Hello,
At 2025-10-21 00:00:59, "Sebastian Reichel" sebastian.reichel@collabora.com wrote:
Hi,
On Mon, Oct 20, 2025 at 02:49:10PM +0200, Heiko Stuebner wrote:
Am Donnerstag, 16. Oktober 2025, 00:57:15 Mitteleuropäische Sommerzeit schrieb Sebastian Reichel:
On Wed, Oct 15, 2025 at 03:27:12PM +0200, Heiko Stübner wrote:
Am Mittwoch, 15. Oktober 2025, 14:58:46 Mitteleuropäische Sommerzeit schrieb Quentin Schulz:
On 10/8/25 3:31 PM, Heiko Stuebner wrote:
dclk_vop2_src currently has CLK_SET_RATE_PARENT | CLK_SET_RATE_NO_REPARENT flags set, which is vastly different than dclk_vop0_src or dclk_vop1_src, which have none of those.
With these flags in dclk_vop2_src, actually setting the clock then results in a lot of other peripherals breaking, because setting the rate results in the PLL source getting changed:
[ 14.898718] clk_core_set_rate_nolock: setting rate for dclk_vop2 to 152840000 [ 15.155017] clk_change_rate: setting rate for pll_gpll to 1680000000 [ clk adjusting every gpll user ]
This includes possibly the other vops, i2s, spdif and even the uarts. Among other possible things, this breaks the uart console on a board I use. Sometimes it recovers later on, but there will be a big block
I can reproduce on the same board as yours and this fixes the issue indeed (note I can only reproduce for now when display the modetest pattern, otherwise after boot the console seems fine to me).
I boot into a Debian rootfs with fbcon on my system, and the serial console produces garbled output when the vop adjusts the clock
Sometimes it recovers after a bit, but other times it doesn't
Reviewed-by: Quentin Schulz quentin.schulz@cherry.de Tested-by: Quentin Schulz quentin.schulz@cherry.de # RK3588 Tiger w/DP carrierboard
I'm pretty sure I've seen this while playing with USB-C DP AltMode on Rock 5B. So far I had no time to investigate further.
What I'm missing in the commit message is the impact on VOP. Also it might be a good idea to have Andy in Cc, so I've added him.
Hmm, it brings VP2 in line with the other two VPs, only VP2 had this special setting - even right from the start, so it could very well have been left there accidentially during submission.
I did the initial upstream submission based on downstream (the TRM is quite bad regading describing the clock trees, so not much validation has been done by me). The old vendor kernel tree had it like this, but that also changed a bit over time afterwards and no longer has any special handling for VP2. OTOH it does set CLK_SET_RATE_NO_REPARENT for all dclk_vop<number>_src, which you are now removing for VP2.
FWIW these are the two flags:
#define CLK_SET_RATE_PARENT BIT(2) /* propagate rate change up one level */ #define CLK_SET_RATE_NO_REPARENT BIT(7) /* don't re-parent on rate change */
So by removing CLK_SET_RATE_NO_REPARENT you are allowing dclk_vop2_src to be switched to a different PLL when a different rate is being requested. That change is completley unrelated to the bug you are seeing right now?
So in the end VP2 will have to deal with this, because when the VP causes a rate change in the GPLL, this changes so many clocks of other possibly running devices. Not only the uart, but also emmc and many more. And all those devices do not like if their clock gets changed under them I think.
It's certainly weird, that VP2 was (and still is in upstream) handled special. Note that GPLL being changed is not really necessary. dclk_vop2_src parent can be GPLL, CPLL, V0PLL or AUPLL. Effects on other hardware IP very much depends on the parent setup. What I try to understand is if there is also a bug in the rockchipdrm driver and/or if removing CLK_SET_RATE_NO_REPARENT is a good idea. That's why I hoped Andy could chime in and provide some background :)
The main limitation is that there are not enough PLLs on the SoC to be used for the display side. In our downstream code implementation, we usually exclusively assign V0PLL to a certain VP. Other VPs generally need to share the PLL with other peripherals , or use the HDMI PHY PLL.
For GPLL and CPLL, they will be set to a fixed frequency during the system startup stage, and they should not be modified again as these two PLL always shared by other peripherals.
When shared with other peripherals, we can not do CLK_SET_RATE_PARENT,. However, when we need a relatively precise frequency in certain scenarios, such as driving an eDP or DSI panel（see what we do for eDP on rk3588s-evb1-v10.dts and rk3588-coolpi-cm5-genbook.dts ）, we tend to use V0PLL. But since V0PLL does not proper initializated at system startup, we then need CLK_SET_RATE_PARENT. This does indeed seem to be a contradiction.
Greetings,
-- Sebastian