> -Original Message-
> From: lng-odp [mailto:lng-odp-boun...@lists.linaro.org] On Behalf Of Bill
> Fischofer
> Sent: Wednesday, November 30, 2016 2:48 AM
> To: lng-odp@lists.linaro.org
> Subject: [lng-odp] [API-NEXT PATCH] linux-generic: pool: reset origin_qe
> on buffer allocation
>
> Res
Seems to resolve my drvshm crash issue.
Reviewed-by: Petri Savolainen
> -Original Message-
> From: Christophe Milard [mailto:christophe.mil...@linaro.org]
> Sent: Tuesday, November 29, 2016 4:46 PM
> To: maxim.uva...@linaro.com; Savolainen, Petri (Nokia - FI/Espoo)
> ; bill.fischo...@li
Thanks for your feedback Petri. Sorry you had to be the first "victim".
Christophe
On 30 November 2016 at 12:02, Savolainen, Petri (Nokia - FI/Espoo)
wrote:
> Seems to resolve my drvshm crash issue.
>
> Reviewed-by: Petri Savolainen
>
>
>> -Original Message-
>> From: Christophe Milard [
On 30 November 2016 at 04:07, Savolainen, Petri (Nokia - FI/Espoo) wrote:
> > -Original Message-
> > From: lng-odp [mailto:lng-odp-boun...@lists.linaro.org] On Behalf Of
> Bill
> > Fischofer
> > Sent: Wednesday, November 30, 2016 2:48 AM
> > To: lng-odp@lists.linaro.org
> > Subject: [lng-
Christophe
Do you have a way I can create the problem and then prove to myself you
fixed it by applying this?
Can I CTRL C any of our apps ?
Mike
On 30 November 2016 at 03:34, Christophe Milard <
christophe.mil...@linaro.org> wrote:
> ping
>
> This should be straightforward to review and would
This is fixing a problem that exists today and has existed since the
revised pool code was merged nine days ago. Matias' patch is already
waiting for a v2 so that can include any changes needed to adjust for
this. But I'd prefer to fix the current code and then work on more
radical improvements.
O
Added definitions for a look-a-side IPSEC offload API. In addition to
IPSEC packet transformations, it also supports:
* inbound SA look up
* outbound IP fragmentation
Signed-off-by: Petri Savolainen
---
Changes in v2:
* Specify that synchronous calls cannot process all packets
if output.num_pk
merged,
Maxim.
On 11/30/16 02:16, Bill Fischofer wrote:
On Fri, Nov 25, 2016 at 8:06 AM, Maxim Uvarov wrote:
exit variable has to be volatile to be visible from thread. Without
that background process is not killed with default optimization level.
Signed-off-by: Maxim Uvarov
Reviewed-by: Bi
Merged,
Maxim.
On 11/30/16 14:38, Christophe Milard wrote:
Thanks for your feedback Petri. Sorry you had to be the first "victim".
Christophe
On 30 November 2016 at 12:02, Savolainen, Petri (Nokia - FI/Espoo)
wrote:
Seems to resolve my drvshm crash issue.
Reviewed-by: Petri Savolainen
-
The internal ordered processing locking functions can be more streamlined
compared to the public API functions.
Signed-off-by: Matias Elo
---
platform/linux-generic/include/odp_schedule_if.h | 4
platform/linux-generic/odp_schedule.c| 12 +++-
platform/linux-generic/odp
Implement ordered locks using per lock atomic counters. The counter values
are compared to the queue’s atomic context to guarantee ordered locking.
Compared to the previous implementation this enables parallel processing of
ordered events outside of the lock context.
Signed-off-by: Matias Elo
---
The number of supported ordered locks may vary between the scheduler
implementations. Add an internal scheduler API call for fetching the
maximum value from currently active scheduler.
Add an internal definition CONFIG_QUEUE_MAX_ORD_LOCKS for the scheduler
independent maximum value.
