| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Patch series "Memory allocation profiling", v6.
Overview:
Low overhead [1] per-callsite memory allocation profiling. Not just for
debug kernels, overhead low enough to be deployed in production.
Example output:
root@moria-kvm:~# sort -rn /proc/allocinfo
127664128 31168 mm/page_ext.c:270 func:alloc_page_ext
56373248 4737 mm/slub.c:2259 func:alloc_slab_page
14880768 3633 mm/readahead.c:247 func:page_cache_ra_unbounded
14417920 3520 mm/mm_init.c:2530 func:alloc_large_system_hash
13377536 234 block/blk-mq.c:3421 func:blk_mq_alloc_rqs
11718656 2861 mm/filemap.c:1919 func:__filemap_get_folio
9192960 2800 kernel/fork.c:307 func:alloc_thread_stack_node
4206592 4 net/netfilter/nf_conntrack_core.c:2567 func:nf_ct_alloc_hashtable
4136960 1010 drivers/staging/ctagmod/ctagmod.c:20 [ctagmod] func:ctagmod_start
3940352 962 mm/memory.c:4214 func:alloc_anon_folio
2894464 22613 fs/kernfs/dir.c:615 func:__kernfs_new_node
...
Usage:
kconfig options:
- CONFIG_MEM_ALLOC_PROFILING
- CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT
- CONFIG_MEM_ALLOC_PROFILING_DEBUG
adds warnings for allocations that weren't accounted because of a
missing annotation
sysctl:
/proc/sys/vm/mem_profiling
Runtime info:
/proc/allocinfo
Notes:
[1]: Overhead
To measure the overhead we are comparing the following configurations:
(1) Baseline with CONFIG_MEMCG_KMEM=n
(2) Disabled by default (CONFIG_MEM_ALLOC_PROFILING=y &&
CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT=n)
(3) Enabled by default (CONFIG_MEM_ALLOC_PROFILING=y &&
CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT=y)
(4) Enabled at runtime (CONFIG_MEM_ALLOC_PROFILING=y &&
CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT=n && /proc/sys/vm/mem_profiling=1)
(5) Baseline with CONFIG_MEMCG_KMEM=y && allocating with __GFP_ACCOUNT
(6) Disabled by default (CONFIG_MEM_ALLOC_PROFILING=y &&
CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT=n) && CONFIG_MEMCG_KMEM=y
(7) Enabled by default (CONFIG_MEM_ALLOC_PROFILING=y &&
CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT=y) && CONFIG_MEMCG_KMEM=y
Performance overhead:
To evaluate performance we implemented an in-kernel test executing
multiple get_free_page/free_page and kmalloc/kfree calls with allocation
sizes growing from 8 to 240 bytes with CPU frequency set to max and CPU
affinity set to a specific CPU to minimize the noise. Below are results
from running the test on Ubuntu 22.04.2 LTS with 6.8.0-rc1 kernel on
56 core Intel Xeon:
kmalloc pgalloc
(1 baseline) 6.764s 16.902s
(2 default disabled) 6.793s (+0.43%) 17.007s (+0.62%)
(3 default enabled) 7.197s (+6.40%) 23.666s (+40.02%)
(4 runtime enabled) 7.405s (+9.48%) 23.901s (+41.41%)
(5 memcg) 13.388s (+97.94%) 48.460s (+186.71%)
(6 def disabled+memcg) 13.332s (+97.10%) 48.105s (+184.61%)
(7 def enabled+memcg) 13.446s (+98.78%) 54.963s (+225.18%)
Memory overhead:
Kernel size:
text data bss dec diff
(1) 26515311 18890222 17018880 62424413
(2) 26524728 19423818 16740352 62688898 264485
(3) 26524724 19423818 16740352 62688894 264481
(4) 26524728 19423818 16740352 62688898 264485
(5) 26541782 18964374 16957440 62463596 39183
Memory consumption on a 56 core Intel CPU with 125GB of memory:
Code tags: 192 kB
PageExts: 262144 kB (256MB)
SlabExts: 9876 kB (9.6MB)
PcpuExts: 512 kB (0.5MB)
Total overhead is 0.2% of total memory.
