Re: sunxi: a83t: does not boot anymore in BigEndian

[Corentin Labbe]

On Tue, Mar 03, 2020 at 05:50:20PM +0000, Robin Murphy wrote:
> On 03/03/2020 2:21 pm, Corentin Labbe wrote:
> > On Tue, Mar 03, 2020 at 01:47:02PM +0000, Robin Murphy wrote:
> >> On 03/03/2020 7:43 am, Corentin Labbe wrote:
> >>> Hello
> >>>
> >>> My sun8i-a83t-bananapi-m3 does not boot anymore in BE.
> >>> Others sunxi platform I have seems not affected (a10, a20, a64, h3, h5, h6)
> >>>
> >>> I have bisected this problem:
> >>> git bisect start
> >>> # bad: [98d54f81e36ba3bf92172791eba5ca5bd813989b] Linux 5.6-rc4
> >>> git bisect bad 98d54f81e36ba3bf92172791eba5ca5bd813989b
> >>> # bad: [d5226fa6dbae0569ee43ecfc08bdcd6770fc4755] Linux 5.5
> >>> git bisect bad d5226fa6dbae0569ee43ecfc08bdcd6770fc4755
> >>> # good: [219d54332a09e8d8741c1e1982f5eae56099de85] Linux 5.4
> >>> git bisect good 219d54332a09e8d8741c1e1982f5eae56099de85
> >>> # bad: [8c39f71ee2019e77ee14f88b1321b2348db51820] Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
> >>> git bisect bad 8c39f71ee2019e77ee14f88b1321b2348db51820
> >>> # bad: [3b397c7ccafe0624018cb09fc96729f8f6165573] Merge tag 'regmap-v5.5' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap
> >>> git bisect bad 3b397c7ccafe0624018cb09fc96729f8f6165573
> >>> # good: [924ea58dadea23cc28b60d02b9c0896b7b168a6f] Merge tag 'mt76-for-kvalo-2019-11-20' of https://github.com/nbd168/wireless
> >>> git bisect good 924ea58dadea23cc28b60d02b9c0896b7b168a6f
> >>> # good: [3f3c8be973af10875cfa1e7b85a535b6ba76b44f] Merge tag 'for-linus-5.5a-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip
> >>> git bisect good 3f3c8be973af10875cfa1e7b85a535b6ba76b44f
> >>> # bad: [642356cb5f4a8c82b5ca5ebac288c327d10df236] Merge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
> >>> git bisect bad 642356cb5f4a8c82b5ca5ebac288c327d10df236
> >>> # good: [57d8154f15e89f53dfb412f4ed32ebe3c3d755a0] crypto: atmel-aes - Change data type for "lastc" buffer
> >>> git bisect good 57d8154f15e89f53dfb412f4ed32ebe3c3d755a0
> >>> # bad: [752272f16dd18f2cac58a583a8673c8e2fb93abb] Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
> >>> git bisect bad 752272f16dd18f2cac58a583a8673c8e2fb93abb
> >>> # good: [9477f4449b0b011ce1d058c09ec450bfcdaab784] KVM: VMX: Add helper to check reserved bits in IA32_PERF_GLOBAL_CTRL
> >>> git bisect good 9477f4449b0b011ce1d058c09ec450bfcdaab784
> >>> # bad: [cd7056ae34af0e9424da97bbc7d2b38246ba8a2c] Merge remote-tracking branch 'kvmarm/misc-5.5' into kvmarm/next
> >>> git bisect bad cd7056ae34af0e9424da97bbc7d2b38246ba8a2c
> >>> # bad: [c7892db5dd6afe921ead502aff7440a1e450d947] KVM: arm64: Select TASK_DELAY_ACCT+TASKSTATS rather than SCHEDSTATS
> >>> git bisect bad c7892db5dd6afe921ead502aff7440a1e450d947
> >>> # bad: [8564d6372a7d8a6d440441b8ed8020f97f744450] KVM: arm64: Support stolen time reporting via shared structure
> >>> git bisect bad 8564d6372a7d8a6d440441b8ed8020f97f744450
> >>> # bad: [55009c6ed2d24fc0f5521ab2482f145d269389ea] KVM: arm/arm64: Factor out hypercall handling from PSCI code
