| Message ID | 20200924123016.13427-1-ionela.voinescu@arm.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | cpufreq,topology,arm: disable FI for BL_SWITCHER | expand |
On Thu, Sep 24, 2020 at 2:30 PM Ionela Voinescu <ionela.voinescu@arm.com> wrote: > > This series is the result of the discussions ([1], [2]) around the > complications that the BL_SWITCHER poses when it comes to Frequency > Invariance (FI) and it aims to restart the discussions. > > To properly scale its per-entity load-tracking signals, the task > scheduler needs to be given a frequency scale factor, i.e. some image of > the current frequency the CPU is running at, relative to its maximum > frequency. > > But (reiterating the message in the changelog of patch 2/2), big.LITTLE > switching complicates the setting of a correct cpufreq-based frequency > invariance scale factor due to (as observed in > drivers/cpufreq/vexpress-spc-cpufreq.c): > - Incorrect current and maximum frequencies as a result of the > exposure of a virtual frequency table to the cpufreq core, > - Missed updates as a result of asynchronous frequency adjustments > caused by frequency changes in other CPU pairs. > More information on this feature can be found at [3]. > > Given that its functionality is atypical in regards to FI and that this > is an old technology, patch 2/2 disable FI for when big.LITTLE switching > is configured in to prevent incorrect scale setting. > > For this purpose patch 1/2 changes the way arch_set_freq_scale() is > defined in architecture code which brings it in line with the logic of > other architectural function definitions while allowing for less invasive > filtering of FI support. > > In the discussions at [2], three possible solutions were suggested: > - (1) conditioning FI by !CONFIG_BL_SWITCHER > - (2) leave as is with note in driver specifying this FI broken > functionality > - (3) removing full BL_SWITCHER support > > This series restructures the solution at (1). The reason for it is that > the new patch limits the ifdef filtering to the arm topology include file, > a location where frequency invariance functions are defined. Therefore, > this seems more appropriate given that the b.L switcher is an arm > technology and that the new FI filtering location seems more natural for > conditioned FI disabling. > > Solutions (2) and (3) were not implemented given that there might be some > remaining users of this technology (Samsung Chromebook 2 - Samsung Exynos > 5 Octa 5420, Samsung Exynos 5 Octa 5800) and therefore leaving this > broken (2) seems equally bad to removing support for it (3). > > [1] https://lore.kernel.org/lkml/20200701090751.7543-5-ionela.voinescu@arm.com/ > [2] https://lore.kernel.org/lkml/20200722093732.14297-4-ionela.voinescu@arm.com/ > [3] https://lwn.net/Articles/481055/ I can take this set with the ACKs from Viresh if that's fine by everyone. Catalin? Sudeep?
On Wed, Oct 07, 2020 at 04:34:44PM +0200, Rafael J. Wysocki wrote: > On Thu, Sep 24, 2020 at 2:30 PM Ionela Voinescu <ionela.voinescu@arm.com> wrote: > > > > This series is the result of the discussions ([1], [2]) around the > > complications that the BL_SWITCHER poses when it comes to Frequency > > Invariance (FI) and it aims to restart the discussions. > > > > To properly scale its per-entity load-tracking signals, the task > > scheduler needs to be given a frequency scale factor, i.e. some image of > > the current frequency the CPU is running at, relative to its maximum > > frequency. > > > > But (reiterating the message in the changelog of patch 2/2), big.LITTLE > > switching complicates the setting of a correct cpufreq-based frequency > > invariance scale factor due to (as observed in > > drivers/cpufreq/vexpress-spc-cpufreq.c): > > - Incorrect current and maximum frequencies as a result of the > > exposure of a virtual frequency table to the cpufreq core, > > - Missed updates as a result of asynchronous frequency adjustments > > caused by frequency changes in other CPU pairs. > > More information on this feature can be found at [3]. > > > > Given that its functionality is atypical in regards to FI and that this > > is an old technology, patch 2/2 disable FI for when big.LITTLE switching > > is configured in to prevent incorrect scale setting. > > > > For this purpose patch 1/2 changes the way arch_set_freq_scale() is > > defined in architecture code which brings it in line with the logic of > > other architectural function definitions while allowing for less invasive > > filtering of FI support. > > > > In the discussions