[PATCH RFC 0/5] arm64: rockchip: enable PCIE3 controller and its phy for Rock5B boards
Peter Geis
pgwipeout at gmail.com
Tue Feb 21 17:25:15 PST 2023
On Tue, Feb 21, 2023 at 4:45 PM Sebastian Reichel
<sebastian.reichel at collabora.com> wrote:
>
> Hi everyone,
>
> On Tue, Feb 21, 2023 at 01:55:48PM -0500, Peter Geis wrote:
> > On Tue, Feb 21, 2023 at 1:04 PM Piotr Oniszczuk
> > <piotr.oniszczuk at gmail.com> wrote:
> > > > Wiadomość napisana przez Qu Wenruo <wqu at suse.com> w dniu 21.02.2023, o godz. 01:14:
> > > > On 2023/2/21 02:25, Piotr Oniszczuk wrote:
> > > >>> Wiadomość napisana przez Qu Wenruo <wqu at suse.com <mailto:wqu at suse.com>> w dniu 04.02.2023, o godz. 09:47:
> > > >>>
> > > >>> This series is based on the existing upstream work from Sebastian
> > > >>> Reichel:
> > > >>> https://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-misc.git/log/?h=rk3588 <https://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-misc.git/log/?h=rk3588>
> > > >>>
> > > >>> And I'm a completely newbie to arm64 world, thus if there is something
> > > >>> wrong, feel free to point out and I'm pretty happy to learn from the
> > > >>> failure.
> > > >>>
> > > >>> [BACKGROUND]
> > > >>> RK3588S and RK3588 have PCIE supports, it's done by the following 3
> > > >>> controllers:
> > > >>>
> > > >>> - PCIE2.0x1 lane @fe180000 (both RK3588S and RK3588)
> > > >>> - PCIE2.0x1 lane @fe190000 (both RK3588S and RK3588)
> > > >>> Thes two are all connected to a naneng combo phy each, normally shared
> > > >>> with SATA or USB.
> > > >>>
> > > >>> - PCIE2.0x1 lane @fe170000 (RK3588 exlusive)
> > > >>> This one is also connected to a naneng combo phy, normally shared
> > > >>> with SATA or USB.
> > > >>>
> > > >>> - PCIE3.0x4 lanes @fe15000 (RK3588 exclusive)
> > > >>>
> > > >>> And unlike other boards, ROCK5B is utilizing PCIE extensively, its
> > > >>> network controller (RTL8125 2.5Gbps Ethernet) is connected to the
> > > >>> PCIE2.0 lane at fe190000, and an M.2 slot is attached to the PCIE3.0x4
> > > >>> lanes.
> > > >>>
> > > >>> [WORKING]
> > > >>> Currently the series is able to bring up the PCIE3.0x4 lanes and
> > > >>> properly boot from an NVME at that M.2 slot of Rock5B boards.
> > > >>>
> > > >>> [NOT WORKING]
> > > >>> All PCIE2.0 lanes connected to naneng combo phy are not working.
> > > >>> I tried forward porting the extra handling from downstream, but it only
> > > >>> results hanging at probing (causing RCU stall).
> > > >>>
> > > >>> [EXTRA WANRING]
> > > >>> - PCI MSI initialization warning
> > > >>> WARNING: CPU: 7 PID: 1 at drivers/pci/msi/msi.h:121 pci_msi_setup_msi_irqs+0x38/0x4c
> > > >>>
> > > >>> This seems to be caused by the fact that we are still using legcacy
> > > >>> msi irqs?
> > > >>>
> > > >>> I checked up the gic and its dts, can not figure out why (all pretty
> > > >>> the same just like rk3399 and rk3568).
> > > >>> Any help would be appreciated.
