Welcome everyone once again,
This is yet another release of the Contiguous Memory Allocator patches.
This version is mainly a result of the discussion on Kernel Summit in
Prague. The main change is completely different code base for the
migration feature of the CMA. Now it shares the code with memory
compaction subsystem, not the hotplug like it was before. This code has
been kindly provided by Michal Nazarewicz. There are also a few fixes
here and there, see changelog for the details.
Please notice that this patch series is aimed to start further
discussion. There are still few issues that need to be resolved before
CMA will be really ready. The most hot problem is the issue with movable
pages that causes migration to fail from time to time. Our investigation
leads us to the point that these rare pages cannot be migrated because
there are some pending io operations on them.
ARM integration code has not been changed since last version, it
provides implementation of all the ideas that has been discussed during
Linaro Sprint meeting. Here are the details:
This version provides a solution for complete integration of CMA to
DMA mapping subsystem on ARM architecture. The issue caused by double
dma pages mapping and possible aliasing in coherent memory mapping has
been finally resolved, both for GFP_ATOMIC case (allocations comes from
coherent memory pool) and non-GFP_ATOMIC case (allocations comes from
CMA managed areas).
For coherent, nommu, ARMv4 and ARMv5 systems the current DMA-mapping
implementation has been kept.
For ARMv6+ systems, CMA has been enabled and a special pool of coherent
memory for atomic allocations has been created. The size of this pool
defaults to DEFAULT_CONSISTEN_DMA_SIZE/8, but can be changed with
coherent_pool kernel parameter (if really required).
All atomic allocations are served from this pool. I've did a little
simplification here, because there is no separate pool for writecombine
memory - such requests are also served from coherent pool. I don't
think that such simplification is a problem here - I found no driver
that use dma_alloc_writecombine with GFP_ATOMIC flags.
All non-atomic allocation are served from CMA area. Kernel mapping is
updated to reflect required memory attributes changes. This is possible
because during early boot, all CMA area are remapped with 4KiB pages in
kernel low-memory.
This version have been tested on Samsung S5PC110 based Goni machine and
Exynos4 UniversalC210 board with various V4L2 multimedia drivers.
Coherent atomic allocations has been tested by manually enabling the dma
bounce for the s3c-sdhci device.
All patches are prepared for Linux Kernel v3.2-rc2.
A few words for these who see CMA for the first time:
The Contiguous Memory Allocator (CMA) makes it possible for device
drivers to allocate big contiguous chunks of memory after the system
has booted.
The main difference from the similar frameworks is the fact that CMA
allows to transparently reuse memory region reserved for the big
chunk allocation as a system memory, so no memory is wasted when no
big chunk is allocated. Once the alloc request is issued, the
framework will migrate system pages to create a required big chunk of
physically contiguous memory.
For more information you can refer to nice LWN articles:
http://lwn.net/Articles/447405/ and http://lwn.net/Articles/450286/
as well as links to previous versions of the CMA framework.
The CMA framework has been initially developed by Michal Nazarewicz
at Samsung Poland R&D Center. Since version 9, I've taken over the
development, because Michal has left the company.
TODO (optional):
- implement support for contiguous memory areas placed in HIGHMEM zone
- resolve issue with movable pages with pending io operations
Best regards
Marek Szyprowski
Samsung Poland R&D Center
Links to previous versions of the patchset:
v16: <http://www.spinics.net/lists/linux-mm/msg25066.html>
v15: <http://www.spinics.net/lists/linux-mm/msg23365.html>
v14: <http://www.spinics.net/lists/linux-media/msg36536.html>
v13: (internal, intentionally not released)
v12: <http://www.spinics.net/lists/linux-media/msg35674.html>
v11: <http://www.spinics.net/lists/linux-mm/msg21868.html>
v10: <http://www.spinics.net/lists/linux-mm/msg20761.html>
v9: <http://article.gmane.org/gmane.linux.kernel.mm/60787>
v8: <http://article.gmane.org/gmane.linux.kernel.mm/56855>
v7: <http://article.gmane.org/gmane.linux.kernel.mm/55626>
v6: <http://article.gmane.org/gmane.linux.kernel.mm/55626>
v5: (intentionally left out as CMA v5 was identical to CMA v4)
v4: <http://article.gmane.org/gmane.linux.kernel.mm/52010>
v3: <http://article.gmane.org/gmane.linux.kernel.mm/51573>
v2: <http://article.gmane.org/gmane.linux.kernel.mm/50986>
v1: <http://article.gmane.org/gmane.linux.kernel.mm/50669>
Changelog:
v17:
1. Replaced whole CMA core memory migration code to the new one kindly
provided by Michal Nazarewicz. The new code is based on memory
compaction framework not the memory hotplug, like it was before. This
change has been suggested by Mel Godman.
2. Addressed most of the comments from Andrew Morton and Mel Gorman in
the rest of the CMA code.
