[PATCH] Introduce mechanism for registering active regions of memory
At a basic level, architectures define structures to record where active
ranges of page frames are located. Once located, the code to calculate zone
sizes and holes in each architecture is very similar. Some of this zone and
hole sizing code is difficult to read for no good reason. This set of patches
eliminates the similar-looking architecture-specific code.
The patches introduce a mechanism where architectures register where the
active ranges of page frames are with add_active_range(). When all areas have
been discovered, free_area_init_nodes() is called to initialise the pgdat and
zones. The zone sizes and holes are then calculated in an architecture
independent manner.
Patch 1 introduces the mechanism for registering and initialising PFN ranges
Patch 2 changes ppc to use the mechanism - 139 arch-specific LOC removed
Patch 3 changes x86 to use the mechanism - 136 arch-specific LOC removed
Patch 4 changes x86_64 to use the mechanism - 74 arch-specific LOC removed
Patch 5 changes ia64 to use the mechanism - 52 arch-specific LOC removed
Patch 6 accounts for mem_map as a memory hole as the pages are not reclaimable.
It adjusts the watermarks slightly
Tony Luck has successfully tested for ia64 on Itanium with tiger_defconfig,
gensparse_defconfig and defconfig. Bob Picco has also tested and debugged on
IA64. Jack Steiner successfully boot tested on a mammoth SGI IA64-based
machine. These were on patches against 2.6.17-rc1 and release 3 of these
patches but there have been no ia64-changes since release 3.
There are differences in the zone sizes for x86_64 as the arch-specific code
for x86_64 accounts the kernel image and the starting mem_maps as memory holes
but the architecture-independent code accounts the memory as present.
The big benefit of this set of patches is a sizable reduction of
architecture-specific code, some of which is very hairy. There should be a
greater reduction when other architectures use the same mechanisms for zone
and hole sizing but I lack the hardware to test on.
Additional credit;
Dave Hansen for the initial suggestion and comments on early patches
Andy Whitcroft for reviewing early versions and catching numerous
errors
Tony Luck for testing and debugging on IA64
Bob Picco for fixing bugs related to pfn registration, reviewing a
number of patch revisions, providing a number of suggestions
on future direction and testing heavily
Jack Steiner and Robin Holt for testing on IA64 and clarifying
issues related to memory holes
Yasunori for testing on IA64
Andi Kleen for reviewing and feeding back about x86_64
Christian Kujau for providing valuable information related to ACPI
problems on x86_64 and testing potential fixes
This patch:
Define the structure to represent an active range of page frames within a node
in an architecture independent manner. Architectures are expected to register
active ranges of PFNs using add_active_range(nid, start_pfn, end_pfn) and call
free_area_init_nodes() passing the PFNs of the end of each zone.
Signed-off-by: Mel Gorman <mel@csn.ul.ie> Signed-off-by: Bob Picco <bob.picco@hp.com> Cc: Dave Hansen <haveblue@us.ibm.com> Cc: Andy Whitcroft <apw@shadowen.org> Cc: Andi Kleen <ak@muc.de> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Cc: "Keith Mannthey" <kmannth@gmail.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Yasunori Goto <y-goto@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
un-, de-, -free, -destroy, -exit, etc functions should in general return
void. Also,
There is very little, say, filesystem driver code can do upon failed
kmem_cache_destroy(). If it will be decided to BUG in this case, BUG
should be put in generic code, instead.
[PATCH] Really ignore kmem_cache_destroy return value
* Rougly half of callers already do it by not checking return value
* Code in drivers/acpi/osl.c does the following to be sure:
(void)kmem_cache_destroy(cache);
* Those who check it printk something, however, slab_error already printed
the name of failed cache.
* XFS BUGs on failed kmem_cache_destroy which is not the decision
low-level filesystem driver should make. Converted to ignore.
SWsoft Virtuozzo/OpenVZ Linux kernel team has discovered that ext3 error
behavior was broken in linux kernels since 2.5.x versions by the following
patch:
2002/10/31 02:15:26-05:00 tytso@snap.thunk.org
Default mount options from superblock for ext2/3 filesystems
http://linux.bkbits.net:8080/linux-2.6/gnupatch@3dc0d88eKbV9ivV4ptRNM8fBuA3JBQ
In case ext3 file system is mounted with errors=continue
(EXT3_ERRORS_CONTINUE) errors should be ignored when possible. However at
present in case of any error kernel aborts journal and remounts filesystem
to read-only. Such behavior was hit number of times and noted to differ
from that of 2.4.x kernels.
