Skip to contentSkip to conversation

RISC-V IPI Improvements

PATCH Superseded Anup Patel 3 participants 12 messages Nov 26, 2022 – Nov 29, 2022

Subsystems (5)

v12 of 18 Superseded Latest is v18 →

lore
Share & tools
lore

Anup Patel

author
#0 lore raw

v12 cover RISC-V IPI Improvements

This series aims to improve IPI support in Linux RISC-V in following ways:
1) Treat IPIs as normal per-CPU interrupts instead of having custom RISC-V specific hooks. This also makes Linux RISC-V IPI support aligned with other architectures.
2) Remote TLB flushes and icache flushes should prefer local IPIs instead of SBI calls whenever we have specialized hardware (such as RISC-V AIA IMSIC and RISC-V SWI) which allows S-mode software to directly inject IPIs without any assistance from M-mode runtime firmware.

These patches were originally part of the "Linux RISC-V ACLINT Support" series but this now a separate series so that it can be merged independently of the "Linux RISC-V ACLINT Support" series.
(Refer, https://lore.kernel.org/lkml/[email protected]/)

These patches are also a preparatory patches for the up-coming:
1) Linux RISC-V AIA support
2) Linux RISC-V SWI support

These patches can also be found in riscv_ipi_imp_v12 branch at:
https://github.com/avpatel/linux.git

Changes since v11:
- Removed ipi_mux_pre/post_handle() callbacks in PATCH3
- Removed sturct ipi_mux_ops in PATCH3
- Removed parent_virq and data pointer from everywhere in PATCH3
- Removed struct ipi_mux_control in PATCH3
- Improved function signature of ipi_mux_send() callback in PATCH3
- Used unsigned type with atomic operation in PATCH3

Changes since v10:
- Rebased on Linux-6.1-rc5
- Drop the "!(pending & ibit)" check in ipi_mux_send_mask() of PATCH3
- Disable local interrupts in ipi_mux_send_mask() of PATCH3 because we can be preempted while using a per-CPU temporary variable.

Changes since v9:
- Rebased on Linux-6.1-rc3
- Updated header comment block of ipi-mux.c in PATCH3
- Use a struct for global data of ipi-mux.c in PATCH3
- Add per-CPU temp cpumask for sending IPIs in PATCH3
- Drop the use of fwspec in PATCH3
- Use static key for ipi_mux_pre_handle() and ipi_mux_post_handle() in PATCH3
- Remove redundant pr_warn_ratelimited() called by ipi_mux_process() in PATCH3
- Remove CPUHP thingy from ipi_mux_create() in PATCH3

Changes since v8:
- Rebased on Linux-6.0-rc3
- Use dummy percpu data as parameter for request_percpu_irq() in PATCH4.

Changes since v7:
- Rebased on Linux-6.0-rc1
- Use atomic operations to track per-CPU pending and enabled IPIs in PATCH3. (Note: this is inspired from IPI muxing implemented in drivers/irqchip/irq-apple-aic.c)
- Made "struct ipi_mux_ops" (added by PATCH3) flexible so that drivers/irqchip/irq-apple-aic.c can adopt it in future.

Changes since v6:
- Rebased on Linux-5.19-rc7
- Added documentation for struct ipi_mux_ops in PATCH3
- Dropped dummy irq_mask()/unmask() in PATCH3
- Added const for "ipi_mux_chip" in PATCH3
- Removed "type" initialization from ipi_mux_domain_alloc() in PATCH3
- Dropped translate() from "ipi_mux_domain_ops" in PATCH3
- Improved barrier documentation in ipi_mux_process() of PATCH3
- Added percpu check in ipi_mux_create() for parent_virq of PATCH3
- Added nr_ipi parameter in ipi_mux_create() of PATCH3

Changes since v5:
- Rebased on Linux-5.18-rc3
- Used kernel doc style in PATCH3
- Removed redundant loop in ipi_mux_process() of PATCH3
- Removed "RISC-V" prefix form ipi_mux_chip.name of PATCH3
- Removed use of "this patch" in PATCH3 commit description
- Addressed few other nit comments in PATCH3

Changes since v4:
- Rebased on Linux-5.17
- Includes new PATCH3 which adds mechanism to multiplex a single HW IPI

Changes since v3:
- Rebased on Linux-5.17-rc6
- Updated PATCH2 to not export riscv_set_intc_hwnode_fn()
- Simplified riscv_intc_hwnode() in PATCH2

Changes since v2:
- Rebased on Linux-5.17-rc4
- Updated PATCH2 to not create synthetic INTC fwnode and instead provide a function which allows drivers to directly discover INTC fwnode

Changes since v1:
- Use synthetic fwnode for INTC instead of irq_set_default_host() in PATCH2

Anup Patel (7):
RISC-V: Clear SIP bit only when using SBI IPI operations irqchip/riscv-intc: Allow drivers to directly discover INTC hwnode
genirq: Add mechanism to multiplex a single HW IPI
RISC-V: Treat IPIs as normal Linux IRQs
RISC-V: Allow marking IPIs as suitable for remote FENCEs
RISC-V: Use IPIs for remote TLB flush when possible
RISC-V: Use IPIs for remote icache flush when possible

 arch/riscv/Kconfig                |   2 +
 arch/riscv/include/asm/irq.h      |   4 +
 arch/riscv/include/asm/sbi.h      |   7 +
 arch/riscv/include/asm/smp.h      |  49 +++++--
 arch/riscv/kernel/Makefile        |   1 +
 arch/riscv/kernel/cpu-hotplug.c   |   3 +-
 arch/riscv/kernel/irq.c           |  21 ++-
 arch/riscv/kernel/sbi-ipi.c       |  86 ++++++++++++
 arch/riscv/kernel/sbi.c           |  11 --
 arch/riscv/kernel/smp.c           | 166 ++++++++++++-----------
 arch/riscv/kernel/smpboot.c       |   5 +-
 arch/riscv/mm/cacheflush.c        |   5 +-
 arch/riscv/mm/tlbflush.c          |  93 ++++++++++---
 drivers/clocksource/timer-clint.c |  58 +++++++--
 drivers/irqchip/Kconfig           |   1 +
 drivers/irqchip/irq-riscv-intc.c  |  60 ++++-----
 include/linux/irq.h               |   4 +
 kernel/irq/Kconfig                |   5 +
 kernel/irq/Makefile               |   1 +
 kernel/irq/ipi-mux.c              | 210 ++++++++++++++++++++++++++++++
 20 files changed, 625 insertions(+), 167 deletions(-)
 create mode 100644 arch/riscv/kernel/sbi-ipi.c
 create mode 100644 kernel/irq/ipi-mux.c
Signature · 7 lines
2.34.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author
#1 lore raw

PATCH v12 1/7 RISC-V: Clear SIP bit only when using SBI IPI operations

The software interrupt pending (i.e. [M|S]SIP) bit is writeable for S-mode but read-only for M-mode so we clear this bit only when using SBI IPI operations.

