ -488,7 +507,8 @@ struct rproc {
>>>>  	const char *name;
>>>>  	char *firmware;
>>>>  	void *priv;
>>>> -	struct rproc_ops *ops;
>>>> +	struct rproc_ops *ops, *sync_ops;
>>>
>>> Nothing wrong with this, but prefer to have the new variable in a new
>>> line for better readability.
> 
> Sure thing.

Thanks,
Suman


> 
>>>
>>> regards
>>> Suman
>>>
>>>> +	struct rproc_sync_states *sync_states;
>>>>  	struct device dev;
>>>>  	atomic_t power;
>>>>  	unsigned int state;
>>>> @@ -512,6 +532,7 @@ struct rproc {
>>>>  	size_t table_sz;
>>>>  	bool has_iommu;
>>>>  	bool auto_boot;
>>>> +	bool sync_with_mcu;
>>>>  	struct list_head dump_segments;
>>>>  	int nb_vdev;
>>>>  };
>>>>
>>>
>>