3. Fixed broken initialization on ARM systems with DMA zone enabled.
4. Rebased onto v3.2-rc2 kernel.
v16:
1. merged a fixup from Michal Nazarewicz to address comments from Dave
Hansen about checking if pfns belong to the same memory zone
2. merged a fix from Michal Nazarewicz for incorrect handling of pages
which belong to page block that is in MIGRATE_ISOLATE state, in very
rare cases the migrate type of page block might have been changed
from MIGRATE_CMA to MIGRATE_MOVABLE because of this bug
3. moved some common code to include/asm-generic
4. added support for x86 DMA-mapping framework for pci-dma hardware,
CMA can be now even more widely tested on KVM/QEMU and a lot of common
x86 boxes
5. rebased onto next-20111005 kernel tree, which includes changes in ARM
DMA-mapping subsystem (CONSISTENT_DMA_SIZE removal)
6. removed patch for CMA s5p-fimc device private regions (served only as
example) and provided the one that matches real life case - s5p-mfc
device
v15:
1. fixed calculation of the total memory after activating CMA area (was
broken from v12)
2. more code cleanup in drivers/base/dma-contiguous.c
3. added address limit for default CMA area
4. rewrote ARM DMA integration:
- removed "ARM: DMA: steal memory for DMA coherent mappings" patch
- kept current DMA mapping implementation for coherent, nommu and
ARMv4/ARMv5 systems
- enabled CMA for all ARMv6+ systems
- added separate, small pool for coherent atomic allocations, defaults
to CONSISTENT_DMA_SIZE/8, but can be changed with kernel parameter
coherent_pool=[size]
v14:
1. Merged with "ARM: DMA: steal memory for DMA coherent mappings"
patch, added support for GFP_ATOMIC allocations.
2. Added checks for NULL device pointer
v13: (internal, intentionally not released)
v12:
1. Fixed 2 nasty bugs in dma-contiguous allocator:
- alignment argument was not passed correctly
- range for dma_release_from_contiguous was not checked correctly
2. Added support for architecture specfic dma_contiguous_early_fixup()
function
3. CMA and DMA-mapping integration for ARM architechture has been
rewritten to take care of the memory aliasing issue that might
happen for newer ARM CPUs (mapping of the same pages with different
cache attributes is forbidden). TODO: add support for GFP_ATOMIC
allocations basing on the "ARM: DMA: steal memory for DMA coherent
mappings" patch and implement support for contiguous memory areas
that are placed in HIGHMEM zone
v11:
1. Removed genalloc usage and replaced it with direct calls to
bitmap_* functions, dropped patches that are not needed
anymore (genalloc extensions)
2. Moved all contiguous area management code from mm/cma.c
to drivers/base/dma-contiguous.c
3. Renamed cm_alloc/free to dma_alloc/release_from_contiguous
4. Introduced global, system wide (default) contiguous area
configured with kernel config and kernel cmdline parameters
5. Simplified initialization to just one function:
dma_declare_contiguous()
6. Added example of device private memory contiguous area
v10:
1. Rebased onto 3.0-rc2 and resolved all conflicts
2. Simplified CMA to be just a pure memory allocator, for use
with platfrom/bus specific subsystems, like dma-mapping.
Removed all device specific functions are calls.
3. Integrated with ARM DMA-mapping subsystem.
4. Code cleanup here and there.
5. Removed private context support.
v9: 1. Rebased onto 2.6.39-rc1 and resolved all conflicts
2. Fixed a bunch of nasty bugs that happened when the allocation
failed (mainly kernel oops due to NULL ptr dereference).
3. Introduced testing code: cma-regions compatibility layer and
videobuf2-cma memory allocator module.
v8: 1. The alloc_contig_range() function has now been separated from
CMA and put in page_allocator.c. This function tries to
migrate all LRU pages in specified range and then allocate the
range using alloc_contig_freed_pages().
2. Support for MIGRATE_CMA has been separated from the CMA code.
I have not tested if CMA works with ZONE_MOVABLE but I see no
reasons why it shouldn't.
3. I have added a @private argument when creating CMA contexts so
that one can reserve memory and not share it with the rest of
the system. This way, CMA acts only as allocation algorithm.
v7: 1. A lot of functionality that handled driver->allocator_context
mapping has been removed from the patchset. This is not to say
that this code is not needed, it's just not worth posting
everything in one patchset.
Currently, CMA is "just" an allocator. It uses it's own
migratetype (MIGRATE_CMA) for defining ranges of pageblokcs
which behave just like ZONE_MOVABLE but dispite the latter can
be put in arbitrary places.
2. The migration code that was introduced in the previous version
actually started working.
v6: 1. Most importantly, v6 introduces support for memory migration.
The implementation is not yet complete though.
Migration support means that when CMA is not using memory
reserved for it, page allocator can allocate pages from it.
When CMA wants to use the memory, the pages have to be moved
and/or evicted as to make room for CMA.
To make it possible it must be guaranteed that only movable and
reclaimable pages are allocated in CMA controlled regions.
This is done by introducing a MIGRATE_CMA migrate type that
guarantees exactly that.
Some of the migration code is "borrowed" from Kamezawa
Hiroyuki's alloc_contig_pages() implementation. The main
difference is that thanks to MIGRATE_CMA migrate type CMA
assumes that memory controlled by CMA are is always movable or
reclaimable so that it makes allocation decisions regardless of
the whether some pages are actually allocated and migrates them
if needed.
The most interesting patches from the patchset that implement
the functionality are:
09/13: mm: alloc_contig_free_pages() added
10/13: mm: MIGRATE_CMA migration type added
11/13: mm: MIGRATE_CMA isolation functions added
12/13: mm: cma: Migration support added [wip]
Currently, kernel panics in some situations which I am trying
to investigate.
