[PATCH] genirq/chip: Don't call add_interrupt_randomness() for NMIs

[Mark Rutland <mark.rutland@arm.com>]

Recently handle_percpu_devid_irq() was changed to call
add_interrupt_randomness(). This introduced a potential deadlock when
handle_percpu_devid_irq() is used to handle an NMI, which can be
detected with lockdep, e.g.

    ================================
    WARNING: inconsistent lock state
    7.1.0-rc2-pnmi #465 Not tainted
    --------------------------------
    inconsistent {INITIAL USE} -> {IN-NMI} usage.
    perf/695 [HC1[1]:SC0[0]:HE0:SE1] takes:
    ffff00837dfd3a18 (&base->lock){-.-.}-{2:2}, at: lock_timer_base+0x6c/0xac
    {INITIAL USE} state was registered at:
      lock_acquire+0x260/0x40c
      _raw_spin_lock_irqsave+0x68/0xb0
      lock_timer_base+0x6c/0xac
      __mod_timer+0x100/0x32c
      add_timer_global+0x2c/0x40
      __queue_delayed_work+0xf0/0x140
      queue_delayed_work_on+0x134/0x138
      mem_cgroup_css_online+0x30c/0x310
      online_css+0x34/0x10c
      cgroup_init_subsys+0x158/0x1c8
      cgroup_init+0x440/0x524
      start_kernel+0x888/0x998
      __primary_switched+0x88/0x90
    irq event stamp: 62068
    hardirqs last  enabled at (62067): [<ffff8000801cc0ec>] seqcount_lockdep_reader_access.constprop.0+0xf0/0xfc
    hardirqs last disabled at (62068): [<ffff80008150e0ac>] _raw_spin_lock_irqsave+0x94/0xb0
    softirqs last  enabled at (62050): [<ffff800080017614>] put_cpu_fpsimd_context+0x1c/0x4c
    softirqs last disabled at (62048): [<ffff8000800175c8>] get_cpu_fpsimd_context+0x1c/0x4c
        other info that might help us debug this:
    Possible unsafe locking scenario:
           CPU0
           ----
      lock(&base->lock);
      <Interrupt>
        lock(&base->lock);
        *** DEADLOCK ***
    3 locks held by perf/695:
     #0: ffff0080146cd2c8 (&type->i_mutex_dir_key#6){++++}-{4:4}, at: lookup_slow+0x30/0x68
     #1: ffff80008332b858 (rcu_read_lock){....}-{1:3}, at: blk_mq_run_hw_queue+0xf4/0x1fc
     #2: ffff008000b6aa18 (&host->lock){-.-.}-{3:3}, at: ata_scsi_queuecmd+0x28/0x88
    stack backtrace:
    CPU: 3 UID: 0 PID: 695 Comm: perf Not tainted 7.1.0-rc2-pnmi #465 PREEMPT
    Hardware name: ARM LTD Morello System Development Platform, BIOS EDK II Mar  7 2024
    Call trace:
     show_stack+0x18/0x24 (C)
     dump_stack_lvl+0xc4/0x148
     dump_stack+0x18/0x24
     print_usage_bug.part.0+0x29c/0x364
     lock_acquire+0x364/0x40c
     _raw_spin_lock_irqsave+0x68/0xb0
     lock_timer_base+0x6c/0xac
     add_timer_on+0x78/0x16c
     add_interrupt_randomness+0x124/0x134
     handle_percpu_devid_irq+0xd4/0x16c
     handle_irq_desc+0x40/0x58
     generic_handle_domain_nmi+0x28/0x50
     __gic_handle_nmi.isra.0+0x4c/0xa0
     gic_handle_irq+0x38/0x2bc
     call_on_irq_stack+0x30/0x48
     do_interrupt_handler+0x80/0x98
     el1_interrupt+0x90/0xac
     el1h_64_irq_handler+0x18/0x24
     el1h_64_irq+0x80/0x84
     [...]

During review, Thomas pointed out it wouldn't be safe for
handle_percpu_devid_irq() to call add_interrupt_randomness() if it was
used to handle NMIs:

  https://lore.kernel.org/lkml/87bjgik042.ffs@tglx/

... but evidently people missed that handle_percpu_devid_irq() *is* used
for NMIs.

While it might seem that we should handle NMIs with a separate
handle_percpu_devid_nmi() function, for various structural reasons this
was impractical, and handle_percpu_devid_irq() has been expected to be
used for NMIs since commits:

  21bbbc50f398f ("irqchip/gic-v3: Switch high priority PPIs over to handle_percpu_devid_irq()")
  5ff78c8de9d83 ("genirq: Kill handle_percpu_devid_fasteoi_nmi()")

Taking the above into account, avoid the deadlock by not calling
add_interrupt_randomness() when handle_percpu_devid_irq() is called in
an NMI context. This is consistent with our other NNI handling flows,
which do not call add_interrupt_randomness().

At the same time, update the kerneldoc comment to make it clear that
handle_percpu_devid_irq() can be called in NMI context. The rest of
handle_percpu_devid_irq() is currently NMI safe and doesn't need to
change.

Fixes: fd7400cfcbaa ("genirq/chip: Invoke add_interrupt_randomness() in handle_percpu_devid_irq()")
Reported-by: Ada Couprie Diaz <ada.coupriediaz@arm.com>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Justin He <justin.he@arm.com>
Cc: Marc Zyngier <maz@kernel.org>
Cc: Michael Kelley <mhklinux@outlook.com>
Cc: Thomas Gleixner <tglx@kernel.org>
Cc: Vladimir Murzin <vladimir.murzin@arm.com>
---
 kernel/irq/chip.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index 6c9b1dc4e7d46..b635e3c5d5b6b 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -14,6 +14,7 @@
 #include <linux/interrupt.h>
 #include <linux/kernel_stat.h>
 #include <linux/irqdomain.h>
+#include <linux/preempt.h>
 #include <linux/random.h>
 
 #include <trace/events/irq.h>
@@ -893,7 +894,10 @@ void handle_percpu_irq(struct irq_desc *desc)
  *
  * action->percpu_dev_id is a pointer to percpu variables which
  * contain the real device id for the cpu on which this handler is
- * called
+ * called.
+ *
+ * May be used for NMI interrupt lines, and so may be called in IRQ or NMI
+ * context.
  */
 void handle_percpu_devid_irq(struct irq_desc *desc)
 {
@@ -930,7 +934,8 @@ void handle_percpu_devid_irq(struct irq_desc *desc)
 			    enabled ? " and unmasked" : "", irq, cpu);
 	}
 
-	add_interrupt_randomness(irq);
+	if (!in_nmi())
+		add_interrupt_randomness(irq);
 
 	if (chip->irq_eoi)
 		chip->irq_eoi(&desc->irq_data);
-- 
2.30.2