> >>> git bisect bad 55009c6ed2d24fc0f5521ab2482f145d269389ea
> >>> # bad: [6a7458485b390f48e481fcd4a0b20e6c5c843d2e] KVM: arm64: Document PV-time interface
> >>> git bisect bad 6a7458485b390f48e481fcd4a0b20e6c5c843d2e
> >>> # bad: [dcac930e9901d765234bc15004db4f7d4416db71] Merge remote-tracking branch 'arm64/for-next/smccc-conduit-cleanup' into kvm-arm64/stolen-time
> >>> git bisect bad dcac930e9901d765234bc15004db4f7d4416db71
> >>> # first bad commit: [dcac930e9901d765234bc15004db4f7d4416db71] Merge remote-tracking branch 'arm64/for-next/smccc-conduit-cleanup' into kvm-arm64/stolen-time
> >>>
> >>> But bisect lead to a merge request.
> >>
> >> FWIW once you've bisected to a merge commit, you can always then try
> >> bisecting down the merged branch itself (i.e. between 5.4-rc3 and
> >> e6ea46511b1a in this case) to narrow things down further.
> >>
> >> Given that that branch is supposed to be a functionally-inert cleanup,
> >> and (judging by the DTS) this platform apparently isn't using PSCI
> >> anyway, it does seem a bit odd. Can you get any earlycon/earlyprintk
> >> output to suggest what the actual cause of the boot failure is?
> >>
> > 
> > Hello
> > 
> > With earlycon I got:
> > 
> > bootz 0x42000000 0x43300000 0x43000000
> > ## Loading init Ramdisk from Legacy Image at 43300000 ...
> >     Image Name:
> >     Image Type:   ARM Linux RAMDisk Image (uncompressed)
> >     Data Size:    32397255 Bytes = 30.9 MiB
> >     Load Address: 00000000
> >     Entry Point:  00000000
> >     Verifying Checksum ... OK
> > ## Flattened Device Tree blob at 43000000
> >     Booting using the fdt blob at 0x43000000
> >     Using Device Tree in place at 43000000, end 43008f07
> > Starting kernel ...
> > [    0.000000] Booting Linux on physical CPU 0x0
> > [    0.000000] Linux version 5.6.0-rc4 (compile@Red) (gcc version 8.3.0 (Gentoo 8.3.0-r1 p1.1)) #21 SMP Tue Mar 3 15:07:20 CET 2020
> > [    0.000000] CPU: ARMv7 Processor [410fc075] revision 5 (ARMv7), cr=12c5387d
> > [    0.000000] CPU: div instructions available: patching division code
> > [    0.000000] CPU: PIPT / VIPT nonaliasing data cache, VIPT aliasing instruction cache
> > [    0.000000] OF: fdt: Machine model: Banana Pi BPI-M3
> > [    0.000000] earlycon: uart0 at MMIO32 0x01c28000 (options '')
> > [    0.000000] printk: bootconsole [uart0] enabled
> > [    0.000000] Memory policy: Data cache writealloc
> > [    0.000000] cma: Reserved 16 MiB at 0xbf000000
> > [    0.000000] percpu: Embedded 15 pages/cpu s31320 r8192 d21928 u61440
> > [    0.000000] Built 1 zonelists, mobility grouping on.  Total pages: 522752
> > [    0.000000] Kernel command line: console=ttyS0,115200n8 root=/dev/ram0 earlycon=uart,mmio32,0x01c28000 ip=dhcp
> > [    0.000000] Dentry cache hash table entries: 131072 (order: 7, 524288 bytes, linear)
> > [    0.000000] Inode-cache hash table entries: 65536 (order: 6, 262144 bytes, linear)
> > [    0.000000] mem auto-init: stack:off, heap alloc:off, heap free:off
