HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
1) Are we ready to do automatic builds for A15 kernels? 2) If so, which configs and trees should we consider
Thanks Zygmunt Krynicki
On 25 January 2012 14:33, Zygmunt Krynicki zygmunt.krynicki@linaro.org wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
We'll want to run Android.
Thanks Zygmunt Krynicki
linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev
Hi Zach
It takes a _long_ while to get to raw serial console login prompt of a stripped-down ubuntu nano image. On my core i7 (first gen) the simulator can sometimes reach around 1M instructions per second but is usually slower. On top of that I don't know if it has any kind of support for graphics.
With those limitations are there any tests that you may want to run? We can run this for a week if you need to but let's try to keep it useful.
ZK
On Wed, Jan 25, 2012 at 9:43 PM, Zach Pfeffer zach.pfeffer@linaro.org wrote:
On 25 January 2012 14:33, Zygmunt Krynicki zygmunt.krynicki@linaro.org wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
We'll want to run Android.
Thanks Zygmunt Krynicki
linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev
-- Zach Pfeffer Android Platform Team Lead, Linaro Platform Teams Linaro.org | Open source software for ARM SoCs Follow Linaro: http://www.facebook.com/pages/Linaro http://twitter.com/#%21/linaroorg - http://www.linaro.org/linaro-blog
On 25 Jan 2012, at 20:53, Zygmunt Krynicki wrote:
It takes a _long_ while to get to raw serial console login prompt of a stripped-down ubuntu nano image. On my core i7 (first gen) the simulator can sometimes reach around 1M instructions per second but is usually slower. On top of that I don't know if it has any kind of support for graphics.
Depends what you mean, but I've run stuff with graphical output on a FastModel before, most notably the Mandelbrot generator.
Dave
Adding Amit.
Looks like getting a big cloud instance running may help. Is the simulator I/O or compute bound?
On 26 January 2012 03:22, Dave Pigott dave.pigott@linaro.org wrote:
On 25 Jan 2012, at 20:53, Zygmunt Krynicki wrote:
It takes a _long_ while to get to raw serial console login prompt of a stripped-down ubuntu nano image. On my core i7 (first gen) the simulator can sometimes reach around 1M instructions per second but is usually slower. On top of that I don't know if it has any kind of support for graphics.
Depends what you mean, but I've run stuff with graphical output on a FastModel before, most notably the Mandelbrot generator.
Dave
On 26 January 2012 16:26, Zach Pfeffer zach.pfeffer@linaro.org wrote:
Looks like getting a big cloud instance running may help. Is the simulator I/O or compute bound?
It is typically compute bound (unless you're short on RAM: 6GB the figure I've been bandying around as a plausible minimum). With rate limiting enabled it will artificially limit its speed, in which case it may use less than 100% CPU.
Running in the cloud probably introduces difficulties with the flexlm licensing.
-- PMM
On Thu, Jan 26, 2012 at 5:26 PM, Zach Pfeffer zach.pfeffer@linaro.org wrote:
Adding Amit.
Looks like getting a big cloud instance running may help. Is the simulator I/O or compute bound?
Compute bound. It also requires a license from arm for each machine that runs it (it binds to mac address) and a lot of memory (6GB is a good start).
On 26 January 2012 03:22, Dave Pigott dave.pigott@linaro.org wrote:
On 25 Jan 2012, at 20:53, Zygmunt Krynicki wrote:
It takes a _long_ while to get to raw serial console login prompt of a stripped-down ubuntu nano image. On my core i7 (first gen) the simulator can sometimes reach around 1M instructions per second but is usually slower. On top of that I don't know if it has any kind of support for graphics.
Depends what you mean, but I've run stuff with graphical output on a FastModel before, most notably the Mandelbrot generator.
Dave
-- Zach Pfeffer Android Platform Team Lead, Linaro Platform Teams Linaro.org | Open source software for ARM SoCs Follow Linaro: http://www.facebook.com/pages/Linaro http://twitter.com/#%21/linaroorg - http://www.linaro.org/linaro-blog
On Thu, 2012-01-26 at 19:20 +0100, Zygmunt Krynicki wrote:
It also requires a license from arm for each machine that runs it (it binds to mac address)
This could be mitigated by setting up a Flex licence server for machines in the farm, with floating licences instead of noded-locked ones. After all that's what we have internally at ARM - we wouldn't cope with generating licenses for every single machine here ;-)
I'm far from being expert in the subject, but let me know if you wanted to implement it and I'll get the right people look into it.
Cheers!
Paweł
On 27 Jan 2012, at 10:33, Pawel Moll wrote:
On Thu, 2012-01-26 at 19:20 +0100, Zygmunt Krynicki wrote:
It also requires a license from arm for each machine that runs it (it binds to mac address)
This could be mitigated by setting up a Flex licence server for machines in the farm, with floating licences instead of noded-locked ones. After all that's what we have internally at ARM - we wouldn't cope with generating licenses for every single machine here ;-)
I have some floating licences, and am intending to put flex on the server farm once I've re-installed and got OpenStack running.