---
 arch/riscv/kernel/sbi.c | 8 +++++++-
 arch/riscv/kernel/smp.c | 2 --
 2 files changed, 7 insertions(+), 3 deletions(-)
arch/riscv/kernel/sbi.c +7 -1
index 775d3322b422..fc614650a2e3 100644
@@ -643,8 +643,14 @@ static void sbi_send_cpumask_ipi(const struct cpumask *target)
 	sbi_send_ipi(target);
 }
 
+static void sbi_ipi_clear(void)
+{
+	csr_clear(CSR_IP, IE_SIE);
+}
+
 static const struct riscv_ipi_ops sbi_ipi_ops = {
-	.ipi_inject = sbi_send_cpumask_ipi
+	.ipi_inject = sbi_send_cpumask_ipi,
+	.ipi_clear = sbi_ipi_clear
 };
 
 void __init sbi_init(void)
arch/riscv/kernel/smp.c +0 -2
index 760a64518c58..c56d67f53ea9 100644
@@ -83,8 +83,6 @@ void riscv_clear_ipi(void)
 {
 	if (ipi_ops && ipi_ops->ipi_clear)
 		ipi_ops->ipi_clear();
-
-	csr_clear(CSR_IP, IE_SIE);
 }
 EXPORT_SYMBOL_GPL(riscv_clear_ipi);
 
Signature · 1 line
2.34.1
author
#2 lore raw

PATCH v12 2/7 irqchip/riscv-intc: Allow drivers to directly discover INTC hwnode

In reply to Anup Patel

Various RISC-V drivers (such as SBI IPI, SBI Timer, SBI PMU, and KVM RISC-V) don't have associated DT node but these drivers need standard per-CPU (local) interrupts defined by the RISC-V privileged specification.

We add riscv_get_intc_hwnode() in arch/riscv which allows RISC-V drivers not having DT node to discover INTC hwnode which in-turn helps these drivers to map per-CPU (local) interrupts provided by the INTC driver.

---
 arch/riscv/include/asm/irq.h     |  4 ++++
 arch/riscv/kernel/irq.c          | 18 ++++++++++++++++++
 drivers/irqchip/irq-riscv-intc.c |  7 +++++++
 3 files changed, 29 insertions(+)
arch/riscv/include/asm/irq.h +4 -0
index e4c435509983..43b9ebfbd943 100644
@@ -12,6 +12,10 @@
 
 #include <asm-generic/irq.h>
 
+void riscv_set_intc_hwnode_fn(struct fwnode_handle *(*fn)(void));
+
+struct fwnode_handle *riscv_get_intc_hwnode(void);
+
 extern void __init init_IRQ(void);
 
 #endif /* _ASM_RISCV_IRQ_H */
arch/riscv/kernel/irq.c +18 -0
index 7207fa08d78f..96d3171f0ca1 100644
@@ -7,9 +7,27 @@
 
 #include <linux/interrupt.h>
 #include <linux/irqchip.h>
+#include <linux/irqdomain.h>
+#include <linux/module.h>
 #include <linux/seq_file.h>
 #include <asm/smp.h>
 
+static struct fwnode_handle *(*__get_intc_node)(void);
+
+void riscv_set_intc_hwnode_fn(struct fwnode_handle *(*fn)(void))
+{
+	__get_intc_node = fn;
+}
+
+struct fwnode_handle *riscv_get_intc_hwnode(void)
+{
+	if (__get_intc_node)
+		return __get_intc_node();
+
+	return NULL;
+}
+EXPORT_SYMBOL_GPL(riscv_get_intc_hwnode);
+
 int arch_show_interrupts(struct seq_file *p, int prec)
 {
 	show_ipi_stats(p, prec);
drivers/irqchip/irq-riscv-intc.c +7 -0
index 499e5f81b3fe..9066467e99e4 100644
@@ -92,6 +92,11 @@ static const struct irq_domain_ops riscv_intc_domain_ops = {
 	.xlate	= irq_domain_xlate_onecell,
 };
 
+static struct fwnode_handle *riscv_intc_hwnode(void)
+{
+	return intc_domain->fwnode;
+}
+
 static int __init riscv_intc_init(struct device_node *node,
 				  struct device_node *parent)
 {
@@ -126,6 +131,8 @@ static int __init riscv_intc_init(struct device_node *node,
 		return rc;
 	}
 
+	riscv_set_intc_hwnode_fn(riscv_intc_hwnode);
+
 	cpuhp_setup_state(CPUHP_AP_IRQ_RISCV_STARTING,
 			  "irqchip/riscv/intc:starting",
 			  riscv_intc_cpu_starting,
Signature · 1 line
2.34.1
author
#3 lore raw

PATCH v12 3/7 genirq: Add mechanism to multiplex a single HW IPI

In reply to Anup Patel

All RISC-V platforms have a single HW IPI provided by the INTC local interrupt controller. The HW method to trigger INTC IPI can be through external irqchip (e.g. RISC-V AIA), through platform specific device (e.g. SiFive CLINT timer), or through firmware (e.g. SBI IPI call).

To support multiple IPIs on RISC-V, we add a generic IPI multiplexing mechanism which help us create multiple virtual IPIs using a single HW IPI. This generic IPI multiplexing is inspired from the Apple AIC irqchip driver and it is shared by various RISC-V irqchip drivers.

---
 include/linux/irq.h  |   4 +
 kernel/irq/Kconfig   |   5 ++
 kernel/irq/Makefile  |   1 +
 kernel/irq/ipi-mux.c | 210 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 220 insertions(+)
 create mode 100644 kernel/irq/ipi-mux.c
include/linux/irq.h +4 -0
index c3eb89606c2b..6024e1ee1257 100644
@@ -1266,6 +1266,10 @@ int __ipi_send_mask(struct irq_desc *desc, const struct cpumask *dest);
 int ipi_send_single(unsigned int virq, unsigned int cpu);
 int ipi_send_mask(unsigned int virq, const struct cpumask *dest);
 
+void ipi_mux_process(void);
+int ipi_mux_create(unsigned int nr_ipi,
+		   void (*mux_send)(const struct cpumask *));
+
 #ifdef CONFIG_GENERIC_IRQ_MULTI_HANDLER
 /*
  * Registers a generic IRQ handling function as the top-level IRQ handler in
kernel/irq/Kconfig +5 -0
index db3d174c53d4..df17dbc54b02 100644
@@ -86,6 +86,11 @@ config GENERIC_IRQ_IPI
 	depends on SMP
 	select IRQ_DOMAIN_HIERARCHY
 
