Patchwork PCI: Don't let mmio fallback to must-only, if ioport fails with must+optional

login
register
mail settings
Submitter Yinghai Lu
Date May 23, 2013, 5:11 p.m.
Message ID <1369329099-8501-1-git-send-email-yinghai@kernel.org>
Download mbox | patch
Permalink /patch/245983/
State Superseded
Headers show

Comments

Yinghai Lu - May 23, 2013, 5:11 p.m.
BenH reported that there is some assign unassigned resource problem
in powerpc.

It turns out after
| commit 0c5be0cb0edfe3b5c4b62eac68aa2aa15ec681af
| Date:   Thu Feb 23 19:23:29 2012 -0800
|
|    PCI: Retry on IORESOURCE_IO type allocations

even the root bus does not have io port range, it will keep retrying
to realloc with mmio.

Current retry logic is : try with must+optional at first, and if
it fails with any ioport or mmio, it will try must then try to extend
must with optional.
That will fail as mmio-non-pref and mmio-pref for bridge will
be next to each other. So we have no chance to extend mmio-non-pref.

We can check fail type and only fall back for io port only, that will
keep mmio type still have must+optional.

This will be become more often when we have x86 8 sockets or 32 sockets
system, and those system will have one root bus per socket.
They will have some root buses do not have ioport range.

-v2: need to remove assigned entries from optional list too.

Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Tested-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>

---
 drivers/pci/setup-bus.c |   23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Bjorn Helgaas - May 24, 2013, 5:25 p.m.
On Thu, May 23, 2013 at 11:11 AM, Yinghai Lu <yinghai@kernel.org> wrote:
> BenH reported that there is some assign unassigned resource problem
> in powerpc.
>
> It turns out after
> | commit 0c5be0cb0edfe3b5c4b62eac68aa2aa15ec681af
> | Date:   Thu Feb 23 19:23:29 2012 -0800
> |
> |    PCI: Retry on IORESOURCE_IO type allocations
>
> even the root bus does not have io port range, it will keep retrying
> to realloc with mmio.
>
> Current retry logic is : try with must+optional at first, and if
> it fails with any ioport or mmio, it will try must then try to extend
> must with optional.
> That will fail as mmio-non-pref and mmio-pref for bridge will
> be next to each other. So we have no chance to extend mmio-non-pref.
>
> We can check fail type and only fall back for io port only, that will
> keep mmio type still have must+optional.
>
> This will be become more often when we have x86 8 sockets or 32 sockets
> system, and those system will have one root bus per socket.
> They will have some root buses do not have ioport range.
>
> -v2: need to remove assigned entries from optional list too.
>
> Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> Tested-by: Gavin Shan <shangw@linux.vnet.ibm.com>
> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
>
> ---
>  drivers/pci/setup-bus.c |   23 +++++++++++++++++++++++
>  1 file changed, 23 insertions(+)
>
> Index: linux-2.6/drivers/pci/setup-bus.c
> ===================================================================
> --- linux-2.6.orig/drivers/pci/setup-bus.c
> +++ linux-2.6/drivers/pci/setup-bus.c
> @@ -317,6 +317,10 @@ static void __assign_resources_sorted(st
>         LIST_HEAD(local_fail_head);
>         struct pci_dev_resource *save_res;
>         struct pci_dev_resource *dev_res;
> +       unsigned long fail_type = 0;
> +       struct pci_dev_resource *fail_res;
> +       unsigned long type_mask = IORESOURCE_IO | IORESOURCE_MEM |
> +                                 IORESOURCE_PREFETCH;
>
>         /* Check if optional add_size is there */
>         if (!realloc_head || list_empty(realloc_head))
> @@ -348,6 +352,25 @@ static void __assign_resources_sorted(st
>                 return;
>         }
>
> +       /* check failed type */
> +       list_for_each_entry(fail_res, &local_fail_head, list)
> +               fail_type |= fail_res->flags & type_mask;
> +       /* only io port fails */
> +       if ((fail_type & type_mask) == IORESOURCE_IO) {
> +               struct pci_dev_resource *tmp_res;
> +
> +               /* remove assigned non ioport from head list etc */
> +               list_for_each_entry_safe(dev_res, tmp_res, head, list)
> +                       if (dev_res->res->parent &&
> +                           !(dev_res->res->flags & IORESOURCE_IO)) {
> +                               /* remove it from realloc_head list */
> +                               remove_from_list(realloc_head, dev_res->res);
> +                               remove_from_list(&save_head, dev_res->res);
> +                               list_del(&dev_res->list);
> +                               kfree(dev_res);
> +                       }
> +       }

The problem we're trying to solve is that when allocation for type X
fails, we retry allocation for type Y.

This patch handles IO specially.  I think it basically says, "if we
only have IO allocation failures, don't retry MEM allocation."  But a
clean strategy would also avoid retrying IO allocation if we only had
MEM allocation failures.

Bjorn

>         free_list(&local_fail_head);
>         /* Release assigned resource */
>         list_for_each_entry(dev_res, head, list)
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Yinghai Lu - May 24, 2013, 11:34 p.m.
On Fri, May 24, 2013 at 10:25 AM, Bjorn Helgaas <bhelgaas@google.com> wrote:
>
> The problem we're trying to solve is that when allocation for type X
> fails, we retry allocation for type Y.
>
> This patch handles IO specially.  I think it basically says, "if we
> only have IO allocation failures, don't retry MEM allocation."  But a
> clean strategy would also avoid retrying IO allocation if we only had
> MEM allocation failures.

Well, that will make it little bit complicated as v3 that is sent in
another mail.

Need to separate ioport, mmio, mmio pref three types.

Yinghai
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

Index: linux-2.6/drivers/pci/setup-bus.c
===================================================================
--- linux-2.6.orig/drivers/pci/setup-bus.c
+++ linux-2.6/drivers/pci/setup-bus.c
@@ -317,6 +317,10 @@  static void __assign_resources_sorted(st
 	LIST_HEAD(local_fail_head);
 	struct pci_dev_resource *save_res;
 	struct pci_dev_resource *dev_res;
+	unsigned long fail_type = 0;
+	struct pci_dev_resource *fail_res;
+	unsigned long type_mask = IORESOURCE_IO | IORESOURCE_MEM |
+				  IORESOURCE_PREFETCH;
 
 	/* Check if optional add_size is there */
 	if (!realloc_head || list_empty(realloc_head))
@@ -348,6 +352,25 @@  static void __assign_resources_sorted(st
 		return;
 	}
 
+	/* check failed type */
+	list_for_each_entry(fail_res, &local_fail_head, list)
+		fail_type |= fail_res->flags & type_mask;
+	/* only io port fails */
+	if ((fail_type & type_mask) == IORESOURCE_IO) {
+		struct pci_dev_resource *tmp_res;
+
+		/* remove assigned non ioport from head list etc */
+		list_for_each_entry_safe(dev_res, tmp_res, head, list)
+			if (dev_res->res->parent &&
+			    !(dev_res->res->flags & IORESOURCE_IO)) {
+				/* remove it from realloc_head list */
+				remove_from_list(realloc_head, dev_res->res);
+				remove_from_list(&save_head, dev_res->res);
+				list_del(&dev_res->list);
+				kfree(dev_res);
+			}
+	}
+
 	free_list(&local_fail_head);
 	/* Release assigned resource */
 	list_for_each_entry(dev_res, head, list)