Re: [PATCH V8 16/21] csky: SMP support
From: Marc Zyngier <hidden>
Date: 2018-10-12 11:09:59
Also in:
linux-arch, lkml
On 12/10/18 05:42, Guo Ren wrote:
quoted hunk ↗ jump to hunk
This patch adds boot, ipi, hotplug code for SMP. Changelog: - remove set_ipi_irq_mapping callback. - Convert the cpumask to an interrupt-controller specific representation in driver's code, and not the SMP code's. - csky: remove irq_mapping from smp.c There are some feedbacks from irqchip, and we need to adjust "smp.c & smp.h" to match the csky_mp_intc modification. - Move IPI_IRQ define into drivers/irqchip/csky_mpintc.c, because it's a interrupt controller specific. - Bugfix request_irq with IPI_IRQ, we must use irq_mapping return value not directly use IPI_IRQ. The modification also involves csky_mp_intc. Signed-off-by: Guo Ren <redacted> Cc: Marc Zyngier <redacted> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> --- arch/csky/include/asm/smp.h | 28 ++++++ arch/csky/kernel/smp.c | 237 ++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 265 insertions(+) create mode 100644 arch/csky/include/asm/smp.h create mode 100644 arch/csky/kernel/smp.cdiff --git a/arch/csky/include/asm/smp.h b/arch/csky/include/asm/smp.h new file mode 100644 index 0000000..31f7b94 --- /dev/null +++ b/arch/csky/include/asm/smp.h@@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +#ifndef __ASM_CSKY_SMP_H +#define __ASM_CSKY_SMP_H + +#include <linux/cpumask.h> +#include <linux/irqreturn.h> +#include <linux/threads.h> + +#ifdef CONFIG_SMP + +void __init setup_smp(void); + +void __init setup_smp_ipi(void); + +void __init enable_smp_ipi(void); + +void arch_send_call_function_ipi_mask(struct cpumask *mask); + +void arch_send_call_function_single_ipi(int cpu); + +void __init set_send_ipi(void (*func)(const struct cpumask *mask), int irq); + +#define raw_smp_processor_id() (current_thread_info()->cpu) + +#endif /* CONFIG_SMP */ + +#endif /* __ASM_CSKY_SMP_H */diff --git a/arch/csky/kernel/smp.c b/arch/csky/kernel/smp.c new file mode 100644 index 0000000..5ea9516 --- /dev/null +++ b/arch/csky/kernel/smp.c@@ -0,0 +1,237 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/kernel.h> +#include <linux/mm.h> +#include <linux/sched.h> +#include <linux/kernel_stat.h> +#include <linux/notifier.h> +#include <linux/cpu.h> +#include <linux/percpu.h> +#include <linux/delay.h> +#include <linux/err.h> +#include <linux/irq.h> +#include <linux/irqdomain.h> +#include <linux/of.h> +#include <linux/sched/task_stack.h> +#include <linux/sched/mm.h> +#include <asm/irq.h> +#include <asm/traps.h> +#include <asm/sections.h> +#include <asm/mmu_context.h> +#include <asm/pgalloc.h> + +static struct { + unsigned long bits ____cacheline_aligned; +} ipi_data[NR_CPUS] __cacheline_aligned;
Why isn't this a per-cpu variable?
+
+enum ipi_message_type {
+ IPI_EMPTY,
+ IPI_RESCHEDULE,
+ IPI_CALL_FUNC,
+ IPI_MAX
+};
+
+static irqreturn_t handle_ipi(int irq, void *dev)
+{
+ unsigned long *pending_ipis = &ipi_data[smp_processor_id()].bits;
+
+ while (true) {
+ unsigned long ops;
+
+ ops = xchg(pending_ipis, 0);
+ if (ops == 0)
+ return IRQ_HANDLED;
+
+ if (ops & (1 << IPI_RESCHEDULE))
+ scheduler_ipi();
+
+ if (ops & (1 << IPI_CALL_FUNC))
+ generic_smp_call_function_interrupt();
+
+ BUG_ON((ops >> IPI_MAX) != 0);
+ }
+
+ return IRQ_HANDLED;
+}
+
+static void (*send_arch_ipi)(const struct cpumask *mask);
+
+static int ipi_irq;
+void __init set_send_ipi(void (*func)(const struct cpumask *mask), int irq)
+{
+ if (send_arch_ipi)
+ return;
+
+ send_arch_ipi = func;
+ ipi_irq = irq;
+}
+
+static void
+send_ipi_message(const struct cpumask *to_whom, enum ipi_message_type operation)
+{
+ int i;
+
+ for_each_cpu(i, to_whom)
+ set_bit(operation, &ipi_data[i].bits);
+
+ smp_mb();
+ send_arch_ipi(to_whom);
+}
+
+void arch_send_call_function_ipi_mask(struct cpumask *mask)
+{
+ send_ipi_message(mask, IPI_CALL_FUNC);
+}
+
+void arch_send_call_function_single_ipi(int cpu)
+{
+ send_ipi_message(cpumask_of(cpu), IPI_CALL_FUNC);
+}
+
+static void ipi_stop(void *unused)
+{
+ while (1);
+}
+
+void smp_send_stop(void)
+{
+ on_each_cpu(ipi_stop, NULL, 1);
+}
+
+void smp_send_reschedule(int cpu)
+{
+ send_ipi_message(cpumask_of(cpu), IPI_RESCHEDULE);
+}
+
+void *__cpu_up_stack_pointer[NR_CPUS];
+void *__cpu_up_task_pointer[NR_CPUS];Why aren't these per-cpu variables? More importantly, what are they used for? None of the patches in this series are using them.
