Re: [PATCH v2 1/7] dt-bindings: clock: Document T-Head TH1520 AP_SUBSYS controller

[Drew Fustini <drew@pdp7.com>]

On Sat, Jul 27, 2024 at 01:21:39PM +0800, Icenowy Zheng wrote:
> 在 2024-07-26星期五的 09:38 -0700，Drew Fustini写道：
> > On Fri, Jul 26, 2024 at 03:45:36AM -0500, Emil Renner Berthing wrote:
> > > Stephen Boyd wrote:
> > > > Quoting Drew Fustini (2024-06-23 19:12:31)
> > > > > Document bindings for the T-Head TH1520 AP sub-system clock
> > > > > controller.
> > > > > 
> > > > > Link:
> > > > > https://openbeagle.org/beaglev-ahead/beaglev-ahead/-/blob/main/docs/TH1520%20System%20User%20Manual.pdf
> > > > > Co-developed-by: Yangtao Li <frank.li@vivo.com>
> > > > > Signed-off-by: Yangtao Li <frank.li@vivo.com>
> > > > > Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
> > > > > Signed-off-by: Drew Fustini <dfustini@tenstorrent.com>
> > > > > ---
> > > > 
> > > > Applied to clk-next
> > > 
> > > Thanks, but this driver seems a bit incomplete. With this applied
> > > the Lichee Pi
> > > 4A no longer boots without the clk_ignore_unused kernel parameter.
> > > 
> > > /Emil
> > 
> > Is this the case when you apply the dts patches from this series?
> > 
> > The dts patches won't go in until 6.12 so I don't think the presence
> > of
> > the clk-th1520-ap.c itself in 6.11 would break existing systems.
> > 
> > That said, I have been using clk_ignore_unused. I had been thinking
> > that
> > made sense because the full set of clock controller drivers like
> > AON_SUBSYS (always on), AUDIO_SUBSYS, DSP_SUBSYS, etc, are not
> > present
> > yet in mainline. However, the T-Head vendor kernel does have drivers
> > for
> > all those clock controllers and I was suprised to see that the vendor
> > kernel fails to boot when I just tested removing clk_ignore_unused.
> > 
> > As for clk-th1520-ap.c in mainline, I'll investigate further which
> > clk
> > disables seem to causing the boot failure when using the dts from
> > this
> > series. I suspect I may need to add nodes that will cause the
> > necessary
> > clks to be enabled by their respective drivers.
> 
> If disabling the clock just leads to system hang, setting
> CLK_IS_CRITICAL should be useful (and needed) here.

Thank you for the suggestion about CLK_IS_CRITICAL.

I have found through process of elimination that the "emmc-sdio" clock
(CLK_EMMC_SDIO) fails to work after the "vp-axi" clock (CLK_VP_AXI) is
disabled. I added the CLK_IGNORE_UNUSED flag to "vp-axi" and the system
is able to boot correctly. I think in that case CLK_IS_CRITICAL is not
needed.

I've been reviewing the TH1520 System User Manual [1] and I am uncertain
why the state of "vp-axi" would affect "emmc-sdio":

  - EMMC_SDIO_REF_CLK_EN is bit 30 in PERI_CLK_CFG (offset 0x204). 
  - VPSYS_AXI_ACLK_EN is bit 15 in VPSYS_CLK_CFG (offsset 0x1e0).

I don't see any linkage between them in the public documentation.
Regardless, the addition of the CLK_IGNORE_UNUSED flag to "vp-axi" fixes
the boot failure, so I will post a patch to that effect.

Thanks,
Drew

[1] https://openbeagle.org/beaglev-ahead/beaglev-ahead/-/blob/main/docs/TH1520%20System%20User%20Manual.pdf