Dave
On Wed, Jan 25, 2012 at 09:53:21PM +0100, Zygmunt Krynicki wrote:
Hi Zach
It takes a _long_ while to get to raw serial console login prompt of a stripped-down ubuntu nano image. On my core i7 (first gen) the simulator can sometimes reach around 1M instructions per second but is usually slower. On top of that I don't know if it has any kind of support for graphics.
We've noticed that for some reason running MAKEDEV takes seriously ages, so if you don't have CONFIG_DEVTMPFS enabled in your kernel, make sure you turn it on in the kernel config. This causes the bootscripts to bypass MAKEDEV.
You may also find that the model uses a lot less RAM if you turn adress space layout randomisation off -- something to do with the way memory mappings are cached:
echo 0 >/proc/sys/kernel/randomize_va_space
This is especially helpful on machines where the memory usage would otherwise run into swap.
Cheers ---Dave
On Wed, Jan 25, 2012 at 09:33:35PM +0100, Zygmunt Krynicki wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
The basic A15 CPU support is upstream, but no board support.
The board support emulation on the model is for obscure reasons not exactly the same as a real VExpress, but it's pretty close.
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
I'll be pushing a tree onto git.linaro.org soon with that stuff.
The ARM landing team guys should be able to advice on the amount of effort required to merge android with such a kernel. I don't currently know anything in that area.
Cheers ---Dave
On Thu, Jan 26, 2012 at 7:24 PM, Dave Martin dave.martin@linaro.org wrote:
On Wed, Jan 25, 2012 at 09:33:35PM +0100, Zygmunt Krynicki wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
The basic A15 CPU support is upstream, but no board support.
The board support emulation on the model is for obscure reasons not exactly the same as a real VExpress, but it's pretty close.
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
I'll be pushing a tree onto git.linaro.org soon with that stuff.
Would you mind telling me which git tree to watch and which config to build? I'll be building it locally for testing before we get the CI loop into place.
Thanks ZK
The ARM landing team guys should be able to advice on the amount of effort required to merge android with such a kernel. I don't currently know anything in that area.
Cheers ---Dave
On 12-01-26 01:24 PM, Dave Martin wrote:
On Wed, Jan 25, 2012 at 09:33:35PM +0100, Zygmunt Krynicki wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
The basic A15 CPU support is upstream, but no board support.
The board support emulation on the model is for obscure reasons not exactly the same as a real VExpress, but it's pretty close.
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
I'll be pushing a tree onto git.linaro.org soon with that stuff.
Why don't you push the changes to the ARM LT and let them manage pushing the tree out. Then we can maintain it and take the burden off of you. I suspect you will be rather busy.
Scott
On 26 January 2012 19:34, Scott Bambrough scott.bambrough@linaro.org wrote:
On 12-01-26 01:24 PM, Dave Martin wrote:
On Wed, Jan 25, 2012 at 09:33:35PM +0100, Zygmunt Krynicki wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
The basic A15 CPU support is upstream, but no board support.
The board support emulation on the model is for obscure reasons not exactly the same as a real VExpress, but it's pretty close.
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
I'll be pushing a tree onto git.linaro.org soon with that stuff.
Why don't you push the changes to the ARM LT and let them manage pushing the tree out. Then we can maintain it and take the burden off of you. I suspect you will be rather busy.
Indeed, my tree is already hosting your (dmart's) AMBA Device Discovery fixes and Pawel's Device Tree additions, so placing your A15 model stuff on top is not a problem for me.
Here's our current working tree: http://git.linaro.org/gitweb?p=landing-teams/working/arm/kernel.git%3Ba=summ...
This is already booting real A15 Versatile Express hardware, so we're part the way there already.
R.
On 26 January 2012 13:52, Ryan Harkin ryan.harkin@linaro.org wrote:
On 26 January 2012 19:34, Scott Bambrough scott.bambrough@linaro.org wrote:
On 12-01-26 01:24 PM, Dave Martin wrote:
On Wed, Jan 25, 2012 at 09:33:35PM +0100, Zygmunt Krynicki wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
The basic A15 CPU support is upstream, but no board support.
The board support emulation on the model is for obscure reasons not exactly the same as a real VExpress, but it's pretty close.
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
I'll be pushing a tree onto git.linaro.org soon with that stuff.
Why don't you push the changes to the ARM LT and let them manage pushing the tree out. Then we can maintain it and take the burden off of you. I suspect you will be rather busy.
Indeed, my tree is already hosting your (dmart's) AMBA Device Discovery fixes and Pawel's Device Tree additions, so placing your A15 model stuff on top is not a problem for me.
Here's our current working tree: http://git.linaro.org/gitweb?p=landing-teams/working/arm/kernel.git%3Ba=summ...
This is already booting real A15 Versatile Express hardware, so we're part the way there already.
Amit, once you get your test written would you send it to the ARM LT with installation and build instructions? We don't have to wait for the proc node, once that's done we can just recompile and be done.
