Re: [PATCH] LoongArch: Only call get_timer_irq() once in constant_clockevent_init()

From: Huacai Chen
Date: Sat Feb 11 2023 - 06:29:38 EST


Hi, Tiezhu,

On Sat, Feb 11, 2023 at 7:16 PM Tiezhu Yang <yangtiezhu@xxxxxxxxxxx> wrote:
>
> Under CONFIG_DEBUG_ATOMIC_SLEEP=y and CONFIG_DEBUG_PREEMPT=y, we can see
> the following messages on LoongArch, this is because using might_sleep()
> in preemption disable context.
>
> [ 0.001127] smp: Bringing up secondary CPUs ...
> [ 0.001222] Booting CPU#1...
> [ 0.001244] 64-bit Loongson Processor probed (LA464 Core)
> [ 0.001247] CPU1 revision is: 0014c012 (Loongson-64bit)
> [ 0.001250] FPU1 revision is: 00000000
> [ 0.001252] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:283
> [ 0.001255] in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 0, name: swapper/1
> [ 0.001257] preempt_count: 1, expected: 0
> [ 0.001258] RCU nest depth: 0, expected: 0
> [ 0.001259] Preemption disabled at:
> [ 0.001261] [<9000000000223800>] arch_dup_task_struct+0x20/0x110
> [ 0.001272] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 6.2.0-rc7+ #43
> [ 0.001275] Hardware name: Loongson Loongson-3A5000-7A1000-1w-A2101/Loongson-LS3A5000-7A1000-1w-A2101, BIOS vUDK2018-LoongArch-V4.0.05132-beta10 12/13/202
> [ 0.001277] Stack : 0072617764726148 0000000000000000 9000000000222f1c 90000001001e0000
> [ 0.001286] 90000001001e3be0 90000001001e3be8 0000000000000000 0000000000000000
> [ 0.001292] 90000001001e3be8 0000000000000040 90000001001e3cb8 90000001001e3a50
> [ 0.001297] 9000000001642000 90000001001e3be8 be694d10ce4139dd 9000000100174500
> [ 0.001303] 0000000000000001 0000000000000001 00000000ffffe0a2 0000000000000020
> [ 0.001309] 000000000000002f 9000000001354116 00000000056b0000 ffffffffffffffff
> [ 0.001314] 0000000000000000 0000000000000000 90000000014f6e90 9000000001642000
> [ 0.001320] 900000000022b69c 0000000000000001 0000000000000000 9000000001736a90
> [ 0.001325] 9000000100038000 0000000000000000 9000000000222f34 0000000000000000
> [ 0.001331] 00000000000000b0 0000000000000004 0000000000000000 0000000000070000
> [ 0.001337] ...
> [ 0.001339] Call Trace:
> [ 0.001342] [<9000000000222f34>] show_stack+0x5c/0x180
> [ 0.001346] [<90000000010bdd80>] dump_stack_lvl+0x60/0x88
> [ 0.001352] [<9000000000266418>] __might_resched+0x180/0x1cc
> [ 0.001356] [<90000000010c742c>] mutex_lock+0x20/0x64
> [ 0.001359] [<90000000002a8ccc>] irq_find_matching_fwspec+0x48/0x124
> [ 0.001364] [<90000000002259c4>] constant_clockevent_init+0x68/0x204
> [ 0.001368] [<900000000022acf4>] start_secondary+0x40/0xa8
> [ 0.001371] [<90000000010c0124>] smpboot_entry+0x60/0x64
>
> Here are the complete call chains:
>
> smpboot_entry()
> start_secondary()
> constant_clockevent_init()
> get_timer_irq()
> irq_find_matching_fwnode()
> irq_find_matching_fwspec()
> mutex_lock()
> might_sleep()
> __might_sleep()
> __might_resched()
>
> In order to avoid the above issue, we should break the call chains,
> using timer_irq_installed variable as check condition to only call
> get_timer_irq() once in constant_clockevent_init() is a simple and
> proper way.
>
> Signed-off-by: Tiezhu Yang <yangtiezhu@xxxxxxxxxxx>
> ---
> arch/loongarch/kernel/time.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/arch/loongarch/kernel/time.c b/arch/loongarch/kernel/time.c
> index a6576de..6263b5a 100644
> --- a/arch/loongarch/kernel/time.c
> +++ b/arch/loongarch/kernel/time.c
> @@ -140,14 +140,15 @@ static int get_timer_irq(void)
>
> int constant_clockevent_init(void)
> {
> - int irq;
> + static int irq;
> unsigned int cpu = smp_processor_id();
> unsigned long min_delta = 0x600;
> unsigned long max_delta = (1UL << 48) - 1;
> struct clock_event_device *cd;
> static int timer_irq_installed = 0;
>
> - irq = get_timer_irq();
> + if (timer_irq_installed == 0)
> + irq = get_timer_irq();
> if (irq < 0)
> pr_err("Failed to map irq %d (timer)\n", irq);
This should also be in the if condition, but you don't need to send a
new version. :)

Huacai
>
> --
> 2.1.0
>