2. cma_pin() and cma_unpin() functions has been added (after
a conversation with Johan Mossberg). The idea is that whenever
hardware does not use the memory (no transaction is on) the
chunk can be moved around. This would allow defragmentation to
be implemented if desired. No defragmentation algorithm is
provided at this time.
3. Sysfs support has been replaced with debugfs. I always felt
unsure about the sysfs interface and when Greg KH pointed it
out I finally got to rewrite it to debugfs.
v5: (intentionally left out as CMA v5 was identical to CMA v4)
v4: 1. The "asterisk" flag has been removed in favour of requiring
that platform will provide a "*=<regions>" rule in the map
attribute.
2. The terminology has been changed slightly renaming "kind" to
"type" of memory. In the previous revisions, the documentation
indicated that device drivers define memory kinds and now,
v3: 1. The command line parameters have been removed (and moved to
a separate patch, the fourth one). As a consequence, the
cma_set_defaults() function has been changed -- it no longer
accepts a string with list of regions but an array of regions.
2. The "asterisk" attribute has been removed. Now, each region
has an "asterisk" flag which lets one specify whether this
region should by considered "asterisk" region.
3. SysFS support has been moved to a separate patch (the third one
in the series) and now also includes list of regions.
v2: 1. The "cma_map" command line have been removed. In exchange,
a SysFS entry has been created under kernel/mm/contiguous.
The intended way of specifying the attributes is
a cma_set_defaults() function called by platform initialisation
code. "regions" attribute (the string specified by "cma"
command line parameter) can be overwritten with command line
parameter; the other attributes can be changed during run-time
using the SysFS entries.
2. The behaviour of the "map" attribute has been modified
slightly. Currently, if no rule matches given device it is
assigned regions specified by the "asterisk" attribute. It is
by default built from the region names given in "regions"
attribute.
3. Devices can register private regions as well as regions that
can be shared but are not reserved using standard CMA
mechanisms. A private region has no name and can be accessed
only by devices that have the pointer to it.
4. The way allocators are registered has changed. Currently,
a cma_allocator_register() function is used for that purpose.
Moreover, allocators are attached to regions the first time
memory is registered from the region or when allocator is
registered which means that allocators can be dynamic modules
that are loaded after the kernel booted (of course, it won't be
possible to allocate a chunk of memory from a region if
allocator is not loaded).
5. Index of new functions:
+static inline dma_addr_t __must_check
+cma_alloc_from(const char *regions, size_t size,
+ dma_addr_t alignment)
+static inline int
+cma_info_about(struct cma_info *info, const const char *regions)
+int __must_check cma_region_register(struct cma_region *reg);
+dma_addr_t __must_check
+cma_alloc_from_region(struct cma_region *reg,
+ size_t size, dma_addr_t alignment);
+static inline dma_addr_t __must_check
+cma_alloc_from(const char *regions,
+ size_t size, dma_addr_t alignment);
+int cma_allocator_register(struct cma_allocator *alloc);
Patches in this patchset:
Marek Szyprowski (4):
drivers: add Contiguous Memory Allocator
X86: integrate CMA with DMA-mapping subsystem
ARM: integrate CMA with DMA-mapping subsystem
ARM: Samsung: use CMA for 2 memory banks for s5p-mfc device
Michal Nazarewicz (7):
mm: page_alloc: handle MIGRATE_ISOLATE in free_pcppages_bulk()
mm: compaction: introduce isolate_{free,migrate}pages_range().
mm: mmzone: introduce zone_pfn_same_memmap()
mm: compaction: export some of the functions
mm: page_alloc: introduce alloc_contig_range()
mm: mmzone: MIGRATE_CMA migration type added
mm: page_isolation: MIGRATE_CMA isolation functions added
Documentation/kernel-parameters.txt | 9 +
arch/Kconfig | 3 +
arch/arm/Kconfig | 2 +
arch/arm/include/asm/dma-contiguous.h | 16 ++
arch/arm/include/asm/mach/map.h | 1 +
arch/arm/kernel/setup.c | 8 +-
arch/arm/mm/dma-mapping.c | 368 +++++++++++++++++++++++++------
arch/arm/mm/init.c | 20 ++-
arch/arm/mm/mm.h | 3 +
arch/arm/mm/mmu.c | 29 ++-
arch/arm/plat-s5p/dev-mfc.c | 51 +----
arch/x86/Kconfig | 1 +
arch/x86/include/asm/dma-contiguous.h | 13 +
arch/x86/include/asm/dma-mapping.h | 4 +
arch/x86/kernel/pci-dma.c | 18 ++-
arch/x86/kernel/pci-nommu.c | 8 +-
arch/x86/kernel/setup.c | 2 +
drivers/base/Kconfig | 89 ++++++++
drivers/base/Makefile | 1 +
drivers/base/dma-contiguous.c | 396 +++++++++++++++++++++++++++++++++
include/asm-generic/dma-contiguous.h | 27 +++
include/linux/device.h | 4 +
include/linux/dma-contiguous.h | 110 +++++++++
include/linux/mmzone.h | 57 ++++-
include/linux/page-isolation.h | 27 ++-
mm/Kconfig | 2 +-
mm/Makefile | 3 +-
mm/compaction.c | 230 +++++++++++--------
mm/internal.h | 35 +++
mm/memory-failure.c | 2 +-
mm/memory_hotplug.c | 6 +-
mm/page_alloc.c | 315 ++++++++++++++++++++++++--
mm/page_isolation.c | 15 +-
33 files changed, 1591 insertions(+), 284 deletions(-)
create mode 100644 arch/arm/include/asm/dma-contiguous.h
create mode 100644 arch/x86/include/asm/dma-contiguous.h
create mode 100644 drivers/base/dma-contiguous.c
create mode 100644 include/asm-generic/dma-contiguous.h
create mode 100644 include/linux/dma-contiguous.h
--
1.7.1.569.g6f426
Hello,
This is another update on my attempt on DMA-mapping framework redesign.
