After 3 days, I managed to reproduce it again with 2 android compilation and firefox with kernel parameter mem=4096m (I have 16GB memory). I found that it's easier to reproduce when the memory is full.

sfc@sfc-DESKTOP:~$ sudo umr -lb
        raven1.gfx91
        raven1.vcn10
        raven1.dcn10
        raven1.nbio70
        raven1.sdma041
        raven1.hdp40
        raven1.oss40
        raven1.mmhub91
        raven1.mp100
sfc@sfc-DESKTOP:~$ sudo umr -O verbose,follow_ib -R gfx[.]
error: Unknown option [follow_ib]
sfc@sfc-DESKTOP:~$ sudo umr -O bits -wa
No active waves!
sfc@sfc-DESKTOP:~$ sudo umr -O many,bits  -r*.*.mmGRBM_STATUS
[ERROR]: Unknown option <-r*.*.mmGRBM_STATUS>
sfc@sfc-DESKTOP:~$ sudo umr -O many,bits  -r *.*.HEADER_DUMP
gfx91.mmCP_MEC_ME1_HEADER_DUMP => 0xc0000e00
        .HEADER_DUMP[0:31]                                               == 3221229056 (0xc0000e00)
gfx91.mmCP_MEC_ME2_HEADER_DUMP => 0xdef0def0
        .HEADER_DUMP[0:31]                                               == 3740327664 (0xdef0def0)
gfx91.mmCP_ME_HEADER_DUMP => 0xc0004200
        .ME_HEADER_DUMP[0:31]                                            == 3221242368 (0xc0004200)
gfx91.mmCP_PFP_HEADER_DUMP => 0xffff1000
        .PFP_HEADER_DUMP[0:31]                                           == 4294905856 (0xffff1000)
gfx91.mmCP_CE_HEADER_DUMP => 0xffff1000
        .CE_HEADER_DUMP[0:31]                                            == 4294905856 (0xffff1000)
sfc@sfc-DESKTOP:~$ sudo umr -O many,bits  -r *.*.CP_EOP
gfx91.mmCP_EOPQ_WAIT_TIME => 0x0000052c
        .WAIT_TIME[0:9]                                                  ==      300 (0x0000012c)
        .SCALE_COUNT[10:17]                                              ==        1 (0x00000001)
gfx91.mmCP_EOP_DONE_ADDR_LO => 0x00609000
        .ADDR_LO[2:31]                                                   ==  1582080 (0x00182400)
gfx91.mmCP_EOP_DONE_ADDR_HI => 0x000000f5
        .ADDR_HI[0:15]                                                   ==      245 (0x000000f5)
gfx91.mmCP_EOP_DONE_DATA_LO => 0x000008e4
        .DATA_LO[0:31]                                                   ==     2276 (0x000008e4)
gfx91.mmCP_EOP_DONE_DATA_HI => 0x00000000
        .DATA_HI[0:31]                                                   ==        0 (0x00000000)
gfx91.mmCP_EOP_LAST_FENCE_LO => 0x000008e4
        .LAST_FENCE_LO[0:31]                                             ==     2276 (0x000008e4)
gfx91.mmCP_EOP_LAST_FENCE_HI => 0x00000000
        .LAST_FENCE_HI[0:31]                                             ==        0 (0x00000000)
gfx91.mmCP_EOP_DONE_EVENT_CNTL => 0x00038060
        .WBINV_TC_OP[0:6]                                                ==       96 (0x00000060)
        .WBINV_ACTION_ENA[12:17]                                         ==       56 (0x00000038)
        .CACHE_POLICY[25:25]                                             ==        0 (0x00000000)
        .EXECUTE[28:28]                                                  ==        0 (0x00000000)
gfx91.mmCP_EOP_DONE_DATA_CNTL => 0x40010000
        .DST_SEL[16:17]                                                  ==        1 (0x00000001)
        .INT_SEL[24:26]                                                  ==        0 (0x00000000)
        .DATA_SEL[29:31]                                                 ==        2 (0x00000002)
gfx91.mmCP_EOP_DONE_CNTX_ID => 0x00000000
        .CNTX_ID[0:31]                                                   ==        0 (0x00000000)

ddebug_dumps:

---
Command: /usr/lib/firefox/firefox 
Driver vendor: X.Org
Device vendor: AMD
Device name: AMD RAVEN (DRM 3.25.0, 4.17.2-041702-generic, LLVM 6.0.0)

Remainder of driver log:
---

I tried netconsole and could not get it working. I bought a serial converter and it's on the way. When it delivered, I will try getting the log from serial port. Then I could try your patches because currently the logs did not get persisted at all. Thanks for your patches.