> > > >>>
> > > >>> - pci_bus 0000:01: busn_res: can not insert [bus 01-ff] under [bus 00-0f] (conflicts with (null) [bus 00-0f])
> > > >>> The vendoer kernel also has this problem, but my RK3399 board with
> > > >>> upstream kernel didn't trigger this at all, but something else like:
> > > >>>
> > > >>> pci 0000:00:00.0: bridge configuration invalid ([bus 00-00]), reconfiguring
> > > >>>
> > > >>> Then:
> > > >>>
> > > >>> pci_bus 0000:01: busn_res: [bus 01-1f] end is updated to 01
> > > >>>
> > > >>> Not sure if it's something missing or can be just ignored.
> > > >>>
> > > >>> [PCI DMESG]
> > > >>> With this patchset, the PCI initialization and nvme would look like this:
> > > >>>
> > > >>> [ 0.142984] rockchip-dw-pcie fe150000.pcie: host bridge /pcie at fe150000 ranges:
> > > >>> [ 0.143653] rockchip-dw-pcie fe150000.pcie: IO 0x00f0100000..0x00f01fffff -> 0x00f0100000
> > > >>> [ 0.144463] rockchip-dw-pcie fe150000.pcie: MEM 0x00f0200000..0x00f0ffffff -> 0x00f0200000
> > > >>> [ 0.145261] rockchip-dw-pcie fe150000.pcie: MEM 0x0900000000..0x093fffffff -> 0x0900000000
> > > >>> [ 0.154022] rockchip-dw-pcie fe150000.pcie: iATU: unroll T, 8 ob, 8 ib, align 64K, limit 8G
> > > >>> [ 0.360415] rockchip-dw-pcie fe150000.pcie: PCIe Gen.3 x4 link up
> > > >>> [ 0.361099] rockchip-dw-pcie fe150000.pcie: PCI host bridge to bus 0000:00
> > > >>> [ 0.361731] pci_bus 0000:00: root bus resource [bus 00-0f]
> > > >>> [ 0.362236] pci_bus 0000:00: root bus resource [io 0x0000-0xfffff] (bus address [0xf0100000-0xf01fffff])
> > > >>> [ 0.363113] pci_bus 0000:00: root bus resource [mem 0xf0200000-0xf0ffffff]
> > > >>> [ 0.363744] pci_bus 0000:00: root bus resource [mem 0x900000000-0x93fffffff pref]
> > > >>> [ 0.364450] pci 0000:00:00.0: [1d87:3588] type 01 class 0x060400
> > > >>> [ 0.365005] pci 0000:00:00.0: reg 0x10: [mem 0x00000000-0x3fffffff]
> > > >>> [ 0.365583] pci 0000:00:00.0: reg 0x14: [mem 0x00000000-0x3fffffff]
> > > >>> [ 0.366159] pci 0000:00:00.0: reg 0x38: [mem 0x00000000-0x0000ffff pref]
> > > >>> [ 0.366801] pci 0000:00:00.0: supports D1 D2
> > > >>> [ 0.367193] pci 0000:00:00.0: PME# supported from D0 D1 D3hot
> > > >>> [ 0.368647] pci_bus 0000:01: busn_res: can not insert [bus 01-ff] under [bus 00-0f] (conflicts with (null) [bus 00-0f])
> > > >>> [ 0.369681] pci 0000:01:00.0: [1d97:5216] type 00 class 0x010802
> > > >>> [ 0.370277] pci 0000:01:00.0: reg 0x10: [mem 0x00000000-0x0000ffff 64bit]
> > > >>> [ 0.370975] pci 0000:01:00.0: reg 0x30: [mem 0x00000000-0x0001ffff pref]
> > > >>> [ 0.372130] pci 0000:00:00.0: BAR 0: no space for [mem size 0x40000000]
> > > >>> [ 0.372742] pci 0000:00:00.0: BAR 0: failed to assign [mem size 0x40000000]
> > > >>> [ 0.373381] pci 0000:00:00.0: BAR 1: no space for [mem size 0x40000000]
> > > >>> [ 0.373988] pci 0000:00:00.0: BAR 1: failed to assign [mem size 0x40000000]
