AM6442: Am64x: Issue replacement of dma_buff_phys with CDev driver

Hi, the domain expert is out of office for the next few days. Please expect delayed response. 

Hi Bin,

  I hope the domain expert is available now.

   Is there any update on this as we wanted to get it resolved at the earliest. 

Thank you

Mamatha B M 

For future readers, Mamatha is using the rpmsg_char_zerocopy project as a starting point:
https://git.ti.com/cgit/rpmsg/rpmsg_char_zerocopy/

Hello Mamatha,

Hmm. It looks like function "dmabuf_heap_alloc" is forming rproc_name by taking rproc_id and putting it at the end of "/dev/remoteproc%d"
https://git.ti.com/cgit/rpmsg/rpmsg_char_zerocopy/tree/linux/src/rpmsg_char_zerocopy.c 

/* Allocate dma-buf from dma-heap, mmap locally, get the phys address
 * and setup the buffer data we want to pass to the remote endpoint
 * in ipc_buf and the local data we want to keep in local_buf.
 */
int dmabuf_heap_alloc(int rproc_id, int dma_heap_fd, int size,
		       struct local_buf *lbuf,
		       struct ipc_buf *ibuf)
{
...
	sprintf(rproc_name, "/dev/remoteproc%d", rproc_id);
	rproc_fd = open(rproc_name, O_RDONLY | O_CLOEXEC);
	if (rproc_fd < 0) {
		printf("Failed to open %s: -%d\n", rproc_name, errno);
		return -1;
	}

But that is not how the code should run, because rproc_id is a fixed value for each core. On the other hand, /dev/remoteprocX is not a fixed value. The X value is assigned by order of which core gets initialized first, which means that which core is associated with a specific remoteprocX number can actually change on each boot.

We can look at function _rpmsg_char_find_rproc() as a reference. rproc_id is used to find the correct remoteprocX instance, and it sets ept->remoteproc_id == X.
https://git.ti.com/cgit/rpmsg/ti-rpmsg-char/tree/src/rpmsg_char.c 

Please try to modify rpmsg_char_zerocopy > dmabuf_heap_alloc to look like _rpmsg_char_find_rproc(). Please let me know if this works, if so I will update the zerocopy project with the new code.

Regards,

Nick

Hello Mamatha,

Thank you for reporting the bug, and for confirming that you are able to move forward! I have filed a bug report so that we make sure to circle back and fix this code in the future.

Regards,

Nick