Benchmarks:
Hackbench tests run 100 times:
hackbench -s 512 -l 200 -g 15 -f 25 -P
baseline disabled profiling enabled profiling
avg 0.3543 0.3559 (+0.0016) 0.3566 (+0.0023)
stdev 0.0137 0.0188 0.0077
hackbench -l 10000
baseline disabled profiling enabled profiling
avg 6.4218 6.4306 (+0.0088) 6.5077 (+0.0859)
stdev 0.0933 0.0286 0.0489
stress-ng tests:
stress-ng --class memory --seq 4 -t 60
stress-ng --class cpu --seq 4 -t 60
Results posted at: https://evilpiepirate.org/~kent/memalloc_prof_v4_stress-ng/
[2] https://lore.kernel.org/all/[email protected]/
This patch (of 37):
The next patch drops vmalloc.h from a system header in order to fix a
circular dependency; this adds it to all the files that were pulling it in
implicitly.
[[email protected]: fix arch/alpha/lib/memcpy.c]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: fix arch/x86/mm/numa_32.c]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: a few places were depending on sizes.h]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: fix mm/kasan/hw_tags.c]
Link: https://lkml.kernel.org/r/[email protected]
[[email protected]: fix arc build]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Kent Overstreet <[email protected]>
Signed-off-by: Suren Baghdasaryan <[email protected]>
Signed-off-by: Arnd Bergmann <[email protected]>
Reviewed-by: Pasha Tatashin <[email protected]>
Tested-by: Kees Cook <[email protected]>
Cc: Alexander Viro <[email protected]>
Cc: Alex Gaynor <[email protected]>
Cc: Alice Ryhl <[email protected]>
Cc: Andreas Hindborg <[email protected]>
Cc: Benno Lossin <[email protected]>
Cc: "Björn Roy Baron" <[email protected]>
Cc: Boqun Feng <[email protected]>
Cc: Christoph Lameter <[email protected]>
Cc: Dennis Zhou <[email protected]>
Cc: Gary Guo <[email protected]>
Cc: Miguel Ojeda <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Tejun Heo <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Wedson Almeida Filho <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
|
|
The __vmalloc_start_set declaration is in a header that is not included
in numa_32.c in current linux-next:
arch/x86/mm/numa_32.c: In function 'initmem_init':
arch/x86/mm/numa_32.c:57:9: error: '__vmalloc_start_set' undeclared (first use in this function)
57 | __vmalloc_start_set = true;
| ^~~~~~~~~~~~~~~~~~~
arch/x86/mm/numa_32.c:57:9: note: each undeclared identifier is reported only once for each function it appears in
Add an explicit #include.
Signed-off-by: Arnd Bergmann <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
|
|
The DISCONTIGMEM support was marked as deprecated in v5.2 and since there
were no complaints about it for almost 5 releases it can be completely
removed.
Signed-off-by: Mike Rapoport <[email protected]>
Signed-off-by: Borislav Petkov <[email protected]>
Acked-by: Dave Hansen <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
|
|
Move remaining definitions and declarations from include/linux/bootmem.h
into include/linux/memblock.h and remove the redundant header.