I focused mainly on the IOMMU mapper for ARM DMA-mapping implementation.
DMA-mapping patches have been rebased onto Linux v3.1-rc9-next kernel
with CMA v16 patches already applied. I've also integrated the code
provided by Krishna Reddy and added the missing methods for IOMMU DMA
mapper. The code has been tested on Samsung Exynos4 board.
Here is the link to the initial version of the DMA-mapping redesign patches:
http://www.spinics.net/lists/linux-mm/msg21241.html
Second version of the patches:
http://lists.linaro.org/pipermail/linaro-mm-sig/2011-September/000571.htmlhttp://lists.linaro.org/pipermail/linaro-mm-sig/2011-September/000577.html
TODO:
- start the discussion about chaning alloc_coherent into alloc_attrs in
dma_map_ops structure.
GIT tree will all the patches:
http://git.infradead.org/users/kmpark/linux-2.6-samsung/shortlog/refs/heads…git://git.infradead.org/users/kmpark/linux-2.6-samsung dma-mapping-v4
Best regards
--
Marek Szyprowski
Samsung Poland R&D Center
Patch summary:
Marek Szyprowski (8):
ARM: dma-mapping: remove offset parameter to prepare for generic
dma_ops
ARM: dma-mapping: use asm-generic/dma-mapping-common.h
ARM: dma-mapping: implement dma sg methods on top of any generic dma
ops
ARM: dma-mapping: move all dma bounce code to separate dma ops
structure
ARM: dma-mapping: remove redundant code and cleanup
common: dma-mapping: change alloc/free_coherent method to more
generic alloc/free_attrs
ARM: dma-mapping: use alloc, mmap, free from dma_ops
ARM: dma-mapping: add support for IOMMU mapper
arch/arm/Kconfig | 9 +
arch/arm/common/dmabounce.c | 78 +++-
arch/arm/include/asm/device.h | 5 +
arch/arm/include/asm/dma-iommu.h | 35 ++
arch/arm/include/asm/dma-mapping.h | 403 +++++------------
arch/arm/mm/dma-mapping.c | 869 +++++++++++++++++++++++++++++++-----
arch/arm/mm/vmregion.h | 2 +-
include/linux/dma-attrs.h | 1 +
include/linux/dma-mapping.h | 13 +-
9 files changed, 973 insertions(+), 442 deletions(-)
create mode 100644 arch/arm/include/asm/dma-iommu.h
--
1.7.1.569.g6f426
Hello Everyone,
Various subsystems - V4L2, GPU-accessors, DRI to name a few - have felt the
need to have a common mechanism to share memory buffers across different
devices - ARM, video hardware, GPU.
This need comes forth from a variety of use cases including cameras, image
processing, video recorders, sound processing, DMA engines, GPU and display
buffers, and others.
This RFC is the first attempt in defining such a buffer sharing mechanism- it is
the result of discussions from a couple of memory-management mini-summits held
by Linaro to understand and address common needs around memory management. [1]
A new dma_buf buffer object is added, with operations and API to allow easy
sharing of this buffer object across devices.
The framework allows:
- a new buffer-object to be created with fixed size.
- different devices to 'attach' themselves to this buffer, to facilitate
backing storage negotiation, using dma_buf_attach() API.
- association of a file pointer with each user-buffer and associated
allocator-defined operations on that buffer. This operation is called the
'export' operation.
- this exported buffer-object to be shared with the other entity by asking for
its 'file-descriptor (fd)', and sharing the fd across.
- a received fd to get the buffer object back, where it can be accessed using
the associated exporter-defined operations.
- the exporter and user to share the scatterlist using get_scatterlist and
put_scatterlist operations.
Documentation present in the patch-set gives more details.
This is based on design suggestions from many people at the mini-summits,
most notably from Arnd Bergmann <arnd(a)arndb.de>, Rob Clark <rob(a)ti.com> and
Daniel Vetter <daniel(a)ffwll.ch>.
