[PATCH 3/9] ARM: MB86S7X: Add MCPM support
Lorenzo Pieralisi
lorenzo.pieralisi at arm.com
Tue Nov 25 10:46:10 PST 2014
On Tue, Nov 25, 2014 at 05:42:32PM +0000, Nicolas Pitre wrote:
> On Thu, 20 Nov 2014, Vincent Yang wrote:
>
> > The remote firmware(SCB) owns the SMP control. This MCPM driver gets
> > CPU/CLUSTER power up/down done by SCB over mailbox.
> >
> > Signed-off-by: Andy Green <andy.green at linaro.org>
> > Signed-off-by: Jassi Brar <jaswinder.singh at linaro.org>
> > Signed-off-by: Vincent Yang <Vincent.Yang at tw.fujitsu.com>
> > Signed-off-by: Tetsuya Nuriya <nuriya.tetsuya at jp.fujitsu.com>
> > ---
> > arch/arm/mach-mb86s7x/Makefile | 2 +-
> > arch/arm/mach-mb86s7x/mcpm.c | 360 +++++++++++++++++++++++++++++++++++++++++
> > arch/arm/mach-mb86s7x/smc.S | 27 ++++
> > 3 files changed, 388 insertions(+), 1 deletion(-)
> > create mode 100644 arch/arm/mach-mb86s7x/mcpm.c
> > create mode 100644 arch/arm/mach-mb86s7x/smc.S
> >
> > diff --git a/arch/arm/mach-mb86s7x/Makefile b/arch/arm/mach-mb86s7x/Makefile
> > index 97640b6..b0fa34b 100644
> > --- a/arch/arm/mach-mb86s7x/Makefile
> > +++ b/arch/arm/mach-mb86s7x/Makefile
> > @@ -1 +1 @@
> > -obj-$(CONFIG_ARCH_MB86S7X) += board.o
> > +obj-$(CONFIG_ARCH_MB86S7X) += board.o mcpm.o smc.o
> > diff --git a/arch/arm/mach-mb86s7x/mcpm.c b/arch/arm/mach-mb86s7x/mcpm.c
> > new file mode 100644
> > index 0000000..bf1b50a
> > --- /dev/null
> > +++ b/arch/arm/mach-mb86s7x/mcpm.c
> > @@ -0,0 +1,360 @@
> > +/*
> > + * arch/arm/mach-mb86s7x/mcpm.c
> > + * Copyright: (C) 2013-2014 Fujitsu Semiconductor Limited
> > + * Copyright: (C) 2014 Linaro Ltd.
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 as
> > + * published by the Free Software Foundation.
> > + */
> > +
> > +#include <linux/io.h>
> > +#include <linux/pm.h>
> > +#include <linux/delay.h>
> > +#include <linux/cpu_pm.h>
> > +#include <linux/kernel.h>
> > +#include <linux/module.h>
> > +#include <linux/arm-cci.h>
> > +#include <linux/spinlock.h>
> > +#include <linux/suspend.h>
> > +#include <linux/of_device.h>
> > +#include <linux/irqchip/arm-gic.h>
> > +#include <linux/platform_device.h>
> > +
> > +#include <soc/mb86s7x/scb_mhu.h>
> > +
> > +#include <asm/mcpm.h>
> > +#include <asm/cp15.h>
> > +#include <asm/cputype.h>
> > +#include <asm/suspend.h>
> > +#include <asm/idmap.h>
> > +
> > +#define S7X_MAX_CLUSTER 2
> > +#define S7X_MAX_CPU 2
> > +
> > +#define MHU_SHM_OFFSET 0x3800
> > +#define WFI_COLOR_OFFSET 0x3f00
> > +#define TRAMPOLINE_OFFSET 0x3c00
> > +#define RESET_OFFSET (TRAMPOLINE_OFFSET + 0x3fc)
> > +
> > +static arch_spinlock_t mb86s7x_pm_lock = __ARCH_SPIN_LOCK_UNLOCKED;
> > +static int mb86s7x_pm_use_count[S7X_MAX_CLUSTER][S7X_MAX_CPU];
> > +extern void __iomem *mb86s7x_shm_base;
> > +
