[Macchiato] EDKII grub boot fails with PCIe init

Tue Mar 13 12:55:46 GMT 2018

On 13 March 2018 at 12:48, Frederik Lotter
<frederik.lotter at netronome.com> wrote:
> On Tue, Mar 13, 2018 at 2:36 PM, Ard Biesheuvel <ard.biesheuvel at linaro.org>
> wrote:
>>
>> On 13 March 2018 at 12:26, Frederik Lotter
>> <frederik.lotter at netronome.com> wrote:
>> > Hi Ard,
>> >
>> > On Mon, Mar 12, 2018 at 6:22 PM, Ard Biesheuvel
>> > <ard.biesheuvel at linaro.org>
>> > wrote:
>> >>
>> >> On 12 March 2018 at 16:15, Frederik Lotter
>> >> <frederik.lotter at netronome.com> wrote:
>> >> > Hi,
>> >> > I am getting CPU stall warnings when booting up using the EFI route.
>> >> > I
>> >> > suspect the PCIe interface, as the stall warning sometimes contain
>> >> > the
>> >> > probe
>> >> > function. Other times is seems to get further than PCIe init, but
>> >> > still
>> >> > stall interrupt handling.
>> >> > Here are some facts around my observation:
>> >> >
>> >> > I have two sdcards for my Machiattobin board. They have identical
>> >> > kernels
>> >> > (4.16 rc5) with Ubuntu 16.04 rootfs. The one sdcard uses a uboot, DT
>> >> > and
>> >> > kernel boot. The second sdcard has EDKII, grub kernel boot. The EDKII
>> >> > build
>> >> > includes the device tree DTB (and DTS which I believe is unused) from
>> >> > the
>> >> > one used on the uboot sdcard.
>> >> >
>> >> > EFI stub: Booting Linux Kernel...
>> >> > EFI stub: Using DTB from configuration table
>> >> > EFI stub: Exiting boot services and installing virtual address map...
>> >> > [    0.000000] Booting Linux on physical CPU 0x0000000000
>> >> > [0x410fd081]
>> >> > [    0.000000] Linux version 4.16.0-rc5-mbcin-netronome-2-dirty
>> >> > (root at mcb1-cpt) (gcc version 5.4.0 20160609 (Ubuntu/Linaro
>> >> > 5.4.0-6ubuntu1~16.04.9)) #2 SMP PREEMPT Mon Mar 12 14:40:25 UTC 2018
>> >> > [    0.000000] Machine model: Marvell 8040 MACHIATOBin
>> >> > [    0.000000] efi: Getting EFI parameters from FDT:
>> >> > [    0.000000] efi: EFI v2.70 by EDK II
>> >> > [    0.000000] efi:  SMBIOS 3.0=0xbfd00000  ACPI 2.0=0xb6760000
>> >> > MEMATTR=0xb8973418  RNG=0xbffdbf98
>> >> > [    0.000000] random: fast init done
>> >> > [    0.000000] efi: seeding entropy pool
>> >> > :
>> >> >
>> >> > (I am using the latest EDKII master, the Marvell edk2-open-platform
>> >> > 17.10
>> >> > banch, with all the latest mv-ddr/ atf /etc....).
>> >> >
>> >> > The DT data appear there in die EFI boot, but the PCIe interface
>> >> > fails,
>> >> > and
>> >> > results (I believe) in the CPU stall warnings:
>> >> >
>> >> > [  717.453025] INFO: rcu_preempt self-detected stall on CPU
>> >> > :
>> >> > :
>> >> > [  717.589783]  armada8k_pcie_probe+0x140/0x240
>> >> > :
>> >> >
>> >> > Other times, the pcie gets further:
>> >> >
>> >> > [    3.312127] PCI: OF: host bridge /cp0/pcie at f2600000 ranges:
>> >> > [    3.317740] PCI: OF:    IO 0xf9000000..0xf900ffff -> 0xf9000000
>> >> > [    3.323692] PCI: OF:   MEM 0xc0000000..0xdfffffff -> 0xc0000000
>> >> > [    3.328915] random: crng init done
>> >> > [    4.326158] armada8k-pcie f2600000.pcie: phy link never came up
>> >> > [    4.332109] armada8k-pcie f2600000.pcie: Link not up after
>> >> > reconfiguration
>> >> > [    4.339056] armada8k-pcie f2600000.pcie: PCI host bridge to bus
>> >> > 0000:00
>> >>
>> >>
>> >> To be brutally honest, the armada8k-pcie driver is a piece of junk,
>> >> and you're much better off using the generic ECAM driver, which now