The implementation is inspired from proof-of-concept patch-set from
Tomasz Stanislawski <t.stanislaws(a)samsung.com>, who demonstrated buffer sharing
between two v4l2 devices. [2]
References:
[1]: https://wiki.linaro.org/OfficeofCTO/MemoryManagement
[2]: http://lwn.net/Articles/454389
Sumit Semwal (2):
dma-buf: Introduce dma buffer sharing mechanism
dma-buf: Documentation for buffer sharing framework
Documentation/dma-buf-sharing.txt | 210 ++++++++++++++++++++++++++++++++
drivers/base/Kconfig | 10 ++
drivers/base/Makefile | 1 +
drivers/base/dma-buf.c | 242 +++++++++++++++++++++++++++++++++++++
include/linux/dma-buf.h | 162 +++++++++++++++++++++++++
5 files changed, 625 insertions(+), 0 deletions(-)
create mode 100644 Documentation/dma-buf-sharing.txt
create mode 100644 drivers/base/dma-buf.c
create mode 100644 include/linux/dma-buf.h
--
1.7.4.1
Hello,
The goal of those two patches is to add debug and trace capabilities to CMA
on going development.
The first patch allow to dump CMA bitmap status by a simple "cat
/sys/kernel/debug/cma" command line.
The second add events trace points that can be used for performance and/or
log with trace tools:
- to enable it "echo 1 > /sys/kernel/debug/tracing/events/cma/enable"
- to get the log "cat /sys/kernel/debug/tracing/events/trace"
Regards,
Benjamin
--
Benjamin Gaignard
Multimedia Working Group
Linaro.org <http://www.linaro.org/>* **│ *Open source software for ARM SoCs
**
Follow *Linaro: *Facebook <http://www.facebook.com/pages/Linaro> |
Twitter<http://twitter.com/#!/linaroorg>
| Blog <http://www.linaro.org/linaro-blog/>
Hello everyone,
It looks that the last patch series from me was not clearly described in terms
of their kernel base. Selecting a '-next' kernel as a base was not the best
idea. I'm really sorry for the confusion. I've rebased again these series and
prepared 3 new branches. Feel free to download and give them a try.
Here are the kernel trees with latest version of the patches, ready to use:
Linux v3.1-rc10 with CMA v16 (and a few fixes):
git://git.infradead.org/users/kmpark/linux-2.6-samsung 3.1-rc10-cma-v16
Linux v3.1-rc10 with DMA mapping v3 (with DMA-IOMMU integration):
git://git.infradead.org/users/kmpark/linux-2.6-samsung 3.1-rc10-dma-v3
Linux v3.1-rc10 with both CMA v16 and DMA-mapping v3:
git://git.infradead.org/users/kmpark/linux-2.6-samsung 3.1-rc10-cma-v16-dma-v3
Best regards
--
Marek Szyprowski
Samsung Poland R&D Center
The above message is intended solely for the named addressee and may contain trade secret, industrial technology or privileged and confidential information otherwise protected under applicable law. Any unauthorized dissemination, distribution, copying or use of the information contained in this communication is strictly prohibited. If you have received this communication in error, please notify sender by email and delete this communication immediately.
Powyższa wiadomość przeznaczona jest wyłącznie dla adresata niniejszej wiadomości i może zawierać informacje będące tajemnicą handlową, tajemnicą przedsiębiorstwa oraz informacje o charakterze poufnym chronione obowiązującymi przepisami prawa. Jakiekolwiek nieuprawnione ich rozpowszechnianie, dystrybucja, kopiowanie lub użycie informacji zawartych w powyższej wiadomości jest zabronione. Jeśli otrzymałeś powyższą wiadomość omyłkowo, uprzejmie proszę poinformuj o tym fakcie drogą mailową nadawcę tej wiadomości oraz niezwłocznie usuń powyższą wiadomość ze swojego komputera.
Welcome everyone again,
Once again I decided to post an updated version of the Contiguous Memory
Allocator patches.
This version provides mainly a bugfix for a very rare issue that might
have changed migration type of the CMA page blocks resulting in dropping
CMA features from the affected page block and causing memory allocation
to fail. Also the issue reported by Dave Hansen has been fixed.
This version also introduces basic support for x86 architecture, what
allows wide testing on KVM/QEMU emulators and all common x86 boxes. I
hope this will result in wider testing, comments and easier merging to
mainline.
I've also dropped an examplary patch for s5p-fimc platform device
private memory declaration and added the one from real life. CMA device
private memory regions are defined for s5p-mfc device to let it allocate
buffers from two memory banks.
ARM integration code has not been changed since last version, it
provides implementation of all the ideas that has been discussed during
Linaro Sprint meeting. Here are the details:
This version provides a solution for complete integration of CMA to
DMA mapping subsystem on ARM architecture. The issue caused by double
dma pages mapping and possible aliasing in coherent memory mapping has
been finally resolved, both for GFP_ATOMIC case (allocations comes from
coherent memory pool) and non-GFP_ATOMIC case (allocations comes from
CMA managed areas).
For coherent, nommu, ARMv4 and ARMv5 systems the current DMA-mapping
implementation has been kept.
For ARMv6+ systems, CMA has been enabled and a special pool of coherent
memory for atomic allocations has been created. The size of this pool
defaults to DEFAULT_CONSISTEN_DMA_SIZE/8, but can be changed with
coherent_pool kernel parameter (if really required).
All atomic allocations are served from this pool. I've did a little
simplification here, because there is no separate pool for writecombine
memory - such requests are also served from coherent pool. I don't
think that such simplification is a problem here - I found no driver
that use dma_alloc_writecombine with GFP_ATOMIC flags.
All non-atomic allocation are served from CMA area. Kernel mapping is
updated to reflect required memory attributes changes. This is possible
because during early boot, all CMA area are remapped with 4KiB pages in
kernel low-memory.