The includes were replaced with the semantic patch below and then
semi-automated removal of duplicated '#include <linux/memblock.h>
@@
@@
- #include <linux/bootmem.h>
+ #include <linux/memblock.h>
[[email protected]: dma-direct: fix up for the removal of linux/bootmem.h]
Link: http://lkml.kernel.org/r/[email protected]
[[email protected]: powerpc: fix up for removal of linux/bootmem.h]
Link: http://lkml.kernel.org/r/[email protected]
[[email protected]: x86/kaslr, ACPI/NUMA: fix for linux/bootmem.h removal]
Link: http://lkml.kernel.org/r/[email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Mike Rapoport <[email protected]>
Signed-off-by: Stephen Rothwell <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Cc: Catalin Marinas <[email protected]>
Cc: Chris Zankel <[email protected]>
Cc: "David S. Miller" <[email protected]>
Cc: Geert Uytterhoeven <[email protected]>
Cc: Greentime Hu <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Guan Xuetao <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Cc: Jonas Bonn <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Ley Foon Tan <[email protected]>
Cc: Mark Salter <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: Matt Turner <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Michal Simek <[email protected]>
Cc: Palmer Dabbelt <[email protected]>
Cc: Paul Burton <[email protected]>
Cc: Richard Kuo <[email protected]>
Cc: Richard Weinberger <[email protected]>
Cc: Rich Felker <[email protected]>
Cc: Russell King <[email protected]>
Cc: Serge Semin <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Vineet Gupta <[email protected]>
Cc: Yoshinori Sato <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
CONFIG_NEED_NODE_MEMMAP_SIZE logic
node_memmap_size_bytes() has been unused since the v3.9 kernel, so remove it.
Signed-off-by: David Rientjes <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Laura Abbott <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Fixes: f03574f2d5b2 ("x86-32, mm: Rip out x86_32 NUMA remapping code")
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
|
|
'__vmalloc_start_set' currently only gets set in initmem_init() when
!CONFIG_NEED_MULTIPLE_NODES. This breaks detection of vmalloc address
with virt_addr_valid() with CONFIG_NEED_MULTIPLE_NODES=y, causing
a kernel crash:
[mm/usercopy] 517e1fbeb6: kernel BUG at arch/x86/mm/physaddr.c:78!
Set '__vmalloc_start_set' appropriately for that case as well.
Reported-by: kbuild test robot <[email protected]>
Signed-off-by: Laura Abbott <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Fixes: dc16ecf7fd1f ("x86-32: use specific __vmalloc_start_set flag in __virt_addr_valid")
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
|
|
Historically a lot of these existed because we did not have
a distinction between what was modular code and what was providing
support to modules via EXPORT_SYMBOL and friends. That changed
when we forked out support for the latter into the export.h file.
This means we should be able to reduce the usage of module.h
in code that is obj-y Makefile or bool Kconfig. The advantage
in doing so is that module.h itself sources about 15 other headers;
adding significantly to what we feed cpp, and it can obscure what
headers we are effectively using.
Since module.h was the source for init.h (for __init) and for
export.h (for EXPORT_SYMBOL) we consider each obj-y/bool instance
for the presence of either and replace accordingly where needed.
Note that some bool/obj-y instances remain since module.h is
the header for some exception table entry stuff, and for things
like __init_or_module (code that is tossed when MODULES=n).
Signed-off-by: Paul Gortmaker <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
|
|
With DISCONTIGMEM, the mapping between a pfn and its owning node is
initialized using data provided by the BIOS. However, the initialization
may fail if the extents are not aligned to section boundary (64M).
The symptom of this bug is an early boot failure in pfn_to_page(),
as it tries to access NODE_DATA(__nid) using index from an unitialized
element of the physnode_map[] array.
While the bug is always present, it is more likely to be hit in kdump
kernels on large machines, because:
1. The memory map for a kdump kernel is specified as exactmap, and
exactmap is more likely to be unaligned.
2. Large reservations are more likely to span across a 64M boundary.
[ hpa: fixed incorrect use of "pfn" instead of "start" ]
Signed-off-by: Petr Tesarik <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Prepare for removing num_physpages and simplify mem_init().
Signed-off-by: Jiang Liu <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: Andreas Herrmann <[email protected]>
Cc: Tang Chen <[email protected]>
Cc: Wen Congyang <[email protected]>
Cc: Jianguo Wu <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
This code was an optimization for 32-bit NUMA systems.
