rpmsg: reserve 0-1023 for predefined services in rpmsg_create_ept

[xiaoxiang781216]

and allocate the dynamic address from 1024 like Linux kernel

Signed-off-by: Xiang Xiao [email protected]

[xiaoxiang781216]

caller can still pass the fixed address from 0 to 1023, but rpmsg_create_ept allocate the dynamic address from 1024.

[arnopo]

I think here it is important that we are aligned first on the Linux implementation.
In rpmsg/virtio_rpmsg_bus.c the reserved address are defined by:

/*
 * Local addresses are dynamically allocated on-demand.
 * We do not dynamically assign addresses from the low 1024 range,
 * in order to reserve that address range for predefined services.
 */
#define RPMSG_RESERVED_ADDRESSES	(1024)

So here the main point here is to understand what is a predefined service

Now let's have look at __rpmsg_create_ept to understand how RPMSG_RESERVED_ADDRESSES is used

	/* do we need to allocate a local address ? */
	if (addr == RPMSG_ADDR_ANY) {
		id_min = RPMSG_RESERVED_ADDRESSES;
		id_max = 0;
	} else {
		id_min = addr;
		id_max = addr + 1;
	}

Regarding this code a predefined service is a service with a fixed address. And as this function can be called through rpmsg_create_ept API, any user can define a "predefined" service by specifying the chinfo.src address.
So if i well understand, the Linux implementation rule is:

• service based on a service name will have an address >= 1024
• address from 0 to 1023 are reserved for services with a fixed address on endpoint creation. fixed address can be used either par a core or a user service.

Did i miss something?

[xiaoxiang781216]

Yes, your understand is correct. This patch align OpenAMP implementation with Linux kernel. The only difference is we trust that caller don't pass the duplicated address inside 0-1023 and then bypass the check to simplify the implemntation.

[arnopo]

This version seems to me a good compromise that would match with Linux implementation, while keeping compatibility with existing implementation:

• address 0 to 1023: predefined address only, no control on source address
• address 1024 to 1151 shared between dynamically allocated and predefined address with a control of the address
• higher than 1151 not supported

I just need to do few more test on my side before approving it.

Then the title of this patch would need to be reworked as rpmsg_create_ept can allocated reserved address.
Some comments in code describing the address management would also be useful to understand the new behavior, especially the fact that address between 0 and 1023 are not checked.

@edmooring: could you review it and give your opinion please?

[xiaoxiang781216]

Then the title of this patch would need to be reworked as rpmsg_create_ept can allocated reserved address.
Some comments in code describing the address management would also be useful to understand the new behavior, especially the fact that address between 0 and 1023 are not checked.

@arnopo done.

[xiaoxiang781216]

@arnopo could you merge the follow patch first:
#209
#207
#204
Then I can rebase other patch on top of them.

[edmooring]

Aside from my comments, looks good to go.

[edmooring]

Good to go.