Re: [PATCH v4 00/12] mm: make userland page table freeing RCU-safe

On Tue, Sep 22, 2026 at 05:42:10PM +0100, Matthew Wilcox wrote:

On Tue, Sep 22, 2026 at 04:35:31PM +0100, Lorenzo Stoakes (ARM) wrote:

quoted

m68k-motorola required the biggest change - since RCU page table freeing
uses call_rcu(), this means page table freeing can arise from softirq
context.

This was fixed with an IRQ-safe spin lock used in both get_pointer_table()
and free_pointer_table().

You don't need to use an irq-safe spinlock for softirqs.  It does work,
but it's overkill since it disables hardirqs.  If you use spin_lock_bh(),
only softirqs are prevented from running (and it's safe to call
spin_lock_bh() when in softirq context, just like it's safe to call
spin_lock_irqsave() in hardirq or softirq context).

Right yup, I know _bh() is fine for softirq, I think this was being overly
cautious (for sparc32 you do actually need irq saving).

But this is m68k :) functionally there's really not that much difference, and
it's over very few instrucitons so the delta here is likely to be functionally
little to nothing.

If I need to respin again I'll fix it up!

--
Cheers, Lorenzo