+
+void __init smp_prepare_boot_cpu(void)
+{
+}
+
+void __init smp_prepare_cpus(unsigned int max_cpus)
+{
+}
+
+void __init enable_smp_ipi(void)
+{
+ enable_percpu_irq(ipi_irq, 0);
+}Why isn't this function static?
+
+static int ipi_dummy_dev;
+void __init setup_smp_ipi(void)
+{
+ int rc;
+
+ if (ipi_irq == 0)
+ panic("%s IRQ mapping failed\n", __func__);
+
+ rc = request_percpu_irq(ipi_irq, handle_ipi, "IPI Interrupt",
+ &ipi_dummy_dev);
+ if (rc)
+ panic("%s IRQ request failed\n", __func__);
+
+ enable_smp_ipi();
+}
+
+void __init setup_smp(void)
+{
+ struct device_node *node = NULL;
+ int cpu;
+
+ while ((node = of_find_node_by_type(node, "cpu"))) {
+ if (!of_device_is_available(node))
+ continue;
+
+ if (of_property_read_u32(node, "reg", &cpu))
+ continue;
+
+ if (cpu >= NR_CPUS)
+ continue;
+
+ set_cpu_possible(cpu, true);
+ set_cpu_present(cpu, true);
+ }
+}
+
+extern void _start_smp_secondary(void);
+
+volatile unsigned int secondary_hint;
+volatile unsigned int secondary_ccr;
+volatile unsigned int secondary_stack;This looks pretty dodgy. Shouldn't you be using READ_ONCE/WRITE_ONCE instead?
+
+int __cpu_up(unsigned int cpu, struct task_struct *tidle)
+{
+ unsigned int tmp;
+
+ secondary_stack = (unsigned int)tidle->stack + THREAD_SIZE;
+
+ secondary_hint = mfcr("cr31");
+
+ secondary_ccr = mfcr("cr18");> +
+ /* Flush dcache */
+ mtcr("cr17", 0x22);
+
+ /* Enable cpu in SMP reset ctrl reg */
+ tmp = mfcr("cr<29, 0>");
+ tmp |= 1 << cpu;
+ mtcr("cr<29, 0>", tmp);
+
+ /* Wait for the cpu online */
+ while (!cpu_online(cpu));
+
+ secondary_stack = 0;
+
+ return 0;
+}
+
+void __init smp_cpus_done(unsigned int max_cpus)
+{
+}
+
+int setup_profiling_timer(unsigned int multiplier)
+{
+ return -EINVAL;
+}
+
+void csky_start_secondary(void)
+{
+ struct mm_struct *mm = &init_mm;
+ unsigned int cpu = smp_processor_id();
+
+ mtcr("cr31", secondary_hint);
+ mtcr("cr18", secondary_ccr);
+
+ mtcr("vbr", vec_base);
+
+ flush_tlb_all();
+ write_mmu_pagemask(0);
+ TLBMISS_HANDLER_SETUP_PGD(swapper_pg_dir);
+ TLBMISS_HANDLER_SETUP_PGD_KERNEL(swapper_pg_dir);
+
+ asid_cache(smp_processor_id()) = ASID_FIRST_VERSION;
+
+#ifdef CONFIG_CPU_HAS_FPU
+ init_fpu();
+#endif
+
+ enable_smp_ipi();
+
+ mmget(mm);
+ mmgrab(mm);
+ current->active_mm = mm;
+ cpumask_set_cpu(cpu, mm_cpumask(mm));
+
+ notify_cpu_starting(cpu);
+ set_cpu_online(cpu, true);
+
+ pr_info("CPU%u Online: %s...\n", cpu, __func__);
+
+ local_irq_enable();
+ preempt_disable();
+ cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
+}Thanks, M. -- Jazz is not dead. It just smells funny...