Yinghai Lu [Fri, 11 Apr 2008 22:14:52 +0000 (15:14 -0700)]
x86_64: don't need set default res if only have one root bus
if there's only one root bus there's no need to split resources.
This patch fixes the issue described at:
http://lkml.org/lkml/2008/4/10/304
Reported-and-bisected-by: Rafael J. Wysocki <rjw@sisk.pl> Tested-by: Rafael J. Wysocki <rjw@sisk.pl> Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Yinghai Lu [Tue, 19 Feb 2008 11:21:20 +0000 (03:21 -0800)]
x86: multi pci root bus with different io resource range, on 64-bit
scan AMD opteron io/mmio routing to make sure every pci root bus get correct
resource range. Thus later pci scan could assign correct resource to device
with unassigned resource.
this can fix a system without _CRS for multi pci root bus.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Yinghai Lu [Tue, 19 Feb 2008 11:20:09 +0000 (03:20 -0800)]
x86: get mp_bus_to_node early
Currently, on an amd k8 system with multi ht chains, the numa_node of
pci devices under /sys/devices/pci0000:80/* is always 0, even if that
chain is on node 1 or 2 or 3.
Workaround: pcibus_to_node(bus) is used when we want to get the node that
pci_device is on.
In struct device, we already have numa_node member, and we could use
dev_to_node()/set_dev_node() to get and set numa_node in the device.
set_dev_node is called in pci_device_add() with pcibus_to_node(bus),
and pcibus_to_node uses bus->sysdata for nodeid.
The problem is when pci_add_device is called, bus->sysdata is not assigned
correct nodeid yet. The result is that numa_node will always be 0.
pcibios_scan_root and pci_scan_root could take sysdata. So we need to get
mp_bus_to_node mapping before these two are called, and thus
get_mp_bus_to_node could get correct node for sysdata in root bus.
In scanning of the root bus, all child busses will take parent bus sysdata.
So all pci_device->dev.numa_node will be assigned correctly and automatically.
Later we could use dev_to_node(&pci_dev->dev) to get numa_node, and we
could also could make other bus specific device get the correct numa_node
too.
This is an updated version of pci_sysdata and Jeff's pci_domain patch.
[ mingo@elte.hu: build fix ]
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Yinghai Lu [Tue, 19 Feb 2008 11:20:41 +0000 (03:20 -0800)]
driver core: try parent numa_node at first before using default
in the device_add, we try to use use parent numa_node.
need to make sure pci root bus's bridge device numa_node is set.
then we could use device->numa_node direclty for all device.
and don't need to call pcibus_to_node().
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Yinghai Lu [Tue, 26 Feb 2008 02:41:35 +0000 (18:41 -0800)]
x86: if acpi=off, force setting the mmconf for fam10h
some BIOS only let AMD fam 10h handle bus0, and nvidia mcp55/ck804
to handle other buses. at that case MCFG will cover all over them.
but with acpi=off, we can not use MCFG. this patch will double check
the busnbits, and if it is less handling 256 bues, and acpi=off
will forcely reset the mmconf in msr, so we still use mmconf in above case.
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Yinghai Lu [Tue, 19 Feb 2008 11:13:43 +0000 (03:13 -0800)]
x86_64: check and enable MMCONFIG for AMD Family 10h
So we can use MMCONF when MMCONF is not set by BIOS
using TOP_MEM2 msr to get memory top, and try to scan fam10h mmio routing to
make sure the range is not conflicted with some prefetch MMIO that is above 4G.
(current only LinuxBIOS assign 64 bit mmio above 4G for some co-processor)
Signed-off-by: Yinghai Lu <yinghai.lu@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Robert Hancock [Fri, 15 Feb 2008 09:27:20 +0000 (01:27 -0800)]
x86: validate against acpi motherboard resources
This path adds validation of the MMCONFIG table against the ACPI reserved
motherboard resources. If the MMCONFIG table is found to be reserved in
ACPI, we don't bother checking the E820 table. The PCI Express firmware
spec apparently tells BIOS developers that reservation in ACPI is required
and E820 reservation is optional, so checking against ACPI first makes
sense. Many BIOSes don't reserve the MMCONFIG region in E820 even though
it is perfectly functional, the existing check needlessly disables MMCONFIG
in these cases.
