The PSC IRQ is requested using request_irq() API and as result it can
be forced to be threaded IRQ in RT-Kernel if PCS_QUIRK_HAS_SHARED_IRQ
is enabled for pinctrl domain.
As result, following 'possible irq lock inversion dependency' report
can be seen:
=========================================================
[ INFO: possible irq lock inversion dependency detected ]
3.14.43-rt42-00360-g96ff499-dirty #24 Not tainted
---------------------------------------------------------
irq/369-pinctrl/927 just changed the state of lock:
(&pcs->lock){+.....}, at: [<
c0375b54>] pcs_irq_handle+0x48/0x9c
but this lock was taken by another, HARDIRQ-safe lock in the past:
(&irq_desc_lock_class){-.....}
and interrupts could create inverse lock ordering between them.
other info that might help us debug this:
Possible interrupt unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&pcs->lock);
local_irq_disable();
lock(&irq_desc_lock_class);
lock(&pcs->lock);
<Interrupt>
lock(&irq_desc_lock_class);
*** DEADLOCK ***
no locks held by irq/369-pinctrl/927.
the shortest dependencies between 2nd lock and 1st lock:
-> (&irq_desc_lock_class){-.....} ops: 58724 {
IN-HARDIRQ-W at:
[<
c0090040>] lock_acquire+0x9c/0x158
[<
c07065c8>] _raw_spin_lock+0x48/0x58
[<
c009edac>] handle_fasteoi_irq+0x24/0x15c
[<
c009abb0>] generic_handle_irq+0x3c/0x4c
[<
c000f83c>] handle_IRQ+0x50/0xa0
[<
c0008674>] gic_handle_irq+0x3c/0x6c
[<
c0707a04>] __irq_svc+0x44/0x8c
[<
c000fc44>] arch_cpu_idle+0x40/0x4c
[<
c009aadc>] cpu_startup_entry+0x270/0x2e0
[<
c06fcbf8>] rest_init+0xd4/0xe4
[<
c0a44bfc>] start_kernel+0x3d0/0x3dc
[<
80008084>] 0x80008084
INITIAL USE at:
[<
c0090040>] lock_acquire+0x9c/0x158
[<
c070674c>] _raw_spin_lock_irqsave+0x54/0x68
[<
c009aff8>] __irq_get_desc_lock+0x64/0xa4
[<
c009e38c>] irq_set_chip+0x30/0x78
[<
c009ec30>] irq_set_chip_and_handler_name+0x24/0x3c
[<
c036ca10>] gic_irq_domain_map+0x48/0xb4
[<
c00a0a80>] irq_domain_associate+0x84/0x1d4
[<
c00a1154>] irq_create_mapping+0x80/0x11c
[<
c00a1270>] irq_create_of_mapping+0x80/0x120
[<
c05cdaa8>] irq_of_parse_and_map+0x34/0x3c
[<
c0a4ea24>] omap_dm_timer_init_one+0x90/0x30c
[<
c0a4eef0>] omap5_realtime_timer_init+0x8c/0x48c
[<
c0a486b0>] time_init+0x28/0x38
[<
c0a44a6c>] start_kernel+0x240/0x3dc
[<
80008084>] 0x80008084
}
... key at: [<
c1049ce0>] irq_desc_lock_class+0x0/0x8
... acquired at:
[<
c07065c8>] _raw_spin_lock+0x48/0x58
[<
c0375a90>] pcs_irq_unmask+0x58/0xa0
[<
c009ea48>] irq_enable+0x38/0x48
[<
c009ead0>] irq_startup+0x78/0x7c
[<
c009d440>] __setup_irq+0x4a8/0x4f4
[<
c009d5dc>] request_threaded_irq+0xb8/0x138
[<
c0415a5c>] omap_8250_startup+0x4c/0x148
[<
c041276c>] serial8250_startup+0x24/0x30
[<
c040d0ec>] uart_startup.part.9+0x5c/0x1b4
[<
c040dbcc>] uart_open+0xf4/0x16c
[<
c03f0540>] tty_open+0x170/0x61c
[<
c0157028>] chrdev_open+0xbc/0x1b4
[<
c0150494>] do_dentry_open+0x1e8/0x2bc
[<
c0150a84>] finish_open+0x44/0x5c
[<
c0160d50>] do_last.isra.47+0x710/0xca0
[<
c01613a4>] path_openat+0xc4/0x640
[<
c0162904>] do_filp_open+0x3c/0x98
[<
c0151bdc>] do_sys_open+0x114/0x1d8
[<
c0151cc8>] SyS_open+0x28/0x2c
[<
