It would be interesting to hear if folks want to do nested open/close… I think 
it a good thing to know the driver was opened/closed reliably but I was not 
thinking there would be nested open/closes but I could see why some folks might 
want that… rest seems fine to me.


> On Sep 10, 2016, at 11:27 AM, Sterling Hughes <[email protected]> wrote:
> 
> For reference, the OS device code in sterly_refactor:
> 
> https://github.com/apache/incubator-mynewt-core/blob/sterly_refactor/libs/os/src/os_dev.c
> 
> I wanted to document my assumptions on OS device locking for folks, and get 
> input as to whether people agree/disagree that this is the right approach.
> 
> Assumptions:
> 
> - Devices are only created _prior_ to the OS running, and devices are _never_ 
> deleted once being created.  Therefore, the OS device list itself does not 
> need to be locked, as it is immutable during system operation.
> 
> - Open & close call the per-device handlers, and those handlers perform 
> locking if they view it as necessary.  An example of a handler that locks: 
> https://github.com/apache/incubator-mynewt-core/blob/sterly_refactor/drivers/adc/adc_nrf52/src/adc_nrf52.c
>  (nrf52_adc_open() - line 96.)
>       - Currently UNSAFE: in cases where locks don’t occur on the device, 
> OS_DEV_STATUS_OPEN is set and released by open and close, without any 
> locking.  This means that they can’t be called from multiple task contexts.
>       - I’m wondering if this should be a reference count, for the cases 
> where multiple opens & closes are allowed, that way the system always knows 
> if the driver is opened reliably.
> 
> - Suspend & resume (TBC - description here), will not look to acquire a lock 
> when suspending a device, but rather its up to the driver implementation to 
> wait for any locks, or provide any housekeeping regarding the driver itself.  
> Suspend/resume are meant to be called outside of the context with which a 
> driver is being used.
> 
> - Suspend & resume are ONLY going to be called on OPEN drivers.
> 
> Sterling

Reply via email to