Re: [PATCH] ACPI: Drop rcu usage for MMIO mappings

[Andy Shevchenko]

On Thu, May 7, 2020 at 3:21 AM Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
>
> Recently a performance problem was reported for a process invoking a
> non-trival ASL program. The method call in this case ends up
> repetitively triggering a call path like:
>
>     acpi_ex_store
>     acpi_ex_store_object_to_node
>     acpi_ex_write_data_to_field
>     acpi_ex_insert_into_field
>     acpi_ex_write_with_update_rule
>     acpi_ex_field_datum_io
>     acpi_ex_access_region
>     acpi_ev_address_space_dispatch
>     acpi_ex_system_memory_space_handler
>     acpi_os_map_cleanup.part.14
>     _synchronize_rcu_expedited.constprop.89
>     schedule
>
> The end result of frequent synchronize_rcu_expedited() invocation is
> tiny sub-millisecond spurts of execution where the scheduler freely
> migrates this apparently sleepy task. The overhead of frequent scheduler
> invocation multiplies the execution time by a factor of 2-3X.
>
> For example, performance improves from 16 minutes to 7 minutes for a
> firmware update procedure across 24 devices.
>
> Perhaps the rcu usage was intended to allow for not taking a sleeping
> lock in the acpi_os_{read,write}_memory() path which ostensibly could be
> called from an APEI NMI error interrupt? Neither rcu_read_lock() nor
> ioremap() are interrupt safe, so add a WARN_ONCE() to validate that rcu
> was not serving as a mechanism to avoid direct calls to ioremap(). Even
> the original implementation had a spin_lock_irqsave(), but that is not
> NMI safe.
>
> APEI itself already has some concept of avoiding ioremap() from
> interrupt context (see erst_exec_move_data()), if the new warning
> triggers it means that APEI either needs more instrumentation like that
> to pre-emptively fail, or more infrastructure to arrange for pre-mapping
> the resources it needs in NMI context.

...

> +static void __iomem *acpi_os_rw_map(acpi_physical_address phys_addr,
> +                                   unsigned int size, bool *did_fallback)
> +{
> +       void __iomem *virt_addr = NULL;

Assignment is not needed as far as I can see.

> +       if (WARN_ONCE(in_interrupt(), "ioremap in interrupt context\n"))
> +               return NULL;
> +
> +       /* Try to use a cached mapping and fallback otherwise */
> +       *did_fallback = false;
> +       mutex_lock(&acpi_ioremap_lock);
> +       virt_addr = acpi_map_vaddr_lookup(phys_addr, size);
> +       if (virt_addr)
> +               return virt_addr;
> +       mutex_unlock(&acpi_ioremap_lock);
> +
> +       virt_addr = acpi_os_ioremap(phys_addr, size);
> +       *did_fallback = true;
> +
> +       return virt_addr;
> +}

I'm wondering if Sparse is okay with this...

> +static void acpi_os_rw_unmap(void __iomem *virt_addr, bool did_fallback)
> +{
> +       if (did_fallback) {
> +               /* in the fallback case no lock is held */
> +               iounmap(virt_addr);
> +               return;
> +       }
> +
> +       mutex_unlock(&acpi_ioremap_lock);
> +}

...and this functions from locking perspective.

-- 
With Best Regards,
Andy Shevchenko