This patch fixes this:
- do nothing in case of EXT3_ERRORS_CONTINUE,
- set EXT3_MOUNT_ABORT and call journal_abort() in all other cases
- panic() should be called after ext3_commit_super() to save
sb marked as EXT3_ERROR_FS
Signed-off-by: Vasily Averin <vvs@sw.ru> Acked-by: Kirill Korotaev <dev@sw.ru> Cc: Theodore Ts'o <tytso@mit.edu> Cc: "Stephen C. Tweedie" <sct@redhat.com> Cc: Mingming Cao <cmm@us.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Mingming Cao [Wed, 27 Sep 2006 08:49:32 +0000 (01:49 -0700)]
[PATCH] ext3: turn on reservation dump on block allocation errors
In the past there were a few kernel panics related to block reservation
tree operations failure (insert/remove etc). It would be very useful to
get the block allocation reservation map info when such error happens.
Signed-off-by: Mingming Cao <cmm@us.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Eric Sandeen [Wed, 27 Sep 2006 08:49:31 +0000 (01:49 -0700)]
[PATCH] JBD: 16T fixes
These are a few places I've found in jbd that look like they may not be
16T-safe, or consistent with the use of unsigned longs for block
containers. Problems here would be somewhat hard to hit, would require
journal blocks past the 8T boundary, which would not be terribly common.
Still, should fix.
(some of these have come from the ext4 work on jbd as well).
I think there's one more possibility that the wrap() function may not be
safe IF your last block in the journal butts right up against the 232 block
boundary, but that seems like a VERY remote possibility, and I'm not
worrying about it at this point.
Signed-off-by: Eric Sandeen <esandeen@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Eric Sandeen [Wed, 27 Sep 2006 08:49:30 +0000 (01:49 -0700)]
[PATCH] ext3: inode numbers are unsigned long
This is primarily format string fixes, with changes to ialloc.c where large
inode counts could overflow, and also pass around journal_inum as an
unsigned long, just to be pedantic about it....
Signed-off-by: Eric Sandeen <esandeen@redhat.com> Cc: Mingming Cao <cmm@us.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Eric Sandeen [Wed, 27 Sep 2006 08:49:29 +0000 (01:49 -0700)]
[PATCH] fix ext3 mounts at 16T
I need to do some actual IO testing now, but this gets things mounting for
a 16T ext3 filesystem. (patched up e2fsprogs is needed too, I'll send that
off the kernel list)
This patch fixes these issues in the kernel:
o sbi->s_groups_count overflows in ext3_fill_super()
at 16T, s_blocks_count is already maxed out; adding
EXT3_BLOCKS_PER_GROUP(sb) overflows it and groups_count comes out to 0.
Not really what we want, and causes a failed mount.
Feel free to check my math (actually, please do!), but changing it this
way should work & avoid the overflow:
(A + B - 1)/B changed to: ((A - 1)/B) + 1
o ext3_check_descriptors() overflows range checks
ext3_check_descriptors() iterates over all block groups making sure
that various bits are within the right block ranges... on the last pass
through, it is checking the error case
[item] >= block + EXT3_BLOCKS_PER_GROUP(sb)
where "block" is the first block in the last block group. The last
block in this group (and the last one that will fit in 32 bits) is block
+ EXT3_BLOCKS_PER_GROUP(sb)- 1. block + EXT3_BLOCKS_PER_GROUP(sb) wraps
back around to 0.
so, make things clearer with "first_block" and "last_block" where those
are first and last, inclusive, and use <, > rather than <, >=.
Finally, the last block group may be smaller than the rest, so account
for this on the last pass through: last_block = sb->s_blocks_count - 1;
(a similar patch could be done for ext2; does anyone in their right mind
use ext2 at 16T? I'll send an ext2 patch doing the same thing if that's
warranted)
Signed-off-by: Eric Sandeen <esandeen@redhat.com> Cc: Mingming Cao <cmm@us.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
jbd_sync_bh releases journal->j_list_lock. Add a lock annotation to this
function so that sparse can check callers for lock pairing, and so that
sparse will not complain about this function since it intentionally uses
the lock in this manner.
