Like was done for read, provide the equivalent for write. Even if
current users are not in the hot path, this should future-proof it.
v2:
- Remove default from _Generic() - callers wanting to write more
than u64 should use iosys_map_memcpy_to()
- Add WRITE_ONCE() cases dereferencing the pointer when using system
memory
v3:
- Fix precedence issue when casting inside WRITE_ONCE(). By not using ()
around vaddr__ the offset was not part of the cast, but rather added
to it, producing a wrong address
- Remove compiletime_assert() as WRITE_ONCE() already contains it
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Reviewed-by: Christian König <christian.koenig@amd.com> # v1
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220628191016.3899428-2-lucas.demarchi@intel.com
Instead of always falling back to memcpy_fromio() for any size, prefer
using read{b,w,l}(). When reading struct members it's common to read
individual integer variables individually. Going through memcpy_fromio()
for each of them poses a high penalty.
Employ a similar trick as __seqprop() by using _Generic() to generate
only the specific call based on a type-compatible variable.
For a pariticular i915 workload producing GPU context switches,
__get_engine_usage_record() is particularly hot since the engine usage
is read from device local memory with dgfx, possibly multiple times
since it's racy. Test execution time for this test shows a ~12.5%
improvement with DG2:
Before:
nrepeats = 1000; min = 7.63243e+06; max = 1.01817e+07;
median = 9.52548e+06; var = 526149;
After:
nrepeats = 1000; min = 7.03402e+06; max = 8.8832e+06;
median = 8.33955e+06; var = 333113;
Other things attempted that didn't prove very useful:
1) Change the _Generic() on x86 to just dereference the memory address
2) Change __get_engine_usage_record() to do just 1 read per loop,
comparing with the previous value read
3) Change __get_engine_usage_record() to access the fields directly as it
was before the conversion to iosys-map
(3) did gave a small improvement (~3%), but doesn't seem to scale well
to other similar cases in the driver.
Additional test by Chris Wilson using gem_create from igt with some
changes to track object creation time. This happens to accidentally
stress this code path:
Pre iosys_map conversion of engine busyness:
lmem0: Creating 262144 4KiB objects took 59274.2ms
Unpatched:
lmem0: Creating 262144 4KiB objects took 108830.2ms
With readl (this patch):
lmem0: Creating 262144 4KiB objects took 61348.6ms
s/readl/READ_ONCE/
lmem0: Creating 262144 4KiB objects took 61333.2ms
So we do take a little bit more time than before the conversion, but
that is due to other factors: bringing the READ_ONCE back would be as
good as just doing this conversion.
v2:
- Remove default from _Generic() - callers wanting to read more
than u64 should use iosys_map_memcpy_from()
- Add READ_ONCE() cases dereferencing the pointer when using system
memory
v3:
- Fix precedence issue when casting inside READ_ONCE(). By not using ()
around vaddr__ the offset was not part of the cast, but rather added
to it, producing a wrong address
- Remove compiletime_assert() as READ_ONCE() already contains it
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Christian König <christian.koenig@amd.com> # v1
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220628191016.3899428-1-lucas.demarchi@intel.com
First the simplest ones:
- iosys_map_memset(): when abstracting system and I/O memory,
just like the memcpy() use case, memset() also has dedicated
functions to be called for using IO memory.
- iosys_map_memcpy_from(): we may need to copy data from I/O
memory, not only to.
In certain situations it's useful to be able to read or write to an
offset that is calculated by having the memory layout given by a struct
declaration. Usually we are going to read/write a u8, u16, u32 or u64.
As a pre-requisite for the implementation, add iosys_map_memcpy_from()
to be the equivalent of iosys_map_memcpy_to(), but in the other
direction. Then add 2 pairs of macros:
- iosys_map_rd() / iosys_map_wr()
- iosys_map_rd_field() / iosys_map_wr_field()
The first pair takes the C-type and offset to read/write. The second
pair uses a struct describing the layout of the mapping in order to
calculate the offset and size being read/written.
We could use readb, readw, readl, readq and the write* counterparts,
however due to alignment issues this may not work on all architectures.
If alignment needs to be checked to call the right function, it's not
possible to decide at compile-time which function to call: so just leave
the decision to the memcpy function that will do exactly that.
Finally, in order to use the above macros with a map derived from
another, add another initializer: IOSYS_MAP_INIT_OFFSET().
v2:
- Rework IOSYS_MAP_INIT_OFFSET() so it doesn't rely on aliasing rules
within the union
- Add offset to both iosys_map_rd_field() and iosys_map_wr_field() to
allow the struct itself to be at an offset from the mapping
- Add documentation to iosys_map_rd_field() with example and expected
memory layout
v3:
- Drop kernel.h include as it's not needed anymore
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Christian König <christian.koenig@amd.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: dri-devel@lists.freedesktop.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220216174147.3073235-3-lucas.demarchi@intel.com
Rename struct dma_buf_map to struct iosys_map and corresponding APIs.
Over time dma-buf-map grew up to more functionality than the one used by
dma-buf: in fact it's just a shim layer to abstract system memory, that
can be accessed via regular load and store, from IO memory that needs to
be acessed via arch helpers.
The idea is to extend this API so it can fulfill other needs, internal
to a single driver. Example: in the i915 driver it's desired to share
the implementation for integrated graphics, which uses mostly system
memory, with discrete graphics, which may need to access IO memory.
The conversion was mostly done with the following semantic patch:
@r1@
@@
- struct dma_buf_map
+ struct iosys_map
@r2@
@@
(
- DMA_BUF_MAP_INIT_VADDR
+ IOSYS_MAP_INIT_VADDR
|
- dma_buf_map_set_vaddr
+ iosys_map_set_vaddr
|
- dma_buf_map_set_vaddr_iomem
+ iosys_map_set_vaddr_iomem
|
- dma_buf_map_is_equal
+ iosys_map_is_equal
|
- dma_buf_map_is_null
+ iosys_map_is_null
|
- dma_buf_map_is_set
+ iosys_map_is_set
|
- dma_buf_map_clear
+ iosys_map_clear
|
- dma_buf_map_memcpy_to
+ iosys_map_memcpy_to
|
- dma_buf_map_incr
+ iosys_map_incr
)
@@
@@
- #include <linux/dma-buf-map.h>
+ #include <linux/iosys-map.h>
Then some files had their includes adjusted and some comments were
update to remove mentions to dma-buf-map.
Since this is not specific to dma-buf anymore, move the documentation to
the "Bus-Independent Device Accesses" section.
v2:
- Squash patches
v3:
- Fix wrong removal of dma-buf.h from MAINTAINERS
- Move documentation from dma-buf.rst to device-io.rst
v4:
- Change documentation title and level
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Sumit Semwal <sumit.semwal@linaro.org>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220204170541.829227-1-lucas.demarchi@intel.com
Randy Dunlap