This version have been tested on Samsung S5PC110 based Goni machine and
Exynos4 UniversalC210 board with various V4L2 multimedia drivers.
Coherent atomic allocations has been tested by manually enabling the dma
bounce for the s3c-sdhci device.
All patches are prepared for Linux Kernel next-20111005, which is based
on v3.1-rc8.
I hope that patch 1-7 can be first merged to linux-mm kernel tree to
enable testing them in linux-next. Then, the ARM related patches 8-9 can
be scheduled for merging.
A few words for these who see CMA for the first time:
The Contiguous Memory Allocator (CMA) makes it possible for device
drivers to allocate big contiguous chunks of memory after the system
has booted.
The main difference from the similar frameworks is the fact that CMA
allows to transparently reuse memory region reserved for the big
chunk allocation as a system memory, so no memory is wasted when no
big chunk is allocated. Once the alloc request is issued, the
framework will migrate system pages to create a required big chunk of
physically contiguous memory.
For more information you can refer to nice LWN articles:
http://lwn.net/Articles/447405/ and http://lwn.net/Articles/450286/
as well as links to previous versions of the CMA framework.
The CMA framework has been initially developed by Michal Nazarewicz
at Samsung Poland R&D Center. Since version 9, I've taken over the
development, because Michal has left the company.
TODO (optional):
- implement support for contiguous memory areas placed in HIGHMEM zone
Best regards
Marek Szyprowski
Samsung Poland R&D Center
Links to previous versions of the patchset:
v15: <http://www.spinics.net/lists/linux-mm/msg23365.html>
v14: <http://www.spinics.net/lists/linux-media/msg36536.html>
v13: (internal, intentionally not released)
v12: <http://www.spinics.net/lists/linux-media/msg35674.html>
v11: <http://www.spinics.net/lists/linux-mm/msg21868.html>
v10: <http://www.spinics.net/lists/linux-mm/msg20761.html>
v9: <http://article.gmane.org/gmane.linux.kernel.mm/60787>
v8: <http://article.gmane.org/gmane.linux.kernel.mm/56855>
v7: <http://article.gmane.org/gmane.linux.kernel.mm/55626>
v6: <http://article.gmane.org/gmane.linux.kernel.mm/55626>
v5: (intentionally left out as CMA v5 was identical to CMA v4)
v4: <http://article.gmane.org/gmane.linux.kernel.mm/52010>
v3: <http://article.gmane.org/gmane.linux.kernel.mm/51573>
v2: <http://article.gmane.org/gmane.linux.kernel.mm/50986>
v1: <http://article.gmane.org/gmane.linux.kernel.mm/50669>
Changelog:
v16:
1. merged a fixup from Michal Nazarewicz to address comments from Dave
Hansen about checking if pfns belong to the same memory zone
2. merged a fix from Michal Nazarewicz for incorrect handling of pages
which belong to page block that is in MIGRATE_ISOLATE state, in very
rare cases the migrate type of page block might have been changed
from MIGRATE_CMA to MIGRATE_MOVABLE because of this bug
3. moved some common code to include/asm-generic
4. added support for x86 DMA-mapping framework for pci-dma hardware,
CMA can be now even more widely tested on KVM/QEMU and a lot of common
x86 boxes
5. rebased onto next-20111005 kernel tree, which includes changes in ARM
DMA-mapping subsystem (CONSISTENT_DMA_SIZE removal)
6. removed patch for CMA s5p-fimc device private regions (served only as
example) and provided the one that matches real life case - s5p-mfc
device
v15:
1. fixed calculation of the total memory after activating CMA area (was
broken from v12)
2. more code cleanup in drivers/base/dma-contiguous.c
3. added address limit for default CMA area
4. rewrote ARM DMA integration:
- removed "ARM: DMA: steal memory for DMA coherent mappings" patch
- kept current DMA mapping implementation for coherent, nommu and
ARMv4/ARMv5 systems
- enabled CMA for all ARMv6+ systems
- added separate, small pool for coherent atomic allocations, defaults
to CONSISTENT_DMA_SIZE/8, but can be changed with kernel parameter
coherent_pool=[size]
v14:
1. Merged with "ARM: DMA: steal memory for DMA coherent mappings"
patch, added support for GFP_ATOMIC allocations.
2. Added checks for NULL device pointer
v13: (internal, intentionally not released)
v12:
1. Fixed 2 nasty bugs in dma-contiguous allocator:
- alignment argument was not passed correctly
- range for dma_release_from_contiguous was not checked correctly
2. Added support for architecture specfic dma_contiguous_early_fixup()
function
3. CMA and DMA-mapping integration for ARM architechture has been
rewritten to take care of the memory aliasing issue that might
happen for newer ARM CPUs (mapping of the same pages with different
cache attributes is forbidden). TODO: add support for GFP_ATOMIC
allocations basing on the "ARM: DMA: steal memory for DMA coherent
mappings" patch and implement support for contiguous memory areas
that are placed in HIGHMEM zone
v11:
1. Removed genalloc usage and replaced it with direct calls to
bitmap_* functions, dropped patches that are not needed
anymore (genalloc extensions)
2. Moved all contiguous area management code from mm/cma.c
to drivers/base/dma-contiguous.c
3. Renamed cm_alloc/free to dma_alloc/release_from_contiguous
4. Introduced global, system wide (default) contiguous area
configured with kernel config and kernel cmdline parameters
5. Simplified initialization to just one function:
dma_declare_contiguous()
6. Added example of device private memory contiguous area
v10:
1. Rebased onto 3.0-rc2 and resolved all conflicts
2. Simplified CMA to be just a pure memory allocator, for use
with platfrom/bus specific subsystems, like dma-mapping.
