Original question:
[FAQ] SK-AM62: How to execute code from external memory using M4F Core?
Tool/software:
Hi Team,
In the related thread, it says that access to DDR from M4F core takes significantly longer. My customer is considering using IPC RPMsg. It says that shared memory can be either DDR or internal memory. Does this mean that when they use IPC RPMsg to load data from DDR, it will take significantly longer as well? Is there some workaround for this?
Also, when it refers to shared internal memory, is it referencing the 64KB OCRAM?
Best regards,
Mari Tsunoda
Hi Mari,
Generally, while using IPC, it is recommended to use non-cached memory regions, since a memory update performed by one core will be reflected in it's cache and when the other core tries to access the memory it will receive a 'stale' value.
Can the internal memory usage selected be different between cores for better performance?
Yes, you can define 2 separate shared memory regions and give access to selective cores to selective regions. If your use case does not require the A53 and M4F IPC then you can create separate regions for A53-M4F and M4F-R5F.
I see that VRING is used. Do you have more detailed documentation on this?
The concept of VRING memory is the same as mentioned in the PRU documentation. The different types of data stored in the VRING could be different in the case of PRU and ARM cores. If you are interested I can look into the VRING structure used in the case of PRU.
On a very basic level, VRING is a shared memory segment, between a pair of CPUs, which holds the messages passed between the two CPUs. The sequence of events as a message is passed from a sender to receiver and back again is shown in below figure:
The local structure to maintain state of a given VRING is as below (from ipc_rpmsg_priv.h)
typedef struct
{
uint16_t lastUsedIdx; /* last read index into used Q */
uint16_t lastAvailIdx; /* last read index into avail Q */
uint16_t vringNumBuf; /* number of buffer in the vring */
struct vring_desc *desc; /* pointer to buffer descriptors in VRING shared memory */
struct vring_avail *avail; /* pointer to avail Q in VRING shared memory */
struct vring_used *used; /* pointer to used Q in VRING shared memory */
uint8_t *bufBaseAddr; /* pointer to message buffer 0 in VRING shared memory */
} RPMessage_Vring;
I do not think there is a documentation explaining VRING in great detail. I will look into it and share it with you if I find any.
Regards,
Nitika