at [2], three possible solutions were suggested: > > - (1) conditioning FI by !CONFIG_BL_SWITCHER > > - (2) leave as is with note in driver specifying this FI broken > > functionality > > - (3) removing full BL_SWITCHER support > > > > This series restructures the solution at (1). The reason for it is that > > the new patch limits the ifdef filtering to the arm topology include file, > > a location where frequency invariance functions are defined. Therefore, > > this seems more appropriate given that the b.L switcher is an arm > > technology and that the new FI filtering location seems more natural for > > conditioned FI disabling. > > > > Solutions (2) and (3) were not implemented given that there might be some > > remaining users of this technology (Samsung Chromebook 2 - Samsung Exynos > > 5 Octa 5420, Samsung Exynos 5 Octa 5800) and therefore leaving this > > broken (2) seems equally bad to removing support for it (3). > > > > [1] https://lore.kernel.org/lkml/20200701090751.7543-5-ionela.voinescu@arm.com/ > > [2] https://lore.kernel.org/lkml/20200722093732.14297-4-ionela.voinescu@arm.com/ > > [3] https://lwn.net/Articles/481055/ > > I can take this set with the ACKs from Viresh if that's fine by > everyone. Catalin? Sudeep? Acked-by: Sudeep Holla <sudeep.holla@arm.com> (BL_SWITCHER and topology parts)
On Thu, Oct 8, 2020 at 4:06 PM Sudeep Holla <sudeep.holla@arm.com> wrote: > > On Wed, Oct 07, 2020 at 04:34:44PM +0200, Rafael J. Wysocki wrote: > > On Thu, Sep 24, 2020 at 2:30 PM Ionela Voinescu <ionela.voinescu@arm.com> wrote: > > > > > > This series is the result of the discussions ([1], [2]) around the > > > complications that the BL_SWITCHER poses when it comes to Frequency > > > Invariance (FI) and it aims to restart the discussions. > > > > > > To properly scale its per-entity load-tracking signals, the task > > > scheduler needs to be given a frequency scale factor, i.e. some image of > > > the current frequency the CPU is running at, relative to its maximum > > > frequency. > > > > > > But (reiterating the message in the changelog of patch 2/2), big.LITTLE > > > switching complicates the setting of a correct cpufreq-based frequency > > > invariance scale factor due to (as observed in > > > drivers/cpufreq/vexpress-spc-cpufreq.c): > > > - Incorrect current and maximum frequencies as a result of the > > > exposure of a virtual frequency table to the cpufreq core, > > > - Missed updates as a result of asynchronous frequency adjustments > > > caused by frequency changes in other CPU pairs. > > > More information on this feature can be found at [3]. > > > > > > Given that its functionality is atypical in regards to FI and that this > > > is an old technology, patch 2/2 disable FI for when big.LITTLE switching > > > is configured in to prevent incorrect scale setting. > > > > > > For this purpose patch 1/2 changes the way arch_set_freq_scale() is > > > defined in architecture code which brings it in line with the logic of > > > other architectural function definitions while allowing for less invasive > > > filtering of FI support. > > > > > > In the discussions at [2], three possible solutions were suggested: > > > - (1) conditioning FI by !CONFIG_BL_SWITCHER > > > - (2) leave as is with note in driver specifying this FI broken > > > functionality > > > - (3) removing full BL_SWITCHER support > > > > > > This series restructures the solution at (1). The reason for it is that > > > the new patch limits the ifdef filtering to the arm topology include file, > > > a location where frequency invariance functions are defined. Therefore, > > > this seems more appropriate given that the b.L switcher is an arm > > > technology and that the new FI filtering location seems more natural for > > > conditioned FI disabling. > > > > > > Solutions (2) and (3) were not implemented given that there might be some > > > remaining users of this technology (Samsung Chromebook 2 - Samsung Exynos > > > 5 Octa 5420, Samsung Exynos 5 Octa 5800) and therefore leaving this > > > broken (2) seems equally bad to removing support for it (3). > > > > > > [1] https://lore.kernel.org/lkml/20200701090751.7543-5-ionela.voinescu@arm.com/ > > > [2] https://lore.kernel.org/lkml/20200722093732.14297-4-ionela.voinescu@arm.com/ > > > [3] https://lwn.net/Articles/481055/ > > > > I can take this set with the ACKs from Viresh if that's fine by > > everyone. Catalin? Sudeep? > > Acked-by: Sudeep Holla <sudeep.holla@arm.com> (BL_SWITCHER and topology parts) OK, the series has been applied as 5.10 material, thanks!
