rockchip/64: fix pl330 cyclic dma transfers

[paolosabatino]

Description

PL330 DMA controller driver in mainline kernel is suboptimal; according to this patch, the driver can be modified to make it more efficient and stable.

This inefficiency causes ALL rockchip SoCs I tested (rk3228, rk3308, rk3328 and rk3399) to produce pops and clicks when playing through I²S (ie: analog or HDMI) or SPDIF outputs even with modest load on memory subsystem, but the DMA controller also serves UART, SPI bus, CAN bus, PWMs, and some other device... so possibly the DMA issues of the mainline driver can affect several other areas, sound is just the most evident one.

In my case, I can steadily reproduce the problem on all the mentioned SoCs this way:

• play a track via aplay, ogg123 or whatever. The issue appears more prominent if at high samplerates (>= 96khz)
• at the same time run mbw, tying the process to a core which is not the same core handling DMA controller interrupts

An example:

ogg123 -d alsa -o dev:hw:1 "fr08 soundtrack.ogg"
taskset -c 3 mbw -t2 -n 256 32M  # use "-c 5" to tie to little core on RK3399

This PR takes the patch above and applies to kernel 6.12 (current) and 6.13 (edge) for both rockchip and rockchip64 families.

After the patch, the issue is much more reduced, but not totally cancelled yet, because if you add a third iperf3 process to the mix, the clicks and pops still appears (at least on rk3308):

ogg123 -d alsa -o dev:hw:1 "fr08 soundtrack.ogg"
iperf3 -c <your_iperf_server_ip> -t 120 -P 4
taskset -c 3 mbw -t2 -n 256 32M  # use "-c 5" to tie to little core on RK3399

Nonetheless, this is a first step that will address the common scenario with "regular" workloads and is probably a good idea to bring in this fix until I progress in completely fixing the issue. As a note, the rockchip vendor 4.4 kernel on rk3308 does not show any issue.

I don't know if rk356x and rk3588 also are affected, but since they use pl330 DMA controllers as well, I guess the issues can appear there too. Other possibly affected SoC may be some Samsung Exynos SoCs.

SPDIF maxburst parameter is also increased from 4 to 8 words, as it is in the vendor kernel.

GitHub issue reference:
Jira reference number AR-2584

How Has This Been Tested?

• current 6.12 kernel built and packaged for both rockchip and rockchip64
• edge 6.13 kernel built and packaged for both rockchip and rockchip64
• tested current 6.12 kernel on rk322x, rk3308, rk3328 and rk3399: kernel boots on all SoCs and a run of ogg123 + mbw has been done to verify there were no artifacts in analog and SPDIF audio

Checklist:

Please delete options that are not relevant.

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• My changes generate no new warnings

[paolosabatino]

I move this to draft, there seems more involved into and I want to dig deeper...