Re: SMP related question
Hello Stavros, On 22/08/18 12:43, Passas, Stavros wrote: Hi all, My main objective was to have a way to run the rtems testsuite and have a script auto-compare the results with the corresponding scn file. Currently, just because core0 starts core1, seems like core1 always starts Init(), all results are flipped, thus this process cannot be automated. Functionality wise, everything runs as expected, I verified individually each core with the debugger, so I am ok with it. Also, this issue comes up in a different form also when the applications exits, because core1 (that had started Init(), exits with the expected return code, while all other cores terminate with RTEMS_FATAL_SOURCE_SMP/SMP_FATAL_SHUTDOWN_RESPONSE, which is handled as an error at the moment from the debugger/loader. For the moment I have added a fatal extension handler to avoid this case, but I am still looking if there is any better solution. I think a long term solution is to use a new test framework which produces machine readable output. https://devel.rtems.org/ticket/3199 The only machine readable output is currently the begin/end of test message. We try currently to get some budget for this. Please let me know if you are interested to contribute to something in this area. -- Sebastian Huber, embedded brains GmbH Address : Dornierstr. 4, D-82178 Puchheim, Germany Phone : +49 89 189 47 41-16 Fax : +49 89 189 47 41-09 E-Mail : sebastian.hu...@embedded-brains.de PGP : Public key available on request. Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG. ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users
RE: SMP related question
Hi all, My main objective was to have a way to run the rtems testsuite and have a script auto-compare the results with the corresponding scn file. Currently, just because core0 starts core1, seems like core1 always starts Init(), all results are flipped, thus this process cannot be automated. Functionality wise, everything runs as expected, I verified individually each core with the debugger, so I am ok with it. Also, this issue comes up in a different form also when the applications exits, because core1 (that had started Init(), exits with the expected return code, while all other cores terminate with RTEMS_FATAL_SOURCE_SMP/SMP_FATAL_SHUTDOWN_RESPONSE, which is handled as an error at the moment from the debugger/loader. For the moment I have added a fatal extension handler to avoid this case, but I am still looking if there is any better solution. Best Regards, Stavros From: users [mailto:users-boun...@rtems.org] On Behalf Of Joel Sherrill Sent: Monday, August 20, 2018 5:39 PM To: Sebastian Huber Cc: rtems-us...@rtems.org Subject: Re: SMP related question On Mon, Aug 20, 2018 at 1:23 AM, Sebastian Huber mailto:sebastian.hu...@embedded-brains.de>> wrote: On 12/08/18 00:41, Chris Johns wrote: On 10/08/2018 22:30, Sebastian Huber wrote: Hello Stavros, you can do this with a custom clustered scheduler configuration: https://docs.rtems.org/branches/master/c-user/configuring_a_system.html#clustered-scheduler-configuration I had a brief look over the documentation you provided but could not see how to achieve the result you are saying. Stavros, asked for two things. 1. A way to change properties of the initialization task which are not covered by configuration options https://docs.rtems.org/branches/master/c-user/configuring_a_system.html#classic-api-initialization-tasks-table-configuration before the initialization task runs. To do this some user provided code must execute after the initialization task is created and before it runs. My suggestion to use a device driver initialization function for this doesn't work. The initialization task is created after the device driver initialization. It can be done via a system initialization handler using the undocumented RTEMS_SYSINIT_ITEM(). Maybe also a high priority constructor, however, in this case the task already runs. The high priority constructor is too late if the goal is to avoid any activity except core 0. RTEMS_SYSINIT_ITEM() would work but adding a confdefs.h item for SMP configurations to pin the thread to a single core seems like a simple solution. 2. A way to pin a task to a particular processor. There are two options to do this in RTEMS. One is a clustered scheduler configuration, the other is the use of a thread to processor affinity. I think being able to pin the Init thread to a single core would be a nice confdefs.h feature. Not specifying a full cpuset, just the core number. I don't think this use case is enough by itself to force a user to clustered scheduling with core 0 an instance by itself. Alternatively, on a recent version the default SMP scheduler supports a one-to-one affinity. You can set it in a device driver. Is there an example of how to do this? As Sebastian points out, you can't do it in a device driver. But no, there is no example specifically for the Init thread. There are smptests which do this but likely no proper example of setting affinity, scheduler, etc. --joel Chris - Am 10. Aug 2018 um 14:10 schrieb Passas, Stavros stavros.pas...@intel.com<mailto:stavros.pas...@intel.com>: Hi, In my local SMP enabled RTEMS, I notice core1 starting the Init() function of the user application. Is there a way to enforce RTEMS to run the Init() in the boot processor, when SMP is enabled? Best Regards, Stavros -- Intel Research and Development Ireland Limited Registered in Ireland Registered Office: Collinstown Industrial Park, Leixlip, County Kildare Registered Number: 308263 This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies. ___ users mailing list users@rtems.org<mailto:users@rtems.org> http://lists.rtems.org/mailman/listinfo/users ___ users mailing list users@rtems.org<mailto:users@rtems.org> http://lists.rtems.org/mailman/listinfo/users ___ users mailing list users@rtems.org<mailto:users@rtems.org> http://lists.rtems.org/mailman/listinfo/users -- Sebastian Huber, embedded brains GmbH Address : Dornierstr. 4, D-82178 Puchheim, Germany Phone : +49 89 189 47 41-16 Fax : +49 89 189 47 41-09 E-Mail : sebastian.hu...@embedded-brains.de&l