> > +#define AT_WFI_DO_NOTHING 0x0
> > +#define AT_WFI_DO_SUSPEND 0x1
> > +#define AT_WFI_DO_POWEROFF 0x2
> > +#define AT_WFI_COLOR_MASK 0x3
> > +
> > +struct mb86s7x_cpu_gate {
> > + u32 payload_size;
> > + u32 cluster_class;
> > + u32 cluster_id;
> > + u32 cpu_id;
> > +#define SCB_CPU_STATE_OFF 0x0
> > +#define SCB_CPU_STATE_ON 0x1
> > +#define SCB_CPU_STATE_SUSP 0x2
> > + u32 cpu_state;
> > +};
> > +
> > +asmlinkage void mb86s70evb_outer_flush_all(void)
> > +{
> > + outer_flush_all();
> > +}
> > +
> > +#define mb86s70evb_exit_coherency_flush(level) { \
> > + asm volatile( \
> > + "stmfd sp!, {fp, ip}\n\t" \
> > + "mrc p15, 0, r0, c1, c0, 0 @ get SCTLR\n\t" \
> > + "bic r0, r0, #"__stringify(CR_C)"\n\t" \
> > + "mcr p15, 0, r0, c1, c0, 0 @ set SCTLR\n\t" \
> > + "isb\n\t" \
> > + "bl v7_flush_dcache_"__stringify(level)"\n\t" \
> > + "bl mb86s70evb_outer_flush_all\n\t" \
>
> This is wrong. As mentioned already, this unconditionally flushes L2 in
> all cases which shouldn't be necessary in the "louis" case.
Is this a bL system with unified and architected L2s ? I think so, so
what's the outercache for ?
> Furthermore, the safety of this macro is ensured by not having any
> memory writes in the middle of the whole sequence. By calling
> mb86s70evb_outer_flush_all() there could be the return address pushed
> onto the stack before calling outer_flush_all() if some tail call
> optimization is not applied. And this is without saying what
> outer_flush_all() actually does.
>
> Why can't you simply do this instead:
>
> v7_exit_coherency_flush(all);
> outer_flush_all();
>
> Of course you'll have to audit everything in the outer_flush_all() path
> to make sure no atomic instructions such as LDREX/STREX are invoked.
> Those have undefined behavior after CR_C is cleared.
What you are saying is correct but first of all I would like to
understand what outercache we are talking about here.
Code snippet above is fragile, as you said, and I'd rather disable
the outercache before executing the power down sequence (which implies
cache is cleaned/invalidated and quiescent before we clear the C bit)
if there is an outercache to be managed.
Waiting in trepidation :)
Lorenzo
>
> > + "mrc p15, 0, r0, c1, c0, 1 @ get ACTLR\n\t" \
> > + "bic r0, r0, #(1 << 6) @ disable local coherency\n\t" \
> > + "mcr p15, 0, r0, c1, c0, 1 @ set ACTLR\n\t" \
> > + "isb\n\t" \
> > + "dsb\n\t" \
> > + "ldmfd sp!, {fp, ip}" \
> > + : : : "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", \
> > + "r9", "r10", "lr", "memory"); \
> > + }
> > +
> > +static void
> > +mb86s7x_set_wficolor(unsigned clstr, unsigned cpu, unsigned clr)
> > +{
> > + u8 val;
> > +
> > + if (clr & ~AT_WFI_COLOR_MASK)
> > + return;
>
> Shouldn't this be a BUG() rather than a return?