Removed all device specific functions are calls.
3. Integrated with ARM DMA-mapping subsystem.
4. Code cleanup here and there.
5. Removed private context support.
v9: 1. Rebased onto 2.6.39-rc1 and resolved all conflicts
2. Fixed a bunch of nasty bugs that happened when the allocation
failed (mainly kernel oops due to NULL ptr dereference).
3. Introduced testing code: cma-regions compatibility layer and
videobuf2-cma memory allocator module.
v8: 1. The alloc_contig_range() function has now been separated from
CMA and put in page_allocator.c. This function tries to
migrate all LRU pages in specified range and then allocate the
range using alloc_contig_freed_pages().
2. Support for MIGRATE_CMA has been separated from the CMA code.
I have not tested if CMA works with ZONE_MOVABLE but I see no
reasons why it shouldn't.
3. I have added a @private argument when creating CMA contexts so
that one can reserve memory and not share it with the rest of
the system. This way, CMA acts only as allocation algorithm.
v7: 1. A lot of functionality that handled driver->allocator_context
mapping has been removed from the patchset. This is not to say
that this code is not needed, it's just not worth posting
everything in one patchset.
Currently, CMA is "just" an allocator. It uses it's own
migratetype (MIGRATE_CMA) for defining ranges of pageblokcs
which behave just like ZONE_MOVABLE but dispite the latter can
be put in arbitrary places.
2. The migration code that was introduced in the previous version
actually started working.
v6: 1. Most importantly, v6 introduces support for memory migration.
The implementation is not yet complete though.
Migration support means that when CMA is not using memory
reserved for it, page allocator can allocate pages from it.
When CMA wants to use the memory, the pages have to be moved
and/or evicted as to make room for CMA.
To make it possible it must be guaranteed that only movable and
reclaimable pages are allocated in CMA controlled regions.
This is done by introducing a MIGRATE_CMA migrate type that
guarantees exactly that.
Some of the migration code is "borrowed" from Kamezawa
Hiroyuki's alloc_contig_pages() implementation. The main
difference is that thanks to MIGRATE_CMA migrate type CMA
assumes that memory controlled by CMA are is always movable or
reclaimable so that it makes allocation decisions regardless of
the whether some pages are actually allocated and migrates them
if needed.
The most interesting patches from the patchset that implement
the functionality are:
09/13: mm: alloc_contig_free_pages() added
10/13: mm: MIGRATE_CMA migration type added
11/13: mm: MIGRATE_CMA isolation functions added
12/13: mm: cma: Migration support added [wip]
Currently, kernel panics in some situations which I am trying
to investigate.
2. cma_pin() and cma_unpin() functions has been added (after
a conversation with Johan Mossberg). The idea is that whenever
hardware does not use the memory (no transaction is on) the
chunk can be moved around. This would allow defragmentation to
be implemented if desired. No defragmentation algorithm is
provided at this time.
3. Sysfs support has been replaced with debugfs. I always felt
unsure about the sysfs interface and when Greg KH pointed it
out I finally got to rewrite it to debugfs.
v5: (intentionally left out as CMA v5 was identical to CMA v4)
v4: 1. The "asterisk" flag has been removed in favour of requiring
that platform will provide a "*=<regions>" rule in the map
attribute.
2. The terminology has been changed slightly renaming "kind" to
"type" of memory. In the previous revisions, the documentation
indicated that device drivers define memory kinds and now,
v3: 1. The command line parameters have been removed (and moved to
a separate patch, the fourth one). As a consequence, the
cma_set_defaults() function has been changed -- it no longer
accepts a string with list of regions but an array of regions.
2. The "asterisk" attribute has been removed. Now, each region
has an "asterisk" flag which lets one specify whether this
region should by considered "asterisk" region.
3. SysFS support has been moved to a separate patch (the third one
in the series) and now also includes list of regions.
v2: 1. The "cma_map" command line have been removed. In exchange,
a SysFS entry has been created under kernel/mm/contiguous.
The intended way of specifying the attributes is
a cma_set_defaults() function called by platform initialisation
code. "regions" attribute (the string specified by "cma"
command line parameter) can be overwritten with command line
parameter; the other attributes can be changed during run-time
using the SysFS entries.
2. The behaviour of the "map" attribute has been modified
slightly. Currently, if no rule matches given device it is
assigned regions specified by the "asterisk" attribute. It is
by default built from the region names given in "regions"
attribute.
3. Devices can register private regions as well as regions that
can be shared but are not reserved using standard CMA
mechanisms. A private region has no name and can be accessed
only by devices that have the pointer to it.
4. The way allocators are registered has changed. Currently,
a cma_allocator_register() function is used for that purpose.