Re: SMP related question
On Mon, Aug 20, 2018 at 1:23 AM, Sebastian Huber < sebastian.hu...@embedded-brains.de> wrote: > > > On 12/08/18 00:41, Chris Johns wrote: > >> On 10/08/2018 22:30, Sebastian Huber wrote: >> >>> Hello Stavros, >>> >>> you can do this with a custom clustered scheduler configuration: >>> >>> https://docs.rtems.org/branches/master/c-user/configuring_a_ >>> system.html#clustered-scheduler-configuration >>> >>> I had a brief look over the documentation you provided but could not see >> how to >> achieve the result you are saying. >> > > Stavros, asked for two things. > > 1. A way to change properties of the initialization task which are not > covered by configuration options > > https://docs.rtems.org/branches/master/c-user/configuring_a_ > system.html#classic-api-initialization-tasks-table-configuration > > before the initialization task runs. To do this some user provided code > must execute after the initialization task is created and before it runs. > My suggestion to use a device driver initialization function for this > doesn't work. The initialization task is created after the device driver > initialization. It can be done via a system initialization handler using > the undocumented RTEMS_SYSINIT_ITEM(). Maybe also a high priority > constructor, however, in this case the task already runs. > The high priority constructor is too late if the goal is to avoid any activity except core 0. RTEMS_SYSINIT_ITEM() would work but adding a confdefs.h item for SMP configurations to pin the thread to a single core seems like a simple solution. > > 2. A way to pin a task to a particular processor. There are two options to > do this in RTEMS. One is a clustered scheduler configuration, the other is > the use of a thread to processor affinity. I think being able to pin the Init thread to a single core would be a nice confdefs.h feature. Not specifying a full cpuset, just the core number. I don't think this use case is enough by itself to force a user to clustered scheduling with core 0 an instance by itself. > > > >> Alternatively, on a recent version the default SMP scheduler supports a >>> one-to-one affinity. You can set it in a device driver. >>> >> Is there an example of how to do this? >> > As Sebastian points out, you can't do it in a device driver. But no, there is no example specifically for the Init thread. There are smptests which do this but likely no proper example of setting affinity, scheduler, etc. --joel > >> Chris >> >> - Am 10. Aug 2018 um 14:10 schrieb Passas, Stavros >>> stavros.pas...@intel.com: >>> >>> Hi, In my local SMP enabled RTEMS, I notice core1 starting the Init() function of the user application. Is there a way to enforce RTEMS to run the Init() in the boot processor, when SMP is enabled? Best Regards, Stavros -- Intel Research and Development Ireland Limited Registered in Ireland Registered Office: Collinstown Industrial Park, Leixlip, County Kildare Registered Number: 308263 This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies. ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users >>> ___ >>> users mailing list >>> users@rtems.org >>> http://lists.rtems.org/mailman/listinfo/users >>> >>> ___ >> users mailing list >> users@rtems.org >> http://lists.rtems.org/mailman/listinfo/users >> > > -- > Sebastian Huber, embedded brains GmbH > > Address : Dornierstr. 4, D-82178 Puchheim, Germany > Phone : +49 89 189 47 41-16 > Fax : +49 89 189 47 41-09 > E-Mail : sebastian.hu...@embedded-brains.de > PGP : Public key available on request. > > Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG. > > > ___ > users mailing list > users@rtems.org > http://lists.rtems.org/mailman/listinfo/users > ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users
Re: SMP related question
On 12/08/18 00:41, Chris Johns wrote: On 10/08/2018 22:30, Sebastian Huber wrote: Hello Stavros, you can do this with a custom clustered scheduler configuration: https://docs.rtems.org/branches/master/c-user/configuring_a_system.html#clustered-scheduler-configuration I had a brief look over the documentation you provided but could not see how to achieve the result you are saying. Stavros, asked for two things. 1. A way to change properties of the initialization task which are not covered by configuration options https://docs.rtems.org/branches/master/c-user/configuring_a_system.html#classic-api-initialization-tasks-table-configuration before the initialization task runs. To do this some user provided code must execute after the initialization task is created and before it runs. My suggestion to use a device driver initialization function for this doesn't work. The initialization task is created after the device driver initialization. It can be done via a system initialization handler using the undocumented RTEMS_SYSINIT_ITEM(). Maybe also a high priority constructor, however, in this case the task already runs. 2. A way to pin a task to a particular processor. There are two options to do this in RTEMS. One is a clustered scheduler configuration, the other is the use of a thread to processor affinity. Alternatively, on a recent version the default SMP scheduler supports a one-to-one affinity. You can set it in a device driver. Is there an example of how to do this? Chris - Am 10. Aug 2018 um 14:10 schrieb Passas, Stavros stavros.pas...