| Message ID | 20210110053615.3594358-1-vanshikonda@os.amperecomputing.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64: Kconfig: Increase NR_CPUS default to 512 | expand |
On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: > From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> > > Increase the default value of NR_CPUS to 512 from 256. This will > enable the defconfig kernel to support platforms that have upto > 512 cores. Do we already support such a platform, and what is it? I'm fine with bumping the number, it's just nice to be able to say specifically _why_ we're dong it. Thanks, Will
On Mon, Jan 11, 2021 at 10:56:36AM +0000, Will Deacon wrote: >On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: >> From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> >> >> Increase the default value of NR_CPUS to 512 from 256. This will >> enable the defconfig kernel to support platforms that have upto >> 512 cores. > >Do we already support such a platform, and what is it? I'm fine with bumping >the number, it's just nice to be able to say specifically _why_ we're dong >it. I'm not aware of any publicly available systems that run into the 256 core limitation. At Ampere we have internal systems that would benefit from this change as they support more than 256 cores. Thanks, Vanshi > >Thanks, > >Will
On 1/11/21 9:57 AM, Vanshidhar Konda wrote: > On Mon, Jan 11, 2021 at 10:56:36AM +0000, Will Deacon wrote: >> On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: >>> From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> >>> >>> Increase the default value of NR_CPUS to 512 from 256. This will >>> enable the defconfig kernel to support platforms that have upto >>> 512 cores. >> >> Do we already support such a platform, and what is it? I'm fine with bumping. >> the number, it's just nice to be able to say specifically _why_ we're dong >> it. > > I'm not aware of any publicly available systems that run into the 256 > core limitation. At Ampere we have internal systems that would benefit > from this change as they support more than 256 cores. But what does that have to do with the default value? Do you expect to run defconfig kernels? I don't ever expect that.
On Mon, Jan 11, 2021 at 10:03:18AM -0800, Randy Dunlap wrote: >[EXTERNAL EMAIL NOTICE: This email originated from an external sender. Please be mindful of safe email handling and proprietary information protection practices.] > > >On 1/11/21 9:57 AM, Vanshidhar Konda wrote: >> On Mon, Jan 11, 2021 at 10:56:36AM +0000, Will Deacon wrote: >>> On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: >>>> From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> >>>> >>>> Increase the default value of NR_CPUS to 512 from 256. This will >>>> enable the defconfig kernel to support platforms that have upto >>>> 512 cores. >>> >>> Do we already support such a platform, and what is it? I'm fine with bumping. >>> the number, it's just nice to be able to say specifically _why_ we're dong >>> it. >> >> I'm not aware of any publicly available systems that run into the 256 >> core limitation. At Ampere we have internal systems that would benefit >> from this change as they support more than 256 cores. > >But what does that have to do with the default value? >Do you expect to run defconfig kernels? >I don't ever expect that. Sorry. I should have been more clear in my earlier statement. We currently have systems in development internally, to be available publicly later, that support more than 256 cores. Given the time it takes for a kernel version to be adopted by distros it makes sense to change the defconfig now rather than later. Thanks, Vanshi > >-- >~Randy >
On Mon, Jan 11, 2021 at 10:03:18AM -0800, Randy Dunlap wrote: > On 1/11/21 9:57 AM, Vanshidhar Konda wrote: > > On Mon, Jan 11, 2021 at 10:56:36AM +0000, Will Deacon wrote: > >> On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: > >>> From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> > >>> > >>> Increase the default value of NR_CPUS to 512 from 256. This will > >>> enable the defconfig kernel to support platforms that have upto > >>> 512 cores. > >> > >> Do we already support such a platform, and what is it? I'm fine with bumping. > >> the number, it's just nice to be able to say specifically _why_ we're dong > >> it. > > > > I'm not aware of any publicly available systems that run into the 256 > > core limitation. At Ampere we have internal systems that would benefit > > from this change as they support more than 256 cores. > > But what does that have to do with the default value? > Do you expect to run defconfig kernels? > I don't ever expect that. We still aim for the arm64 defconfig to run on all supported SoCs, even if not optimally. Distros indeed tweak the config to their needs.
On Mon, Jan 11, 2021 at 06:25:27PM +0000, Catalin Marinas wrote: >On Mon, Jan 11, 2021 at 10:03:18AM -0800, Randy Dunlap wrote: >> On 1/11/21 9:57 AM, Vanshidhar Konda wrote: >> > On Mon, Jan 11, 2021 at 10:56:36AM +0000, Will Deacon wrote: >> >> On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: >> >>> From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> >> >>> >> >>> Increase the default value of NR_CPUS to 512 from 256. This will >> >>> enable the defconfig kernel to support platforms that have upto >> >>> 512 cores. >> >> >> >> Do we already support such a platform, and what is it? I'm fine with bumping. >> >> the number, it's just nice to be able to say specifically _why_ we're dong >> >> it. >> > >> > I'm not aware of any publicly available systems that run into the 256 >> > core limitation. At Ampere we have internal systems that would benefit >> > from this change as they support more than 256 cores. >> >> But what does that have to do with the default value? >> Do you expect to run defconfig kernels? >> I don't ever expect that. > >We still aim for the arm64 defconfig to run on all supported SoCs, even >if not optimally. Distros indeed tweak the config to their needs. Would "all supported SoCs" mean only SoCs that are currently available publicly? Could we include support for SoCs/systems in development but to be available publicly in the next few years? Thanks, Vanshi > >-- >Catalin
On Mon, Jan 11, 2021 at 09:20:03PM -0800, Vanshidhar Konda wrote: > On Mon, Jan 11, 2021 at 06:25:27PM +0000, Catalin Marinas wrote: > > On Mon, Jan 11, 2021 at 10:03:18AM -0800, Randy Dunlap wrote: > > > On 1/11/21 9:57 AM, Vanshidhar Konda wrote: > > > > On Mon, Jan 11, 2021 at 10:56:36AM +0000, Will Deacon wrote: > > > >> On Sat, Jan 09, 2021 at 09:36:15PM -0800, vanshikonda@os.amperecomputing.com wrote: > > > >>> From: Vanshidhar Konda <vanshikonda@os.amperecomputing.com> > > > >>> > > > >>> Increase the default value of NR_CPUS to 512 from 256. This will > > > >>> enable the defconfig kernel to support platforms that have upto > > > >>> 512 cores. > > > >> > > > >> Do we already support such a platform, and what is it? I'm fine with bumping. > > > >> the number, it's just nice to be able to say specifically _why_ we're dong > > > >> it. > > > > > > > > I'm not aware of any publicly available systems that run into the 256 > > > > core limitation. At Ampere we have internal systems that would benefit > > > > from this change as they support more than 256 cores. > > > > > > But what does that have to do with the default value? > > > Do you expect to run defconfig kernels? > > > I don't ever expect that. > > > > We still aim for the arm64 defconfig to run on all supported SoCs, even > > if not optimally. Distros indeed tweak the config to their needs. > > Would "all supported SoCs" mean only SoCs that are currently available > publicly? Could we include support for SoCs/systems in development but > to be available publicly in the next few years? I don't really see the need to rush that into defconfig, so I'd prefer to make the update when something actually exists otherwise it's really hard to keep track of why we made the decision (especially as unreleased hardware is liable to change). Will
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index 05e17351e4f3..23fbbf413f58 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -983,7 +983,7 @@ config SCHED_SMT config NR_CPUS int "Maximum number of CPUs (2-4096)" range 2 4096 - default "256" + default "512" config HOTPLUG_CPU bool "Support for hot-pluggable CPUs"