Re: [PATCH] ACPI / OSL: Fix rcu synchronization logic

[Rafael J. Wysocki]

On Mon, Jan 9, 2017 at 10:56 AM, Lv Zheng <lv.zheng@xxxxxxxxx> wrote:
> The rcu synchronization logic is originally provided to protect
> apei_read()/apei_write() as in the APEI drivers, there is NMI event source
> requiring non spinlock based synchronization mechanism.
>
> After that, ACPI developers think FADT registers may also require same
> facility, so they moved the RCU stuffs to generic ACPI layer.
>
> So now non-task-context ACPI map lookup is only protected by RCU.
>
> This triggers problem as acpi_os_map_memory()/acpi_os_unmap_memory() can be
> used to map/unmap tables as long as to map/unmap ACPI registers. When it is
> used for the ACPI tables, the caller could invoke this very early. When it
> is invoked earlier than workqueue_init() and later than
> check_early_ioremp_leak(), invoking synchronize_rcu_expedited() can cause a
> kernel hang.
>
> Actually this facility is only used to protect non-task-context ACPI map
> lookup,

That doesn't sound quite right.

acpi_os_read/write_memory() use RCU-protected list lookups, so it's
not just non-task-context AFAICS.

> and such mappings are only introduced by
> acpi_os_map_generic_address(). So before it is invoked, there is no need to
> invoke synchronize_rcu_expedited().

That said it may be fine to start actually synchronize RCU after
acpi_os_map_generic_address() has been called for the first time.  I
need a better (or more detailed) explanation why it is fine, though.

Thanks,
Rafael