Drivers: DAI: Intel: DMIC: Program start symmetrically for PDMx

[singalsu]

This PR contains proposed improvements to ACE and cAVS DMIC start sequence to avoid possible PDMx controllers out-of-sync start.

[singalsu]

I rebased this patch and dropped commit for spinlock set move. A SOF PR test for this patch runs now in thesofproject/sof#9859 .

[singalsu]

I dropped the cAVS patch. The issue is more difficult to repeat than the issue with PTL. I'm using this Octave script for stress testing the issue.

function ret = dmic_noise_stress_test()
dev = "hw:0,6";
nch = 4;
test_count = 100;
level_thr = -20;

fn1 = "/tmp/dmic_test_mute_1.wav";
fn2 = "/tmp/dmic_test_mute_2.wav";
redir = "2> /dev/null";

rec_cmd = sprintf('arecord -D%s -fS32_LE -r48000 -c%d -d 5 %s %s; sleep 0.5; arecord -D%s -fS32_LE -r48000 -c%d -d 5 %s %s', ...
                  dev, nch, fn1, redir, dev, nch, fn2, redir);

sleep_cmd = "sleep 1";

n = 0;
fail_count = 0;
for i = 1:test_count
        [ret, out] = system(rec_cmd);
        if ret
                fprintf(1, "Failed: %s\n", rec_cmd);
                return
        end

        % Check 1st capture
        n = n + 1;
        levels = channel_levels(fn1);
        fail = check_levels(levels, level_thr);
        fail_count = fail_count + fail;
        print_stats(n, nch, levels, fail, fail_count);

        n = n + 1;
        levels = channel_levels(fn2);
        fail = check_levels(levels, level_thr);
        fail_count = fail_count + fail;
        print_stats(n, nch, levels, fail, fail_count);

        [ret, out] = system(sleep_cmd);
        if ret
                fprintf(1, "Failed: %s\n", rec_cmd);
                return
        end
end
end

function print_stats(n, nch, levels, fail, fail_count)
        printf("%8d,", n);
        for j = 1:nch
                printf(" %8.1f,", levels(j));
        end
        printf("%8d, %8d, %8.1f\n", fail, fail_count, 100*fail_count/n);
end

function fail = check_levels(levels, level_thr)
        idx = find(levels > level_thr);
        if isempty(idx)
                fail = 0;
        else
                fail = 1;
        end
end

function levels = channel_levels(fn)
        [x, fs] = audioread(fn);
        s = size(x);
        nch = s(2);
        for i = 1:nch
                levels(i) = level_dbfs(x(:,i));
        end
end

function dbfs = level_dbfs(x)
level_ms = mean(x.^2);
dbfs = 10*log10(level_ms + 1e-20) + 20*log10(sqrt(2));
end

[singalsu]

With the failing version, with xt-clang, the trimmed output looks like this. The -132.2 dB levels are from disconnected microphones in DUT for channels 2 and 4. In the end there is 43.5% failures.

       1,    -49.6,   -132.2,    -49.5,   -132.2,       0,        0,      0.0
       2,    -49.2,   -134.9,     -0.9,   -134.9,       1,        1,     50.0
       3,    -49.4,   -135.1,     -0.9,   -135.1,       1,        2,     66.7
       4,    -49.3,   -135.0,    -49.2,   -135.0,       0,        2,     50.0
...
     197,    -49.3,   -132.2,     -0.9,   -132.2,       1,       85,     43.1
     198,    -49.2,   -132.2,     -0.9,   -132.2,       1,       86,     43.4
     199,    -49.6,   -134.9,     -0.9,   -134.9,       1,       87,     43.7
     200,    -49.2,   -133.2,    -49.2,   -133.2,       0,       87,     43.5

With gcc build, there is channels swapping happening. It can be seen with channels 3-4 levels . The disconnected level of -73 dB is due to different version of equalizer for gcc build. In the end 5.5% are failed. The channels swaps are not counted in the script.

       1,    -47.5,    -73.3,    -73.0,    -47.0,       0,        0,      0.0
       2,    -48.8,    -73.3,    -48.8,    -73.1,       0,        0,      0.0
       3,    -47.3,    -73.1,    -47.1,    -73.1,       0,        0,      0.0
       4,    -47.8,    -73.3,    -73.0,    -47.8,       0,        0,      0.0
---
     197,    -49.7,    -73.2,    -73.0,    -49.7,       0,       10,      5.1
     198,    -49.9,    -73.1,     -1.3,    -73.1,       1,       11,      5.6
     199,    -49.7,    -73.3,    -49.6,    -73.1,       0,       11,      5.5
     200,    -50.1,    -73.2,    -50.1,   -104.4,       0,       11,      5.5

[singalsu]

I lost PTL device while stress testing this patch with xt-clang build. The gcc build test completed for 100 loops without corruptions and channels swaps.

...
     197,    -49.9,    -73.1,    -50.0,    -73.1,       0,        0,      0.0
     198,    -50.0,    -73.1,    -50.3,    -73.1,       0,        0,      0.0
     199,    -50.1,    -73.3,    -50.4,    -73.1,       0,        0,      0.0
     200,    -50.0,    -73.1,    -50.4,    -73.1,       0,        0,      0.0

[singalsu]

The PTL xt-clang build of this patch passed 500 loops without corruption or channels swap.

---
     997,    -49.8,   -132.2,    -50.5,   -132.2,       0,        0,      0.0
     998,    -49.9,   -132.2,    -50.5,   -132.2,       0,        0,      0.0
     999,    -49.9,   -132.2,    -50.5,   -132.2,       0,        0,      0.0
    1000,    -49.6,   -135.2,    -50.3,   -134.5,       0,        0,      0.0