Signed-off-by
V2:
- Support for multiple ordered locks (Bill)
- New ordered lock implementation
Add new implementation for ordered queues. Compared to the old
implementation this is much simpler and improves performance ~1-4x
depending on the test case. Some performance numbers are provided below.
The implemen
Add new implementation for ordered queues. Compared to the old
implementation this is much simpler and improves performance ~1-4x
depending on the test case.
The implementation is based on an atomic ordered context, which only a
single thread may possess at a time. Only the thread owning the atomi
Remove old ordered queue code. Replaced temporarily by atomic handling.
Signed-off-by: Matias Elo
---
V2:
- Refactored code to better obey the internal scheduler API
- Don't cache ordered events in scheduler (Bill)
platform/linux-generic/Makefile.am | 3 -
.../linux-generic/i
On 11/30/16 15:48, Mike Holmes wrote:
Christophe
Do you have a way I can create the problem and then prove to myself
you fixed it by applying this?
Can I CTRL C any of our apps ?
Mike
yes, Mike, I asked for that change.
run something like make check and watch for background processes wi
On 11/30/16 03:48, Bill Fischofer wrote:
Resolve bug https://bugs.linaro.org/show_bug.cgi?id=2622 by
re-initializing origin_qe to NULL when a buffer is allocated. This step
was omitted in the switch to ring pool allocation introduced in
commit ID c8cf1d87783d4b4c628f219803b78731b8d4ade4
Signed-o
original configure.ac enables abi compat mode by default,
even without --enable-abi-compat provided. And has broken
logic to disable abi compat mode. Correct logic to build abi
compat mode and option to disable it. Shared library is not
needed for non abi compat mode, so turn it off.
Signed-off-by
v9: update helper so version to 111.1.2, hope we red autotools docs correctly
(Mike)
Bill Fischofer (1):
changelog: summary of changes for odp v1.12.0.0
Maxim Uvarov (2):
configure.ac: disable shared library for non abi compat mode
configure.ac update version numbers
CHANGELOG| 177
Default is abi compat mode, all interface functions changed,
so increase first number of .so
Signed-off-by: Maxim Uvarov
---
configure.ac | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/configure.ac b/configure.ac
index b460a65..5da42eb 100644
--- a/configure.ac
+++ b/c
From: Bill Fischofer
Signed-off-by: Bill Fischofer
Signed-off-by: Maxim Uvarov
---
CHANGELOG | 177 ++
1 file changed, 177 insertions(+)
diff --git a/CHANGELOG b/CHANGELOG
index 1d652a8..17afe44 100644
--- a/CHANGELOG
+++ b/CHANGELOG
On Wed, Nov 30, 2016 at 8:09 AM, Maxim Uvarov wrote:
> On 11/30/16 03:48, Bill Fischofer wrote:
>>
>> Resolve bug https://bugs.linaro.org/show_bug.cgi?id=2622 by
>> re-initializing origin_qe to NULL when a buffer is allocated. This step
>> was omitted in the switch to ring pool allocation introduc
I think there is an issue, can anyone confirm
./configure --enable-test-vald --enable-user-guides --disable-abi-compat
make distcheck
make[4]: Entering directory
'/home/mike/git/odp/opendataplane-1.12.0.0/_build/sub/test/common_plat/performance'
CC odp_crypto-odp_crypto.o
CCLD odp
On 11/30/16 19:24, Mike Holmes wrote:
I think there is an issue, can anyone confirm
./configure --enable-test-vald --enable-user-guides --disable-abi-compat
make distcheck
make[4]: Entering directory
'/home/mike/git/odp/opendataplane-1.12.0.0/_build/sub/test/common_plat/performance'
CC
I tried again after git clean -xdf and it failed, it also failed in a new
docker instance so its not my env as far as I can easily tell.
Mike
On 30 November 2016 at 13:50, Maxim Uvarov wrote:
> On 11/30/16 19:24, Mike Holmes wrote:
>
>> I think there is an issue, can anyone confirm
>>
>> ./conf
Ok, worked on this with Anders, the bug was there before your patch, we can
look into it afterwards IMHO
On 30 November 2016 at 14:12, Mike Holmes wrote:
> I tried again after git clean -xdf and it failed, it also failed in a new
> docker instance so its not my env as far as I can easily tell.