>
> > +
> > + val = readb_relaxed(mb86s7x_shm_base
> > + + WFI_COLOR_OFFSET + clstr * 2 + cpu);
> > + val &= ~AT_WFI_COLOR_MASK;
> > + val |= clr;
> > + writeb_relaxed(val, mb86s7x_shm_base
> > + + WFI_COLOR_OFFSET + clstr * 2 + cpu);
> > +}
> > +
> > +static int mb86s7x_pm_power_up(unsigned int cpu, unsigned int cluster)
> > +{
> > + int ret = 0;
> > +
> > + if (cluster >= S7X_MAX_CLUSTER || cpu >= S7X_MAX_CPU)
> > + return -EINVAL;
> > +
> > + pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
> > +
> > + local_irq_disable();
> > + arch_spin_lock(&mb86s7x_pm_lock);
> > +
> > + mb86s7x_pm_use_count[cluster][cpu]++;
> > +
> > + if (mb86s7x_pm_use_count[cluster][cpu] == 1) {
> > + struct mb86s7x_cpu_gate cmd;
> > +
> > + arch_spin_unlock(&mb86s7x_pm_lock);
> > + local_irq_enable();
>
> From that point onward, nothing prevents concurrent execution in
> mb86s7x_pm_suspend() racing to call mb86s7x_set_wficolor() resulting in
> an incoherent state depending on who wins the race.
>
> > + cmd.payload_size = sizeof(cmd);
> > + cmd.cluster_class = 0;
> > + cmd.cluster_id = cluster;
> > + cmd.cpu_id = cpu;
> > + cmd.cpu_state = SCB_CPU_STATE_ON;
> > +
> > + pr_debug("%s:%d CMD Cl_Class-%u CL_ID-%u CPU_ID-%u STATE-%u}\n",
> > + __func__, __LINE__, cmd.cluster_class,
> > + cmd.cluster_id, cmd.cpu_id, cmd.cpu_state);
> > +
> > + mb86s7x_set_wficolor(cluster, cpu, AT_WFI_DO_NOTHING);
> > + ret = mb86s7x_send_packet(CMD_CPU_CLOCK_GATE_SET_REQ,
> > + &cmd, sizeof(cmd));
> > + if (ret < 0) {
> > + pr_err("%s:%d failed!\n", __func__, __LINE__);
> > + return ret;
> > + }
> > +
> > + pr_debug("%s:%d REP Cl_Class-%u CL_ID-%u CPU_ID-%u STATE-%u}\n",
> > + __func__, __LINE__, cmd.cluster_class,
> > + cmd.cluster_id, cmd.cpu_id, cmd.cpu_state);
> > +
> > + if (cmd.cpu_state != SCB_CPU_STATE_ON)
> > + return -ENODEV;
> > + } else if (mb86s7x_pm_use_count[cluster][cpu] != 2) {
> > + /*
> > + * The only possible values are:
> > + * 0 = CPU down
> > + * 1 = CPU (still) up
> > + * 2 = CPU requested to be up before it had a chance
> > + * to actually make itself down.
> > + * Any other value is a bug.
> > + */
> > + BUG();
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +static void mb86s7x_pm_suspend(u64 ignored)
> > +{
> > + unsigned int mpidr, cpu, cluster;
> > + bool last_man = false, skip_wfi = false;
> > +
> > + mpidr = read_cpuid_mpidr();
> > + cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
> > + cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
> > +
> > + pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
> > + __mcpm_cpu_going_down(cpu, cluster);
> > +
> > + arch_spin_lock(&mb86s7x_pm_lock);
> > + BUG_ON(__mcpm_cluster_state(cluster) != CLUSTER_UP);
> > +
> > + mb86s7x_pm_use_count[cluster][cpu]--;
> > +
> > + if (mb86s7x_pm_use_count[cluster][cpu] == 0) {
> > + if (!mb86s7x_pm_use_count[cluster][0] &&
> > + !mb86s7x_pm_use_count[cluster][1])
> > + last_man = true;
> > + mb86s7x_set_wficolor(cluster, cpu, AT_WFI_DO_POWEROFF);
> > + } else if (mb86s7x_pm_use_count[cluster][cpu] == 1) {
> > + skip_wfi = true; /* Overtaken by a power up */
> > + } else {
> > + BUG();
> > + }
> > +
> > + if (!skip_wfi)
> > + gic_cpu_if_down();
> > +
> > + if (last_man && __mcpm_outbound_enter_critical(cpu, cluster)) {
> > + arch_spin_unlock(&mb86s7x_pm_lock);
> > +
> > + if (read_cpuid_part() == ARM_CPU_PART_CORTEX_A15) {
> > + /*
> > + * On the Cortex-A15 we need to disable
> > + * L2 prefetching before flushing the cache.