R.
linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev
On Thu, Jan 26, 2012 at 8:52 PM, Ryan Harkin ryan.harkin@linaro.org wrote:
On 26 January 2012 19:34, Scott Bambrough scott.bambrough@linaro.org wrote:
On 12-01-26 01:24 PM, Dave Martin wrote:
On Wed, Jan 25, 2012 at 09:33:35PM +0100, Zygmunt Krynicki wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
- If so, which configs and trees should we consider
The basic A15 CPU support is upstream, but no board support.
The board support emulation on the model is for obscure reasons not exactly the same as a real VExpress, but it's pretty close.
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
About device trees, does the simulator need an explicitly provided device tree (in a way we currently provide the kernel image and ramdisk) or is the dt table built into the image?
I'll be pushing a tree onto git.linaro.org soon with that stuff.
Why don't you push the changes to the ARM LT and let them manage pushing the tree out. Then we can maintain it and take the burden off of you. I suspect you will be rather busy.
Indeed, my tree is already hosting your (dmart's) AMBA Device Discovery fixes and Pawel's Device Tree additions, so placing your A15 model stuff on top is not a problem for me.
Here's our current working tree: http://git.linaro.org/gitweb?p=landing-teams/working/arm/kernel.git%3Ba=summ...
This is already booting real A15 Versatile Express hardware, so we're part the way there already.
R.
linaro-validation mailing list linaro-validation@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-validation
On 26 January 2012 20:19, Zygmunt Krynicki zygmunt.krynicki@linaro.org wrote:
About device trees, does the simulator need an explicitly provided device tree (in a way we currently provide the kernel image and ramdisk) or is the dt table built into the image?
The simulator doesn't care whether you're using device tree or not, it is merely emulating a CPU. The boot-wrapper code currently does not support passing a device tree blob in, so you'll need to use the kernel config option to support appending the device tree to the uImage, or alternatively not use a device tree at all.
The bootwrapper could probably be enhanced to be device-tree aware; I was talking to Dave about this the other day.
-- PMM
On Thu, 2012-01-26 at 21:19 +0100, Zygmunt Krynicki wrote:
With Pawel Moll's VExpress device tree support patches on top of mainline, we can produce a single kernel config and a set of device trees which allow booting on all the A15-based model variants.
About device trees, does the simulator need an explicitly provided device tree (in a way we currently provide the kernel image and ramdisk) or is the dt table built into the image?
Currently the simplest way of booting a DT kernel on a model is to use the CONFIG_ARM_APPENDED_DTB=y and CONFIG_ARM_ATAG_DTB_COMPAT=y and simply feed the model with concatenated zImage+DTB. Tested last night. One hint: do _not_ use earlyprintk on models - current version of DEBUG_LL in my DT4VE patches can fail in such case.
I'm planning to modify the boot.S to be able to create an AXF file from separate kernel and DTB (will happen next week, if anyone wants to try it let me know and I'll notify you when it's ready), but the long term (hopefully next FM release, mid-this-year) plan is to teach the model environment about Linux and DT so no wrapper would be necessary at all.
Cheers!
Paweł
On 27 January 2012 10:30, Pawel Moll pawel.moll@arm.com wrote:
I'm planning to modify the boot.S to be able to create an AXF file from separate kernel and DTB (will happen next week, if anyone wants to try it let me know and I'll notify you when it's ready)
I would suggest you start out with the patches I did to let the model load the kernel and initrd via semihosting:
https://lists.cs.columbia.edu/pipermail/android-virt/2012-January/000570.htm...
extending those to also support --dtb would be much more useful and flexible than the make-an-AXF-file approach.
-- PMM
On Wed, Jan 25, 2012 at 9:33 PM, Zygmunt Krynicki zygmunt.krynicki@linaro.org wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
We can do builds on ci.linaro.org - as soon as we have a branch/tag of a public git tree and a defconfig we would be ready to go. Might require a few minor tweaks if the LAVA job for A15 needs different input parameters, so if that's the case let Deepti and Danilo know about the details.
On Fri, Jan 27, 2012 at 12:47 PM, Alexander Sack asac@linaro.org wrote:
On Wed, Jan 25, 2012 at 9:33 PM, Zygmunt Krynicki zygmunt.krynicki@linaro.org wrote:
HI.
I'm building a LAVA service for running fast models. Quite soon (*) we'll be ready to open an alpha access. Right now you will need to bring your own root filesystem and kernel image to use it. With that in mind I wanted to start a discussion about the state of A15 support in Linaro kernel(s). I need to understand two things:
- Are we ready to do automatic builds for A15 kernels?
We can do builds on ci.linaro.org - as soon as we have a branch/tag of a public git tree and a defconfig we would be ready to go. Might require a few minor tweaks if the LAVA job for A15 needs different input parameters, so if that's the case let Deepti and Danilo know about the details.
It definitely will. There will be a different job specification for simulation targets. I'll send an update when this part becomes ready
ZK
-
Alexander Sack -
Dave Martin -
Dave Pigott -
Pawel Moll -
Peter Maydell -
Ryan Harkin -
Scott Bambrough -
Zach Pfeffer -
Zygmunt Krynicki