In order to do this, MMCONFIG setup has been split into two phases. If PCI
configuration type 1 is not available then MMCONFIG is enabled early as
before. Otherwise, it is enabled later after the ACPI interpreter is
enabled, since we need to be able to execute control methods in order to
check the ACPI reserved resources. Presently this is just triggered off
the end of ACPI interpreter initialization.
There are a few other behavioral changes here:
- Validate all MMCONFIG configurations provided, not just the first one.
- Validate the entire required length of each configuration according to
the provided ending bus number is reserved, not just the minimum required
allocation.
- Validate that the area is reserved even if we read it from the chipset
directly and not from the MCFG table. This catches the case where the
BIOS didn't set the location properly in the chipset and has mapped it
over other things it shouldn't have.
This also cleans up the MMCONFIG initialization functions so that they
simply do nothing if MMCONFIG is not compiled in.
Based on an original patch by Rajesh Shah from Intel.
[akpm@linux-foundation.org: many fixes and cleanups] Signed-off-by: Robert Hancock <hancockr@shaw.ca> Signed-off-by: Andi Kleen <ak@suse.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Greg KH <greg@kroah.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Andi Kleen <ak@suse.de> Cc: Rajesh Shah <rajesh.shah@intel.com> Cc: Jesse Barnes <jbarnes@virtuousgeek.org> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andi Kleen <ak@suse.de> Cc: Greg KH <greg@kroah.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-x86-bigbox-bootmem-v3
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-x86-bigbox-bootmem-v3:
x86_64/mm: check and print vmemmap allocation continuous
x86_64: fix setup_node_bootmem to support big mem excluding with memmap
x86_64: make reserve_bootmem_generic() use new reserve_bootmem()
mm: allow reserve_bootmem() cross nodes
mm: offset align in alloc_bootmem()
mm: fix alloc_bootmem_core to use fast searching for all nodes
mm: make mem_map allocation continuous
Yinghai Lu [Tue, 18 Mar 2008 19:52:37 +0000 (12:52 -0700)]
x86_64: fix setup_node_bootmem to support big mem excluding with memmap
typical case: four sockets system, every node has 4g ram, and we are using:
memmap=10g$4g
to mask out memory on node1 and node2
when numa is enabled, early_node_mem is used to get node_data and node_bootmap.
if it can not get memory from the same node with find_e820_area(), it will
use alloc_bootmem to get buff from previous nodes.
so check it and print out some info about it.
need to move early_res_to_bootmem into every setup_node_bootmem.
and it takes range that node has. otherwise alloc_bootmem could return addr
that reserved early.
depends on "mm: make reserve_bootmem can crossed the nodes".
Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Yinghai Lu [Tue, 18 Mar 2008 19:50:21 +0000 (12:50 -0700)]
x86_64: make reserve_bootmem_generic() use new reserve_bootmem()
"mm: make reserve_bootmem can crossed the nodes" provides new
reserve_bootmem(), let reserve_bootmem_generic() use that.
reserve_bootmem_generic() is used to reserve initramdisk, so this way
we can make sure even when bootloader or kexec load ranges cross the
node memory boundaries, reserve_bootmem still works.
Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-generic-bitops-v3
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-generic-bitops-v3:
x86, bitops: select the generic bitmap search functions
x86: include/asm-x86/pgalloc.h/bitops.h: checkpatch cleanups - formatting only
x86: finalize bitops unification
x86, UML: remove x86-specific implementations of find_first_bit
x86: optimize find_first_bit for small bitmaps
x86: switch 64-bit to generic find_first_bit
x86: generic versions of find_first_(zero_)bit, convert i386
bitops: use __fls for fls64 on 64-bit archs
generic: implement __fls on all 64-bit archs
generic: introduce a generic __fls implementation
x86: merge the simple bitops and move them to bitops.h
x86, generic: optimize find_next_(zero_)bit for small constant-size bitmaps
x86, uml: fix uml with generic find_next_bit for x86
x86: change x86 to use generic find_next_bit
uml: Kconfig cleanup
uml: fix build error
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-x86-bigbox-bootparam
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-x86-bigbox-bootparam:
x86, boot: Document for linked list of struct setup_data
x86, boot: export linked list of struct setup_data via debugfs
x86, boot: add linked list of struct setup_data
x86, boot: add free_early to early reservation machanism
* Export ide_dma_exec_cmd() and __ide_dma_test_irq().