Bill Fischofer (1):
changelog: summary of changes for odp v1.12.0.0
Maxim Uvarov (2):
configure.ac: disable shared library for non abi compat mode
configure.ac update version numbers
CHANGELOG| 177 +++
configure.ac | 15 +++--
original configure.ac enables abi compat mode by default,
even without --enable-abi-compat provided. And has broken
logic to disable abi compat mode. Correct logic to build abi
compat mode and option to disable it. Shared library is not
needed for non abi compat mode, so turn it off.
Signed-off-by
Default is abi compat mode, all interface functions changed,
so increase first number of .so
Signed-off-by: Maxim Uvarov
---
configure.ac | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/configure.ac b/configure.ac
index b460a65..fe7e47d 100644
--- a/configure.ac
+++ b/c
From: Bill Fischofer
Signed-off-by: Bill Fischofer
Signed-off-by: Maxim Uvarov
---
CHANGELOG | 177 ++
1 file changed, 177 insertions(+)
diff --git a/CHANGELOG b/CHANGELOG
index 1d652a8..17afe44 100644
--- a/CHANGELOG
+++ b/CHANGELOG
I just CC'ed you in Steve.
My head is spinning but I think we have this straight now, perhaps you have
time to sync with Maxim and possibly Anders if he has time to check this
from ytour Debian background?
I think if we can get the next couple of release out correctly the pattern
will establish
On 30 November 2016 at 22:02, Mike Holmes wrote:
> I just CC'ed you in Steve.
>
> My head is spinning but I think we have this straight now, perhaps you have
> time to sync with Maxim and possibly Anders if he has time to check this
> from ytour Debian background?
>
> I think if we can get the ne
Resolve bug https://bugs.linaro.org/show_bug.cgi?id=2622 by
re-initializing origin_qe to NULL when a buffer is allocated. This step
was omitted in the switch to ring pool allocation introduced in
commit ID c8cf1d87783d4b4c628f219803b78731b8d4ade4
Signed-off-by: Bill Fischofer
---
Changes in v2:
-
https://bugs.linaro.org/show_bug.cgi?id=2622
--- Comment #6 from Bill Fischofer ---
v2 patch submitted: http://patches.opendataplane.org/patch/7492/ incorporating
Maxim's review comments.
--
You are receiving this mail because:
You are on the CC list for the bug.
On Wed, Nov 30, 2016 at 7:56 AM, Matias Elo wrote:
> The number of supported ordered locks may vary between the scheduler
> implementations. Add an internal scheduler API call for fetching the
> maximum value from currently active scheduler.
>
> Add an internal definition CONFIG_QUEUE_MAX_ORD_LOCK
On Wed, Nov 30, 2016 at 7:56 AM, Matias Elo wrote:
> Add new implementation for ordered queues. Compared to the old
> implementation this is much simpler and improves performance ~1-4x
> depending on the test case.
>
> The implementation is based on an atomic ordered context, which only a
> single
yes, I agree this fix can help pass make check in recent development and
future bisect maybe.
Reviewed-and-tested-by: Yi He
On 1 December 2016 at 07:08, Bill Fischofer
wrote:
> Resolve bug https://bugs.linaro.org/show_bug.cgi?id=2622 by
> re-initializing origin_qe to NULL when a buffer is allo
Hi, thanks Matias
The example is very helpful, one question follow it:
Previously I understand atomic queue like multiple producer/single consumer
behaviour, so producers (enqueue threads) can still run in parallel?
But in the example the two producer threads behaves like sequentially while
enqu
On 1 Dec 2016, at 6:58, Yi He mailto:yi...@linaro.org>> wrote:
Hi, thanks Matias
The example is very helpful, one question follow it:
Previously I understand atomic queue like multiple producer/single consumer
behaviour, so producers (enqueue threads) can still run in parallel?
Yes. This actu
39 matches
Mail list logo