It has probably been the cause of a number of subtle bugs over
the years, although the conditions to excite them would have
been hard to trigger. Essentially, we remap part of the kernel
linear mapping area, and then sometimes part of that area gets
freed back in to the bootmem allocator. If those pages get
used by kernel data structures (say mem_map[] or a dentry),
there's no big deal. But, if anyone ever tried to use the
linear mapping for these pages _and_ cared about their physical
address, bad things happen.
For instance, say you passed __GFP_ZERO to the page allocator
and then happened to get handed one of these pages, it zero the
remapped page, but it would make a pte to the _old_ page.
There are probably a hundred other ways that it could screw
with things.
We don't need to hang on to performance optimizations for
these old boxes any more. All my 32-bit NUMA systems are long
dead and buried, and I probably had access to more than most
people.
This code is causing real things to break today:
https://lkml.org/lkml/2013/1/9/376
I looked in to actually fixing this, but it requires surgery
to way too much brittle code, as well as stuff like
per_cpu_ptr_to_phys().
[ hpa: Cc: this for -stable, since it is a memory corruption issue.
However, an alternative is to simply mark NUMA as depends BROKEN
rather than EXPERIMENTAL in the X86_32 subclause... ]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: H. Peter Anvin <[email protected]>
Cc: <[email protected]>
|
|
Other than sanity check and debug message, the x86 specific version of
memblock reserve/free functions are simple wrappers around the generic
versions - memblock_reserve/free().
This patch adds debug messages with caller identification to the
generic versions and replaces x86 specific ones and kills them.
arch/x86/include/asm/memblock.h and arch/x86/mm/memblock.c are empty
after this change and removed.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Cc: Yinghai Lu <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
25818f0f28 (memblock: Make MEMBLOCK_ERROR be 0) thankfully made
MEMBLOCK_ERROR 0 and there already are codes which expect error return
to be 0. There's no point in keeping MEMBLOCK_ERROR around. End its
misery.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Cc: Yinghai Lu <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
DISCONTIGMEM on x86-32 implements pfn -> nid mapping similarly to
SPARSEMEM; however, it calls each mapping unit ELEMENT instead of
SECTION. This patch renames it to SECTION so that PAGES_PER_SECTION
is valid for both DISCONTIGMEM and SPARSEMEM. This will be used by
the next patch to implement mapping granularity check.
This patch is trivial constant rename.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Cc: Hans Rosenfeld <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
With both _numa_init() methods converted and the rest of init code
adjusted, numa_32.c now can switch from the 32bit only init code to
the common one in numa.c.
* Shim get_memcfg_*()'s are dropped and initmem_init() calls
x86_numa_init(), which is updated to handle NUMAQ.
* All boilerplate operations including node range limiting, pgdat
alloc/init are handled by numa_init(). 32bit only implementation is
removed.
* 32bit numa_add_memblk(), numa_set_distance() and
memory_add_physaddr_to_nid() removed and common versions in
numa_32.c enabled for 32bit.
This change causes the following behavior changes.
* NODE_DATA()->node_start_pfn/node_spanned_pages properly initialized
for 32bit too.
* Much more sanity checks and configuration cleanups.
* Proper handling of node distances.
* The same NUMA init messages as 64bit.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
setup_node_bootmem() is taken from 64bit and doesn't use remap
allocator. It's about to be shared with 32bit so add support for it.
If NODE_DATA is remapped, it's noted in the debug message and node
locality check is skipped as the __pa() of the remapped address
doesn't reflect the actual physical address.
On 64bit, remap allocator becomes noop and doesn't affect the
behavior.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
Instead of dereferencing node_start/end_pfn[] directly, make
init_alloc_remap() take @start and @end and let the caller be
responsible for making sure the range is sane. This is to prepare for
use from unified NUMA init code.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
Generic NUMA init code was moved to numa.c from numa_64.c but is still
guaraded by CONFIG_X86_64. This patch removes the compile guard and
enables compiling on 32bit.
* numa_add_memblk() and numa_set_distance() clash with the shim
implementation in numa_32.c and are left out.