* master.kernel.org:/pub/scm/linux/kernel/git/gregkh/i2c-2.6: (30 commits)
i2c: Drop unimplemented slave functions
i2c: Constify i2c_algorithm declarations, part 2
i2c: Constify i2c_algorithm declarations, part 1
i2c: Let drivers constify i2c_algorithm data
i2c-isa: Restore driver owner
i2c-viapro: Add support for the VT8237A and VT8251
i2c: Warn on i2c client creation failure
i2c-core: Drop useless bitmaskings
i2c-algo-pcf: Discard the mdelay data struct member
i2c-algo-bit: Cleanups
i2c-isa: Fail adding driver on attach_adapter error
i2c: __must_check fixes (chip drivers)
i2c-dev: attach/detach_adapter cleanups
i2c-stub: Chip address as a module parameter
i2c: Plan i2c-isa for removal
i2c: New bus driver for TI OMAP boards
i2c-algo-bit: Discard the mdelay data struct member
i2c-matroxfb: Struct init conversion
i2c: Fix copy-n-paste in subsystem Kconfig
i2c-au1550: Add I2C support for Au1200
...
Franck Bui-Huu [Fri, 11 Aug 2006 15:51:53 +0000 (17:51 +0200)]
[MIPS] setup.c: use early_param() for early command line parsing
There's no point to rewrite some logic to parse command line
to pass initrd parameters or to declare a user memory area.
We could use instead parse_early_param() that does the same
thing.
Franck Bui-Huu [Fri, 18 Aug 2006 14:18:09 +0000 (16:18 +0200)]
[MIPS] get_wchan(): remove uses of mfinfo[64]
This array was used to 'cache' some frame info about scheduler
functions to speed up get_wchan(). This array was 1Ko size and
was only used when CONFIG_KALLSYMS was set but declared for all
configs.
Rather than make the array statement conditional, this patches
removes this array and its uses. Indeed the common case doesn't
seem to use this array and get_wchan() is not a critical path
anyways.
It results in a smaller bss and a smaller/cleaner code:
text data bss dec hex filename 2543808 254148 139296 2937252 2cd1a4 vmlinux-new-get-wchan 2544080 254148 143392 2941620 2ce2b4 vmlinux~old
Franck Bui-Huu [Fri, 18 Aug 2006 14:18:08 +0000 (16:18 +0200)]
[MIPS] get_frame_info(): null function size means size is unknown
This patch adds 2 sanity checks.
The first one test that the start address of the function to analyze has been
set by the caller. If not return an error since nothing usefull can be done
without.
The second one checks that the function's size has been set. A null size can
happen if CONFIG_KALLSYMS is not set and it means that we don't know the size
of the function to analyze. In this case, we make it equal to 128 instructions
by default.
MIPS is the only port to call its fstatat()-related syscalls
"__NR_fstatat". Now I can see why that might be seen as every
other port being wrong, but I think for o32, it is at best confusing.
__NR_fstat provides a plain (32-bit) stat while __NR_fstatat provides a
64-bit stat. Changing the name to __NR_fstatat64 would make things more
explicit, match x86, and make the glibc port slightly easier.
The current name is more appropriate for n32 and n64, but it would be
appropriate for other 64-bit targets too, and those targets have chosen
to call it __NR_newfstatat instead. Using the same name for MIPS would
again be more consistent and make the glibc port slightly easier.
I'm not wedded to this idea if the current names are preferred,
but FWIW...
Signed-off-by: Richard Sandiford <richard@codesourcery.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
[MIPS] The o32 fstatat syscall behaves differently on 32 and 64 bit kernels
While working on a glibc patch to support the fstatat() functions[1],
I noticed that the o32 implementation behaves differently on 32-bit and
64-bit kernels; the former provides a stat64 while the latter provides
a plain (o32) stat. I think the former is what's intended, as there is
no separate fstatat64. It's also what x86 does.
I think this is just a case of a compat too far.
[1] I've seen Khem's patch, but I don't think it's right.
