| Message ID | 20090527194144.GA7933@us.ibm.com (mailing list archive) |
|---|---|
| State | RFC, archived |
| Headers | show |
On Wed, 27 May 2009 12:41:44 -0700 Gary Hade <garyhade@us.ibm.com> wrote: > On Thu, May 21, 2009 at 09:37:37AM -0700, Gary Hade wrote: > > On Thu, May 21, 2009 at 08:46:17AM -0600, Bjorn Helgaas wrote: > > > On Wednesday 20 May 2009 5:49:10 pm Jesse Barnes wrote: > > > > On Fri, 8 May 2009 16:40:10 -0600 > > > > Bjorn Helgaas <bjorn.helgaas@hp.com> wrote: > > > > > Did anything happen with this? > > > > > > > > > > The longer we wait to make "use_crs" the default, the harder > > > > > it will be, so I'd like to push ahead. > > > > > > > > Here's a patch to make CRS the default. If it looks ok I can > > > > push it into my linux-next branch. I'm all for using reliable > > > > data from the BIOS. I guess we'll find out fairly quickly if > > > > this stuff isn't... > > > > > > Thanks for following up on this, Jesse. It was on my to-do list > > > for yesterday, but I didn't get to it. > > > > > > Yinghai mentioned a specific box where we might have trouble, but > > > we never got enough details to really debug it. So I think we > > > might as well give it a shot and fine-tune it as we need to. > > > > I just remembered that Andrew Morton once raised some concerns > > related to the number of _CRS returned resources exceeding the > > fixed resource array size -> http://lkml.org/lkml/2007/11/1/49 > > > > I believe PCI_BUS_NUM_RESOURCES was later increased but if > > that increase was insufficient to cover all systems that the > > code will now be exposed to, it seems like the "And should we > > really be silently ignoring this problem? Should we at least > > report it?" comment that was not addressed could become relevent. > > Not only do we not warn, if we did warn based on the current > 'if (info->res_num >= PCI_BUS_NUM_RESOURCES)' > resource array overrun avoidance criteria it would be incorrect > for a root bus that is connected to a subordinate bus with a > transparent bridge. We need to avoid the last 3 slots of the > resource array since they do not get mapped to a transparent > bridge connected subordinate bus. I believe the below patch > addresses these issues. > > With PCI_BUS_NUM_RESOURCES currently at 16 we still have space > to accomodate 13 _CRS returned resource descriptors which will > hopefully be sufficient. Ok, applied the patch (with Yinghai's suggested name change) to my linux-next branch. We'll see now it goes and revert it if there's trouble.
On Wed, 27 May 2009 12:41:44 -0700 Gary Hade <garyhade@us.ibm.com> wrote: > On Thu, May 21, 2009 at 09:37:37AM -0700, Gary Hade wrote: > > On Thu, May 21, 2009 at 08:46:17AM -0600, Bjorn Helgaas wrote: > > > On Wednesday 20 May 2009 5:49:10 pm Jesse Barnes wrote: > > > > On Fri, 8 May 2009 16:40:10 -0600 > > > > Bjorn Helgaas <bjorn.helgaas@hp.com> wrote: > > > > > Did anything happen with this? > > > > > > > > > > The longer we wait to make "use_crs" the default, the harder > > > > > it will be, so I'd like to push ahead. > > > > > > > > Here's a patch to make CRS the default. If it looks ok I can > > > > push it into my linux-next branch. I'm all for using reliable > > > > data from the BIOS. I guess we'll find out fairly quickly if > > > > this stuff isn't... > > > > > > Thanks for following up on this, Jesse. It was on my to-do list > > > for yesterday, but I didn't get to it. > > > > > > Yinghai mentioned a specific box where we might have trouble, but > > > we never got enough details to really debug it. So I think we > > > might as well give it a shot and fine-tune it as we need to. > > > > I just remembered that Andrew Morton once raised some concerns > > related to the number of _CRS returned resources exceeding the > > fixed resource array size -> http://lkml.org/lkml/2007/11/1/49 > > > > I believe PCI_BUS_NUM_RESOURCES was later increased but if > > that increase was insufficient to cover all systems that the > > code will now be exposed to, it seems like the "And should we > > really be silently ignoring this problem? Should we at least > > report it?" comment that was not addressed could become relevent. > > Not only do we not warn, if we did warn based on the current > 'if (info->res_num >= PCI_BUS_NUM_RESOURCES)' > resource array overrun avoidance criteria it would be incorrect > for a root bus that is connected to a subordinate bus with a > transparent bridge. We need to avoid the last 3 slots of the > resource array since they do not get mapped to a transparent > bridge connected subordinate bus. I believe the below patch > addresses these issues. > > With PCI_BUS_NUM_RESOURCES currently at 16 we still have space > to accomodate 13 _CRS returned resource descriptors which will > hopefully be sufficient. Ok, just applied this one to my linux-next tree, since we'll want it now that _CRS is used by default. Len, is this ok with you? It touches x86 ACPI files rather than PCI files, but it's pretty tightly related... Thanks,
Index: linux-2.6.28-rc7/arch/x86/pci/acpi.c =================================================================== --- linux-2.6.28-rc7.orig/arch/x86/pci/acpi.c 2009-05-27 10:42:10.000000000 -0700 +++ linux-2.6.28-rc7/arch/x86/pci/acpi.c 2009-05-27 10:43:35.000000000 -0700 @@ -38,15 +38,26 @@ count_resource(struct acpi_resource *acp struct acpi_resource_address64 addr; acpi_status status; - if (info->res_num >= PCI_BUS_NUM_RESOURCES) - return AE_OK; - status = resource_to_addr(acpi_res, &addr); if (ACPI_SUCCESS(status)) info->res_num++; return AE_OK; } +static int +bus_has_transparent_bridge(struct pci_bus *bus) +{ + struct pci_dev *dev; + + list_for_each_entry(dev, &bus->devices, bus_list) { + u16 class = dev->class >> 8; + + if (class == PCI_CLASS_BRIDGE_PCI && dev->transparent) + return true; + } + return false; +} + static acpi_status setup_resource(struct acpi_resource *acpi_res, void *data) { @@ -56,9 +67,7 @@ setup_resource(struct acpi_resource *acp acpi_status status; unsigned long flags; struct resource *root; - - if (info->res_num >= PCI_BUS_NUM_RESOURCES) - return AE_OK; + int max_root_bus_resources = PCI_BUS_NUM_RESOURCES; status = resource_to_addr(acpi_res, &addr); if (!ACPI_SUCCESS(status)) @@ -82,6 +91,18 @@ setup_resource(struct acpi_resource *acp res->end = res->start + addr.address_length - 1; res->child = NULL; + if (bus_has_transparent_bridge(info->bus)) + max_root_bus_resources -= 3; + if (info->res_num >= max_root_bus_resources) { + printk(KERN_WARNING "PCI: Failed to allocate 0x%lx-0x%lx " + "from %s for %s due to _CRS returning more than " + "%d resource descriptors\n", (unsigned long) res->start, + (unsigned long) res->end, root->name, info->name, + max_root_bus_resources); + info->res_num++; + return AE_OK; + } + if (insert_resource(root, res)) { printk(KERN_ERR "PCI: Failed to allocate 0x%lx-0x%lx " "from %s for %s\n", (unsigned long) res->start,