>> >> includes special handling for the missing root port on Synopsys IP.
>> >>
>> >> It also allows you to have both MMIO32 and MMIO64 regions, which can
>> >> be useful with some PCIe cards with large BARs
>> >>
>> >> Could you try
>> >>
>> >> compatible = "marvell,armada8k-pcie-ecam";
>> >>
>> >> in the DT node, please?
>> >>
>> >> (Before you do that, please check whether UEFI recognizes your PCI
>> >> hardware using the 'pci' command in the shell)
>> >
>> >
>> > This exercise help a lot. Thank you for the proposal.
>> >
>> > So now I can consistently boot using uboot and efi.
>> >
>> > However, the pcie driver init fails. I have provided boot logs and also
>> > my
>> > DT entry - we need custom BAR ranges, and I am not sure if this driver
>> > understand everything.
>> >
>> >  cp0_pcie0: pcie at f2600000 {
>> >   compatible = "marvell,armada8k-pcie-ecam", "snps,dw-pcie";
>> >   reg = <0 0xf2600000 0 0x10000>,
>> >         <0 ((0xf6000000 + (0 * 0x1000000)) + 0xf00000) 0 0x80000>;
>> >   reg-names = "ctrl", "config";
>> >   #address-cells = <3>;
>> >   #size-cells = <2>;
>> >   #interrupt-cells = <1>;
>> >   device_type = "pci";
>> >   dma-coherent;
>> >   msi-parent = <&gic_v2m0>;
>> >
>> >   bus-range = <0 0xff>;
>> >   ranges =
>> >
>> >   <0x81000000 0 (0xf9000000 + (0 * 0x10000)) 0 (0xf9000000 + (0 *
>> > 0x10000))
>> > 0 0x10000
>> >
>> >   0x82000000 0 (0xf6000000 + (0 * 0x1000000)) 0 (0xf6000000 + (0 *
>> > 0x1000000)) 0 0xf00000>;
>> >   interrupt-map-mask = <0 0 0 0>;
>> >   interrupt-map = <0 0 0 0 &cp0_icu 0x0 22 4>;
>> >   interrupts = <0x0 22 4>;
>> >   num-lanes = <1>;
>> >   clocks = <&cp0_clk 1 13>;
>> >   status = "disabled";
>> >  };
>> >
>> >
>> > Error:
>> >
>> > [    1.396968] PCI: OF: host bridge /cp0/pcie at f2600000 ranges:
>> > [    1.396979] PCI: OF:    IO 0xf9000000..0xf900ffff -> 0xf9000000
>> > [    1.396984] PCI: OF:   MEM 0xc0000000..0xdfffffff -> 0xc0000000
>> > [    1.396998] pci-host-generic f2600000.pcie: ECAM area [mem
>> > 0xf2600000-0xf260ffff] can only accommodate [bus 00-ffffffffffffffff]
>> > (reduced from [bus 00-ff] desired)
>> > [    1.397002] pci-host-generic f2600000.pcie: ECAM ioremap failed
>> > [    1.397011] pci-host-generic: probe of f2600000.pcie failed with
>> > error
>> > -12
>> >
>> >
>> > Thanks for the support.
>> >
>>
>> Please try the following config
>>
>> cp0_pcie0: pcie at e0000000 {
>>    compatible = "marvell,armada8k-pcie-ecam", "snps,dw-pcie";
>>    reg = <0 0xe0000000 0 0xff00000>;
>>    #address-cells = <3>;
>>    #size-cells = <2>;
>>    #interrupt-cells = <1>;
>>    device_type = "pci";
>>    dma-coherent;
>>    msi-parent = <&gic_v2m0>;
>>
>>    bus-range = <0 0xfe>;
>>    ranges = <0x1000000 0x0 0x00000000 0x0 0xeff00000 0x0 0x00010000>,
>>             <0x2000000 0x0 0xc0000000 0x0 0xc0000000 0x0 0x20000000>,
>>             <0x3000000 0x8 0x00000000 0x8 0x00000000 0x1 0x00000000>;
>>
>>    interrupt-map-mask = <0 0 0 0>;
>>    interrupt-map = <0 0 0 0 &cp0_icu 0x0 22 4>;
>> };
>
>
> I am trying it now.
>
> Could you just give me some insight on how the peripheral base address can
> be just modified like that ?
>
> Is there a mapping change somewhere?
>

All those addresses are configurable, and the default armada8k-pcie
driver sets up all the translation windows from scratch (in a rather
limited way, mind you)

The armada8k-pcie-ecam driver just reuses the configuration set by the
firmware, allowing for a larger bus range and an additional 4 GB
window for 64-bit MMIO