* memory_add_physaddr_to_nid() clashes with 32bit implementation and
is left out.
* MAX_DMA_PFN definition in dma.h moved out of !CONFIG_X86_32.
* node_data definition in numa_32.c removed in favor of the one in
numa.c.
There are places where ulong is assumed to be 64bit. The next patch
will fix them up. Note that although the code is compiled it isn't
used yet and this patch doesn't cause any functional change.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
Update numaq such that it calls numa_add_memblk() and sets
numa_nodes_parsed instead of directly diddling with NUMA states. The
original get_memcfg_numaq() is renamed to numaq_numa_init() and new
get_memcfg_numaq() is created in numa_32.c.
The shim numa_add_memblk() implementation handles node_start/end_pfn[]
and node_set_online() for nodes with memory. The new
get_memcfg_numaq() exactly the same with get_memcfg_from_srat() other
than calling the numaq init function. Things get_memcfgs_numaq() do
are not strictly necessary for numaq but added for consistency and to
help unifying NUMA init handling.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
SRAT support implementation in srat_32.c and srat.c are generally
similar; however, there are some differences.
First of all, 64bit implementation supports more types of SRAT
entries. 64bit supports x2apic, affinity, memory and SLIT. 32bit
only supports processor and memory.
Most other differences stem from different initialization protocols
employed by 64bit and 32bit NUMA init paths.
On 64bit,
* Mappings among PXM, node and apicid are directly done in each SRAT
entry callback.
* Memory affinity information is passed to numa_add_memblk() which
takes care of all interfacing with NUMA init.
* Doesn't directly initialize NUMA configurations. All the
information is recorded in numa_nodes_parsed and memblks.
On 32bit,
* Checks numa_off.
* Things go through one more level of indirection via private tables
but eventually end up initializing the same mappings.
* node_start/end_pfn[] are initialized and
memblock_x86_register_active_regions() is called for each memory
chunk.
* node_set_online() is called for each online node.
* sort_node_map() is called.
There are also other minor differences in sanity checking and messages
but taking 64bit version should be good enough.
This patch drops the 32bit specific implementation and makes the 64bit
implementation common for both 32 and 64bit.
The init protocol differences are dealt with in two places - the
numa_add_memblk() shim added in the previous patch and new temporary
numa_32.c:get_memcfg_from_srat() which wraps invocation of
x86_acpi_numa_init().
The shim numa_add_memblk() handles the folowings.
* node_start/end_pfn[] initialization.
* node_set_online() for memory nodes.
* Invocation of memblock_x86_register_active_regions().
The shim get_memcfg_from_srat() handles the followings.
* numa_off check.
* node_set_online() for CPU nodes.
* sort_node_map() invocation.
* Clearing of numa_nodes_parsed and active_ranges on failure.
The shims are temporary and will be removed as the generic NUMA init
path in 32bit is replaced with 64bit one.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
To help transition to common NUMA init, implement temporary 32bit
shims for numa_add_memblk() and numa_set_distance().
numa_add_memblk() registers the memblk and adjusts
node_start/end_pfn[]. numa_set_distance() is noop.
These shims will allow using 64bit NUMA init functions on 32bit and
gradual transition to common NUMA init path.
For detailed description, please read description of commits which
make use of the shim functions.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
There's no reason get_memcfg_numa() to be implemented inline in
mmzone_32.h. Move it to numa_32.c and also make
get_memcfg_numa_flag() static.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
Instead of calling memory_present() for each region from NUMA init,
call sparse_memory_present_with_active_regions() from paging_init()
similarly to x86-64.
For flat and numaq, this results in exactly the same memory_present()
calls. For srat, if there are multiple memory chunks for a node,
after this change, memory_present() will be called separately for each
chunk instead of being called once to encompass the whole range, which
doesn't cause any harm and actually is the better behavior.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
Currently, the only meaningful user of apic->x86_32_numa_cpu_node() is
NUMAQ which returns valid mapping only after CPU is initialized during
SMP bringup; thus, the previous patch to set apicid -> node in
setup_local_APIC() makes __apicid_to_node[] always contain the correct
mapping whether custom apic->x86_32_numa_cpu_node() is used or not.
