[PATCH 4/5] phy: rockchip-snps-pcie3: Check more sram init status for RK3588
Shawn Lin
shawn.lin at rock-chips.com
Fri Nov 13 01:55:27 PST 2026
Hi Alexey
在 2026/11/13 星期五 16:46, Alexey Charkov 写道:
> Hi Shawn,
>
> On Wed, Dec 24, 2025 at 11:10 AM Shawn Lin <shawn.lin at rock-chips.com> wrote:
>>
>> All the lower 4 bits should be checked which shows the mpllx_state.
>>
>> Signed-off-by: Shawn Lin <shawn.lin at rock-chips.com>
>> ---
>>
>> drivers/phy/rockchip/phy-rockchip-snps-pcie3.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/phy/rockchip/phy-rockchip-snps-pcie3.c b/drivers/phy/rockchip/phy-rockchip-snps-pcie3.c
>> index f5a5d0af..6cc38e3 100644
>> --- a/drivers/phy/rockchip/phy-rockchip-snps-pcie3.c
>> +++ b/drivers/phy/rockchip/phy-rockchip-snps-pcie3.c
>> @@ -43,7 +43,7 @@
>> #define RK3588_PCIE3PHY_GRF_PHY0_LN1_CON1 0x1104
>> #define RK3588_PCIE3PHY_GRF_PHY1_LN0_CON1 0x2004
>> #define RK3588_PCIE3PHY_GRF_PHY1_LN1_CON1 0x2104
>> -#define RK3588_SRAM_INIT_DONE(reg) (reg & BIT(0))
>> +#define RK3588_SRAM_INIT_DONE(reg) ((reg & 0xf) == 0xf)
>
> This breaks PCIe3x4 on my Radxa ROCK 5B, which then cannot find its
> NVMe root filesystem. Sebastian's rockchip-devel branch carries this
> series, and kernels built from it fail with:
>
> phy phy-fee80000.phy.13: lock failed 0x7, check input refclk and power supply
> rockchip-dw-pcie a40000000.pcie: phy lock failed
> rockchip-dw-pcie a40000000.pcie: probe with driver rockchip-dw-pcie
> failed with error -110
>
> The RK3588 TRM (V1.0, Part 1) describes PCIe3PHY_GRF_PHY{0,1}_STATUS1
> as follows:
>
> bit 0 sram_init_done
> bit 1 pcs_pwr_en (reset value 1)
> bit 2 pma_pwr_en (reset value 1)
> bit 3 mplla_state
> bits 5:4 reserved
> bit 6 mpllb_state
>
> So the lower four bits are not all PLL states: bits 1 and 2 are power
> switch enables, and the only PLL among them is MPLLA, while MPLLB is
> bit 6.
>
> On this board MPLLA is off even on a working link. With this patch
> reverted, PHY0_STATUS1 and PHY1_STATUS1 both read 0x47 once the link
> has trained at 8 GT/s x4: SRAM initialised, MPLLB running, MPLLA off.
> When the check fails they read 0x7, so neither PLL is running yet at
> that point, and checking for MPLLB instead would fail too.
>
> Reverting only this patch, with the rest of the series applied
> (rockchip-devel at d7436346fc31, v7.4-rc2), brings PCIe3x4 and the
> NVMe back.
>
Each port has a MPLLA + a MPLLB. In aggregation mode, MPLLA should
always be 1 once done for both of port0 and port1. But in bifurcation
mode, MPLLB will change. So basiclly, Rockchip PHY team suggests to
check mplla instead of mpllb for differnt use cases, during calibration.
That was done for vendor kernel for quite a long time. So I was
surprising to know in your case, rock 5b using it in aggregation mode,
it didn't work that way. Please let me ask interanlly...
> What is the check meant to catch? If it is PLL lock, which bits, and
> at what point, would be right for a link that runs on MPLLB?
> Otherwise, sram_init_done (bit 0) alone looks like the right condition
> here.
>
> Best regards,
> Alexey
>
> _______________________________________________
> Linux-rockchip mailing list
> Linux-rockchip at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-rockchip
More information about the Linux-rockchip
mailing list