[src/trunk]: src/sys provide some semblance of valid cpu topology for big.lit...

[mrg <mrg%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/c5df3844b3d4
branches:  trunk
changeset: 467052:c5df3844b3d4
user:      mrg <mrg%NetBSD.org@localhost>
date:      Sun Jan 12 09:29:18 2020 +0000

description:
provide some semblance of valid cpu topology for big.little systems.

while attaching cpus, if the FDT provides "capacity-dmips-mhz" track
the fastest set, and call cpu_topology_set() with slow=true for any
cpus that are not the fastest.

bug fix for cpu_topology_set(): actually set ci_is_slow for slow cpus.

with this change, and -current's recent scheduler changes, this means
that long running processes run on the faster cores.  on RK3399 based
systems, i am seeing 20-50% speed ups for many tasks.


XXX: all this can be made common with armv7 big.little.

diffstat:

 sys/arch/aarch64/aarch64/cpu.c     |  50 +++++++++++++++++++++++++++++++++----
 sys/arch/aarch64/aarch64/cpufunc.c |  10 +++---
 sys/arch/aarch64/aarch64/locore.S  |   7 +++--
 sys/arch/aarch64/include/cpu.h     |   5 +++-
 sys/arch/aarch64/include/cpufunc.h |   4 +-
 sys/arch/arm/fdt/cpu_fdt.c         |  15 +++++++++-
 sys/kern/subr_cpu.c                |   5 ++-
 7 files changed, 75 insertions(+), 21 deletions(-)

diffs (268 lines):

diff -r 066f161febe7 -r c5df3844b3d4 sys/arch/aarch64/aarch64/cpu.c
--- a/sys/arch/aarch64/aarch64/cpu.c    Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/arch/aarch64/aarch64/cpu.c    Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu.c,v 1.32 2020/01/09 16:23:41 martin Exp $ */
+/* $NetBSD: cpu.c,v 1.33 2020/01/12 09:29:18 mrg Exp $ */
 
 /*
  * Copyright (c) 2017 Ryo Shimizu <ryo%nerv.org@localhost>
@@ -27,7 +27,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(1, "$NetBSD: cpu.c,v 1.32 2020/01/09 16:23:41 martin Exp $");
+__KERNEL_RCSID(1, "$NetBSD: cpu.c,v 1.33 2020/01/12 09:29:18 mrg Exp $");
 
 #include "locators.h"
 #include "opt_arm_debug.h"
@@ -65,6 +65,7 @@
 static void cpu_identify2(device_t self, struct cpu_info *);
 static void cpu_setup_id(struct cpu_info *);
 static void cpu_setup_sysctl(device_t, struct cpu_info *);
+static void cpu_do_topology(struct cpu_info *);
 
 #ifdef MULTIPROCESSOR
 uint64_t cpu_mpidr[MAXCPUS];
@@ -144,9 +145,9 @@
        ci->ci_dev = dv;
        dv->dv_private = ci;
 
-       aarch64_gettopology(ci, ci->ci_id.ac_mpidr);
+       cpu_do_topology(ci);
+       cpu_identify(ci->ci_dev, ci);
 
-       cpu_identify(ci->ci_dev, ci);
 #ifdef MULTIPROCESSOR
        if (unit != 0) {
                mi_cpu_attach(ci);
@@ -308,8 +309,8 @@
 
        dfr0 = reg_id_aa64dfr0_el1_read();
 
-       aprint_debug_dev(self, "midr=0x%" PRIx32 "\n",
-           (uint32_t)ci->ci_id.ac_midr);
+       aprint_debug_dev(self, "midr=0x%" PRIx32 " mpidr=0x%" PRIx32 "\n",
+           (uint32_t)ci->ci_id.ac_midr, (uint32_t)ci->ci_id.ac_mpidr);
        aprint_normal_dev(self, "revID=0x%" PRIx64, id->ac_revidr);
 