+# Generic IRQ IPI Mux support
+config GENERIC_IRQ_IPI_MUX
+	bool
+	depends on SMP
+
 # Generic MSI interrupt support
 config GENERIC_MSI_IRQ
 	bool
kernel/irq/Makefile +1 -0
index b4f53717d143..f19d3080bf11 100644
@@ -15,6 +15,7 @@ obj-$(CONFIG_GENERIC_IRQ_MIGRATION) += cpuhotplug.o
 obj-$(CONFIG_PM_SLEEP) += pm.o
 obj-$(CONFIG_GENERIC_MSI_IRQ) += msi.o
 obj-$(CONFIG_GENERIC_IRQ_IPI) += ipi.o
+obj-$(CONFIG_GENERIC_IRQ_IPI_MUX) += ipi-mux.o
 obj-$(CONFIG_SMP) += affinity.o
 obj-$(CONFIG_GENERIC_IRQ_DEBUGFS) += debugfs.o
 obj-$(CONFIG_GENERIC_IRQ_MATRIX_ALLOCATOR) += matrix.o
kernel/irq/ipi-mux.c new file +210 -0

210 lines not on this page · raw message

Signature · 7 lines
2.34.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
#4 lore raw

Re: [PATCH v12 3/7] genirq: Add mechanism to multiplex a single HW IPI

Quoted 126 lines · Anup Patel

All RISC-V platforms have a single HW IPI provided by the INTC local interrupt controller. The HW method to trigger INTC IPI can be through external irqchip (e.g. RISC-V AIA), through platform specific device (e.g. SiFive CLINT timer), or through firmware (e.g. SBI IPI call).

To support multiple IPIs on RISC-V, we add a generic IPI multiplexing mechanism which help us create multiple virtual IPIs using a single HW IPI. This generic IPI multiplexing is inspired from the Apple AIC irqchip driver and it is shared by various RISC-V irqchip drivers.

---
 include/linux/irq.h  |   4 +
 kernel/irq/Kconfig   |   5 ++
 kernel/irq/Makefile  |   1 +
 kernel/irq/ipi-mux.c | 210 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 220 insertions(+)
 create mode 100644 kernel/irq/ipi-mux.c
include/linux/irq.h +4 -0
index c3eb89606c2b..6024e1ee1257 100644
@@ -1266,6 +1266,10 @@ int __ipi_send_mask(struct irq_desc *desc, const struct cpumask *dest);
 int ipi_send_single(unsigned int virq, unsigned int cpu);
 int ipi_send_mask(unsigned int virq, const struct cpumask *dest);
 
+void ipi_mux_process(void);
+int ipi_mux_create(unsigned int nr_ipi,
+		   void (*mux_send)(const struct cpumask *));
+
 #ifdef CONFIG_GENERIC_IRQ_MULTI_HANDLER
 /*
  * Registers a generic IRQ handling function as the top-level IRQ handler in
kernel/irq/Kconfig +5 -0
index db3d174c53d4..df17dbc54b02 100644
@@ -86,6 +86,11 @@ config GENERIC_IRQ_IPI
 	depends on SMP
 	select IRQ_DOMAIN_HIERARCHY
 
+# Generic IRQ IPI Mux support
+config GENERIC_IRQ_IPI_MUX
+	bool
+	depends on SMP
+
 # Generic MSI interrupt support
 config GENERIC_MSI_IRQ
 	bool
kernel/irq/Makefile +1 -0
index b4f53717d143..f19d3080bf11 100644
@@ -15,6 +15,7 @@ obj-$(CONFIG_GENERIC_IRQ_MIGRATION) += cpuhotplug.o
 obj-$(CONFIG_PM_SLEEP) += pm.o
 obj-$(CONFIG_GENERIC_MSI_IRQ) += msi.o
 obj-$(CONFIG_GENERIC_IRQ_IPI) += ipi.o
+obj-$(CONFIG_GENERIC_IRQ_IPI_MUX) += ipi-mux.o
 obj-$(CONFIG_SMP) += affinity.o
 obj-$(CONFIG_GENERIC_IRQ_DEBUGFS) += debugfs.o
 obj-$(CONFIG_GENERIC_IRQ_MATRIX_ALLOCATOR) += matrix.o
kernel/irq/ipi-mux.c new file +57 -0
new file mode 100644
index 000000000000..366d8cd5320b
@@ -0,0 +1,210 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Multiplex several virtual IPIs over a single HW IPI.
+ *
+ * Copyright The Asahi Linux Contributors
+ * Copyright (c) 2022 Ventana Micro Systems Inc.
+ */
+
+#define pr_fmt(fmt) "ipi-mux: " fmt
+#include <linux/cpu.h>
+#include <linux/init.h>
+#include <linux/irq.h>
+#include <linux/irqchip.h>
+#include <linux/irqchip/chained_irq.h>
+#include <linux/irqdomain.h>
+#include <linux/jump_label.h>
+#include <linux/percpu.h>
+#include <linux/smp.h>
+
+struct ipi_mux_cpu {
+	atomic_t			enable;
+	atomic_t			bits;
+	struct cpumask			send_mask;
+};
+
+static struct ipi_mux_cpu __percpu *ipi_mux_pcpu;
+static struct irq_domain *ipi_mux_domain;
+static void (*ipi_mux_send)(const struct cpumask *mask);
+
+static void ipi_mux_mask(struct irq_data *d)
+{
+	struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+
+	atomic_andnot(BIT(irqd_to_hwirq(d)), &icpu->enable);
+}
+
+static void ipi_mux_unmask(struct irq_data *d)
+{
+	u32 ibit = BIT(irqd_to_hwirq(d));
+	struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+
+	atomic_or(ibit, &icpu->enable);
+
+	/*
+	 * The atomic_or() above must complete before the atomic_read()
+	 * below to avoid racing ipi_mux_send_mask().
+	 */
+	smp_mb__after_atomic();
+
+	/* If a pending IPI was unmasked, raise a parent IPI immediately. */
+	if (atomic_read(&icpu->bits) & ibit)
+		ipi_mux_send(cpumask_of(smp_processor_id()));
+}
+
+static void ipi_mux_send_mask(struct irq_data *d, const struct cpumask *mask)
+{
+	u32 ibit = BIT(irqd_to_hwirq(d));

kernel/irq/ipi-mux.c line 67

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+	struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+	struct cpumask *send_mask = &icpu->send_mask;
+	unsigned long flags;
+	int cpu;
+
+	/*
+	 * We use send_mask as a per-CPU variable so disable local
+	 * interrupts to avoid being preempted.
+	 */
+	local_irq_save(flags);

The correct way to avoid preemption is to use preempt_disable(), which is a lot cheaper than disabling interrupt on most architectures.

kernel/irq/ipi-mux.c line 69

Quoted 2 lines
kernel/irq/ipi-mux.c new file +2 -0
+
+	cpumask_clear(send_mask);

This thing is likely to be unnecessarily expensive on very large systems, as it is proportional to the number of CPUs.

kernel/irq/ipi-mux.c line 73

Quoted 4 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+	for_each_cpu(cpu, mask) {
+		icpu = per_cpu_ptr(ipi_mux_pcpu, cpu);
+		atomic_or(ibit, &icpu->bits);

The original code had an atomic_fetch_or_release() to allow eliding the IPI if the target interrupt was already pending. Why is that code gone? This is a pretty cheap and efficient optimisation.

