On Thu, Sep 9, 2021 at 8:15 AM Rob Herring <robh+dt@xxxxxxxxxx> wrote: > > On Thu, Sep 9, 2021 at 9:09 AM Andre Muller <andre.muller@xxxxxx> wrote: > > > > On 09/09/2021 00.31, Rob Herring wrote: > > > On Tue, Sep 7, 2021 at 10:15 PM Saravana Kannan <saravanak@xxxxxxxxxx> wrote: > > >> > > >> On Tue, Sep 7, 2021 at 7:12 PM Andre Muller <andre.muller@xxxxxx> wrote: > > >>> > > >>> On 08/09/2021 00.05, Saravana Kannan wrote: > > >>>> On Sun, Sep 5, 2021 at 1:15 AM Andre Muller <andre.muller@xxxxxx> wrote: > > >>>>> > > >>>>> With linux-5.13 and linux-5.14, the internal drive and SD card reader are gone from the XO-1.5. I bisected the issue to come up with ea718c699055: > > >>>>> > > >>>>> # first bad commit: [ea718c699055c8566eb64432388a04974c43b2ea] Revert "Revert "driver core: Set fw_devlink=on by default"" > > >>>>> > > >>>>> The /dev/mmcblk* nodes are not generated since this patch. > > >>>>> > > >>>>> Please find the output of lspsi -vv and lshw below. > > >>>>> > > >>>>> I will be happy to provide more info and/or test patches. > > >>>> > > >>>> Hi Andre, > > >>>> > > >>>> Can you point me to the dts file in upstream that corresponds to this system? > > >>>> > > >>>> Also, if you can give the output of: > > >>>> cat /sys/kernel/debug/devices_deferred > > >>> > > >>> Hi Saravana, > > >>> > > >>> > > >>> /sys/kernel/debug/devices_deferred is empty. > > >>> I used the last good commit b6f617. > > >> > > >> Sorry, I wanted that with the bad commit. > > > > Uh-oh, my bad... > > > > The bad case says > > # cat devices_deferred > > 0000:00:0c.0 > > > > That's the SD Host controller. > > > > >> > > >>> > > >>> The XO-1.5 has an x86 compatible VIA C7 processor. > > >>> It uses the VX855 chip for about all I/O tasks, including SDIO. > > >>> I am not aware of a device tree file for it. > > >>> > > >>> It is a bit of a strange beast, it uses OFW to initialize the hardware and provide a FORTH shell. > > >>> Which also is the boot manager, configured via FORTH scripts. > > >>> > > >>> From the linux side of the fence, dmesg's line 2 is: > > >>> > > >>> "OFW detected in memory, cif @ 0xff83ae68 (reserving top 8MB)" > > >>> > > >>> AIUI, this mechanism is used in lieu of a device tree file, like UEFI on most x86 hardware. > > >>> But my understanding of device trees is severely limited, I might be allwrong. > > >> > > >> Uhh... I'm so confused. If Linux doesn't use OF, then none of the code > > >> enabled by fw_devlink=on should be executed. > > > > > > Linux does, but maybe not for memory (like UEFI on arm64). > > > > > >> The only thing that might remotely even execute is: > > >> efifb_add_links() in drivers/firmware/efi/efi-init.c > > >> > > >> If you want you can just do an early return 0; in that to see if it > > >> makes a difference (unlikely). > > >> > > >> Rob, Do you know what's going on with OLPC and DT? > > > > > > Not really. I have an XO-1 DT dump[1]. It's probably a similar looking > > > DT though. It's pretty ancient lacking anything we've invented for DT > > > in the last 10 years. There's not really much to it as about the only > > > phandle I see is for interrupts. > > > > > >>> Anyway, the firmware source is here: > > >>> http://dev.laptop.org/git/users/quozl/openfirmware/ > > >>> > > >>> This file is the closest dt-analogous thing for the XO-1.5 I can find therein: > > >>> cpu/x86/pc/olpc/via/devices.fth > > >> > > >> That file is all gibberish to me. > > > > > > Running this on a booted system would help: > > > > > > dtc -f -I fs -O dts /proc/device-tree > dump.dts > > > > Ah, thanks. I never knew about the DT in there... > > XO-1.5_dump.dts is attached. > > > > > > > > If you don't have dtc on the system, then you'll have to zip up > > > /proc/device-tree contents and run dtc elsewhere (or just post that). > > > > > >>> My machine runs the latest version: > > >>> http://wiki.laptop.org/go/OLPC_Firmware_q3c17 > > >>> > > >>> The XO-1.5 hardware specs are here: > > >>> http://wiki.laptop.org/images/f/f0/CL1B_Hdwe_Design_Spec.pdf > > >>> http://wiki.laptop.org/go/Hardware_specification_1.5 > > >>> > > >>> Would the .config or dmesg help? > > >> > > >> At this point, why not? When you do send them, please send them as > > >> attachments and not inline. > > >> > > >> Also, when you collect the dmesg logs, the following could help: > > >> Enable the existing dev_dbg logs in these functions: > > >> device_link_add() > > >> device_links_check_suppliers() > > >> > > >> And add the following log to fwnode_link_add(): > > >> +++ b/drivers/base/core.c > > >> @@ -87,6 +87,8 @@ int fwnode_link_add(struct fwnode_handle *con, > > >> struct fwnode_handle *sup) > > >> goto out; > > >> } > > >> > > >> + pr_info("Link fwnode %pfwP as a consumer of fwnode %pfwP\n", con, sup); > > >> + > > > > > > > OK. The dmesg with debug info is attached as well (for the broken case). > > Humm, ACPI and DT together... > > Looks to me like it's waiting for the wrong interrupt-parent. The log > says it is waiting for 'interrupt-controller@i20' which is the only > interrupt-controller found in the DT, but the parent is the PCI bridge > with whatever interrupt-map is pointing to. That's not clear as the > phandle (0x767a4) doesn't exist in the DT. I suppose the parent is > defined in ACPI? After staring at it for a while, I realized that interrupt-controller@i20 is indeed the right node. Looks like we need to do endian conversion of the ".node" property in the interrupt controller and it would match with 0x767a4. > pci 0000:00:0c.0: probe deferral - wait for supplier interrupt-controller@i20 The SD controller is waiting forever on interrupt-controller@i20 to be added as a device. Rob, My guess is that the fwnode value is not getting set for ISA devices populated when isa@11 is added. Any idea how/where those child devices are populated? I thought they'd be platform devices, but it doesn't look like that's the case? > If there's not an easy fix, just disable devlinks for x86. There's > only one other DT platform, ce4100, and I really doubt it is even used > at all. I think the easy fix is to set the ISA device's fwnode when it's added, but I can't tell how they are getting added. But yeah, if that turns out to be hard, then I'd vote for disabling it for x86 too. -Saravana