> > + */
> > + asm volatile(
> > + "mcr p15, 1, %0, c15, c0, 3\n\t"
> > + "isb\n\t"
> > + "dsb"
> > + : : "r" (0x400));
> > + }
> > +
> > + mb86s70evb_exit_coherency_flush(all);
> > +
> > + cci_disable_port_by_cpu(mpidr);
> > +
> > + __mcpm_outbound_leave_critical(cluster, CLUSTER_DOWN);
> > + } else {
> > + arch_spin_unlock(&mb86s7x_pm_lock);
> > + mb86s70evb_exit_coherency_flush(louis);
> > + }
> > +
> > + __mcpm_cpu_down(cpu, cluster);
> > +
> > + /* Now we are prepared for power-down, do it: */
> > + if (!skip_wfi)
> > + wfi();
> > +}
> > +
> > +static void mb86s7x_pm_power_down(void)
> > +{
> > + mb86s7x_pm_suspend(0);
> > +}
> > +
> > +static int mb86s7x_wait_for_powerdown(unsigned int cpu, unsigned int cluster)
> > +{
> > + struct mb86s7x_cpu_gate cmd;
> > + int i, ret;
> > +
> > + if (cluster >= S7X_MAX_CLUSTER || cpu >= S7X_MAX_CPU)
> > + return 0;
>
> This should return an error or BUG().
>
> > + cmd.payload_size = sizeof(cmd);
> > + cmd.cluster_class = 0;
> > + cmd.cluster_id = cluster;
> > + cmd.cpu_id = cpu;
> > + cmd.cpu_state = SCB_CPU_STATE_ON;
> > +
> > + for (i = 0; i < 50; i++) {
> > + ret = mb86s7x_send_packet(CMD_CPU_CLOCK_GATE_GET_REQ,
> > + &cmd, sizeof(cmd));
> > + if (ret < 0) {
> > + pr_err("%s:%d failed to get CPU status\n",
> > + __func__, __LINE__);
> > + return ret;
> > + }
> > +
> > + pr_debug("%s:%d Cl_Class-%u CL_ID-%u CPU_ID-%u STATE-%u\n",
> > + __func__, __LINE__,
> > + cmd.cluster_class, cmd.cluster_id,
> > + cmd.cpu_id, cmd.cpu_state);
> > +
> > + if (cmd.cpu_state == SCB_CPU_STATE_OFF)
> > + return 0;
> > +
> > + msleep(20);
> > + }
> > +
> > + return -ETIMEDOUT;
> > +}
> > +
> > +static void mb86s7x_pm_powered_up(void)
> > +{
> > + unsigned int mpidr, cpu, cluster;
> > +
> > + mpidr = read_cpuid_mpidr();
> > + cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
> > + cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
> > +
> > + arch_spin_lock(&mb86s7x_pm_lock);
> > + if (!mb86s7x_pm_use_count[cluster][cpu])
> > + mb86s7x_pm_use_count[cluster][cpu] = 1;
> > + arch_spin_unlock(&mb86s7x_pm_lock);
> > +}
> > +
> > +static const struct mcpm_platform_ops mb86s7x_pm_power_ops = {
> > + .power_up = mb86s7x_pm_power_up,
> > + .power_down = mb86s7x_pm_power_down,
> > + .wait_for_powerdown = mb86s7x_wait_for_powerdown,
> > + .suspend = mb86s7x_pm_suspend,
> > + .powered_up = mb86s7x_pm_powered_up,
> > +};
> > +
> > +/*
> > + * Enable cluster-level coherency, in preparation for turning on the MMU.