        /* ID_AA64DFR0_EL1 */
@@ -497,6 +498,43 @@
                       CTL_CREATE, CTL_EOL);
 }
 
+static void
+cpu_do_topology(struct cpu_info *newci)
+{
+       struct cpu_info *ci;
+       CPU_INFO_ITERATOR cii;
+       prop_dictionary_t dict;
+       uint32_t capacity_dmips_mhz;
+       static uint32_t best_cap = 0;
+
+       dict = device_properties(newci->ci_dev);
+       if (prop_dictionary_get_uint32(dict, "capacity_dmips_mhz",
+           &capacity_dmips_mhz)) {
+               newci->ci_capacity_dmips_mhz = capacity_dmips_mhz;
+       } else {
+               newci->ci_capacity_dmips_mhz = 0;
+       }
+
+       if (newci->ci_capacity_dmips_mhz > best_cap)
+               best_cap = newci->ci_capacity_dmips_mhz;
+
+       /*
+        * CPU_INFO_FOREACH() doesn't work for this CPU until mi_cpu_attach()
+        * is called and ncpu is bumped, so call it directly here.
+        */
+       aarch64_set_topology(newci, newci->ci_id.ac_mpidr,
+           newci->ci_capacity_dmips_mhz < best_cap);
+
+       /*
+        * Using saved largest capacity, refresh previous topology info.
+        * It's supposed to be OK to re-set topology.
+        */
+       for (CPU_INFO_FOREACH(cii, ci)) {
+               aarch64_set_topology(ci, ci->ci_id.ac_mpidr,
+                   ci->ci_capacity_dmips_mhz < best_cap);
+       }
+}
+
 #ifdef MULTIPROCESSOR
 void
 cpu_boot_secondary_processors(void)
diff -r 066f161febe7 -r c5df3844b3d4 sys/arch/aarch64/aarch64/cpufunc.c
--- a/sys/arch/aarch64/aarch64/cpufunc.c        Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/arch/aarch64/aarch64/cpufunc.c        Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: cpufunc.c,v 1.13 2020/01/09 16:35:03 ad Exp $  */
+/*     $NetBSD: cpufunc.c,v 1.14 2020/01/12 09:29:18 mrg Exp $ */
 
 /*
  * Copyright (c) 2017 Ryo Shimizu <ryo%nerv.org@localhost>
@@ -29,7 +29,7 @@
 #include "opt_multiprocessor.h"
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cpufunc.c,v 1.13 2020/01/09 16:35:03 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cpufunc.c,v 1.14 2020/01/12 09:29:18 mrg Exp $");
 
 #include <sys/param.h>
 #include <sys/types.h>
@@ -89,7 +89,7 @@
 }
 
 void
-aarch64_gettopology(struct cpu_info * const ci, uint64_t mpidr)
+aarch64_set_topology(struct cpu_info * const ci, uint64_t mpidr, bool slow)
 {
 
        if (mpidr & MPIDR_MT) {
@@ -98,14 +98,14 @@
                    __SHIFTOUT(mpidr, MPIDR_AFF1),
                    __SHIFTOUT(mpidr, MPIDR_AFF0),
                    0,
-                   false);
+                   slow);
        } else {
                cpu_topology_set(ci,
                    __SHIFTOUT(mpidr, MPIDR_AFF1),
                    __SHIFTOUT(mpidr, MPIDR_AFF0),
                    0,
                    0,
-                   false);
+                   slow);
        }
 }
 
diff -r 066f161febe7 -r c5df3844b3d4 sys/arch/aarch64/aarch64/locore.S
--- a/sys/arch/aarch64/aarch64/locore.S Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/arch/aarch64/aarch64/locore.S Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: locore.S,v 1.50 2020/01/08 05:41:07 ryo Exp $  */
+/*     $NetBSD: locore.S,v 1.51 2020/01/12 09:29:18 mrg Exp $  */
 
 /*
  * Copyright (c) 2017 Ryo Shimizu <ryo%nerv.org@localhost>
@@ -38,7 +38,7 @@
 #include <aarch64/hypervisor.h>
 #include "assym.h"
 
-RCSID("$NetBSD: locore.S,v 1.50 2020/01/08 05:41:07 ryo Exp $")
+RCSID("$NetBSD: locore.S,v 1.51 2020/01/12 09:29:18 mrg Exp $")
 
 #ifdef AARCH64_DEVICE_MEM_STRONGLY_ORDERED
 #define        MAIR_DEVICE_MEM         MAIR_DEVICE_nGnRnE
@@ -187,7 +187,8 @@
        /* get cache configuration */
        mrs     x0, tpidr_el1           /* curcpu */
        mrs     x1, mpidr_el1
-       bl      aarch64_gettopology
+       mov     x2, #0
+       bl      aarch64_set_topology
        mov     x0, xzr
        bl      aarch64_getcacheinfo
 
diff -r 066f161febe7 -r c5df3844b3d4 sys/arch/aarch64/include/cpu.h
--- a/sys/arch/aarch64/include/cpu.h    Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/arch/aarch64/include/cpu.h    Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu.h,v 1.17 2020/01/05 20:17:43 ad Exp $ */
+/* $NetBSD: cpu.h,v 1.18 2020/01/12 09:29:18 mrg Exp $ */
 
