powerpc: Don't select HAVE_IRQ_EXIT_ON_IRQ_STACK
authorMichael Ellerman <mpe@ellerman.id.au>
Wed, 25 May 2022 03:26:39 +0000 (13:26 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Sun, 29 May 2022 00:30:42 +0000 (10:30 +1000)
commit1346d00e1bdfd4067f92bc14e8a6131a01de4190
treeafda759af0080a4de91abe219a6a97d044ea4eea
parenta1b29ba2f2c171b9bea73be993bfdf0a62d37d15
powerpc: Don't select HAVE_IRQ_EXIT_ON_IRQ_STACK

The HAVE_IRQ_EXIT_ON_IRQ_STACK option tells generic code that irq_exit()
is called while still running on the hard irq stack (hardirq_ctx[] in
the powerpc code).

Selecting the option means the generic code will *not* switch to the
softirq stack before running softirqs, because the code is already
running on the (mostly empty) hard irq stack.

But since commit 1b1b6a6f4cc0 ("powerpc: handle irq_enter/irq_exit in
interrupt handler wrappers"), irq_exit() is now called on the regular task
stack, not the hard irq stack.

That's because previously irq_exit() was called in __do_irq() which is
run on the hard irq stack, but now it is called in
interrupt_async_exit_prepare() which is called from do_irq() constructed
by the wrapper macro, which is after the switch back to the task stack.

So drop HAVE_IRQ_EXIT_ON_IRQ_STACK from the Kconfig. This will mean an
extra stack switch when processing some interrupts, but should
significantly reduce the likelihood of stack overflow.

It also means the softirq stack will be used for running softirqs from
other interrupts that don't use the hard irq stack, eg. timer interrupts.

Fixes: 1b1b6a6f4cc0 ("powerpc: handle irq_enter/irq_exit in interrupt handler wrappers")
Cc: stable@vger.kernel.org # v5.12+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220525032639.1947280-1-mpe@ellerman.id.au
arch/powerpc/Kconfig