Re: [PATCH v1 2/5] misc: fastrpc: Move all remote heap allocations to a new list

[Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>]

On 12/06/2025 08:13, Ekansh Gupta wrote:
> 
> 
> On 5/22/2025 5:39 PM, Dmitry Baryshkov wrote:
>> On Thu, 22 May 2025 at 07:54, Ekansh Gupta
>> <ekansh.gupta@oss.qualcomm.com> wrote:
>>>
>>>
>>> On 5/19/2025 6:59 PM, Dmitry Baryshkov wrote:
>>>> On Mon, May 19, 2025 at 04:36:13PM +0530, Ekansh Gupta wrote:
>>>>> On 5/19/2025 3:46 PM, Dmitry Baryshkov wrote:
>>>>>> On Tue, May 13, 2025 at 09:58:22AM +0530, Ekansh Gupta wrote:
>>>>>>> Remote heap allocations are not organized in a maintainable manner,
>>>>>>> leading to potential issues with memory management. As the remote
>>>>>> Which issues? I think I have been asking this question previously.
>>>>>> Please expand the commit message here.
>>>>> This is mostly related to the memory clean-up and the other patch where
>>>>> unmap request was added, I'll try to pull more details about the issue
>>>>> scenario.
>>>> Thanks.
>>>>
>>>>>>> heap allocations are maintained in fl mmaps list, the allocations
>>>>>>> will go away if the audio daemon process is killed but there are
>>>>>> What is audio daemon process?
>>>>> As audio PD on DSP is static, there is HLOS process(audio daemon) required to
>>>>> attach to audio PD to fulfill it's memory and file operation requirements.
>>>>>
>>>>> This daemon can be thought of to be somewhat similar to rootPD(adsprpcd) or
>>>>> sensorsPD(sscrpcd) daemons. Although, there is a slight difference in case of audio
>>>>> daemon as it is required to take additional information and resources to audio PD
>>>>> while attaching.
>>>> I find it a little bit strange to see 'required' here, while we have
>>>> working audio setup on all up platforms up to and including SM8750
>>>> without any additional daemons. This is the primary reason for my
>>>> question: what is it, why is it necessary, when is it necessary, etc.
>>> This daemon is critical to facilitate dynamic loading and memory
>>> requirement for audio PD(running on DSP for audio processing). Even
>>> for audio testing on SM8750, I believe Alexey was enabling this daemon.
>> Could you please point out the daemon sources?
>>
>> As far as I remember, we didn't need it on any of the platforms up to
>> and including SM8650, that's why I'm asking.
> This source was used for testing audio use case on SM8750:
> https://github.com/quic/fastrpc/blob/development/src/adsprpcd.c
> 
> The use case tried by Alexey as per my knowledge is audio playback where dynamic
> loading was needed but he can give more details on the use case.

Okay.
You need to be more specific in the commit messages.

- It is a normal adsprpcd.
- It is only required for compressed audio playback.

> 
> He was observing failures and panic which got resolved after picking this patch series.

Which failures? Panic in which driver?

>>
>>> What is it?
>>> - HLOS process to attached to audio PD to fulfill the requirements that
>>> cannot be met by DSP alone(like file operations, memory etc.)
>>>
>>> Why is it necessary?
>>> - There are limitation on DSP for which the PD requirements needs to be
>>> taken to HLOS. For example, DSP does not have it's own file system, so
>>> any file operation request it PD(say for dynamic loading) needs to be
>>> taken to HLOS(using listener/reverse calls) and is fulfilled there.
>>> Similarly memory requirement is another example.
>>>
>>> When is it necessary?
>>> - When audio PD needs to perform any task that requires HLOS relying
>>> operations like dynamic loading etc.
>> This doesn't exactly answer the question. I can play and capture audio
>> on most of the platforms that I tested without using extra daemon. So,
>> when is it necessary?
> For use case details, I'll let Alexey comment here.
> 
> The daemons major requirement is to facilitate any dynamic loading or memory
> requirements from DSP audio PD. The daemons are already supported for
> different types of static PDs to facilitate these requirements(fops and memory).

So... compressed audio only or a normal playback / capture too?

> 
>>
> 


-- 
With best wishes
Dmitry