| Commit message (Collapse) | Author | Age |
| |
|
| |
|
|\
| |
| |
| | |
[dgit --overwrite]
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Only check online if local copy is not found.
Author: Sudip Mukherjee <sudipm.mukherjee@gmail.com>
Bug-Debian: https://bugs.debian.org/963364
Forwarded: not-needed
Gbp-Pq: Name use_local.patch
|
| |\
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
* Non-maintainer upload.
* Fix FTBFS. (Closes: #963364)
- Use local header file.
[dgit import unpatched libocxl 1.1.0-1.1]
|
| |
| |
| |
| | |
[dgit import tarball libocxl 1.1.0-1.1 libocxl_1.1.0-1.1.debian.tar.xz]
|
|
|
|
| |
[dgit import orig libocxl_1.1.0.orig.tar.gz]
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
There no path reference in the binary except in this part :
$ strings ./debian/libocxl1/usr/lib/powerpc64le-linux-gnu/libocxl.so.1.2 |grep build
CFLAGS: -g -O2 -ffile-prefix-map=/build/libocxl-EMICZo/libocxl-1.2.1=. -fstack-protector-strong -Wformat -Werror=format-security -I src/include -I kernel/include -fPIC -D_FILE_OFFSET_BITS=64
So, the reproducible options introduce that .. not sure if this is
supposed to happen
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
[git-debrebase changelog: new upstream 1.2.1]
|
|\
| |
| |
| | |
[git-debrebase anchor: new upstream 1.2.1, merge]
|
| |\
| | |
| | | |
Fix test build + prepare release 1.2.1
|
| | |
| | |
| | |
| | | |
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| | |
| | |
| | |
| | |
| | |
| | | |
I missed the update when tagging 1.2.0. So going straight to 1.2.1
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |/
| |
| |
| |
| |
| |
| |
| | |
We had a problem with the redefinition of _POSIX_C_SOURCE, which is
set in libocxl_internal.h. It needs to be defined before including
system header files.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |\
| | |
| | | |
A few fixes
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Remove a bunch of obsolete (and unused) variables from the doxygen
configuration files, since they were generating warnings with recent
versions of the tools.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| | |
| | |
| | |
| | | |
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |/
| |
| |
| |
| |
| |
| | |
The LIBOCXL_SYSPATH env variable can override the sys path. Can be
useful for containers
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Gnu make 4.3 introduced a backward-incompatible change in its handling
of the escape character before # when calling a function. And it is
used when writing some test code to check the latest version of the
ocxl header file (#include <misc/ocxl.h>).
This patch reworks how the test code is passed to the compiler to
avoid complications and be compatible with all version of make.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
It is perfectly legal to override CFLAGS on the command line,
but this doesn't work with libocxl:
$ make V=1 CFLAGS='-g'
VERSION_MAJOR=1 VERSION_MINOR=1 \
VERSION_PATCH=0 CC="gcc" CFLAGS="-g" \
./version.pl > src/libocxl_info.h
mkdir obj
gcc -g -c -o obj/afu.o src/afu.c
In file included from src/afu.c:17:
src/libocxl_internal.h:25:10: fatal error: libocxl.h: No such file or directory
25 | #include "libocxl.h"
| ^~~~~~~~~~~
compilation terminated.
This is especially painful for some building frameworks that rely on
that, eg. buildroot.
The GNU make documentation clearly indicates that overriding a variable
on the command line causes ordinary assignments to this variable in the
makefile to be ignored.
Use the override directive so that we can add the required items to
CFLAGS.
https://ftp.gnu.org/old-gnu/Manuals/make-3.79.1/html_chapter/make_6.html#SEC66
Signed-off-by: Greg Kurz <groug@kaod.org>
|
| |
| |
| |
| |
| |
| |
| | |
The memcpy AFU has been fixed. We don't need to insert a COPY before a
series of INCREMENT any more.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| | |
The memcpy AFU supports memory copy commands of size 64B to 2KB in 64B
increments.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The test does:
1. clear memory location
2. trigger the AFU to update memory location
3. poll memory location
The core may re-order 1 and 2 so we need an explicit memory barrier
between them.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| | |
* afutests: Add reset tests for AFUs IBM,MEMCPY3 and IBM,AFP3
Signed-off-by: Philippe Bergheaud <felix@linux.ibm.com>
|
| |\
| | |
| | | |
Add afuobj directory to .gitignore
|
| |/
| |
| |
| | |
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |\
| | |
| | | |
ocxl_afp3: Remove unused timeout parameter
|
| |/
| |
| |
| |
| |
| | |
Parameter -t (timeout) is not used by ocxl_afp3, so removing it.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |\
| | |
| | | |
new AFP latency test
|
| | |
| | |
| | |
| | | |
default value changed + add print of params values
|
| |/
| |
| |
| |
| |
| | |
Import of internal test.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
It's theoretically possible for the core to reorder some instructions
and start loading the source and destination buffers for the
comparison before the work element status is set to complete. The
hardware cannot guess the data dependency if we don't spell it out.
It seems a bit far-fetched here, as it would require some amount of
branch predictions. But it was apparently hit on a slightly simpler
version of this test used by HTX.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| |
| | |
curl doesn't follow link redirection by default. So pass the option to
do it from now on. No change needed for wget as it appears to follow
link redirection by default.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| | |
Link currently used has a redirection.
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| |
| | |
When ioctl() fails it always returns -1 and it puts the actual error in
errno.
Signed-off-by: Greg Kurz <groug@kaod.org>
|
| |
| |
| | |
Signed-off-by: Philippe Bergheaud <felix@linux.ibm.com>
|
| |
| |
| |
| |
| |
| | |
* 'ocxl_memcpy -p0' forks as many of processes as permitted by AFU
Signed-off-by: Philippe Bergheaud <felix@linux.ibm.com>
|
| |
| |
| |
| |
| |
| | |
Can be useful in case multiple cards with AFP are on the system
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| | |
libocxl doesn't return an errno value, so perror() shouldn't be used
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|
| |
| |
| |
| |
| |
| | |
This fixes issue #24.
Signed-off-by: Philippe Bergheaud <felix@linux.ibm.com>
|
| |
| |
| |
| | |
Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
|
| |
| |
| |
| | |
Signed-off-by: Frederic Barrat <fbarrat@linux.ibm.com>
|