> > > >>> [ 0.374628] pci 0000:00:00.0: BAR 14: assigned [mem 0xf0200000-0xf02fffff]
> > > >>> [ 0.375259] pci 0000:00:00.0: BAR 6: assigned [mem 0xf0300000-0xf030ffff pref]
> > > >>> [ 0.375923] pci 0000:01:00.0: BAR 6: assigned [mem 0xf0200000-0xf021ffff pref]
> > > >>> [ 0.376590] pci 0000:01:00.0: BAR 0: assigned [mem 0xf0220000-0xf022ffff 64bit]
> > > >>> [ 0.377281] pci 0000:00:00.0: PCI bridge to [bus 01-ff]
> > > >>> [ 0.377762] pci 0000:00:00.0: bridge window [mem 0xf0200000-0xf02fffff]
> > > >>> [ 0.426841] pcieport 0000:00:00.0: PME: Signaling with IRQ 33
> > > >>> [ 0.427487] pcieport 0000:00:00.0: AER: enabled with IRQ 33
> > > >>> [ 0.625353] nvme nvme0: pci function 0000:01:00.0
> > > >>> [ 0.625774] nvme 0000:01:00.0: enabling device (0000 -> 0002)
> > > >>> [ 0.717069] nvme nvme0: Shutdown timeout set to 8 seconds
> > > >>> [ 0.723025] nvme nvme0: allocated 64 MiB host memory buffer.
> > > >>> [ 0.816820] nvme nvme0: 1/0/0 default/read/poll queues
> > > >>> [ 0.818079] sdhci-pltfm: SDHCI platform and OF driver helper
> > > >>> [ 0.820678] nvme0n1: p1 p2
> > > >>>
> > > >>>
> > > >> Qu, all
> > > >> I’m playing with your work on my rock5b as I want to have working Eth on rock5b.
> > > >> My code is from https://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-misc.git/log/?h=rk3588 <https://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-misc.git/log/?h=rk3588> + your’s PCIE3 patches.
> > > >> SBC boots from sd card, I see PCIE related logs in dmesg. but no rtl8125 is detected.
> > > >> PCIE logs are like this:
> > > >> 8.207810] rockchip-dw-pcie fe150000.pcie: host bridge /pcie at fe150000 ranges:
> > > >> [ 8.208501] rockchip-dw-pcie fe150000.pcie: Parsing ranges property...
> > > >> [ 8.209089] rockchip-dw-pcie fe150000.pcie: IO 0x00f0100000..0x00f01fffff -> 0x00f0100000
> > > >> [ 8.209944] rockchip-dw-pcie fe150000.pcie: MEM 0x00f0200000..0x00f0ffffff -> 0x00f0200000
> > > >> [ 8.210740] rockchip-dw-pcie fe150000.pcie: MEM 0x0900000000..0x093fffffff -> 0x0900000000
> > > >> [ 8.218918] rockchip-dw-pcie fe150000.pcie: iATU: unroll T, 8 ob, 8 ib, align 64K, limit 8G
> > > >> [ 9.324473] rockchip-dw-pcie fe150000.pcie: Phy link never came up
> > > >> [ 9.325186] rockchip-dw-pcie fe150000.pcie: PCI host bridge to bus 0000:00
> > > >> [ 9.325786] pci_bus 0000:00: root bus resource [bus 00-0f]
> > > >> [ 9.326266] pci_bus 0000:00: root bus resource [io 0x0000-0xfffff] (bus address [0xf0100000-0xf01fffff])
> > > >> [ 9.327097] pci_bus 0000:00: root bus resource [mem 0xf0200000-0xf0ffffff]
> > > >> [ 9.327713] pci_bus 0000:00: root bus resource [mem 0x900000000-0x93fffffff pref]
> > > >> [ 9.328364] pci_bus 0000:00: scanning bus
> > > >> [ 9.328729] pci 0000:00:00.0: [1d87:3588] type 01 class 0x060400
> > > >> [ 9.329258] pci 0000:00:00.0: reg 0x10: [mem 0x00000000-0x3fffffff]
> > > >> [ 9.329807] pci 0000:00:00.0: reg 0x14: [mem 0x00000000-0x3fffffff]