* Constify struct ide_dma_ops.
* Always set hwif->dma_ops to &sff_dma_ops in ide_setup_dma()
(it is later overriden by ide_init_port() if needed) and drop
'const struct ide_port_info *d' argument.
While at it:
* Rename __ide_dma_test_irq() to ide_dma_test_irq().
Acked-by: Sergei Shtylyov <sshtylyov@ru.mvista.com> Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
ide: do complete DMA setup in ->init_dma method (take 2)
* Make ide_hwif_setup_dma() return an error value.
* Pass 'const struct ide_port_info *d' instead of 'unsigned long dmabase'
to ->init_dma method and make it return an error value.
* Rename ide_get_or_set_dma_base() to ide_pci_dma_base(),
change ordering of its arguments and then export it.
* Export ide_pci_set_master().
* Do complete DMA setup inside ->init_dma method and update ->init_dma
users accordingly.
* Sanitize code for DMA setup in ide_init_port().
v2:
* Fix for CONFIG_BLK_DEV_IDEDMA_PCI=n configs
(from Jiri Slaby <jirislaby@gmail.com>):
Fix following compiler warning by returning EINVAL:
In file included from ANYTHING-INCLUDING-IDE.H:45:
include/linux/ide.h: In function ‘ide_hwif_setup_dma’:
include/linux/ide.h:1022: warning: no return statement in function returning non-void
Cc: Jiri Slaby <jirislaby@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Acked-by: Sergei Shtylyov <sshtylyov@ru.mvista.com> Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Always use "fast" MWDMA support and remove dma_{black,white}_list
(they were based on completely bogus ->ide_dma_check implementation
which didn't set neither the host controller timings nor the device
for the desired transfer mode).
alim15x3: skip DMA initialization completely on revs < 0x20
Skip DMA initialization completely on revs < 0x20 by setting IDE_HFLAG_NO_DMA
host flag and resetting DMA host masks in alim15x3_init_one() (currently
ide_hwif_setup_dma() will try to obtain DMA base and setup PCI bus-mastering
but init_dma_ali15x3() will fail).
It is no longer needed to set hwif->mmio flag to tell IDE layer to not
manage resources so cleanup host drivers that used hwif->mmio flag only
for this purpose.
ide: manage resources for PCI devices in ide_pci_enable() (take 3)
* Reserve PCI BARs 0-3 (0-1 for single port controllers) in
ide_pci_enable() and remove ide_hwif_request_regions() call
from ide_device_add_all() (also cleanup resource management
in scc_pata host driver).
* Fix handling of PCI BAR 4 in ide_pci_enable(), then cleanup
ide_iomio_dma() (+ init_hwif_trm290() in trm290 host driver)
and remove ide_release[_iomio]_dma().
v2:
* Fixup trm290 host driver.
v3:
* Because of scc_pata host driver changes we need to call
pci_request_selected_regions() also in setup_mmio_scc().
Some of them are:
WARNING: braces {} are not necessary for single statement blocks
CHECK: multiple assignments should be avoided
WARNING: printk() should include KERN_ facility level
WARNING: no space between function name and open parenthesis '('
Huang, Ying [Fri, 28 Mar 2008 02:49:44 +0000 (10:49 +0800)]
x86, boot: add linked list of struct setup_data
This patch adds a field of 64-bit physical pointer to NULL terminated
single linked list of struct setup_data to real-mode kernel
header. This is used as a more extensible boot parameters passing
mechanism.
disable /dev/mem mmap of RAM with PAT. It makes things safer and
eliminates aliasing. A future improvement would be to avoid the
range_is_allowed duplication.
x86, bitops: select the generic bitmap search functions
Introduce GENERIC_FIND_FIRST_BIT and GENERIC_FIND_NEXT_BIT in
lib/Kconfig, defaulting to off. An arch that wants to use the
generic implementation now only has to use a select statement
to include them.
I added an always-y option (X86_CPU) to arch/x86/Kconfig.cpu
and used that to select the generic search functions. This
way ARCH=um SUBARCH=i386 automatically picks up the change
too, and arch/um/Kconfig.i386 can therefore be simplified a
bit. ARCH=um SUBARCH=x86_64 does things differently, but
still compiles fine. It seems that a "def_bool y" always
wins over a "def_bool n"?