So, there is no reason to keep separate 32bit implementation. We can
always consult __apicid_to_node[]. Move 64bit implementation from
numa_64.c to numa.c and remove 32bit implementation from numa_32.c.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
|
|
Now that remap allocator is cleaned up, update comments such that they
are in docbook function description format and reflect the actual
implementation.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Remap area size can be determined from node_remap_start_vaddr[] and
node_remap_end_vaddr[] making node_remap_size[] redundant. Remove it.
While at it, make resume_map_numa_kva() use @nr_pages for number of
pages instead of @size.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
With lowmem address reservation moved into init_alloc_remap(),
node_remap_offset[] is no longer useful. Remove it and related offset
handling code.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
pgdat allocation is handled differnetly from other remap allocations -
it's reserved during initialization. There's no reason to handle this
any differnetly. Remap allocator is initialized for every node and if
init failed the allocation will fail and pgdat allocation can fall
back to generic code like anyone else.
Remove special init-time pgdat reservation and make allocate_pgdat()
use alloc_remap() like everyone else.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
There's no reason to perform the actual remapping separately.
Collapse remap_numa_kva() into init_alloc_remap() and, while at it,
make it less verbose.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Remap alloc init is done in the following stages.
1. init_alloc_remap() calculates how much memory is necessary for each
node and reserves node local memory.
2. initmem_init() collects how much each node needs and reserves a
single contiguous lowmem area which can contain all.
3. init_remap_allocator() initializes allocator parameters from the
determined lowmem address and per-node offsets.
4. Actual remap happens.
There is no reason for the lowmem remap area to be reserved as a
single contiguous area at one go. They don't interact with each other
and the memblock allocator will put them side-by-side anyway.
This patch breaks up the single lowmem address reservation and put
per-node lowmem address reservation into init_alloc_remap() and
initializes allocator parameters directly in the function as all the
addresses are determined there. This merges steps 2 and 3 into 1.
While at it, remove now largely irrelevant comments in
init_alloc_remap().
This change causes the following behavior changes.
* Remap lowmem areas are allocated in smaller per-node chunks.
* Remap lowmem area reservation failure fail future remap allocations
instead of panicking.
* Remap allocator initialization is less verbose.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Remap allocator failure isn't fatal. The callers are required to fall
back to regular early memory allocation mechanisms on failure anyway,
so there's no reason to panic on remap init failure. Whining and
returning are enough.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Only pgdat and memmap use remap area and there isn't much benefit in
allowing per-node override. In addition, the use of node_remap_size[]
is confusing in that it contains number of bytes before remap
initialization and then number of pages afterwards.
Move remap size calculation for memap from specific NUMA config
implementations to init_alloc_remap() and make node_remap_size[]
static.
The only behavior difference is that, before this patch, numaq_32
didn't consider max_pfn when calculating the memmap size but it's
enforced after this patch, which is the right thing to do.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
@size variable in init_alloc_remap() is confusing in that it starts as
number of bytes as its name implies and then becomes number of pages.
Make it consistently represent bytes.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
init_alloc_remap() is about to do more and using _kva suffix for
physical address becomes confusing because the function will be
handling both physical and virtual addresses. Rename @node_kva to
@node_pa.
This is trivial rename and doesn't cause any behavior difference.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Separate the outer node walking loop and per-node logic from
calculate_numa_remap_pages(). The outer loop is collapsed into
initmem_init() and the per-node logic is moved into a new function -
init_alloc_remap().
The new function name is confusing with the existing
init_remap_allocator() and the behavior is the function isn't very
clean either at this point, but this is to prepare for further
cleanups and it will become prettier.
This function doesn't introduce any behavior change.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
memblock_find_in_range() now does top-down allocation by default, so
there's no reason for its callers to explicitly implement it by
gradually lowering the start address.