> > [    0.000000] Memory: 2019352K/2097152K available (7168K kernel code, 485K rwdata, 2832K rodata, 1024K init, 248K bss, 61416K reserved, 16384K cma-reserved, 1294336K highmem)
> > [    0.000000] SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=8, Nodes=1
> > [    0.000000] rcu: Hierarchical RCU implementation.
> > [    0.000000] rcu: 	RCU event tracing is enabled.
> > [    0.000000] rcu: RCU calculated value of scheduler-enlistment delay is 10 jiffies.
> > [    0.000000] NR_IRQS: 16, nr_irqs: 16, preallocated irqs: 16
> > [    0.000000] random: get_random_bytes called from start_kernel+0x2a4/0x440 with crng_init=0
> > [    0.000000] arch_timer: cp15 timer(s) running at 24.00MHz (virt).
> > [    0.000000] clocksource: arch_sys_counter: mask: 0xffffffffffffff max_cycles: 0x588fe9dc0, max_idle_ns: 440795202592 ns
> > [    0.000008] sched_clock: 56 bits at 24MHz, resolution 41ns, wraps every 4398046511097ns
> > [    0.008636] Switching to timer-based delay loop, resolution 41ns
> > [    0.015439] clocksource: timer: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
> > [    0.025876] Console: colour dummy device 80x30
> > [    0.030743] Calibrating delay loop (skipped), value calculated using timer frequency.. 48.00 BogoMIPS (lpj=240000)
> > [    0.041927] pid_max: default: 32768 minimum: 301
> > [    0.047093] Mount-cache hash table entries: 2048 (order: 1, 8192 bytes, linear)
> > [    0.055003] Mountpoint-cache hash table entries: 2048 (order: 1, 8192 bytes, linear)
> > [    0.064553] CPU: Testing write buffer coherency: ok
> > [    0.070534] /cpus/cpu@0 missing clock-frequency property
> > [    0.076282] /cpus/cpu@1 missing clock-frequency property
> > [    0.082071] /cpus/cpu@2 missing clock-frequency property
> > [    0.087816] /cpus/cpu@3 missing clock-frequency property
> > [    0.093560] /cpus/cpu@100 missing clock-frequency property
> > [    0.099498] /cpus/cpu@101 missing clock-frequency property
> > [    0.105441] /cpus/cpu@102 missing clock-frequency property
> > [    0.111363] /cpus/cpu@103 missing clock-frequency property
> > [    0.117292] CPU0: thread -1, cpu 0, socket 0, mpidr 80000000
> > [    0.124164] Setting up static identity map for 0x40100000 - 0x40100060
> > [    0.131496] ARM CCI driver probed
> > [    0.135832] sunxi multi cluster SMP support installed
> > [    0.141784] rcu: Hierarchical SRCU implementation.
> > [    0.147537] smp: Bringing up secondary CPUs ...
> > 
> > I will try to redo the bisect
> 
> Hmm, if there was a horribly subtle endianness/uninitialised data bug 
> (or combination thereof) that led to somehow falling into the middle of 
> the PSCI or spectre code and issuing a spurious SMC/HVC to non-existent 
> firmware at secondary bringup, it might well look exactly like that...
> 
> If different values of CONFIG_PSCI and CONFIG_HARDEN_BRANCH_PREDICTOR 
> make any difference to a failing build then that might lend weight to 
> that theory. Assuming the initial bisection is reliable then 
> 6848253ddeae and/or a5520eac4d2d would seem like the most likely 
> suspects for perturbing whatever subtlety is at play in this case.
> 

