PCI: revert to allocating bottom-up, avoid E820 areas
Get series with b4
b4 am [email protected]Versions newest first
- v1 Latest9 patches 10 repliesIn review
Patches in v1 No applies recorded yet
Reviews trailers per patch
None of the 9 patches has a Reviewed-by or Acked-by of their own; the series has one for all of them.
| Patch | H. Peter Anvin |
|---|---|
| Series | A-b |
| 1/9 | |
| 2/9 | |
| 3/9 | |
| 4/9 | |
| 5/9 | |
| 6/9 | |
| 7/9 | |
| 8/9 | |
| 9/9 |
Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.
Cover letter [PATCH 0/9] PCI: revert to allocating bottom-up, avoid E820 areas
[This is 2.6.37 material]
In 2.6.34 and later, we automatically use "pci=use_crs" on recent (2008 or newer) machines, which means we assign PCI resources from the host bridge windows described by ACPI _CRS.
In many cases, _CRS claims the windows are bigger than they really are, e.g., they may overlap system memory, ROM, or other ACPI devices.
Linux allocates PCI memory bottom-up, ignores most ACPI devices, and does an incomplete job of avoiding E820 reserved areas, which caused problems like
https://bugzilla.kernel.org/show_bug.cgi?id=16228 .
As far as I can tell, Windows deals with this situation by removing all ACPI resources and allocating the remaining space top-down.
I tried to make Linux allocate top-down as well, but since we don't have a good mechanism for avoiding the ACPI devices that clutter the top of memory, we tripped over several of those, which resulted in these regressions:
https://bugzilla.kernel.org/show_bug.cgi?id=23132 (Dell 1545) https://bugzilla.kernel.org/show_bug.cgi?id=23332 (HP nx6325) https://bugzilla.kernel.org/show_bug.cgi?id=23542 (HP 2530p) https://bugzilla.kernel.org/show_bug.cgi?id=23802 (PCI / TPM conflict)
So I think the best band-aid for now is to revert to bottom-up allocation and do a better job of avoiding E820 reserved areas. Obviously, it's not complete because we could still trip over ACPI devices (as in 23802), but that's a problem we've always had and it's not as likely if we go bottom-up.
Patches 1-5 of this series are straight reverts ("git revert -n") of these patches which were merged in 2.6.37-rc1:
82e3e76 PCI: fix pci_bus_alloc_resource() hang, prefer positive decode 1af3c2e x86: allocate space within a region top-down
dc9887d x86/PCI: allocate space from the end of a region, not the beginning b126b47 PCI: allocate bus resources from the top down
e7f8567 resources: support allocating space within a region from the top down
Patches 6-9 add the ability to avoid E820 regions and the top 2MB below 4GB. The last patch (avoid top 2MB) is not strictly necessary in that it doesn't fix any current issues I'm aware of.
Bjorn
---
Bjorn Helgaas (9):
Revert "PCI: fix pci_bus_alloc_resource() hang, prefer positive decode" Revert "x86: allocate space within a region top-down"
Revert "x86/PCI: allocate space from the end of a region, not the beginning" Revert "PCI: allocate bus resources from the top down"
Revert "resources: support allocating space within a region from the top down"
resources: add arch hook for preventing allocation in reserved areas
x86: avoid low BIOS area when allocating address space
x86: avoid E820 regions when allocating address space
x86: avoid high BIOS area when allocating address space
Documentation/kernel-parameters.txt | 5 -- arch/x86/include/asm/e820.h | 3 + arch/x86/kernel/Makefile | 1 arch/x86/kernel/resource.c | 48 ++++++++++++++++ arch/x86/kernel/setup.c | 1 arch/x86/pci/i386.c | 18 ++---- drivers/pci/bus.c | 81 ++------------------------- include/linux/ioport.h | 2 - kernel/resource.c | 104 +++-------------------------------- 9 files changed, 73 insertions(+), 190 deletions(-) create mode 100644 arch/x86/kernel/resource.c