> > + */
> > +static void __naked mb86s7x_pm_power_up_setup(unsigned int affinity_level)
> > +{
> > + asm volatile ("\n"
> > +" cmp r0, #1\n"
> > +" bxne lr\n"
> > +" b cci_enable_port_for_self");
> > +}
> > +
> > +static void __init mb86s7x_cache_off(void)
> > +{
> > + if (read_cpuid_part() == ARM_CPU_PART_CORTEX_A15) {
> > + /* disable L2 prefetching on the Cortex-A15 */
> > + asm volatile(
> > + "mcr p15, 1, %0, c15, c0, 3\n\t"
> > + "isb\n\t"
> > + "dsb"
> > + : : "r" (0x400));
> > + }
> > + mb86s70evb_exit_coherency_flush(all);
> > +}
> > +
> > +struct mb86s7x_scb_version {
> > + u32 payload_size;
> > + u32 version;
> > + u32 config_version;
> > +};
> > +
> > +extern void mb86s7x_cpu_entry(unsigned long secondary_entry);
> > +
> > +static int __init mb86s7x_mcpm_init(void)
> > +{
> > + unsigned int mpidr, cpu, cluster;
> > + struct mb86s7x_scb_version cmd;
> > + int ret;
> > +
> > + if (!cci_probed())
> > + return -ENODEV;
> > +
> > + mpidr = read_cpuid_mpidr();
> > + cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
> > + cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
> > +
> > + pr_info("Booting on cpu_%u cluster_%u\n", cpu, cluster);
> > + mb86s7x_pm_use_count[cluster][cpu] = 1;
> > +
> > + /* reset the wfi 'color' for primary cpu */
> > + mb86s7x_set_wficolor(cluster, cpu, AT_WFI_DO_NOTHING);
> > +
> > + /* Do SMC to set entry address for CPUs coming online */
> > + mb86s7x_cpu_entry(virt_to_phys(mcpm_entry_point));
> > +
> > + cmd.payload_size = sizeof(cmd);
> > + cmd.version = 0;
> > + cmd.config_version = 0;
> > + ret = mb86s7x_send_packet(CMD_SCB_CAPABILITY_GET_REQ,
> > + &cmd, sizeof(cmd));
> > + if (ret < 0)
> > + pr_err("%s:%d failed to get SCB version\n",
> > + __func__, __LINE__);
> > +
> > + pr_info("MB86S7x MCPM initialized: SCB version 0x%x:0x%x\n",
> > + cmd.version, cmd.config_version);
> > +
> > + ret = mcpm_platform_register(&mb86s7x_pm_power_ops);
> > + if (!ret)
> > + ret = mcpm_sync_init(mb86s7x_pm_power_up_setup);
> > + if (!ret)
> > + ret = mcpm_loopback(mb86s7x_cache_off); /* turn on the CCI */
> > + mcpm_smp_set_ops();
> > +
> > + return ret;
> > +}
> > +early_initcall(mb86s7x_mcpm_init);
> > diff --git a/arch/arm/mach-mb86s7x/smc.S b/arch/arm/mach-mb86s7x/smc.S
> > new file mode 100644
> > index 0000000..a14330b
> > --- /dev/null
> > +++ b/arch/arm/mach-mb86s7x/smc.S
> > @@ -0,0 +1,27 @@
> > +/*
> > + * SMC command interface to set secondary entry point
> > + * Copyright: (C) 2013-2014 Fujitsu Semiconductor Limited
> > + * Copyright: (C) 2014 Linaro Ltd.
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 as
> > + * published by the Free Software Foundation.
> > + */
> > +
> > +#include <linux/linkage.h>
> > +
> > +.arch_extension sec
> > +
> > +/* void mb86s7x_cpu_entry(unsigned long secondary_entry); */
> > +ENTRY(mb86s7x_cpu_entry)
> > + stmfd sp!, {r1-r11, lr}
> > + mov r1, r0
> > + ldr r0, =1
> > + mrc p15, 0, r3, c1, c0, 0
> > + mov r4, r3
> > + and r3, #0xbfffffff
> > + mcr p15, 0, r3, c1, c0, 0
> > + smc #0
> > + mcr p15, 0, r4, c1, c0, 0
> > + ldmfd sp!, {r1-r11, pc}
> > +ENDPROC(mb86s7x_cpu_entry)
> > --
> > 1.9.0
> >
> >
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>
More information about the linux-arm-kernel
mailing list