ACK: [PATCH 1/1] acpi_init: move to full RCU free for IO mappings

[Andy Whitcroft]

On Fri, Sep 23, 2011 at 10:23:03AM -0600, Tim Gardner wrote:
> On 09/23/2011 09:50 AM, Andy Whitcroft wrote:
> >During acpi_init we map and unmap a large number of mappings.  The existing
> >code uses 'simple' RCU based free, triggering a full sychronize_rcu()
> >against the caller.  If the machine is busy (for example unpacking the
> >initramfs) at the time this triggers a significant delay until all cpus
> >rendevous.  Move those to a delayed RCU free callback.
> >
> >For our reference platform this cuts the average acpi_init time in half,
> >dropping boot time by .25s.
> >
> >Signed-off-by: Andy Whitcroft<apw at canonical.com>
> >---
> >  drivers/acpi/atomicio.c |   12 +++++++++---
> >  drivers/acpi/osl.c      |   16 +++++++++++-----
> >  2 files changed, 20 insertions(+), 8 deletions(-)
> >
> >diff --git a/drivers/acpi/atomicio.c b/drivers/acpi/atomicio.c
> >index 7489b89..0a7a554 100644
> >--- a/drivers/acpi/atomicio.c
> >+++ b/drivers/acpi/atomicio.c
> >@@ -49,6 +49,7 @@ struct acpi_iomap {
> >  	unsigned long size;
> >  	phys_addr_t paddr;
> >  	struct kref ref;
> >+	struct rcu_head iomap_rcu;
> >  };
> >
> >  /* acpi_iomaps_lock or RCU read lock must be held before calling */
> >@@ -161,6 +162,13 @@ static void __acpi_kref_del_iomap(struct kref *ref)
> >   * Used to post-unmap the specified IO memory area. The iounmap is
> >   * done only if the reference count goes zero.
> >   */
> >+static void acpi_post_unmap_rcu(struct rcu_head *rcu)
> >+{
> >+	struct acpi_iomap *map = container_of(rcu,
> >+					struct acpi_iomap, iomap_rcu);
> >+	iounmap(map->vaddr);
> >+	kfree(map);
> >+}
> >  static void acpi_post_unmap(phys_addr_t paddr, unsigned long size)
> >  {
> >  	struct acpi_iomap *map;
> >@@ -176,9 +184,7 @@ static void acpi_post_unmap(phys_addr_t paddr, unsigned long size)
> >  	if (!del)
> >  		return;
> >
> >-	synchronize_rcu();
> >-	iounmap(map->vaddr);
> >-	kfree(map);
> >+	call_rcu(&map->iomap_rcu, acpi_post_unmap_rcu);
> >  }
> >
> >  /* In NMI handler, should set silent = 1 */
> >diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
> >index 372f9b7..9b3a520 100644
> >--- a/drivers/acpi/osl.c
> >+++ b/drivers/acpi/osl.c
> >@@ -105,6 +105,7 @@ struct acpi_ioremap {
> >  	acpi_physical_address phys;
> >  	acpi_size size;
> >  	unsigned long refcount;
> >+	struct rcu_head map_rcu;
> >  };
> >
> >  static LIST_HEAD(acpi_ioremaps);
> >@@ -373,13 +374,18 @@ static void acpi_os_drop_map_ref(struct acpi_ioremap *map)
> >  		list_del_rcu(&map->list);
> >  }
> >
> >+static void acpi_os_map_cleanup_rcu(struct rcu_head *rcu)
> >+{
> >+	struct acpi_ioremap *map = container_of(rcu,
> >+					struct acpi_ioremap, map_rcu);
> >+	iounmap(map->virt);
> >+	kfree(map);
> >+}
> >+
> >  static void acpi_os_map_cleanup(struct acpi_ioremap *map)
> >  {
> >-	if (!map->refcount) {
> >-		synchronize_rcu();
> >-		iounmap(map->virt);
> >-		kfree(map);
> >-	}
> >+	if (!map->refcount)
> >+		call_rcu(&map->map_rcu, acpi_os_map_cleanup_rcu);
> >  }
> >
> >  void __ref acpi_os_unmap_memory(void __iomem *virt, acpi_size size)
> 
> Should this be 'UBUNTU: SAUCE:' until its upstreamed (which I think
> is a reasonable thing to do)?

Yes it should.

> Acked-by: Tim Gardner <tim.gardner at canonical.com>

-apw