| Message ID | 20241022061004.62812-1-e@freeshell.de (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | riscv: dts: starfive: Update ethernet phy0 delay parameter values for Star64 | expand |
On Mon, Oct 21, 2024 at 11:09:51PM -0700, E Shattow wrote: > Improve function of Star64 bottom network port phy0 with updated delay values. > Initial upstream patches supporting Star64 use the same vendor board support > package parameters known to result in an unreliable bottom network port. Should I add: Fixes: 2606bf583b962 ("riscv: dts: starfive: add Star64 board devicetree") CC: stable@vger.kernel.org ? "unreliable" sounds to me like something that is worthy of going to fixes/stable
E Shattow wrote: > Improve function of Star64 bottom network port phy0 with updated delay values. > Initial upstream patches supporting Star64 use the same vendor board support > package parameters known to result in an unreliable bottom network port. > > Success acquiring DHCP lease and no dropped packets to ping LAN address: > rx 900: tx 1500 1650 1800 1950 > rx 750: tx 1650 1800 1950 > rx 600: tx 1800 1950 > rx 1050: tx 1650 1800 1950 > rx 1200: tx 1500 1650 1800 1950 > rx 1350: tx 1500 1650 1800 1950 > rx 1500: tx 1500 1650 1800 1950 > rx 1650: tx 1500 1650 1800 1950 > rx 1800: tx 1500 1650 1800 1950 > rx 1900: tx 1950 > rx 1950: tx 1950 > > Failure acquiring DHCP lease or many dropped packets: > rx 450: tx 1500 1800 1950 > rx 600: tx 1200 1350 1650 > rx 750: tx 1350 1500 > rx 900: tx 1200 1350 > rx 1050: tx 1050 1200 1350 1500 > rx 1200: tx 1350 > rx 1350: tx 1350 > rx 1500: tx 1200 1350 > rx 1650: tx 1050 1200 1350 > rx 1800: tx 1050 1200 1350 > rx 1900: tx 1500 1650 1800 > rx 1950: tx 1200 1350 > > Non-functional: > rx 0: tx 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 > rx 150: tx 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 > rx 300: tx 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 > rx 450: tx 0 150 300 450 600 750 900 1050 1200 1350 1650 > rx 600: tx 0 150 300 450 600 750 900 1050 > rx 750: tx 0 150 300 450 600 750 900 1050 1200 > rx 900: tx 0 150 300 450 600 750 900 1050 > rx 1050: tx 0 150 300 450 600 750 900 > rx 1200: tx 0 150 300 450 600 750 900 1050 1200 > rx 1350: tx 0 150 300 450 600 750 900 1050 1200 > rx 1500: tx 0 150 300 450 600 750 900 1050 > rx 1650: tx 0 150 300 450 600 750 900 > rx 1800: tx 0 150 300 450 600 750 900 > rx 1900: tx 0 150 300 450 600 750 900 1050 1200 1350 > rx 1950: tx 0 150 300 450 600 750 900 1050 > > Selecting the median of all working rx delay values 1500 combined with tx delay > values 1500, 1650, 1800, and 1950 only the tx delay value of 1950 (default) is > reliable as tested in both Linux 6.11.2 and U-Boot v2024.10 > > Signed-off-by: E Shattow <e@freeshell.de> Thanks for working this out. Acked-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
On 10/22/24 09:41, Conor Dooley wrote: > On Mon, Oct 21, 2024 at 11:09:51PM -0700, E Shattow wrote: >> Improve function of Star64 bottom network port phy0 with updated delay values. >> Initial upstream patches supporting Star64 use the same vendor board support >> package parameters known to result in an unreliable bottom network port. > Should I add: > Fixes: 2606bf583b962 ("riscv: dts: starfive: add Star64 board devicetree") > CC: stable@vger.kernel.org > ? > > "unreliable" sounds to me like something that is worthy of going to > fixes/stable Applying as a fix to stable sounds reasonable, thanks. The bottom network port has always been known by Star64 users in reviews and discussions to be affected by dropped packets and low network throughput. If we want to prove correctness does this require expertise and use of an oscilloscope to characterize the signal timing? Though I am not sure I got it right, it's not worse than previously was on any of these Star64 boards in the wild and probably is better for at least some (if not all). Notable aside is to mention the re-worked motorcomm driver of more-recent Linux kernel releases (when compared to the vendor board support package) dropped the Fast Ethernet configuration parameters on the reasoning that Fast Ethernet (as compared to Gigabit Ethernet) is relatively slow enough of a signal that a default delay parameter is good enough for all use cases. The non-default Fast Ethernet delay parameter values missing from the upstream effort are not possible to implement or test for in my effort here, but are no worse or better for having this patch applied. -E
From: Conor Dooley <conor.dooley@microchip.com> On Mon, 21 Oct 2024 23:09:51 -0700, E Shattow wrote: > Improve function of Star64 bottom network port phy0 with updated delay values. > Initial upstream patches supporting Star64 use the same vendor board support > package parameters known to result in an unreliable bottom network port. > > Success acquiring DHCP lease and no dropped packets to ping LAN address: > rx 900: tx 1500 1650 1800 1950 > rx 750: tx 1650 1800 1950 > rx 600: tx 1800 1950 > rx 1050: tx 1650 1800 1950 > rx 1200: tx 1500 1650 1800 1950 > rx 1350: tx 1500 1650 1800 1950 > rx 1500: tx 1500 1650 1800 1950 > rx 1650: tx 1500 1650 1800 1950 > rx 1800: tx 1500 1650 1800 1950 > rx 1900: tx 1950 > rx 1950: tx 1950 > > [...] Applied to riscv-soc-fixes, thanks! [1/1] riscv: dts: starfive: Update ethernet phy0 delay parameter values for Star64 https://git.kernel.org/conor/c/825bb69228c8 Thanks, Conor.