c0a44d70>] kernel_init_freeable+0x168/0x1e4
[<
c06fcc24>] kernel_init+0x1c/0xf8
[<
c000eee8>] ret_from_fork+0x14/0x20
-> (&pcs->lock){+.....} ops: 65 {
HARDIRQ-ON-W at:
[<
c0090040>] lock_acquire+0x9c/0x158
[<
c07065c8>] _raw_spin_lock+0x48/0x58
[<
c0375b54>] pcs_irq_handle+0x48/0x9c
[<
c0375c5c>] pcs_irq_handler+0x1c/0x28
[<
c009c458>] irq_forced_thread_fn+0x30/0x74
[<
c009c784>] irq_thread+0x158/0x1c4
[<
c0063fc4>] kthread+0xd4/0xe8
[<
c000eee8>] ret_from_fork+0x14/0x20
INITIAL USE at:
[<
c0090040>] lock_acquire+0x9c/0x158
[<
c070674c>] _raw_spin_lock_irqsave+0x54/0x68
[<
c0375344>] pcs_enable+0x7c/0xe8
[<
c0372a44>] pinmux_enable_setting+0x178/0x220
[<
c036fecc>] pinctrl_select_state+0x110/0x194
[<
c04732dc>] pinctrl_bind_pins+0x7c/0x108
[<
c045853c>] driver_probe_device+0x70/0x254
[<
c0458810>] __driver_attach+0x9c/0xa0
[<
c045674c>] bus_for_each_dev+0x78/0xac
[<
c0458030>] driver_attach+0x2c/0x30
[<
c0457c78>] bus_add_driver+0x15c/0x204
[<
c0458ee0>] driver_register+0x88/0x108
[<
c045a168>] __platform_driver_register+0x64/0x6c
[<
c0a8170c>] omap_hsmmc_driver_init+0x1c/0x20
[<
c0008a94>] do_one_initcall+0x110/0x170
[<
c0a44d48>] kernel_init_freeable+0x140/0x1e4
[<
c06fcc24>] kernel_init+0x1c/0xf8
[<
c000eee8>] ret_from_fork+0x14/0x20
}
... key at: [<
c1088a8c>] __key.18572+0x0/0x8
... acquired at:
[<
c008cdd4>] mark_lock+0x388/0x76c
[<
c008df40>] __lock_acquire+0x6d0/0x1f98
[<
c0090040>] lock_acquire+0x9c/0x158
[<
c07065c8>] _raw_spin_lock+0x48/0x58
[<
c0375b54>] pcs_irq_handle+0x48/0x9c
[<
c0375c5c>] pcs_irq_handler+0x1c/0x28
[<
c009c458>] irq_forced_thread_fn+0x30/0x74
[<
c009c784>] irq_thread+0x158/0x1c4
[<
c0063fc4>] kthread+0xd4/0xe8
[<
c000eee8>] ret_from_fork+0x14/0x20
stack backtrace:
CPU: 1 PID: 927 Comm: irq/369-pinctrl Not tainted
3.14.43-rt42-00360-g96ff499-dirty #24
[<
c00177e0>] (unwind_backtrace) from [<
c00130b0>] (show_stack+0x20/0x24)
[<
c00130b0>] (show_stack) from [<
c0702958>] (dump_stack+0x84/0xd0)
[<
c0702958>] (dump_stack) from [<
c008bcfc>] (print_irq_inversion_bug+0x1d0/0x21c)
[<
c008bcfc>] (print_irq_inversion_bug) from [<
c008bf18>] (check_usage_backwards+0xb4/0x11c)
[<
c008bf18>] (check_usage_backwards) from [<
c008cdd4>] (mark_lock+0x388/0x76c)
[<
c008cdd4>] (mark_lock) from [<
c008df40>] (__lock_acquire+0x6d0/0x1f98)
[<
c008df40>] (__lock_acquire) from [<
c0090040>] (lock_acquire+0x9c/0x158)
[<
c0090040>] (lock_acquire) from [<
c07065c8>] (_raw_spin_lock+0x48/0x58)
[<
c07065c8>] (_raw_spin_lock) from [<
c0375b54>] (pcs_irq_handle+0x48/0x9c)
[<
c0375b54>] (pcs_irq_handle) from [<
c0375c5c>] (pcs_irq_handler+0x1c/0x28)
[<
c0375c5c>] (pcs_irq_handler) from [<
c009c458>] (irq_forced_thread_fn+0x30/0x74)
[<
c009c458>] (irq_forced_thread_fn) from [<
c009c784>] (irq_thread+0x158/0x1c4)
[<
c009c784>] (irq_thread) from [<
c0063fc4>] (kthread+0xd4/0xe8)
[<
c0063fc4>] (kthread) from [<
c000eee8>] (ret_from_fork+0x14/0x20)
To fix it use IRQF_NO_THREAD to ensure that pcs irq will not be forced threaded.
Cc: Tony Lindgren <tony@atomide.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Acked-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>