Quoted 4 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+		/*
+		 * The atomic_or() above must complete before
+		 * the atomic_read() below to avoid racing with

kernel/irq/ipi-mux.c line 87

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+		 * ipi_mux_unmask().
+		 */
+		smp_mb__after_atomic();
+
+		if (atomic_read(&icpu->enable) & ibit)
+			cpumask_set_cpu(cpu, send_mask);
+	}
+
+	/* Trigger the parent IPI */
+	ipi_mux_send(send_mask);

IPIs are very rarely made pending on more than a single CPU at a time. The overwhelming majority of them are targeting a single CPU. So accumulating bits to avoid doing two or more "send" actions only penalises the generic case.

My conclusion is that this "send_mask" can probably be removed, together with the preemption fiddling.

kernel/irq/ipi-mux.c line 97

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+
+	local_irq_restore(flags);
+}
+
+static const struct irq_chip ipi_mux_chip = {
+	.name		= "IPI Mux",
+	.irq_mask	= ipi_mux_mask,
+	.irq_unmask	= ipi_mux_unmask,
+	.ipi_send_mask	= ipi_mux_send_mask,
+};

OK, you have now dropped the superfluous pre/post handlers. But the need still exists. Case in point, the aic_handle_ipi() prologue and epilogue to the interrupt handling. I have suggested last time that the driver could provide the actual struct irq_chip in order to provide the callbacks it requires.

Please realise that I will not take this patch if this cannot be made to work with the single existing in-tree instance of an IPI MUX. 90% of the code having been lifted from there, I think this is a pretty fair ask.

kernel/irq/ipi-mux.c line 107

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+
+static int ipi_mux_domain_alloc(struct irq_domain *d, unsigned int virq,
+				unsigned int nr_irqs, void *arg)
+{
+	int i;
+
+	for (i = 0; i < nr_irqs; i++) {
+		irq_set_percpu_devid(virq + i);
+		irq_domain_set_info(d, virq + i, i,
+				    &ipi_mux_chip, d->host_data,

What does d->host_data represent here?

Quoted 25 lines
kernel/irq/ipi-mux.c new file +25 -0
+				    handle_percpu_devid_irq, NULL, NULL);
+	}
+
+	return 0;
+}
+
+static const struct irq_domain_ops ipi_mux_domain_ops = {
+	.alloc		= ipi_mux_domain_alloc,
+	.free		= irq_domain_free_irqs_top,
+};
+
+/**
+ * ipi_mux_process - Process multiplexed virtual IPIs
+ */
+void ipi_mux_process(void)
+{
+	struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+	irq_hw_number_t hwirq;
+	unsigned long ipis;
+	unsigned int en;
+
+	/*
+	 * Reading enable mask does not need to be ordered as long as
+	 * this function called from interrupt handler because only
+	 * the CPU itself can change it's own enable mask.

kernel/irq/ipi-mux.c line 142

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+	 */
+	en = atomic_read(&icpu->enable);
+
+	/*
+	 * Clear the IPIs we are about to handle. This pairs with the
+	 * atomic_fetch_or_release() in ipi_mux_send_mask().
+	 */
+	ipis = atomic_fetch_andnot(en, &icpu->bits) & en;
+
+	for_each_set_bit(hwirq, &ipis, BITS_PER_LONG)

BITS_PER_LONG...

Quoted 14 lines
kernel/irq/ipi-mux.c new file +14 -0
+		generic_handle_domain_irq(ipi_mux_domain, hwirq);
+}
+
+/**
+ * ipi_mux_create - Create virtual IPIs multiplexed on top of a single
+ * parent IPI.
+ * @nr_ipi:		number of virtual IPIs to create. This should
+ *			be <= BITS_PER_TYPE(int)
+ * @mux_send:		callback to trigger parent IPI
+ *
+ * Returns first virq of the newly created virtual IPIs upon success
+ * or <=0 upon failure
+ */
+int ipi_mux_create(unsigned int nr_ipi,

kernel/irq/ipi-mux.c line 166

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+		   void (*mux_send)(const struct cpumask *))
+{
+	struct fwnode_handle *fwnode;
+	struct irq_domain *domain;
+	int rc;
+
+	if (ipi_mux_domain)
+		return -EEXIST;
+
+	if (BITS_PER_TYPE(int) < nr_ipi || !mux_send)

... vs BITS_PER_TYPE(int) ...

	M.
Signature · 6 lines
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author
#5 lore raw

Re: [PATCH v12 3/7] genirq: Add mechanism to multiplex a single HW IPI

Quoted 134 lines · Marc Zyngier
Quoted 131 lines · Anup Patel

All RISC-V platforms have a single HW IPI provided by the INTC local interrupt controller. The HW method to trigger INTC IPI can be through external irqchip (e.g. RISC-V AIA), through platform specific device (e.g. SiFive CLINT timer), or through firmware (e.g. SBI IPI call).

To support multiple IPIs on RISC-V, we add a generic IPI multiplexing mechanism which help us create multiple virtual IPIs using a single HW IPI. This generic IPI multiplexing is inspired from the Apple AIC irqchip driver and it is shared by various RISC-V irqchip drivers.