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm> Signed-off-by: Ingo Molnar <mingo@elte.hu>
include/asm-x86/bitops_32.h and include/asm-x86/bitops_64.h are now
almost identical. The 64-bit version sets ARCH_HAS_FAST_MULTIPLIER
and has an extra inline function set_bit_string. The define currently
has no influence on the generated code, but it can be argued that
setting it on i386 is the right thing to do anyhow. The addition
of the extra inline function on i386 does not hurt either.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm> Signed-off-by: Ingo Molnar <mingo@elte.hu>
x86, UML: remove x86-specific implementations of find_first_bit
x86 has been switched to the generic versions of find_first_bit
and find_first_zero_bit, but the original versions were retained.
This patch just removes the now unused x86-specific versions.
also update UML.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Avoid a call to find_first_bit if the bitmap size is know at
compile time and small enough to fit in a single long integer.
Modeled after an optimization in the original x86_64-specific
code.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm> Signed-off-by: Ingo Molnar <mingo@elte.hu>
x86: generic versions of find_first_(zero_)bit, convert i386
Generic versions of __find_first_bit and __find_first_zero_bit
are introduced as simplified versions of __find_next_bit and
__find_next_zero_bit. Their compilation and use are guarded by
a new config variable GENERIC_FIND_FIRST_BIT.
The generic versions of find_first_bit and find_first_zero_bit
are implemented in terms of the newly introduced __find_first_bit
and __find_first_zero_bit.
This patch does not remove the i386-specific implementation,
but it does switch i386 to use the generic functions by setting
GENERIC_FIND_FIRST_BIT=y for X86_32.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm> Signed-off-by: Ingo Molnar <mingo@elte.hu>
x86: merge the simple bitops and move them to bitops.h
Some of those can be written in such a way that the same
inline assembly can be used to generate both 32 bit and
64 bit code.
For ffs and fls, x86_64 unconditionally used the cmov
instruction and i386 unconditionally used a conditional
branch over a mov instruction. In the current patch I
chose to select the version based on the availability
of the cmov instruction instead. A small detail here is
that x86_64 did not previously set CONFIG_X86_CMOV=y.
Improved comments for ffs, ffz, fls and variations.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm> Signed-off-by: Ingo Molnar <mingo@elte.hu>
x86, generic: optimize find_next_(zero_)bit for small constant-size bitmaps
This moves an optimization for searching constant-sized small
bitmaps form x86_64-specific to generic code.
On an i386 defconfig (the x86#testing one), the size of vmlinux hardly
changes with this applied. I have observed only four places where this
optimization avoids a call into find_next_bit:
In the functions return_unused_surplus_pages, alloc_fresh_huge_page,
and adjust_pool_surplus, this patch avoids a call for a 1-bit bitmap.
In __next_cpu a call is avoided for a 32-bit bitmap. That's it.
On x86_64, 52 locations are optimized with a minimal increase in
code size:
Current #testing defconfig:
146 x bsf, 27 x find_next_*bit
text data bss dec hex filename 5392637 846592 724424 6963653 6a41c5 vmlinux
After removing the x86_64 specific optimization for find_next_*bit:
94 x bsf, 79 x find_next_*bit
text data bss dec hex filename 5392358 846592 724424 6963374 6a40ae vmlinux
After this patch (making the optimization generic):
146 x bsf, 27 x find_next_*bit
text data bss dec hex filename 5392396 846592 724424 6963412 6a40d4 vmlinux
The versions with inline assembly are in fact slower on the machines I
tested them on (in userspace) (Athlon XP 2800+, p4-like Xeon 2.8GHz, AMD
Opteron 270). The i386-version needed a fix similar to 06024f21 to avoid
crashing the benchmark.
Benchmark using: gcc -fomit-frame-pointer -Os. For each bitmap size
1...512, for each possible bitmap with one bit set, for each possible
offset: find the position of the first bit starting at offset. If you
follow ;). Times include setup of the bitmap and checking of the
results.
If the bitmap size is not a multiple of BITS_PER_LONG, and no set
(cleared) bit is found, find_next_bit (find_next_zero_bit) returns a
value outside of the range [0, size]. The generic version always returns
exactly size. The generic version also uses unsigned long everywhere,
while the x86 versions use a mishmash of int, unsigned (int), long and
unsigned long.
Using the generic version does give a slightly bigger kernel, though.
projects / linux-2.6 / log