On Thursday 08 Oct 2020 at 17:18:25 (+0200), Rafael J. Wysocki wrote: > On Thu, Oct 8, 2020 at 4:06 PM Sudeep Holla <sudeep.holla@arm.com> wrote: > > > > On Wed, Oct 07, 2020 at 04:34:44PM +0200, Rafael J. Wysocki wrote: > > > On Thu, Sep 24, 2020 at 2:30 PM Ionela Voinescu <ionela.voinescu@arm.com> wrote: > > > > > > > > This series is the result of the discussions ([1], [2]) around the > > > > complications that the BL_SWITCHER poses when it comes to Frequency > > > > Invariance (FI) and it aims to restart the discussions. > > > > > > > > To properly scale its per-entity load-tracking signals, the task > > > > scheduler needs to be given a frequency scale factor, i.e. some image of > > > > the current frequency the CPU is running at, relative to its maximum > > > > frequency. > > > > > > > > But (reiterating the message in the changelog of patch 2/2), big.LITTLE > > > > switching complicates the setting of a correct cpufreq-based frequency > > > > invariance scale factor due to (as observed in > > > > drivers/cpufreq/vexpress-spc-cpufreq.c): > > > > - Incorrect current and maximum frequencies as a result of the > > > > exposure of a virtual frequency table to the cpufreq core, > > > > - Missed updates as a result of asynchronous frequency adjustments > > > > caused by frequency changes in other CPU pairs. > > > > More information on this feature can be found at [3]. > > > > > > > > Given that its functionality is atypical in regards to FI and that this > > > > is an old technology, patch 2/2 disable FI for when big.LITTLE switching > > > > is configured in to prevent incorrect scale setting. > > > > > > > > For this purpose patch 1/2 changes the way arch_set_freq_scale() is > > > > defined in architecture code which brings it in line with the logic of > > > > other architectural function definitions while allowing for less invasive > > > > filtering of FI support. > > > > > > > > In the discussions at [2], three possible solutions were suggested: > > > > - (1) conditioning FI by !CONFIG_BL_SWITCHER > > > > - (2) leave as is with note in driver specifying this FI broken > > > > functionality > > > > - (3) removing full BL_SWITCHER support > > > > > > > > This series restructures the solution at (1). The reason for it is that > > > > the new patch limits the ifdef filtering to the arm topology include file, > > > > a location where frequency invariance functions are defined. Therefore, > > > > this seems more appropriate given that the b.L switcher is an arm > > > > technology and that the new FI filtering location seems more natural for > > > > conditioned FI disabling. > > > > > > > > Solutions (2) and (3) were not implemented given that there might be some > > > > remaining users of this technology (Samsung Chromebook 2 - Samsung Exynos > > > > 5 Octa 5420, Samsung Exynos 5 Octa 5800) and therefore leaving this > > > > broken (2) seems equally bad to removing support for it (3). > > > > > > > > [1] https://lore.kernel.org/lkml/20200701090751.7543-5-ionela.voinescu@arm.com/ > > > > [2] https://lore.kernel.org/lkml/20200722093732.14297-4-ionela.voinescu@arm.com/ > > > > [3] https://lwn.net/Articles/481055/ > > > > > > I can take this set with the ACKs from Viresh if that's fine by > > > everyone. Catalin? Sudeep? > > > > Acked-by: Sudeep Holla <sudeep.holla@arm.com> (BL_SWITCHER and topology parts) > > OK, the series has been applied as 5.10 material, thanks! Many thanks, guys! Ionela.