rtw89: WOW_CPUIO_RX_CTRL is never answered on RTL8851B

rtw89: WOW_CPUIO_RX_CTRL is never answered on RTL8851B

[Maxim Skokov]

On RTL8851BE every suspend/resume cycle with WoWLAN enabled spends about
20 seconds in firmware timeouts:

  rtw89_8851be 0000:02:00.0: c2h reg timeout          (x3)
  rtw89_8851be 0000:02:00.0: Unknown wakeup reason 0
  rtw89_8851be 0000:02:00.0: c2h reg timeout          (x3)

With WoWLAN disabled the same suspend/resume path is silent. Both results
are 3 out of 3, using rtcwake so the cycles are identical.

A kretprobe on rtw89_fw_msg_reg() shows all six timeouts are the same
command:

  h2c=10  ret=-110  3355 ms   RTW89_FWCMD_H2CREG_FUNC_WOW_CPUIO_RX_CTRL
  h2c=10  ret=-110  3345 ms
  h2c=10  ret=-110  3344 ms
  h2c=10  ret=-110  3315 ms
  h2c=10  ret=-110  3305 ms
  h2c=10  ret=-110  3307 ms
  h2c=7   ret=0        0 ms   AOAC_RPT_1
  h2c=8   ret=0        1 ms   AOAC_RPT_2
  h2c=5   ret=0        0 ms   SCH_TX_EN
  h2c=5   ret=0        1 ms   SCH_TX_EN

Other commands on the same h2creg channel are answered in 0 to 1 ms, so
the channel itself works. The firmware just never answers this one.

Six calls because rtw89_mac_cpu_io_rx() retries CPU_IO_RX_RETRY_CNT = 3
times and is called once on suspend and once on resume. Its return value
is ignored by rtw89_wow_config_mac_ax(), so WoWLAN setup is not aborted
and the only cost is the delay.

_rtw89_mac_cpu_io_rx() already has a way out:

	if (RTW89_CHK_FW_FEATURE(NO_WOW_CPU_IO_RX, &rtwdev->fw))
		return 0;

but rtw89_fw_feat_list only carries

	__CFG_FW_FEAT(RTL8852A, lt, 0, 13, 37, 0, NO_WOW_CPU_IO_RX),
	__CFG_FW_FEAT(RTL8852B, lt, 0, 29, 30, 0, NO_WOW_CPU_IO_RX),

and nothing for RTL8851B.

Should RTL8851B get an entry, and with what version boundary? The newest
firmware published for this chip in linux-firmware is 0.29.41.5, and it
does not answer the command either, so from here I cannot tell whether
support is planned for a later version. I am happy to send and test a
patch once I know which version to use.

One unrelated note from the same trace: "Unknown wakeup reason 0" is not
a failed AoAC read. AOAC_RPT_1 and AOAC_RPT_2 both succeed, and
rtw89_wow_show_wakeup_reason() reads the reason straight from
wow_reason_reg, which holds 0 because the card was not the wake source
here, the RTC alarm was. The switch has no case for that and falls
through to the warning.

Tested on:
  RTL8851BE [10ec:b851], subsystem AzureWave [1a3b:6111]
  wireless-next 13e51269b6, built with KASAN, UBSAN, PROVE_LOCKING and
  DEBUG_OBJECTS; none of them fire during these cycles
  firmware 0.29.41.3 (65cefb31), also reproduced with 0.29.41.5

RE: rtw89: WOW_CPUIO_RX_CTRL is never answered on RTL8851B

[Ping-Ke Shih]

Maxim Skokov <skokovmaksimevg@gmail.com> wrote:
> 
> but rtw89_fw_feat_list only carries
> 
>         __CFG_FW_FEAT(RTL8852A, lt, 0, 13, 37, 0, NO_WOW_CPU_IO_RX),
>         __CFG_FW_FEAT(RTL8852B, lt, 0, 29, 30, 0, NO_WOW_CPU_IO_RX),
> 
> and nothing for RTL8851B.

I'll send a patch to add this:

__CFG_FW_FEAT(RTL8851B, lt, 0, 29, 57, 0, NO_WOW_CPU_IO_RX),

> 
> One unrelated note from the same trace: "Unknown wakeup reason 0" is not
> a failed AoAC read. AOAC_RPT_1 and AOAC_RPT_2 both succeed, and
> rtw89_wow_show_wakeup_reason() reads the reason straight from
> wow_reason_reg, which holds 0 because the card was not the wake source
> here, the RTC alarm was. The switch has no case for that and falls
> through to the warning.

I'll check this internally if it is possible to avoid the "reason 0" 
message.