Hello

I have tried to disable CONFIG_PSCI and CONFIG_HARDEN_BRANCH_PREDICTOR without any visible change.

Furthermore, the problem seems to happen on random place but replicable.
For example I hit the problem on v5.3 v5.1 and v5.1-rc4 but not on v5.4 v5.2 and v5.0.

But when the kernel boot, at least SMP is still non-functionnal:
[    0.000000] Booting Linux on physical CPU 0x0
[    0.000000] Linux version 5.2.0 (compile@Red) (gcc version 8.3.0 (Gentoo 8.3.0-r1 p1.1)) #74 SMP Thu Mar 5 18:16:17 CET 2020
[    0.000000] CPU: ARMv7 Processor [410fc075] revision 5 (ARMv7), cr=12c5387d
[    0.000000] CPU: div instructions available: patching division code
[    0.000000] CPU: PIPT / VIPT nonaliasing data cache, VIPT aliasing instruction cache
[    0.000000] OF: fdt: Machine model: Banana Pi BPI-M3
[    0.000000] earlycon: uart0 at MMIO32 0x01c28000 (options '')
[    0.000000] printk: bootconsole [uart0] enabled
[    0.000000] Memory policy: Data cache writealloc
[    0.000000] cma: Reserved 16 MiB at 0x6f000000
[    0.000000] percpu: Embedded 16 pages/cpu s35288 r8192 d22056 u65536
[    0.000000] Built 1 zonelists, mobility grouping on.  Total pages: 522752
[    0.000000] Kernel command line: console=ttyS0,115200n8 root=/dev/ram0 earlycon=uart,mmio32,0x01c28000 ip=dhcp
[    0.000000] Dentry cache hash table entries: 131072 (order: 7, 524288 bytes)
[    0.000000] Inode-cache hash table entries: 65536 (order: 6, 262144 bytes)
[    0.000000] Memory: 2022028K/2097152K available (6144K kernel code, 460K rwdata, 2512K rodata, 1024K init, 243K bss, 58740K reserved, 16384K cma-reserved, 1310720K highmem)
[    0.000000] SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=8, Nodes=1
[    0.000000] rcu: Hierarchical RCU implementation.
[    0.000000] rcu: 	RCU event tracing is enabled.
[    0.000000] rcu: RCU calculated value of scheduler-enlistment delay is 10 jiffies.
[    0.000000] NR_IRQS: 16, nr_irqs: 16, preallocated irqs: 16
[    0.000000] random: get_random_bytes called from start_kernel+0x29c/0x434 with crng_init=0
[    0.000000] arch_timer: cp15 timer(s) running at 24.00MHz (virt).
[    0.000000] clocksource: arch_sys_counter: mask: 0xffffffffffffff max_cycles: 0x588fe9dc0, max_idle_ns: 440795202592 ns
[    0.000010] sched_clock: 56 bits at 24MHz, resolution 41ns, wraps every 4398046511097ns
[    0.008648] Switching to timer-based delay loop, resolution 41ns
[    0.015394] clocksource: timer: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
[    0.025772] Console: colour dummy device 80x30
[    0.030645] Calibrating delay loop (skipped), value calculated using timer frequency.. 48.00 BogoMIPS (lpj=240000)
[    0.041811] pid_max: default: 32768 minimum: 301
[    0.046975] Mount-cache hash table entries: 2048 (order: 1, 8192 bytes)
[    0.054128] Mountpoint-cache hash table entries: 2048 (order: 1, 8192 bytes)
[    0.062813] CPU: Testing write buffer coherency: ok
[    0.068770] /cpus/cpu@0 missing clock-frequency property
[    0.074519] /cpus/cpu@1 missing clock-frequency property
[    0.080284] /cpus/cpu@2 missing clock-frequency property
[    0.086015] /cpus/cpu@3 missing clock-frequency property
[    0.091762] /cpus/cpu@100 missing clock-frequency property
[    0.097695] /cpus/cpu@101 missing clock-frequency property
[    0.103626] /cpus/cpu@102 missing clock-frequency property
[    0.109563] /cpus/cpu@103 missing clock-frequency property
[    0.115493] CPU0: thread -1, cpu 0, socket 0, mpidr 80000000
[    0.122353] Setting up static identity map for 0x40100000 - 0x40100060
[    0.129704] ARM CCI driver probed
[    0.133988] sunxi multi cluster SMP support installed
[    0.139930] rcu: Hierarchical SRCU implementation.
[    0.145669] smp: Bringing up secondary CPUs ...
[    1.161271] CPU1: failed to come online
[    2.201772] CPU2: failed to come online
[    3.242379] CPU3: failed to come online
[    4.282996] CPU4: failed to come online
[    5.323613] CPU5: failed to come online
[    6.364233] CPU6: failed to come online
[    7.404842] CPU7: failed to come online
[    7.409038] smp: Brought up 1 node, 1 CPU
[    7.413357] SMP: Total of 1 processors activated (48.00 BogoMIPS).
[    7.420045] CPU: All CPU(s) started in SVC mode.

Sometime(5.3 and v5.4-rc3 for example) instead of behing stuck, it reboot with
BUG: spinlock cpu recursion on CPU#0, swapper/0/0

This replicable alternate of working/non-working is very strange.
I have started to run boot multiple times to be sure it is really replicable (2 3 boots is not enough).