[RFC PATCH 0/2] arm64: run softirqs on the per-CPU IRQ stack

[Qi Zheng <zhengqi.arch@bytedance.com>]

Hi all,

Currently arm64 supports per-CPU IRQ stack, but softirqs are
still handled in the task context.

Since any call to local_bh_enable() at any level in the task's
call stack may trigger a softirq processing run, which could
potentially cause a task stack overflow if the combined stack
footprints exceed the stack's size. And we did encounter this
situation in the real environment:

Call trace:
 dump_backtrace+0x0/0x1cc,
 show_stack+0x14/0x1c,
 dump_stack+0xc4/0xfc,
 panic+0x150/0x2c8,
 panic+0x0/0x2c8,
 handle_bad_stack+0x11c/0x130,
 __bad_stack+0x88/0x8c,
 vsnprintf+0x2c/0x524,
 vscnprintf+0x38/0x7c,
 scnprintf+0x6c/0x90,
 /* ... */
 __do_softirq+0x1e0/0x370,
 do_softirq+0x40/0x50,
 __local_bh_enable_ip+0x8c/0x90,
 _raw_spin_unlock_bh+0x1c/0x24,
 /* ... */
 process_one_work+0x1dc/0x3e4,
 worker_thread+0x260/0x360,
 kthread+0x118/0x128,
 ret_from_fork+0x10/0x18,

So let's run these softirqs on the IRQ stack as well.

This series is based on next-20220705.

Comments and suggestions are welcome.

Thanks,
Qi.

Qi Zheng (2):
  arm64: run softirqs on the per-CPU IRQ stack
  arm64: support HAVE_IRQ_EXIT_ON_IRQ_STACK

 arch/arm64/Kconfig                 |  2 ++
 arch/arm64/include/asm/exception.h |  4 +++-
 arch/arm64/kernel/entry-common.c   | 30 ++++++++++++++++++++----------
 arch/arm64/kernel/entry.S          |  6 ++++--
 arch/arm64/kernel/irq.c            | 12 ++++++++++++
 5 files changed, 41 insertions(+), 13 deletions(-)

-- 
2.20.1