Some information in case someone tries to reproduce it.

I applied this patch additionally to the revert:
https://bugs.freedesktop.org/show_bug.cgi?id=80880#c10

Many demos work now. But the Cave Demo does not work, it hangs the GPU.

I can not reproduce the original bug I posted here anymore with the patch, just
the hang. It looks like this:

ring 0 stalled for more than 10146msec
GPU lockup (waiting for 0x000000000000a921 last fence id 0x000000000000a920 on
ring 0)
failed to get a new IB (-35)
Saved 205 dwords of commands on ring 0.
GPU softreset: 0x0000004D
  GRBM_STATUS               = 0xE7DE5028
  GRBM_STATUS_SE0           = 0xEF800002
  GRBM_STATUS_SE1           = 0xEFC00002
  SRBM_STATUS               = 0x200000C0
  SRBM_STATUS2              = 0x00000000
  R_008674_CP_STALLED_STAT1 = 0x00000000
  R_008678_CP_STALLED_STAT2 = 0x00010800
  R_00867C_CP_BUSY_STAT     = 0x00408006
  R_008680_CP_STAT          = 0x84038647
  R_00D034_DMA_STATUS_REG   = 0x44C83146
  R_00D834_DMA_STATUS_REG   = 0x44C83D57
  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x00000000
  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x00000000
GRBM_SOFT_RESET=0x0000DDFF
SRBM_SOFT_RESET=0x00100100
  GRBM_STATUS               = 0x00003028
  GRBM_STATUS_SE0           = 0x00000006
  GRBM_STATUS_SE1           = 0x00000006
  SRBM_STATUS               = 0x200000C0
  SRBM_STATUS2              = 0x00000000
  R_008674_CP_STALLED_STAT1 = 0x00000000
  R_008678_CP_STALLED_STAT2 = 0x00000000
  R_00867C_CP_BUSY_STAT     = 0x00000000
  R_008680_CP_STAT          = 0x00000000
  R_00D034_DMA_STATUS_REG   = 0x44C83D57
  R_00D834_DMA_STATUS_REG   = 0x44C83D57
GPU reset succeeded, trying to resume

I'll leave this bug report open, because I have no idea if this bug will
reappear once the hang is fixed.