---
 include/linux/irq.h  |   4 +
 kernel/irq/Kconfig   |   5 ++
 kernel/irq/Makefile  |   1 +
 kernel/irq/ipi-mux.c | 210 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 220 insertions(+)
 create mode 100644 kernel/irq/ipi-mux.c
include/linux/irq.h +4 -0
index c3eb89606c2b..6024e1ee1257 100644
@@ -1266,6 +1266,10 @@ int __ipi_send_mask(struct irq_desc *desc, const struct cpumask *dest);
 int ipi_send_single(unsigned int virq, unsigned int cpu);
 int ipi_send_mask(unsigned int virq, const struct cpumask *dest);
 
+void ipi_mux_process(void);
+int ipi_mux_create(unsigned int nr_ipi,
+                void (*mux_send)(const struct cpumask *));
+
 #ifdef CONFIG_GENERIC_IRQ_MULTI_HANDLER
 /*
  * Registers a generic IRQ handling function as the top-level IRQ handler in
kernel/irq/Kconfig +5 -0
index db3d174c53d4..df17dbc54b02 100644
@@ -86,6 +86,11 @@ config GENERIC_IRQ_IPI
      depends on SMP
      select IRQ_DOMAIN_HIERARCHY
 
+# Generic IRQ IPI Mux support
+config GENERIC_IRQ_IPI_MUX
+     bool
+     depends on SMP
+
 # Generic MSI interrupt support
 config GENERIC_MSI_IRQ
      bool
kernel/irq/Makefile +1 -0
index b4f53717d143..f19d3080bf11 100644
@@ -15,6 +15,7 @@ obj-$(CONFIG_GENERIC_IRQ_MIGRATION) += cpuhotplug.o
 obj-$(CONFIG_PM_SLEEP) += pm.o
 obj-$(CONFIG_GENERIC_MSI_IRQ) += msi.o
 obj-$(CONFIG_GENERIC_IRQ_IPI) += ipi.o
+obj-$(CONFIG_GENERIC_IRQ_IPI_MUX) += ipi-mux.o
 obj-$(CONFIG_SMP) += affinity.o
 obj-$(CONFIG_GENERIC_IRQ_DEBUGFS) += debugfs.o
 obj-$(CONFIG_GENERIC_IRQ_MATRIX_ALLOCATOR) += matrix.o
kernel/irq/ipi-mux.c new file +62 -0
new file mode 100644
index 000000000000..366d8cd5320b
@@ -0,0 +1,210 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Multiplex several virtual IPIs over a single HW IPI.
+ *
+ * Copyright The Asahi Linux Contributors
+ * Copyright (c) 2022 Ventana Micro Systems Inc.
+ */
+
+#define pr_fmt(fmt) "ipi-mux: " fmt
+#include <linux/cpu.h>
+#include <linux/init.h>
+#include <linux/irq.h>
+#include <linux/irqchip.h>
+#include <linux/irqchip/chained_irq.h>
+#include <linux/irqdomain.h>
+#include <linux/jump_label.h>
+#include <linux/percpu.h>
+#include <linux/smp.h>
+
+struct ipi_mux_cpu {
+     atomic_t                        enable;
+     atomic_t                        bits;
+     struct cpumask                  send_mask;
+};
+
+static struct ipi_mux_cpu __percpu *ipi_mux_pcpu;
+static struct irq_domain *ipi_mux_domain;
+static void (*ipi_mux_send)(const struct cpumask *mask);
+
+static void ipi_mux_mask(struct irq_data *d)
+{
+     struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+
+     atomic_andnot(BIT(irqd_to_hwirq(d)), &icpu->enable);
+}
+
+static void ipi_mux_unmask(struct irq_data *d)
+{
+     u32 ibit = BIT(irqd_to_hwirq(d));
+     struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+
+     atomic_or(ibit, &icpu->enable);
+
+     /*
+      * The atomic_or() above must complete before the atomic_read()
+      * below to avoid racing ipi_mux_send_mask().
+      */
+     smp_mb__after_atomic();
+
+     /* If a pending IPI was unmasked, raise a parent IPI immediately. */
+     if (atomic_read(&icpu->bits) & ibit)
+             ipi_mux_send(cpumask_of(smp_processor_id()));
+}
+
+static void ipi_mux_send_mask(struct irq_data *d, const struct cpumask *mask)
+{
+     u32 ibit = BIT(irqd_to_hwirq(d));
+     struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+     struct cpumask *send_mask = &icpu->send_mask;
+     unsigned long flags;
+     int cpu;
+
Quoted 8 lines

kernel/irq/ipi-mux.c line 67

Quoted 5 lines
kernel/irq/ipi-mux.c new file +5 -0
+     /*
+      * We use send_mask as a per-CPU variable so disable local
+      * interrupts to avoid being preempted.
+      */
+     local_irq_save(flags);

The correct way to avoid preemption is to use preempt_disable(), which is a lot cheaper than disabling interrupt on most architectures.

Okay, I will update.

Quoted 7 lines

kernel/irq/ipi-mux.c line 69

Quoted 3 lines
kernel/irq/ipi-mux.c new file +2 -0
+
+     cpumask_clear(send_mask);

This thing is likely to be unnecessarily expensive on very large systems, as it is proportional to the number of CPUs.

Quoted 8 lines

kernel/irq/ipi-mux.c line 73

Quoted 4 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+     for_each_cpu(cpu, mask) {
+             icpu = per_cpu_ptr(ipi_mux_pcpu, cpu);
+             atomic_or(ibit, &icpu->bits);

The original code had an atomic_fetch_or_release() to allow eliding the IPI if the target interrupt was already pending. Why is that code gone? This is a pretty cheap and efficient optimisation.

That optimization is causing RCU stalls on QEMU RISC-V virt machine with large number of CPUs.

Quoted 21 lines
Quoted 5 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+             /*
+              * The atomic_or() above must complete before
+              * the atomic_read() below to avoid racing with

kernel/irq/ipi-mux.c line 87

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+              * ipi_mux_unmask().
+              */
+             smp_mb__after_atomic();
+
+             if (atomic_read(&icpu->enable) & ibit)
+                     cpumask_set_cpu(cpu, send_mask);
+     }
+
+     /* Trigger the parent IPI */
+     ipi_mux_send(send_mask);

IPIs are very rarely made pending on more than a single CPU at a time. The overwhelming majority of them are targeting a single CPU. So accumulating bits to avoid doing two or more "send" actions only penalises the generic case.

Quoted 2 lines

My conclusion is that this "send_mask" can probably be removed, together with the preemption fiddling.

So, we should call ipi_mux_send() for one target CPU at a time ?

Quoted 17 lines

kernel/irq/ipi-mux.c line 97

Quoted 11 lines
kernel/irq/ipi-mux.c new file +10 -0
+
+     local_irq_restore(flags);
+}
+
+static const struct irq_chip ipi_mux_chip = {
+     .name           = "IPI Mux",
+     .irq_mask       = ipi_mux_mask,
+     .irq_unmask     = ipi_mux_unmask,
+     .ipi_send_mask  = ipi_mux_send_mask,
+};

OK, you have now dropped the superfluous pre/post handlers. But the need still exists. Case in point, the aic_handle_ipi() prologue and epilogue to the interrupt handling. I have suggested last time that the driver could provide the actual struct irq_chip in order to provide the callbacks it requires.