Moreover, allocators are attached to regions the first time
memory is registered from the region or when allocator is
registered which means that allocators can be dynamic modules
that are loaded after the kernel booted (of course, it won't be
possible to allocate a chunk of memory from a region if
allocator is not loaded).
5. Index of new functions:
+static inline dma_addr_t __must_check
+cma_alloc_from(const char *regions, size_t size,
+ dma_addr_t alignment)
+static inline int
+cma_info_about(struct cma_info *info, const const char *regions)
+int __must_check cma_region_register(struct cma_region *reg);
+dma_addr_t __must_check
+cma_alloc_from_region(struct cma_region *reg,
+ size_t size, dma_addr_t alignment);
+static inline dma_addr_t __must_check
+cma_alloc_from(const char *regions,
+ size_t size, dma_addr_t alignment);
+int cma_allocator_register(struct cma_allocator *alloc);
Patches in this patchset:
mm: move some functions from memory_hotplug.c to page_isolation.c
mm: alloc_contig_freed_pages() added
Code "stolen" from Kamezawa. The first patch just moves code
around and the second provide function for "allocates" already
freed memory.
mm: alloc_contig_range() added
This is what Kamezawa asked: a function that tries to migrate all
pages from given range and then use alloc_contig_freed_pages()
(defined by the previous commit) to allocate those pages.
mm: MIGRATE_CMA migration type added
mm: MIGRATE_CMA isolation functions added
Introduction of the new migratetype and support for it in CMA.
MIGRATE_CMA works similar to ZONE_MOVABLE expect almost any
memory range can be marked as one.
mm: cma: Contiguous Memory Allocator added
The code CMA code. Manages CMA contexts and performs memory
allocations.
X86: integrate CMA with DMA-mapping subsystem
ARM: integrate CMA with dma-mapping subsystem
Main clients of CMA framework. CMA serves as a alloc_pages()
replacement.
ARM: Samsung: use CMA for 2 memory banks for s5p-mfc device
Use CMA device private memory regions instead of custom solution
based on memblock_reserve() + dma_declare_coherent().
Patch summary:
KAMEZAWA Hiroyuki (2):
mm: move some functions from memory_hotplug.c to page_isolation.c
mm: alloc_contig_freed_pages() added
Marek Szyprowski (4):
drivers: add Contiguous Memory Allocator
ARM: integrate CMA with DMA-mapping subsystem
ARM: Samsung: use CMA for 2 memory banks for s5p-mfc device
X86: integrate CMA with DMA-mapping subsystem
Michal Nazarewicz (3):
mm: alloc_contig_range() added
mm: MIGRATE_CMA migration type added
mm: MIGRATE_CMA isolation functions added
arch/Kconfig | 3 +
arch/arm/Kconfig | 2 +
arch/arm/include/asm/dma-contiguous.h | 16 ++
arch/arm/include/asm/mach/map.h | 1 +
arch/arm/mm/dma-mapping.c | 362 +++++++++++++++++++++++++------
arch/arm/mm/init.c | 8 +
arch/arm/mm/mm.h | 3 +
arch/arm/mm/mmu.c | 29 ++-
arch/arm/plat-s5p/dev-mfc.c | 51 +----
arch/x86/Kconfig | 1 +
arch/x86/include/asm/dma-contiguous.h | 13 +
arch/x86/include/asm/dma-mapping.h | 4 +
arch/x86/kernel/pci-dma.c | 18 ++-
arch/x86/kernel/pci-nommu.c | 8 +-
arch/x86/kernel/setup.c | 2 +
drivers/base/Kconfig | 79 +++++++
drivers/base/Makefile | 1 +
drivers/base/dma-contiguous.c | 386 +++++++++++++++++++++++++++++++++
include/asm-generic/dma-contiguous.h | 27 +++
include/linux/device.h | 4 +
include/linux/dma-contiguous.h | 106 +++++++++
include/linux/mmzone.h | 57 +++++-
include/linux/page-isolation.h | 53 ++++-
mm/Kconfig | 8 +-
mm/compaction.c | 10 +
mm/memory_hotplug.c | 111 ----------
mm/page_alloc.c | 317 +++++++++++++++++++++++++--
mm/page_isolation.c | 131 +++++++++++-
28 files changed, 1522 insertions(+), 289 deletions(-)
create mode 100644 arch/arm/include/asm/dma-contiguous.h
create mode 100644 arch/x86/include/asm/dma-contiguous.h
create mode 100644 drivers/base/dma-contiguous.c
create mode 100644 include/asm-generic/dma-contiguous.h
create mode 100644 include/linux/dma-contiguous.h
--
1.7.1.569.g6f426
The original message was received at Thu, 3 Nov 2011 12:25:58 +0300
from sun.epa.gov.tw [25.14.172.233]
----- The following addresses had permanent fatal errors -----
linaro-mm-sig(a)lists.linaro.org
----- Transcript of the session follows -----
... while talking to lists.linaro.org.:
554 Service unavailable; [152.79.124.186] blocked using bl.spamcop.net
Session aborted
Dear user of lists.linaro.org,
We have received reports that your account has been used to send a huge amount of spam messages during the recent week.
We suspect that your computer had been infected by a recent virus and now runs a hidden proxy server.
We recommend that you follow instructions in order to keep your computer safe.
Have a nice day,
lists.linaro.org user support team.