@intel.com: Hi, In my local SMP enabled RTEMS, I notice core1 starting the Init() function of the user application. Is there a way to enforce RTEMS to run the Init() in the boot processor, when SMP is enabled? Best Regards, Stavros -- Intel Research and Development Ireland Limited Registered in Ireland Registered Office: Collinstown Industrial Park, Leixlip, County Kildare Registered Number: 308263 This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies. ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users -- Sebastian Huber, embedded brains GmbH Address : Dornierstr. 4, D-82178 Puchheim, Germany Phone : +49 89 189 47 41-16 Fax : +49 89 189 47 41-09 E-Mail : sebastian.hu...@embedded-brains.de PGP : Public key available on request. Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG. ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users
Re: SMP related question
On 10/08/2018 22:30, Sebastian Huber wrote: > Hello Stavros, > > you can do this with a custom clustered scheduler configuration: > > https://docs.rtems.org/branches/master/c-user/configuring_a_system.html#clustered-scheduler-configuration > I had a brief look over the documentation you provided but could not see how to achieve the result you are saying. > Alternatively, on a recent version the default SMP scheduler supports a > one-to-one affinity. You can set it in a device driver. Is there an example of how to do this? Chris > > - Am 10. Aug 2018 um 14:10 schrieb Passas, Stavros > stavros.pas...@intel.com: > >> Hi, >> >> In my local SMP enabled RTEMS, I notice core1 starting the Init() function of >> the user application. >> Is there a way to enforce RTEMS to run the Init() in the boot processor, when >> SMP is enabled? >> >> Best Regards, >> Stavros >> -- >> Intel Research and Development Ireland Limited >> Registered in Ireland >> Registered Office: Collinstown Industrial Park, Leixlip, County Kildare >> Registered Number: 308263 >> >> >> This e-mail and any attachments may contain confidential material for the >> sole >> use of the intended recipient(s). Any review or distribution by others is >> strictly prohibited. If you are not the intended recipient, please contact >> the >> sender and delete all copies. >> >> ___ >> users mailing list >> users@rtems.org >> http://lists.rtems.org/mailman/listinfo/users > ___ > users mailing list > users@rtems.org > http://lists.rtems.org/mailman/listinfo/users > ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users
Re: SMP related question
Hello Stavros, you can do this with a custom clustered scheduler configuration: https://docs.rtems.org/branches/master/c-user/configuring_a_system.html#clustered-scheduler-configuration Alternatively, on a recent version the default SMP scheduler supports a one-to-one affinity. You can set it in a device driver. - Am 10. Aug 2018 um 14:10 schrieb Passas, Stavros stavros.pas...@intel.com: > Hi, > > In my local SMP enabled RTEMS, I notice core1 starting the Init() function of > the user application. > Is there a way to enforce RTEMS to run the Init() in the boot processor, when > SMP is enabled? > > Best Regards, > Stavros > -- > Intel Research and Development Ireland Limited > Registered in Ireland > Registered Office: Collinstown Industrial Park, Leixlip, County Kildare > Registered Number: 308263 > > > This e-mail and any attachments may contain confidential material for the sole > use of the intended recipient(s). Any review or distribution by others is > strictly prohibited. If you are not the intended recipient, please contact the > sender and delete all copies. > > ___ > users mailing list > users@rtems.org > http://lists.rtems.org/mailman/listinfo/users ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users
Re: SMP related question
On Fri, Aug 10, 2018, 7:11 AM Passas, Stavros wrote: > Hi, > > > > In my local SMP enabled RTEMS, I notice core1 starting the Init() function > of the user application. > I assume the boot processor is on core 0. How many cores? > Is there a way to enforce RTEMS to run the Init() in the boot processor, > when SMP is enabled? > Off the top of my head, all I can think of is using clustered scheduling and have a scheduler instance just for core 0. A potential alternative would be to have a way to set the affinity for the Init thread via a configuration setting like you can do stack size now. A work around would be to have the Init thread create another thread pinned to core 0 to do the initialization but that seems like a hack. > > Best Regards, > >Stavros > > -- > Intel Research and Development Ireland Limited > Registered in Ireland > Registered Office: Collinstown Industrial Park, Leixlip, County Kildare > Registered Number: 308263 > > This e-mail and any attachments may contain confidential material for the > sole use of the intended recipient(s). Any review or distribution by others > is strictly prohibited. If you are not the intended recipient, please > contact the sender and delete all copies. > > ___ > users mailing list > users@rtems.org > http://lists.rtems.org/mailman/listinfo/users ___ users mailing list users@rtems.org http://lists.rtems.org/mailman/listinfo/users