The aic_handle_ipi() can simply call ipi_mux_process() between the prologue and epilogue.

Quoted 5 lines

Please realise that I will not take this patch if this cannot be made to work with the single existing in-tree instance of an IPI MUX. 90% of the code having been lifted from there, I think this is a pretty fair ask.

Only the muxing part of AIC has been factored out. All the register programming of AIC will remain in the AIA irqchip driver without any change in sequence.

Quoted 13 lines

kernel/irq/ipi-mux.c line 107

Quoted 11 lines
kernel/irq/ipi-mux.c new file +10 -0
+
+static int ipi_mux_domain_alloc(struct irq_domain *d, unsigned int virq,
+                             unsigned int nr_irqs, void *arg)
+{
+     int i;
+
+     for (i = 0; i < nr_irqs; i++) {
+             irq_set_percpu_devid(virq + i);
+             irq_domain_set_info(d, virq + i, i,
+                                 &ipi_mux_chip, d->host_data,

What does d->host_data represent here?

It's always NULL so we don't need to pass it. I will update.

Quoted 30 lines
Quoted 30 lines
kernel/irq/ipi-mux.c new file +29 -0
+                                 handle_percpu_devid_irq, NULL, NULL);
+     }
+
+     return 0;
+}
+
+static const struct irq_domain_ops ipi_mux_domain_ops = {
+     .alloc          = ipi_mux_domain_alloc,
+     .free           = irq_domain_free_irqs_top,
+};
+
+/**
+ * ipi_mux_process - Process multiplexed virtual IPIs
+ */
+void ipi_mux_process(void)
+{
+     struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+     irq_hw_number_t hwirq;
+     unsigned long ipis;
+     unsigned int en;
+
+     /*
+      * Reading enable mask does not need to be ordered as long as
+      * this function called from interrupt handler because only
+      * the CPU itself can change it's own enable mask.
+      */
+     en = atomic_read(&icpu->enable);
+
+     /*
Quoted 8 lines

kernel/irq/ipi-mux.c line 142

Quoted 6 lines
kernel/irq/ipi-mux.c new file +6 -0
+      * Clear the IPIs we are about to handle. This pairs with the
+      * atomic_fetch_or_release() in ipi_mux_send_mask().
+      */
+     ipis = atomic_fetch_andnot(en, &icpu->bits) & en;
+
+     for_each_set_bit(hwirq, &ipis, BITS_PER_LONG)

BITS_PER_LONG...

Argh, I should have used BITS_PER_TYPE(int) here. I will update.

Quoted 28 lines
Quoted 15 lines
kernel/irq/ipi-mux.c new file +14 -0
+             generic_handle_domain_irq(ipi_mux_domain, hwirq);
+}
+
+/**
+ * ipi_mux_create - Create virtual IPIs multiplexed on top of a single
+ * parent IPI.
+ * @nr_ipi:          number of virtual IPIs to create. This should
+ *                   be <= BITS_PER_TYPE(int)
+ * @mux_send:                callback to trigger parent IPI
+ *
+ * Returns first virq of the newly created virtual IPIs upon success
+ * or <=0 upon failure
+ */
+int ipi_mux_create(unsigned int nr_ipi,

kernel/irq/ipi-mux.c line 166

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+                void (*mux_send)(const struct cpumask *))
+{
+     struct fwnode_handle *fwnode;
+     struct irq_domain *domain;
+     int rc;
+
+     if (ipi_mux_domain)
+             return -EEXIST;
+
+     if (BITS_PER_TYPE(int) < nr_ipi || !mux_send)

... vs BITS_PER_TYPE(int) ...

Quoted 4 lines
        M.

--
Without deviation from the norm, progress is not possible.

Regards,
Anup

#6 lore raw

Re: [PATCH v12 3/7] genirq: Add mechanism to multiplex a single HW IPI

Quoted 37 lines · Anup Patel
Quoted 4 lines · Marc Zyngier
Quoted 1 line · Anup Patel
Quoted 16 lines

kernel/irq/ipi-mux.c line 67

Quoted 13 lines
kernel/irq/ipi-mux.c new file +13 -0
+static void ipi_mux_send_mask(struct irq_data *d, const struct cpumask *mask)
+{
+     u32 ibit = BIT(irqd_to_hwirq(d));
+     struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+     struct cpumask *send_mask = &icpu->send_mask;
+     unsigned long flags;
+     int cpu;
+
+     /*
+      * We use send_mask as a per-CPU variable so disable local
+      * interrupts to avoid being preempted.
+      */
+     local_irq_save(flags);

The correct way to avoid preemption is to use preempt_disable(), which is a lot cheaper than disabling interrupt on most architectures.

Okay, I will update.

Quoted 12 lines

Review

Quoted 3 lines
(quoted diff) +2 -0
+
+     cpumask_clear(send_mask);

This thing is likely to be unnecessarily expensive on very large systems, as it is proportional to the number of CPUs.

Quoted 4 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+     for_each_cpu(cpu, mask) {
+             icpu = per_cpu_ptr(ipi_mux_pcpu, cpu);
+             atomic_or(ibit, &icpu->bits);
Quoted 6 lines
Quoted 3 lines

The original code had an atomic_fetch_or_release() to allow eliding the IPI if the target interrupt was already pending. Why is that code gone? This is a pretty cheap and efficient optimisation.

That optimization is causing RCU stalls on QEMU RISC-V virt machine with large number of CPUs.

Then there is a bug somewhere, either in the implementation of the atomic operations or in QEMU. Or maybe even in the original code (though this looks unlikely given how heavily this is used on actual HW - I'm typing this email from one of these machines, and I'd be pretty annoyed if I was missing IPIs).

In any case, please don't paper over this.

Quoted 22 lines
Quoted 22 lines
Quoted 6 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+             /*
+              * The atomic_or() above must complete before
+              * the atomic_read() below to avoid racing with

kernel/irq/ipi-mux.c line 87

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+              * ipi_mux_unmask().
+              */
+             smp_mb__after_atomic();
+
+             if (atomic_read(&icpu->enable) & ibit)
+                     cpumask_set_cpu(cpu, send_mask);
+     }
+
+     /* Trigger the parent IPI */
+     ipi_mux_send(send_mask);

IPIs are very rarely made pending on more than a single CPU at a time. The overwhelming majority of them are targeting a single CPU. So accumulating bits to avoid doing two or more "send" actions only penalises the generic case.

Quoted 4 lines
Quoted 2 lines

My conclusion is that this "send_mask" can probably be removed, together with the preemption fiddling.