> > > >> [ 9.330354] pci 0000:00:00.0: reg 0x38: [mem 0x00000000-0x0000ffff pref]
> > > >> [ 9.330984] pci 0000:00:00.0: supports D1 D2
> > > >> [ 9.331356] pci 0000:00:00.0: PME# supported from D0 D1 D3hot
> > > >> [ 9.331858] pci 0000:00:00.0: PME# disabled
> > > >> [ 9.332309] pci 0000:00:00.0: vgaarb: pci_notify
> > > >> [ 9.333735] pci_bus 0000:00: fixups for bus
> > > >> [ 9.334106] pci 0000:00:00.0: scanning [bus 01-ff] behind bridge, pass 0
> > > >> [ 9.334731] pci_bus 0000:01: busn_res: can not insert [bus 01-ff] under [bus 00-0f] (conflicts with (null) [bus 00-0f])
> > > >> [ 9.335668] pci_bus 0000:01: scanning bus
> > > >> [ 9.336052] pci_bus 0000:01: fixups for bus
> > > >> [ 9.336416] pci_bus 0000:01: bus scan returning with max=01
> > > >> [ 9.336903] pci 0000:00:00.0: scanning [bus 01-ff] behind bridge, pass 1
> > > >> [ 9.337503] pci_bus 0000:00: bus scan returning with max=ff
> > > >> [ 9.337994] pci 0000:00:00.0: BAR 0: no space for [mem size 0x40000000]
> > > >> [ 9.338570] pci 0000:00:00.0: BAR 0: failed to assign [mem size 0x40000000]
> > > >> [ 9.339175] pci 0000:00:00.0: BAR 1: no space for [mem size 0x40000000]
> > > >> [ 9.339749] pci 0000:00:00.0: BAR 1: failed to assign [mem size 0x40000000]
> > > >> [ 9.340356] pci 0000:00:00.0: BAR 6: assigned [mem 0xf0200000-0xf020ffff pref]
> > > >> [ 9.340991] pci 0000:00:00.0: PCI bridge to [bus 01-ff]
> > > >> and nothing more :-(
> > > >> Are you progressing maybe with pcie on rock5b?
> > > > fe150000 is the PCIE3.0 controller, on Rock5B, that's the M.2 slot.
> > > >
> > > > But for R8125, it's on a PCIE2.0 controller, which is using naneng combo phy.
> > > >
> > > > I'm not able to bring the PCIE2.0 part up yet, it always hangs at PCIE2.0 initialization, thus only the PCIE3.0 part is submitted to the list.
> > > >
> > > > Thanks,
> > > > Qu
> > >
> > > Yes. Indeed.
> > > I'm trying to add pci2.0 and it looks i meet the same problem probably.
> > >
> > > I backported (from well working neggles quartz64pro repo I assume):
> > > https://github.com/neggles/linux-quartz64/commit/2ad84e89fc75b82c783345b72c97f5d7e3d45723
> > > https://github.com/neggles/linux-quartz64/commit/4ac44c2b53758eca671d695d19b456d1849d7e14
> > > https://github.com/neggles/linux-quartz64/commit/714c5e01d8f0f73b3a49cbdee29e3ffe0f3353dd
> > > https://github.com/neggles/linux-quartz64/commit/64971f9c85f27e29c44b31a0c326ea4bb8ec3c56
> > >
> > > then i added rock5b pcie dt enablements (basing on radxa BSP):
> > > https://github.com/warpme/minimyth2/blob/master/script/kernel/linux-6.2/files/1058-arm64-dts-rockchip-enable-pcie-rock5b.patch
> > >
> > >
> > > this gives me quite clean 6.2 mainline boot log with hang at pcie2 init (pls see below)
> > >
> > > I assume https://github.com/neggles/linux-quartz64/commits/linux-quartz64 repo works well on Quartsz64 board - so I indirectly assume above backports should give us working pice2.0.