Remove redundant top-down allocation logic from init_meminit() and
calculate_numa_remap_pages().
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
When pgdat is reserved in init_remap_allocator(), PAGE_SIZE aligned
size will be used. Match the size alignment in initialization to
avoid allocation failure down the road.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
node_remap_{start|end}_vaddr[] describe [start, end) ranges; however,
alloc_remap() incorrectly failed when the current allocation + size
equaled the end but it should fail only when it goes over. Fix it.
Signed-off-by: Tejun Heo <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Acked-by: Yinghai Lu <[email protected]>
Cc: David Rientjes <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
There's no reason for these to live in setup_arch(). Move them inside
initmem_init().
- v2: x86-32 initmem_init() weren't updated breaking 32bit builds.
Fixed. Found by Ankita.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Ankita Garg <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: Brian Gerst <[email protected]>
Cc: Cyrill Gorcunov <[email protected]>
Cc: Shaohui Zheng <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: H. Peter Anvin <[email protected]>
|
|
initmem_init() extensively accesses and modifies global data
structures and the parameters aren't even followed depending on which
path is being used. Drop @start/last_pfn and let it deal with
@max_pfn directly. This is in preparation for further NUMA init
cleanups.
- v2: x86-32 initmem_init() weren't updated breaking 32bit builds.
Fixed. Found by Yinghai.
Signed-off-by: Tejun Heo <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: Brian Gerst <[email protected]>
Cc: Cyrill Gorcunov <[email protected]>
Cc: Shaohui Zheng <[email protected]>
Cc: David Rientjes <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: H. Peter Anvin <[email protected]>
|
|
Now that everything else is unified, NUMA initialization can be
unified too.
* numa_init_array() and init_cpu_to_node() are moved from
numa_64 to numa.
* numa_32::initmem_init() is updated to call numa_init_array()
and setup_arch() to call init_cpu_to_node() on 32bit too.
* x86_cpu_to_node_map is now initialized to NUMA_NO_NODE on
32bit too. This is safe now as numa_init_array() will initialize
it early during boot.
This makes NUMA mapping fully initialized before
setup_per_cpu_areas() on 32bit too and thus makes the first
percpu chunk which contains all the static variables and some of
dynamic area allocated with NUMA affinity correctly considered.
Signed-off-by: Tejun Heo <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Reported-by: Eric Dumazet <[email protected]>
Reviewed-by: Pekka Enberg <[email protected]>
|
|
The mapping between cpu/apicid and node is done via
apicid_to_node[] on 64bit and apicid_2_node[] +
apic->x86_32_numa_cpu_node() on 32bit. This difference makes it
difficult to further unify 32 and 64bit NUMA handling.
This patch unifies it by replacing both apicid_to_node[] and
apicid_2_node[] with __apicid_to_node[] array, which is accessed
by two accessors - set_apicid_to_node() and numa_cpu_node(). On
64bit, numa_cpu_node() always consults __apicid_to_node[]
directly while 32bit goes through apic->numa_cpu_node() method
to allow apic implementations to override it.
srat_detect_node() for amd cpus contains workaround for broken
NUMA configuration which assumes relationship between APIC ID,
HT node ID and NUMA topology. Leave it to access
__apicid_to_node[] directly as mapping through CPU might result
in undesirable behavior change. The comment is reformatted and
updated to note the ugliness.
Signed-off-by: Tejun Heo <[email protected]>
Reviewed-by: Pekka Enberg <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Cc: David Rientjes <[email protected]>
|
|
Originally the only early reserved range that is overlapped with high
pages is "KVA RAM", but we already do remove that from the active ranges.
However, It turns out Xen could have that kind of overlapping to support memory
ballooning.x
So we need to make add_highpage_with_active_regions() to subtract
memblock reserved just like low ram; this is the proper design anyway.
In this patch, refactering get_freel_all_memory_range() to make it can
be used by add_highpage_with_active_regions(). Also we don't need to
remove "KVA RAM" from active ranges.