So, we should call ipi_mux_send() for one target CPU at a time ?

I think so, as it matches my measurements from a few years ago. It also simplifies things significantly, leading to better performance for the common case. Add some instrumentation and see whether this is still the case though.

Quoted 19 lines
Quoted 18 lines

kernel/irq/ipi-mux.c line 97

Quoted 12 lines
kernel/irq/ipi-mux.c new file +10 -0
+
+     local_irq_restore(flags);
+}
+
+static const struct irq_chip ipi_mux_chip = {
+     .name           = "IPI Mux",
+     .irq_mask       = ipi_mux_mask,
+     .irq_unmask     = ipi_mux_unmask,
+     .ipi_send_mask  = ipi_mux_send_mask,
+};

OK, you have now dropped the superfluous pre/post handlers. But the need still exists. Case in point, the aic_handle_ipi() prologue and epilogue to the interrupt handling. I have suggested last time that the driver could provide the actual struct irq_chip in order to provide the callbacks it requires.

Quoted 2 lines

The aic_handle_ipi() can simply call ipi_mux_process() between the prologue and epilogue.

Hmm. OK. That's not what I had in mind, but fair enough.

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
maintainer
#7 lore raw

Re: [PATCH v12 3/7] genirq: Add mechanism to multiplex a single HW IPI

Quoted 47 lines · Marc Zyngier
Quoted 37 lines · Anup Patel
Quoted 4 lines · Marc Zyngier
Quoted 1 line · Anup Patel
Quoted 16 lines

kernel/irq/ipi-mux.c line 67

Quoted 13 lines
kernel/irq/ipi-mux.c new file +13 -0
+static void ipi_mux_send_mask(struct irq_data *d, const struct cpumask *mask)
+{
+     u32 ibit = BIT(irqd_to_hwirq(d));
+     struct ipi_mux_cpu *icpu = this_cpu_ptr(ipi_mux_pcpu);
+     struct cpumask *send_mask = &icpu->send_mask;
+     unsigned long flags;
+     int cpu;
+
+     /*
+      * We use send_mask as a per-CPU variable so disable local
+      * interrupts to avoid being preempted.
+      */
+     local_irq_save(flags);

The correct way to avoid preemption is to use preempt_disable(), which is a lot cheaper than disabling interrupt on most architectures.

Okay, I will update.

Quoted 12 lines

Review

Quoted 3 lines
(quoted diff) +2 -0
+
+     cpumask_clear(send_mask);

This thing is likely to be unnecessarily expensive on very large systems, as it is proportional to the number of CPUs.

Quoted 4 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+     for_each_cpu(cpu, mask) {
+             icpu = per_cpu_ptr(ipi_mux_pcpu, cpu);
+             atomic_or(ibit, &icpu->bits);
Quoted 6 lines
Quoted 3 lines

The original code had an atomic_fetch_or_release() to allow eliding the IPI if the target interrupt was already pending. Why is that code gone? This is a pretty cheap and efficient optimisation.

That optimization is causing RCU stalls on QEMU RISC-V virt machine with large number of CPUs.

Quoted 7 lines

Then there is a bug somewhere, either in the implementation of the atomic operations or in QEMU. Or maybe even in the original code (though this looks unlikely given how heavily this is used on actual HW - I'm typing this email from one of these machines, and I'd be pretty annoyed if I was missing IPIs).

In any case, please don't paper over this.

I was trying to defer the optimization to a later stage until this issue was fixed for RISC-V.

Anyways, I found the root cause. This turned out to be missing broadcast timer initialization in time_init() for RISC-V. Removing the optimization over here was simply hiding the issue.

I will bring back the optimization in the next patch revision.

Quoted 26 lines
Quoted 26 lines
Quoted 25 lines
Quoted 7 lines
kernel/irq/ipi-mux.c new file +4 -0
+
+             /*
+              * The atomic_or() above must complete before
+              * the atomic_read() below to avoid racing with

kernel/irq/ipi-mux.c line 87

Quoted 10 lines
kernel/irq/ipi-mux.c new file +10 -0
+              * ipi_mux_unmask().
+              */
+             smp_mb__after_atomic();
+
+             if (atomic_read(&icpu->enable) & ibit)
+                     cpumask_set_cpu(cpu, send_mask);
+     }
+
+     /* Trigger the parent IPI */
+     ipi_mux_send(send_mask);

IPIs are very rarely made pending on more than a single CPU at a time. The overwhelming majority of them are targeting a single CPU. So accumulating bits to avoid doing two or more "send" actions only penalises the generic case.

My conclusion is that this "send_mask" can probably be removed, together with the preemption fiddling.

Quoted 6 lines
Quoted 1 line

So, we should call ipi_mux_send() for one target CPU at a time ?

I think so, as it matches my measurements from a few years ago. It also simplifies things significantly, leading to better performance for the common case. Add some instrumentation and see whether this is still the case though.

I did not see any difference in the hackbench running on QEMU RISC-V. I will simplify ipi_mux_send() like you suggested.

Quoted 23 lines
Quoted 20 lines
Quoted 19 lines

kernel/irq/ipi-mux.c line 97

Quoted 13 lines
kernel/irq/ipi-mux.c new file +10 -0
+
+     local_irq_restore(flags);
+}
+
+static const struct irq_chip ipi_mux_chip = {
+     .name           = "IPI Mux",
+     .irq_mask       = ipi_mux_mask,
+     .irq_unmask     = ipi_mux_unmask,
+     .ipi_send_mask  = ipi_mux_send_mask,
+};

OK, you have now dropped the superfluous pre/post handlers. But the need still exists. Case in point, the aic_handle_ipi() prologue and epilogue to the interrupt handling. I have suggested last time that the driver could provide the actual struct irq_chip in order to provide the callbacks it requires.

Quoted 2 lines

The aic_handle_ipi() can simply call ipi_mux_process() between the prologue and epilogue.

Quoted 6 lines

Hmm. OK. That's not what I had in mind, but fair enough.

        M.

--
Without deviation from the norm, progress is not possible.

Regards,
Anup

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

author
#8 lore raw

PATCH v12 4/7 RISC-V: Treat IPIs as normal Linux IRQs

In reply to Anup Patel

Currently, the RISC-V kernel provides arch specific hooks (i.e. struct riscv_ipi_ops) to register IPI handling methods. The stats gathering of IPIs is also arch specific in the RISC-V kernel.

Other architectures (such as ARM, ARM64, and MIPS) have moved away from custom arch specific IPI handling methods. Currently, these architectures have Linux irqchip drivers providing a range of Linux IRQ numbers to be used as IPIs and IPI triggering is done using generic IPI APIs. This approach allows architectures to treat IPIs as normal Linux IRQs and IPI stats gathering is done by the generic Linux IRQ subsystem.