Signed-off-by: Richard Sandiford <richard@codesourcery.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Peter Watkins [Wed, 23 Aug 2006 15:15:49 +0000 (11:15 -0400)]
[MIPS] Fix USER_PTRS_PER_PGD for 64K page size.
The code in pgtable-64.h assumes TASK_SIZE is always bigger than a first
level PGDIR_SIZE. This is not the case for 64K pages, where task size is
40 bits (1TB) and a pgd entry can map 42 bits. This leads to
USER_PTRS_PER_PGD being zero for 64K pages.
Signed-off-by: Peter Watkins <treestem@gmail.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
[MIPS] Add configuration variables for RM9xxx processor
This patch introduces a number of configuration variables. These allow to
specify presence/absence of integrated peripherals found on the MIPS
RM9xxx processor family, based on the particular processor model used.
Signed-off-by: Thomas Koeller <thomas.koeller@baslerweb.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The excite platform exports hardware resources for device drivers to use.
Any driver wanting to use these resources will look up them by their names.
Since these resources are declared to have static linkage, but are not
used in the source file defining them, the compiler used to emit an
'unused' warning, which this patch suppresses.
Signed-off-by: Thomas Koeller <thomas.koeller@baslerweb.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Ralf Baechle [Tue, 1 Aug 2006 22:39:42 +0000 (23:39 +0100)]
[MIPS] c-r4k: Convert init functions from inline to __init.
With more recent compilers inline doesn't necessarily means a function
will always be inlined. So leave that decission to the compiler and
make the function as __init.
Atlas maps its RTC chip in the host mmio space rather than using the
"traditional" location in the PCI/ISA port space. A change that has
happened to the generic RTC header requires to define ARCH_RTC_LOCATION
now.
Signed-off-by: Maciej W. Rozycki <macro@mips.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Kevin D. Kissell [Tue, 12 Sep 2006 10:08:08 +0000 (12:08 +0200)]
[MIPS] Patch to arch/mips/mips-boards/generic/time.c
In hooking up the perf counter overflow interrupt to the experimental
deprecated-real-soon-now /proc/perf interface last night, I had to
revisit arch/mips/mips-boards/generic/time.c, and discovered that
when the 2.6.9-based SMTC prototype was merged with the more
recent tree, it was missed that arch/mips/kernel/time.c had changed
so that even in SMP kernels, timer_interrupt() calls
local_timer_interrupt(), so there is no longer a need to invoke it
directly from mips_timer_interrupt() in those cases where
timer_interrupt() has been called. So I got rid of that, and added the
invocation of perf_irq() if Cause.PCI is set, more-or-less following the
same logic as in the non-SMTC case, with the modifications that (a) a
runtime check for Release 2 isn't done, because it's redundant in SMTC),
and (b) we check for a clock interrupt regardless of the value returned
by the perf counter service - I don't understand why we'd want to control
that with perf_irq(), but maybe one of you knows the story. I also got
rid of the stupid warning about the unused variable when compiled for
SMTC (another artifact of the merge). The result hasn't been beaten to
death, but boots, seems stable, and supports extended precision event
counting.
Signed-off-by: Kevin D. Kissell <kevink@mips.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Atsushi Nemoto [Wed, 7 Jun 2006 16:09:01 +0000 (01:09 +0900)]
[MIPS] Reduce race between cpu_wait() and need_resched() checking
If a thread became runnable between need_resched() and the WAIT
instruction, switching to the thread will delay until a next interrupt.
Some CPUs can execute the WAIT instruction with interrupt disabled, so
we can get rid of this race on them (at least UP case).
Original Patch by Atsushi with fixing up for MIPS Technology's cores by
Ralf based on feedback from the RTL designers.
Alexander Bigga [Tue, 29 Aug 2006 14:48:34 +0000 (16:48 +0200)]
[MIPS] Fix for pci config_access on alchemy au1x000
I've encountered a serious problem with PCI config space access on Au1x000
platforms with recent 2.6.x-kernel. With 2.4.31 the same hardware works fine.
So I was looking for the differences:
Symptoms:
- no PCI-device is seen on bootup though two or three cards are present
- lspci output is empty
- OR: lspci shows 20 times the same device
(- OR: in some slot-configurations it worked anyhow)
System(s):
1. platform with Au1500 and three PCI-devices (actually a mycable XXS1500
with backplane for three PCI-devices)
2. platform with Au1550 and two PCI-devices (custom board)
Debugging:
I digged down to the config_access() of the au1xxx-processors in
arch/mips/pci/ops-au1000.c and switched on DEBUG.
