Messages in this thread | | | Subject | Re: [PATCH 0/7] arm64: capabilities: Optimize checking and enabling | From | Vladimir Murzin <> | Date | Mon, 5 Nov 2018 12:14:16 +0000 |
| |
Hi Suzuki,
On 05/11/18 11:55, Suzuki K Poulose wrote: > We maintain two separate tables (i.e, arm64_features and arm64_errata) of > struct arm64_cpu_capabilities which decide the capabilities of the system. > We iterate over the two tables for detecting/verifying/enabling the capabilities. > e.g, this_cpu_has_cap() needs to iterate over the two tables one by one to > find the "capability" structure corresponding to the cap number and then > check it on the CPU. > > Also, we enable all the non-boot scoped capabilities after all the SMP cpus > are brought up by the kernel, using stop_machine() for each available > capability. We could batch all the "enabling" activity to a single > stop_machine() callback. But that implies you need a way to map > a given capability number to the corresponding capability entry > to finish the operation quickly. > > So we need a quicker way to access the entry for a given capability. > We have two choices : > > 1) Unify both the tables to a static/dynamic sorted entry based on > the capability number. This has the following drawbacks : > a) The entries must be unique. i.e, no duplicate entries for a > capability. > b) Looses the separation of "features" vs. "errata" classification > c) Statically sorting the list is error prone. Runtime sorting the > array means more time for booting. > > 2) Keep an array of pointers to the capability sorted at boot time > based on the capability. > a) As for (1), the entries must be unique for a capability. > > This series implements (2) and uses the new list for optimizing the > operations on the entries. As a prepatory step, we remove the > duplicate entries for the same capabilities (Patch 1-3). >
Thanks a lot for getting it sorted out! In case it'd help:
Reviewed-by: Vladimir Murzin <vladimir.murzin@arm.com> Tested-by: Vladimir Murzin <vladimir.murzin@arm.com>
Cheers Vladimir
> > Suzuki K Poulose (7): > arm64: capabilities: Merge entries for ARM64_WORKAROUND_CLEAN_CACHE > arm64: capabilities: Merge duplicate Cavium erratum entries > arm64: capabilities: Merge duplicate entries for Qualcomm erratum 1003 > arm64: capabilities: Speed up capability lookup > arm64: capabilities: Optimize this_cpu_has_cap > arm64: capabilities: Use linear array for detection and verification > arm64: capabilities: Batch cpu_enable callbacks > > arch/arm64/include/asm/cpufeature.h | 3 + > arch/arm64/include/asm/cputype.h | 2 + > arch/arm64/kernel/cpu_errata.c | 94 ++++++++++---------- > arch/arm64/kernel/cpufeature.c | 165 ++++++++++++++++++++---------------- > 4 files changed, 145 insertions(+), 119 deletions(-) >
| |