Signed-off-by: Yinghai Lu <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Requested by Ingo, Thomas and HPA.
The old bootmem code is no longer necessary, and the transition is
complete. Remove it.
Signed-off-by: Yinghai Lu <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
1.include linux/memblock.h directly. so later could reduce e820.h reference.
2 this patch is done by sed scripts mainly
-v2: use MEMBLOCK_ERROR instead of -1ULL or -1UL
Signed-off-by: Yinghai Lu <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
Let's make 32bit consistent with 64bit.
-v2: Andrew pointed out for 32bit that we should use -1ULL
Signed-off-by: Yinghai Lu <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: H. Peter Anvin <[email protected]>
|
|
To eventually interleave emulated nodes over physical nodes, we
need to know the physical topology of the machine without actually
registering it. This does the k8 node setup in two parts:
detection and registration. NUMA emulation can then used the
physical topology detected to setup the address ranges of emulated
nodes accordingly. If emulation isn't used, the k8 nodes are
registered as normal.
Two formals are added to the x86 NUMA setup functions: `acpi' and
`k8'. These represent whether ACPI or K8 NUMA has been detected;
both cannot be true at the same time. This specifies to the NUMA
emulation code whether an underlying physical NUMA topology exists
and which interface to use.
This patch deals solely with separating the k8 setup path into
Northbridge detection and registration steps and leaves the ACPI
changes for a subsequent patch. The `acpi' formal is added here,
however, to avoid touching all the header files again in the next
patch.
This approach also ensures emulated nodes will not span physical
nodes so the true memory latency is not misrepresented.
k8_get_nodes() may now be used to export the k8 physical topology
of the machine for NUMA emulation.
Signed-off-by: David Rientjes <[email protected]>
Cc: Andreas Herrmann <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: Balbir Singh <[email protected]>
Cc: Ankita Garg <[email protected]>
Cc: Len Brown <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
|
|
calculate_numa_remap_pages() is called only by __init initmem_init()
further calculate_numa_remap_pages is calling:
__init find_e820_area() and __init reserve_early()
So calculate_numa_remap_pages() should be __init calculate_numa_remap_pages().
WARNING: arch/x86/built-in.o(.text+0x82ea3): Section mismatch in reference from the function calculate_numa_remap_pages() to the function .init.text:find_e820_area()
The function calculate_numa_remap_pages() references
the function __init find_e820_area().
This is often because calculate_numa_remap_pages lacks a __init
annotation or the annotation of find_e820_area is wrong.
WARNING: arch/x86/built-in.o(.text+0x82f5f): Section mismatch in reference from the function calculate_numa_remap_pages() to the function .init.text:reserve_early()
The function calculate_numa_remap_pages() references
the function __init reserve_early().
This is often because calculate_numa_remap_pages lacks a __init
annotation or the annotation of reserve_early is wrong.
[ Impact: save memory, address Section mismatch warning ]
Signed-off-by: Jaswinder Singh Rajput <[email protected]>
Cc: Sam Ravnborg <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
|
|
Impact: fix panic on system 2g x4 sockets
Found one system with 4 sockets and every sockets has 2g can not boot
with numa32 because boot mem is crossing nodes.
So try to have numa version of setup_bootmem_allocator().
Signed-off-by: Yinghai Lu <[email protected]>
Cc: Andrew Morton <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
|
|
'x86/setup-lzma', 'x86/signal' and 'x86/urgent' into x86/core
|
|
Impact: cleanup
This patch moves set_highmem_pages_init() to arch/x86/mm/highmem_32.c.
The declaration of the function is kept in asm/numa_32.h because
asm/highmem.h is included only if CONFIG_HIGHMEM is enabled so we
can't put the empty static inline function there.
Signed-off-by: Pekka Enberg <[email protected]>
LKML-Reference: <1236082212.2675.24.camel@penberg-laptop>
Signed-off-by: Ingo Molnar <[email protected]>
|