The code of config_access() seems to be almost the same as of the
2.4.x-kernel. But the "pci_cfg_vm->addr" returned by get_vm_area(0x2000, 0)
once on booting is different. That's of course not forbidden. But the
alignment seems to be wrong. In my case, I received:
To make it short: With 2.6.x it fails on the first config-access with:
"PCI ERR detected: status 83a00356".
Fixup:
My fix is now, to use the VM_IOREMAP-flag in the get_vm_area call. This flag
seems to be introduced in mm/vmalloc.c a long time ago (in 2.6.7-bk13, I
found in gitweb).
Now, the returned address is pci_cfg_vm->addr = c0104000 and everything works
fine.
Atsushi Nemoto [Tue, 29 Aug 2006 03:10:22 +0000 (12:10 +0900)]
[MIPS] Make prepare_frametrace() not clobber v0
Since lmo commit 323a380bf9e1a1679a774a2b053e3c1f2aa3f179 ("Simplify
dump_stack()") made prepare_frametrace() always inlined, using $2 (v0)
in __asm__ is not safe anymore. We can use $1 (at) instead. Also we
should use "dla" instead of "la" for 64-bit kernel.
Atsushi Nemoto [Fri, 25 Aug 2006 08:55:31 +0000 (17:55 +0900)]
[MIPS] Do not use drop_mmu_context to flusing other task's VIPT I-cache.
c-r4k.c and c-sb1.c use drop_mmu_context() to flush virtually tagged
I-caches, but this does not work for flushing other task's icache. This
is for example triggered by copy_to_user_page() called from ptrace(2).
Use indexed flush for such cases.
Ralf Baechle [Sat, 12 Aug 2006 15:40:08 +0000 (16:40 +0100)]
[MIPS] Retire flush_icache_page from mm use.
On the 34K the redundant cache operations were causing excessive stalls
resulting in realtime code running on the second VPE missing its deadline.
For all other platforms this patch is just a significant performance
improvment as illustrated by below benchmark numbers.
Ralf Baechle [Tue, 8 Aug 2006 02:47:01 +0000 (03:47 +0100)]
[MIPS] Avoid double signal restarting.
In entry.S resume_userspace ... jal do_notify_resume form a loop through
which the kernel will iterate as long as work is pending. If we
iterate through this loop more than once with no signal pending for at
least one but the last iteration we will take do the syscall restarting
multiple times resulting in a syscall return prior to the the syscall
instruction in userspace. This may happen when debugging a multithreaded
program.
Debugging and original fix by Maciej; extended to other ABIs by me.
Signed-off-by: Maciej W. Rozycki <macro@mips.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:21 +0000 (09:29 +0200)]
[MIPS] Improve unwind_stack()
This patch allows unwind_stack() to return ra for leaf function.
But it tries to detects cases where get_frame_info() wrongly
consider nested function as a leaf one.
It also pass 'unsinged long *sp' instead of 'unsigned long **sp'
as second parameter. The code looks cleaner.
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:20 +0000 (09:29 +0200)]
[MIPS] Make get_frame_info() more robust
Now get_frame_info() wants to detect move sp instruction first. It
assumes that the save ra in the stack instruction can't happen
before allocating frame size space into the stack.
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:17 +0000 (09:29 +0200)]
[MIPS] Miscellaneous cleanup in prologue analysis code
We usually use backtrace term for dumping a call tree during
debug. Therefore this patch renames show_frametrace() into
show_backtrace() and show_trace() into show_raw_backtrace().
[MIPS] dump_stack() based on prologue code analysis
Instead of dump all possible address in the stack, unwind the stack frame
based on prologue code analysis, as like as get_wchan() does. While the
code analysis might fail for some reason, there is a new kernel option
"raw_show_trace" to disable this feature.
This patch fixes the problem that trying to enable already enabled
slot disables the slot by returning the proper value from
pciehp_enable_slot()/pciehp_disable_slot().
We need to assign resources to ioapics being hot-added. This patch
changes pbus_assign_resources_sorted() to assign resources if the
ioapic has no assigned resources.