[linux-yocto] [PATCH 1/1] xilinx-zyqn: Move disable_nonboot_cpus() in front of local_irq_disable()

[Quanyang Wang]

Hi Michal,

On 10/21/19 4:16 PM, Michal Simek wrote:
> On 21. 10. 19 7:50, quanyang.wang at windriver.com wrote:
>> From: Quanyang Wang <quanyang.wang at windriver.com>
>>
>> When run kdump with enabling CONFIG_DEBUG_PREEMPT, there is a calltrace
>> as below:
>>
>> BUG: using smp_processor_id() in preemptible [00000000] code: sh/303
>> caller is machine_crash_shutdown+0x2c/0xe8
>> CPU: 0 PID: 303 Comm: sh Kdump: loaded Not tainted 5.2.20-yocto-standard #1
>> Hardware name: Xilinx Zynq Platform
>> [<80112ff4>] (unwind_backtrace) from [<8010ca4c>] (show_stack+0x18/0x1c)
>> [<8010ca4c>] (show_stack) from [<809b000c>] (dump_stack+0x70/0x8c)
>> [<809b000c>] (dump_stack) from [<80549a14>] (debug_smp_processor_id+0xd4/0x118)
>> [<80549a14>] (debug_smp_processor_id) from [<80111428>] (machine_crash_shutdown+0x2c/0xe8)
>> [<80111428>] (machine_crash_shutdown) from [<801afe24>] (__crash_kexec+0x70/0xd0)
>> [<801afe24>] (__crash_kexec) from [<801259b4>] (panic+0x110/0x324)
>> [<801259b4>] (panic) from [<805f7018>] (sysrq_handle_crash+0x18/0x1c)
>> [<805f7018>] (sysrq_handle_crash) from [<805f7584>] (__handle_sysrq+0x9c/0x14c)
>> [<805f7584>] (__handle_sysrq) from [<805f79e8>] (write_sysrq_trigger+0x5c/0x6c)
>> [<805f79e8>] (write_sysrq_trigger) from [<8031e850>] (proc_reg_write+0x78/0x8c)
>> [<8031e850>] (proc_reg_write) from [<802b1b28>] (vfs_write+0xc0/0x154)
>> [<802b1b28>] (vfs_write) from [<802b2a64>] (ksys_write+0x6c/0xd4)
>> [<802b2a64>] (ksys_write) from [<80101000>] (ret_fast_syscall+0x0/0x54)
>> Exception stack(0xba157fa8 to 0xba157ff0)
>> 7fa0: 00000002 005ab930 00000001 005ab930 00000002 00000000
>> 7fc0: 00000002 005ab930 76fa2290 00000004 76f3d124 76f3cc8c 00000000 00000000
>> 7fe0: 00000004 7edec940 76edbfff 76e67d16
>>
>> This is because that the function disable_nonboot_cpus is called in
>> order to make sure that the crash kernel runs in the boot CPU(cpu0).
>> And it will enable local irq by calling as below:
>>
>> disable_nonboot_cpus
>>   -> freeze_secondary_cpus
>>    -> _cpu_down
>>     -> percpu_down_write
>>      -> rcu_sync_enter
>>       -> spin_unlock_irq(&rsp->rss_lock)
>>        -> local_irq_enable()
>>
>> Then the functions including smp_processor_id() behind disable_nonboot_cpus
>> will run at the irq-enabled context, and this will trigger the calltrace.
>>
>> So move disable_nonboot_cpus() in front of local_irq_disable() to avoid
>> it since disable_nonboot_cpus() not need run at an atomic context.
>>
>> Signed-off-by: Quanyang Wang <quanyang.wang at windriver.com>
>> ---
>>   arch/arm/kernel/machine_kexec.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/arm/kernel/machine_kexec.c b/arch/arm/kernel/machine_kexec.c
>> index 654f2b1f9ac0..83d2025a4ab1 100644
>> --- a/arch/arm/kernel/machine_kexec.c
>> +++ b/arch/arm/kernel/machine_kexec.c
>> @@ -145,9 +145,10 @@ static void machine_kexec_mask_interrupts(void)
>>   
>>   void machine_crash_shutdown(struct pt_regs *regs)
>>   {
>> -	local_irq_disable();
>>   	disable_nonboot_cpus();
>>   
>> +	local_irq_disable();
>> +
>>   	crash_smp_send_stop();
>>   
>>   	crash_save_cpu(regs, smp_processor_id());
>>
> ok. Can you please check before this if your usecases work without
> disable_nonboot_cpus(). This patch was done pretty long time ago where
> there was an issue with kexec. Long time ago I was talking to arm-soc
> maintainers about this and they told me that mainline code should work
> fine without any need to call disable_nonboot_cpus().
> It means if kexec is working fine we can revert origin patch and use
> what mainline is using.

It seems that the issue is still there. When crash at cpu1 and crash 
kernel runs at cpu1,

it will hang, the log is as below:

root at xilinx-zynq:~# sh 1.sh
syscall kexec_file_load not available.
sysrq: Trigger a crash
Kernel panic - not syncing: sysrq triggered crash
CPU: 1 PID: 308 Comm: sh Kdump: loaded Not tainted 5.2.20-yocto-standard #4
Hardware name: Xilinx Zynq Platform
[<80112eb0>] (unwind_backtrace) from [<8010cc04>] (show_stack+0x18/0x1c)
[<8010cc04>] (show_stack) from [<8094f8f4>] (dump_stack+0x70/0x8c)
[<8094f8f4>] (dump_stack) from [<801256f4>] (panic+0xf8/0x320)
[<801256f4>] (panic) from [<805dbeb0>] (sysrq_handle_crash+0x18/0x1c)
[<805dbeb0>] (sysrq_handle_crash) from [<805dc3b8>] 
(__handle_sysrq+0x9c/0x148)
[<805dc3b8>] (__handle_sysrq) from [<805dc804>] 
(write_sysrq_trigger+0x5c/0x6c)
[<805dc804>] (write_sysrq_trigger) from [<8031b040>] 
(proc_reg_write+0x78/0x8c)
[<8031b040>] (proc_reg_write) from [<802aeec4>] (vfs_write+0xc0/0x154)
[<802aeec4>] (vfs_write) from [<802afd18>] (ksys_write+0x64/0xc8)
[<802afd18>] (ksys_write) from [<80101000>] (ret_fast_syscall+0x0/0x54)
Exception stack(0xb905bfa8 to 0xb905bff0)
bfa0:                   00000002 0059afa0 00000001 0059afa0 00000002 
00000000
bfc0: 00000002 0059afa0 76f8e290 00000004 76f29124 76f28c8c 00000000 
00000000
bfe0: 00000004 7eb858c0 76ec7fff 76e53d16
CPU 0 will stop doing anything useful since another CPU has crashed
Loading crashdump kernel...
Bye!
Booting Linux on physical CPU 0x1
Linux version 5.2.20-yocto-standard (oe-user at oe-host) (gcc version 9.2.0 
(GCC)) #1 SMP PREEMPT Thu Oct 17 08:15:14 UTC 2019
CPU: ARMv7 Processor [413fc090] revision 0 (ARMv7), cr=18c5387d
CPU: PIPT / VIPT nonaliasing data cache, VIPT aliasing instruction cache
OF: fdt: Machine model: Xilinx ZC706 board
OF: fdt: Ignoring memory range 0x0 - 0x8000000
printk: debug: ignoring loglevel setting.
printk: bootconsole [earlycon0] enabled
Memory policy: Data cache writealloc
cma: Reserved 16 MiB at 0x16c00000
On node 0 totalpages: 65280
   Normal zone: 574 pages used for memmap
   Normal zone: 0 pages reserved
   Normal zone: 65280 pages, LIFO batch:15
percpu: Embedded 19 pages/cpu s47756 r8192 d21876 u77824
pcpu-alloc: s47756 r8192 d21876 u77824 alloc=19*4096
pcpu-alloc: [0] 0 [0] 1
Built 1 zonelists, mobility grouping on.  Total pages: 64706
Kernel command line: console=ttyPS0,115200n8 root=/dev/nfs rw 
nfsroot=128.224.165.20:/export/pxeboot/vlm-boards/22009/rootfs,v3,tcp 
ip=128.224.179.217:128.224.165.20:128.224.178.1:255.255.254.0:zc702:eth0:off 
ignore_loglevel earlyprintk noinitrd selinux=0 enforcing=0 kmemleak=on 
elfcorehdr=0x17f00000 mem=261120K
Dentry cache hash table entries: 32768 (order: 5, 131072 bytes)
Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
Memory: 227332K/261120K available (9216K kernel code, 725K rwdata, 2284K 
rodata, 1024K init, 567K bss, 17404K reserved, 16384K cma-reserved)
SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=2, Nodes=1
ftrace: allocating 35203 entries in 69 pages
rcu: Preemptible hierarchical RCU implementation.
rcu:    RCU restricting CPUs from NR_CPUS=4 to nr_cpu_ids=2.
         Tasks RCU enabled.
rcu: RCU calculated value of scheduler-enlistment delay is 10 jiffies.
rcu: Adjusting geometry for rcu_fanout_leaf=16, nr_cpu_ids=2
NR_IRQS: 16, nr_irqs: 16, preallocated irqs: 16
efuse mapped to (ptrval)
slcr mapped to (ptrval)
L2C: platform provided aux values match the hardware, so have no 
effect.  Please remove them.
L2C-310 erratum 769419 enabled
L2C-310 enabling early BRESP for Cortex-A9
L2C-310: enabling full line of zeros but not enabled in Cortex-A9
L2C-310 ID prefetch enabled, offset 1 lines
L2C-310 dynamic clock gating enabled, standby mode enabled
L2C-310 cache controller enabled, 8 ways, 512 kB
L2C-310: CACHE_ID 0x410000c8, AUX_CTRL 0x76760001
random: get_random_bytes called from start_kernel+0x2b0/0x4c4 with 
crng_init=0
zynq_clock_init: clkc starts at (ptrval)
Zynq clock init
sched_clock: 64 bits at 333MHz, resolution 3ns, wraps every 4398046511103ns
clocksource: arm_global_timer: mask: 0xffffffffffffffff max_cycles: 
0x4ce07af025, max_idle_ns: 440795209040 ns
Switching to timer-based delay loop, resolution 3ns
clocksource: ttc_clocksource: mask: 0xffff max_cycles: 0xffff, 
max_idle_ns: 537538477 ns
timer #0 at (ptrval), irq=17
Console: colour dummy device 80x30
Calibrating delay loop (skipped), value calculated using timer 
frequency.. 666.66 BogoMIPS (lpj=3333333)
pid_max: default: 32768 minimum: 301
LSM: Security Framework initializing
Mount-cache hash table entries: 1024 (order: 0, 4096 bytes)
Mountpoint-cache hash table entries: 1024 (order: 0, 4096 bytes)
CPU: Testing write buffer coherency: ok
CPU0: Spectre v2: using BPIALL workaround
CPU0: thread -1, cpu 1, socket 0, mpidr 80000001
Setting up static identity map for 0x8100000 - 0x8100060
rcu: Hierarchical SRCU implementation.
smp: Bringing up secondary CPUs ...


Thanks,

Quanyang

>
> Thanks,
> Michal
>
>