> > >
> > > It fails on rock5b - so I conclude: issue/error is on my side and is related to my rock5b specifics.
> > > Unfortunately I don’t owe Quartz64 board so can't verify correctness of my backports by testing on Quartz64pro.
> > >
> > > I’m curious about opinion of smarter than me guys…
> >
> > If the phy is misconfigured, not powered, or the clocks aren't going
> > active, you'll hang when the controller tries to touch it. Unless
> > someone has completed the combophy rk3588 bits the driver is not
> > functional yet for rk3588.
> >
> > Looking at the current 6.2 release, I only see the rk3568 compatible,
> > so you'll need to add support for rk3588 before it will work.
> >
> > Very Respectfully,
> > Peter Geis
>
> Sorry for being late to the game. Life is busy right now :)
>
> I haven't looked into PCIe myself so far, but some of my colleagues
> are looking into native network support on Rock 5B (and thus PCIe2
> controller).
>
> Apart from the obvious (missing rk3588 support in the combophy driver),
> the clocks will need some work. The clock tree implementation I upstreamed
> is different from the downstream implementation. Downstream has some
> clocks that have two parent clocks using a hacked implementation
> that's obviously not upstreamable as is. The upstream implementation
> currently only describes the first parent. More details are in the
> following comment:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/clk/rockchip/clk-rk3588.c#n16
>
> Back than I wrote that I do not understand the exact need (the TRM
> does not describe the clock tree unfortuantely), I found this
> explanation:
>
> https://lore.kernel.org/dri-devel/20220309094139198367142@rock-chips.com/
>
> To get the advanced blocks properly running upstream this needs a
> solution for this. Note that trying to access registers that are
> not clocked properly will result in a hang (as Peter already wrote).
TLDR: There are now hardware blocks called a "Native Interface Unit"
(NIU) that gate the clocks to devices behind them. Essentially it
makes certain device clocks dependent on multiple parents being active
at the same time, which the current clock structure does not support.
It was decided that until the clock structure is updated to support
this, the NIU gate clocks would be marked critical and left always on.
Better to burn a miniscule more power than risk locking the board up.
Though the issue here is much simpler than that, the controller simply
can't operate without a functional phy attached to it. They are
physically tied together and it's the controller < - > phy link that
doesn't function without the phy being configured. The combophy config
for rk3588 isn't significantly more complicated than rk356x, but one
can't mainline code that can't be tested and rk3588 wasn't available
at that time.
>
> Apart from that the power-domain controller might need some of the
> extra bits downstream has.
If PCIe 3 works, then this likely won't affect PCIe 2, though it
limits our ability to implement runtime power management.
>
> Last but not least the GIC controller is handled differently in
> downstream. For that following the workaround that has been used for
> rk356x should also work for rk3588.
The hack for the GIC controller that I maintain for rk356x is not
upstreamable. For the ITS to be supported in mainline we still require
an official errata from Rockchip documenting the issues here. The
mbi-alias method, while it works for simple configs, isn't scalable.
>
> TLDR: This is not trivial. It's really unfortunate, that the board
> is not just using the native ethernet :(
>
> P.S.: We try to keep a rk3588 / Rock 5 status matrix maintained here:
>
> https://gitlab.collabora.com/hardware-enablement/rockchip-3588/notes-for-rockchip-3588/-/blob/main/mainline-status.md
>
> Greetings,
>
> -- Sebastian
More information about the linux-arm-kernel
mailing list