On Wed, Oct 5, 2022 at 4:54 AM Ondrej Jirman <megi@xxxxxx> wrote: > Good Morning, > I have two Realtek PCIe wifi cards connected over the 4 port PCIe swtich > to Quartz64-A. The cards fail to work, when nvme SSD is connected at the > same time to the bridge. Without nvme connected, cards work fine. The > issue seems to be related to mixed use of devices which make use of I/O > ranges and memory ranges. > > This patch changes I/O, MEM and config mappings so that config and I/O > mappings use the 0xf4000000 outbound address space, and MEM range uses > the whole 0x300000000 outbound space. > > This is simialar to how BSP does the mappings. This change was very recent in the BSP stuff (Jan 2022): https://github.com/rockchip-linux/kernel/commit/cfab7abefc4093daa379fbd90a1e7ac1a484332b A few other interesting changes there as well. They added a 32 bit window in the lower range and made the entire upper range a 64 bit relocatable (why?) and prefetchable window. They also set the viewport number to 8. The dt-binding says this is autodetected, but I wonder if the value is being detected correctly. It looks like it is dependent in BSP on a backported change from mainline: https://github.com/rockchip-linux/kernel/commit/50a01d3c10a6212f66364575a3c8f66c07f41591 Can someone weigh in why the dw core has config in the reg node instead of ranges? > > I changed num-ob-windows to value detected by the kernel so if for whatever > reason the kernel ever starts respecting this DT property, it would not > switch to sharing I/O and CFG spaces via a single iATU mapping for > no reason. This worries me that this value may be being detected incorrectly, they set it to this for a reason. It's not unheard of for Rockchip to need to override what they encode in the silicon. Very Respectfully, Peter Geis > > This change to the regs/ranges makes the issue go away and both nvme and > wifi cards work when connected at the same time to the bridge. I tested > the nvme with large amount of reads/writes, both behind the PCIe bridge > and when directly connected to Quartz64-A board. > > Signed-off-by: Ondrej Jirman <megi@xxxxxx> > --- > BSP for reference: https://github.com/rockchip-linux/kernel/blob/develop-4.19/arch/arm64/boot/dts/rockchip/rk3568.dtsi#L2370 > > v2: > - change ranges to use 0x300000000 fully for MEM and make use of > the 0xf4000000 outbound range for IO and config > - full retest with/without the switch > - if lscpi/dmesg is useful in the future for comparison, see: > https://xff.cz/kernels/random/quartz64a-pcie/ > > I used this script for the tests: > > #!/bin/bash > > OUT=/mnt/data > n=8 > > test -f /tmp/test.dat || \ > dd if=/dev/urandom of=/tmp/test.dat bs=1M count=1024 > md5sum /tmp/test.dat > > i=0 > while test $i -lt $n > do > dd if=/tmp/test.dat of=$OUT/test$i.dat bs=4M oflag=direct > > i=$(($i+1)) > done > > i=0 > while test $i -lt $n > do > dd if=$OUT/test$i.dat bs=4M iflag=direct | md5sum > > i=$(($i+1)) > done > > > arch/arm64/boot/dts/rockchip/rk356x.dtsi | 9 +++++---- > 1 file changed, 5 insertions(+), 4 deletions(-) > > diff --git a/arch/arm64/boot/dts/rockchip/rk356x.dtsi b/arch/arm64/boot/dts/rockchip/rk356x.dtsi > index 319981c3e9f7..99fd9543fc6f 100644 > --- a/arch/arm64/boot/dts/rockchip/rk356x.dtsi > +++ b/arch/arm64/boot/dts/rockchip/rk356x.dtsi > @@ -855,7 +855,8 @@ pcie2x1: pcie@fe260000 { > compatible = "rockchip,rk3568-pcie"; > reg = <0x3 0xc0000000 0x0 0x00400000>, > <0x0 0xfe260000 0x0 0x00010000>, > - <0x3 0x3f000000 0x0 0x01000000>; > + <0x0 0xf4000000 0x0 0x01f00000>; > + > reg-names = "dbi", "apb", "config"; > interrupts = <GIC_SPI 75 IRQ_TYPE_LEVEL_HIGH>, > <GIC_SPI 74 IRQ_TYPE_LEVEL_HIGH>, > @@ -877,15 +878,15 @@ pcie2x1: pcie@fe260000 { > <0 0 0 4 &pcie_intc 3>; > linux,pci-domain = <0>; > num-ib-windows = <6>; > - num-ob-windows = <2>; > + num-ob-windows = <8>; > max-link-speed = <2>; > msi-map = <0x0 &gic 0x0 0x1000>; > num-lanes = <1>; > phys = <&combphy2 PHY_TYPE_PCIE>; > phy-names = "pcie-phy"; > power-domains = <&power RK3568_PD_PIPE>; > - ranges = <0x01000000 0x0 0x3ef00000 0x3 0x3ef00000 0x0 0x00100000 > - 0x02000000 0x0 0x00000000 0x3 0x00000000 0x0 0x3ef00000>; > + ranges = <0x01000000 0x0 0x00000000 0x0 0xf5f00000 0x0 0x00100000 > + 0x02000000 0x0 0x40000000 0x3 0x00000000 0x0 0x40000000>; > resets = <&cru SRST_PCIE20_POWERUP>; > reset-names = "pipe"; > #address-cells = <3>; > -- > 2.37.3 >