[DO NOT MERGE] I.MX8 MU support

[dianacretu]

Add implementation for MU write in order to communicate with the host.

[dbaluta]

I think it is OK to leave the initial comment here ' set value via SHM' because we are actually setting the value of register via SHM. Then move the comment with 'Interrupt arrived' just above the switch.

[lgirdwood]

Good to merge once @dbaluta approves

[dbaluta]

Will keep this open for few days. I need to do some tests and more thinking, because MU (Messaging Unit) it is an interesting piece of hardware from emulation pow.

It has two sides: A and B one for Host and for DSP. The thing is that writing a register from A (lets say Generic Interrupt Request) has the effect of enabling a bit from side B (Generic Interrupt Pending).

I'm not sure we emulated this correctly. I need to think about it.

[lgirdwood]

We have the same with the SHIM doorbell on BYT. We set the Bit on A and it's immediately visible on B (via SHM). However B is unaware of the change until we send the IRQ message.

[dbaluta]

Yes, this is what we want. @dianacretu can you have a look on BYT implementation?

On qemu MU_A and MU_B sides are not in a continuous memory area.

So, when SOF writes on MU_A the write handler gets a reference to MU_A area but we couldn't figure out how to get a reference to MU_B also.

/* SHIM IO from ADSP */
static void mu_write(void *opaque, hwaddr addr,
        uint64_t val, unsigned size)
{
    struct adsp_io_info *info = opaque;
    struct adsp_dev *adsp = info->adsp;
    struct adsp_reg_space *space = info->space;
    struct qemu_io_msg_irq irq;
    uint64_t aux;

Here info->region[] points to MU_A but we would also want to have a reference to a similar object for MU_B.

[crazoes]

@dianacretu I am trying to make sof boot with fuzzer in imx8...in my investigation I noticed that the merged code doesn't write anything to parent yet which led me to your PR, seems like we introduce some writing to parent queue in this PR. Can you tell us what's left?
I can continue working on this if you can't.

[dianacretu]

@dianacretu I am trying to make sof boot with fuzzer in imx8...in my investigation I noticed that the merged code doesn't write anything to parent yet which led me to your PR, seems like we introduce some writing to parent queue in this PR. Can you tell us what's left?
I can continue working on this if you can't.

The problem with this PR is that the interrupts are not activated, so the interrupt handler from SOF does not receive any interrupts. More exactly, we need an equivalent to "adsp_set_lvl1_irq" function from the byt implementation. This function is used in byt-shim.c file when qemu received a message (which can be a message sent by the fuzzer), so we need to do the same for imx8.