diff --git a/arch/riscv/boot/dts/starfive/jh7110-pine64-star64.dts b/arch/riscv/boot/dts/starfive/jh7110-pine64-star64.dts index b720cdd15ed6..8e39fdc73ecb 100644 --- a/arch/riscv/boot/dts/starfive/jh7110-pine64-star64.dts +++ b/arch/riscv/boot/dts/starfive/jh7110-pine64-star64.dts @@ -44,8 +44,7 @@ &pcie1 { }; &phy0 { - rx-internal-delay-ps = <1900>; - tx-internal-delay-ps = <1500>; + rx-internal-delay-ps = <1500>; motorcomm,rx-clk-drv-microamp = <2910>; motorcomm,rx-data-drv-microamp = <2910>; motorcomm,tx-clk-adj-enabled;
Improve function of Star64 bottom network port phy0 with updated delay values. Initial upstream patches supporting Star64 use the same vendor board support package parameters known to result in an unreliable bottom network port. Success acquiring DHCP lease and no dropped packets to ping LAN address: rx 900: tx 1500 1650 1800 1950 rx 750: tx 1650 1800 1950 rx 600: tx 1800 1950 rx 1050: tx 1650 1800 1950 rx 1200: tx 1500 1650 1800 1950 rx 1350: tx 1500 1650 1800 1950 rx 1500: tx 1500 1650 1800 1950 rx 1650: tx 1500 1650 1800 1950 rx 1800: tx 1500 1650 1800 1950 rx 1900: tx 1950 rx 1950: tx 1950 Failure acquiring DHCP lease or many dropped packets: rx 450: tx 1500 1800 1950 rx 600: tx 1200 1350 1650 rx 750: tx 1350 1500 rx 900: tx 1200 1350 rx 1050: tx 1050 1200 1350 1500 rx 1200: tx 1350 rx 1350: tx 1350 rx 1500: tx 1200 1350 rx 1650: tx 1050 1200 1350 rx 1800: tx 1050 1200 1350 rx 1900: tx 1500 1650 1800 rx 1950: tx 1200 1350 Non-functional: rx 0: tx 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 rx 150: tx 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 rx 300: tx 0 150 300 450 600 750 900 1050 1200 1350 1500 1650 1800 1950 rx 450: tx 0 150 300 450 600 750 900 1050 1200 1350 1650 rx 600: tx 0 150 300 450 600 750 900 1050 rx 750: tx 0 150 300 450 600 750 900 1050 1200 rx 900: tx 0 150 300 450 600 750 900 1050 rx 1050: tx 0 150 300 450 600 750 900 rx 1200: tx 0 150 300 450 600 750 900 1050 1200 rx 1350: tx 0 150 300 450 600 750 900 1050 1200 rx 1500: tx 0 150 300 450 600 750 900 1050 rx 1650: tx 0 150 300 450 600 750 900 rx 1800: tx 0 150 300 450 600 750 900 rx 1900: tx 0 150 300 450 600 750 900 1050 1200 1350 rx 1950: tx 0 150 300 450 600 750 900 1050 Selecting the median of all working rx delay values 1500 combined with tx delay values 1500, 1650, 1800, and 1950 only the tx delay value of 1950 (default) is reliable as tested in both Linux 6.11.2 and U-Boot v2024.10 Signed-off-by: E Shattow <e@freeshell.de> --- arch/riscv/boot/dts/starfive/jh7110-pine64-star64.dts | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)