 /*-
  * Copyright (c) 2014 The NetBSD Foundation, Inc.
@@ -93,6 +93,9 @@
        struct evcnt ci_vfp_save;
        struct evcnt ci_vfp_release;
 
+       /* FDT or similar supplied "cpu capacity" */
+       uint32_t ci_capacity_dmips_mhz;
+
        /* interrupt controller */
        u_int ci_gic_redist;    /* GICv3 redistributor index */
        uint64_t ci_gic_sgir;   /* GICv3 SGIR target */
diff -r 066f161febe7 -r c5df3844b3d4 sys/arch/aarch64/include/cpufunc.h
--- a/sys/arch/aarch64/include/cpufunc.h        Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/arch/aarch64/include/cpufunc.h        Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: cpufunc.h,v 1.9 2019/12/19 09:47:42 ryo Exp $  */
+/*     $NetBSD: cpufunc.h,v 1.10 2020/01/12 09:29:18 mrg Exp $ */
 
 /*
  * Copyright (c) 2017 Ryo Shimizu <ryo%nerv.org@localhost>
@@ -69,7 +69,7 @@
 extern u_int cputype;                  /* compat arm */
 
 int set_cpufuncs(void);
-void aarch64_gettopology(struct cpu_info *, uint64_t);
+void aarch64_set_topology(struct cpu_info *, uint64_t, bool);
 void aarch64_getcacheinfo(int);
 void aarch64_printcacheinfo(device_t);
 
diff -r 066f161febe7 -r c5df3844b3d4 sys/arch/arm/fdt/cpu_fdt.c
--- a/sys/arch/arm/fdt/cpu_fdt.c        Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/arch/arm/fdt/cpu_fdt.c        Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu_fdt.c,v 1.30 2019/11/01 13:22:08 bad Exp $ */
+/* $NetBSD: cpu_fdt.c,v 1.31 2020/01/12 09:29:18 mrg Exp $ */
 
 /*-
  * Copyright (c) 2017 Jared McNeill <jmcneill%invisible.ca@localhost>
@@ -30,7 +30,7 @@
 #include "psci_fdt.h"
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cpu_fdt.c,v 1.30 2019/11/01 13:22:08 bad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cpu_fdt.c,v 1.31 2020/01/12 09:29:18 mrg Exp $");
 
 #include <sys/param.h>
 #include <sys/atomic.h>
@@ -126,10 +126,21 @@
        enum cpu_fdt_type type;
        bus_addr_t mpidr;
        cpuid_t cpuid;
+       const uint32_t *cap_ptr;
+       int len;
 
        sc->sc_dev = self;
        sc->sc_phandle = phandle;
 
+       cap_ptr = fdtbus_get_prop(phandle, "capacity-dmips-mhz", &len);
+       if (cap_ptr && len == 4) {
+               prop_dictionary_t dict = device_properties(self);
+               uint32_t capacity_dmips_mhz = be32toh(*cap_ptr);
+
+               prop_dictionary_set_uint32(dict, "capacity_dmips_mhz",
+                   capacity_dmips_mhz);
+       }
+
        type = of_search_compatible(phandle, compat_data)->data;
 
        switch (type) {
diff -r 066f161febe7 -r c5df3844b3d4 sys/kern/subr_cpu.c
--- a/sys/kern/subr_cpu.c       Sun Jan 12 06:41:02 2020 +0000
+++ b/sys/kern/subr_cpu.c       Sun Jan 12 09:29:18 2020 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: subr_cpu.c,v 1.6 2020/01/09 16:35:03 ad Exp $  */
+/*     $NetBSD: subr_cpu.c,v 1.7 2020/01/12 09:29:18 mrg Exp $ */
 
 /*-
  * Copyright (c) 2007, 2008, 2009, 2010, 2012, 2019, 2020
@@ -61,7 +61,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: subr_cpu.c,v 1.6 2020/01/09 16:35:03 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: subr_cpu.c,v 1.7 2020/01/12 09:29:18 mrg Exp $");
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -152,6 +152,7 @@
        ci->ci_core_id = core_id;
        ci->ci_smt_id = smt_id;
        ci->ci_numa_id = numa_id;
+       ci->ci_is_slow = slow;
        for (rel = 0; rel < __arraycount(ci->ci_sibling); rel++) {
                ci->ci_sibling[rel] = ci;
                ci->ci_nsibling[rel] = 1;