We extend the RISC-V IPI handling as-per above approach so that arch specific IPI handling methods (struct riscv_ipi_ops) can be removed and the IPI handling is done through the Linux IRQ subsystem.

---
 arch/riscv/Kconfig                |   2 +
 arch/riscv/include/asm/sbi.h      |   7 ++
 arch/riscv/include/asm/smp.h      |  35 ++++---
 arch/riscv/kernel/Makefile        |   1 +
 arch/riscv/kernel/cpu-hotplug.c   |   3 +-
 arch/riscv/kernel/irq.c           |   3 +-
 arch/riscv/kernel/sbi-ipi.c       |  86 +++++++++++++++++
 arch/riscv/kernel/sbi.c           |  17 ----
 arch/riscv/kernel/smp.c           | 155 +++++++++++++++---------------
 arch/riscv/kernel/smpboot.c       |   5 +-
 drivers/clocksource/timer-clint.c |  58 ++++++++---
 drivers/irqchip/Kconfig           |   1 +
 drivers/irqchip/irq-riscv-intc.c  |  55 +++++------
 13 files changed, 273 insertions(+), 155 deletions(-)
 create mode 100644 arch/riscv/kernel/sbi-ipi.c
arch/riscv/Kconfig +2 -0
index c0e22648bd16..2fc20a189425 100644
@@ -61,6 +61,8 @@ config RISCV
 	select GENERIC_GETTIMEOFDAY if HAVE_GENERIC_VDSO
 	select GENERIC_IDLE_POLL_SETUP
 	select GENERIC_IOREMAP if MMU
+	select GENERIC_IRQ_IPI if SMP
+	select GENERIC_IRQ_IPI_MUX if SMP
 	select GENERIC_IRQ_MULTI_HANDLER
 	select GENERIC_IRQ_SHOW
 	select GENERIC_IRQ_SHOW_LEVEL
arch/riscv/include/asm/sbi.h +7 -0
index 2a0ef738695e..c43f115714a0 100644
@@ -327,4 +327,11 @@ int sbi_err_map_linux_errno(int err);
 static inline int sbi_remote_fence_i(const struct cpumask *cpu_mask) { return -1; }
 static inline void sbi_init(void) {}
 #endif /* CONFIG_RISCV_SBI */
+
+#if IS_ENABLED(CONFIG_SMP) && IS_ENABLED(CONFIG_RISCV_SBI)
+void sbi_ipi_init(void);
+#else
+static inline void sbi_ipi_init(void) { }
+#endif
+
 #endif /* _ASM_RISCV_SBI_H */
arch/riscv/include/asm/smp.h +21 -14
index d3443be7eedc..79ed0b73cd4e 100644
@@ -15,11 +15,6 @@
 struct seq_file;
 extern unsigned long boot_cpu_hartid;
 
-struct riscv_ipi_ops {
-	void (*ipi_inject)(const struct cpumask *target);
-	void (*ipi_clear)(void);
-};
-
 #ifdef CONFIG_SMP
 /*
  * Mapping between linux logical cpu index and hartid.
@@ -33,9 +28,6 @@ void show_ipi_stats(struct seq_file *p, int prec);
 /* SMP initialization hook for setup_arch */
 void __init setup_smp(void);
 
-/* Called from C code, this handles an IPI. */
-void handle_IPI(struct pt_regs *regs);
-
 /* Hook for the generic smp_call_function_many() routine. */
 void arch_send_call_function_ipi_mask(struct cpumask *mask);
 
@@ -44,11 +36,17 @@ void arch_send_call_function_single_ipi(int cpu);
 
 int riscv_hartid_to_cpuid(unsigned long hartid);
 
-/* Set custom IPI operations */
-void riscv_set_ipi_ops(const struct riscv_ipi_ops *ops);
+/* Enable IPI for CPU hotplug */
+void riscv_ipi_enable(void);
+
+/* Disable IPI for CPU hotplug */
+void riscv_ipi_disable(void);
 
-/* Clear IPI for current CPU */
-void riscv_clear_ipi(void);
+/* Check if IPI interrupt numbers are available */
+bool riscv_ipi_have_virq_range(void);
+
+/* Set the IPI interrupt numbers for arch (called by irqchip drivers) */
+void riscv_ipi_set_virq_range(int virq, int nr);
 
 /* Secondary hart entry */
 asmlinkage void smp_callin(void);
@@ -82,11 +80,20 @@ static inline unsigned long cpuid_to_hartid_map(int cpu)
 	return boot_cpu_hartid;
 }
 
-static inline void riscv_set_ipi_ops(const struct riscv_ipi_ops *ops)
+static inline void riscv_ipi_enable(void)
 {
 }
 
-static inline void riscv_clear_ipi(void)
+static inline void riscv_ipi_disable(void)
+{
+}
+
+static inline bool riscv_ipi_have_virq_range(void)
+{
+	return false;
+}
+
+static inline void riscv_ipi_set_virq_range(int virq, int nr)
 {
 }
 
arch/riscv/kernel/Makefile +1 -0
index db6e4b1294ba..939f60f971a4 100644
@@ -74,6 +74,7 @@ obj-$(CONFIG_PERF_EVENTS)	+= perf_callchain.o
 obj-$(CONFIG_HAVE_PERF_REGS)	+= perf_regs.o
 obj-$(CONFIG_RISCV_SBI)		+= sbi.o
 ifeq ($(CONFIG_RISCV_SBI), y)
+obj-$(CONFIG_SMP)		+= sbi-ipi.o
 obj-$(CONFIG_SMP) += cpu_ops_sbi.o
 endif
 obj-$(CONFIG_HOTPLUG_CPU)	+= cpu-hotplug.o
arch/riscv/kernel/cpu-hotplug.c +2 -1
index f7a832e3a1d1..39235cf50652 100644
@@ -13,7 +13,7 @@
 #include <asm/irq.h>
 #include <asm/cpu_ops.h>
 #include <asm/numa.h>
-#include <asm/sbi.h>
+#include <asm/smp.h>
 
 bool cpu_has_hotplug(unsigned int cpu)
 {
@@ -43,6 +43,7 @@ int __cpu_disable(void)
 	remove_cpu_topology(cpu);
 	numa_remove_cpu(cpu);
 	set_cpu_online(cpu, false);
+	riscv_ipi_disable();
 	irq_migrate_all_off_this_cpu();
 
 	return ret;
arch/riscv/kernel/irq.c +2 -1
index 96d3171f0ca1..eb9a68a539e6 100644
@@ -10,7 +10,7 @@
 #include <linux/irqdomain.h>
 #include <linux/module.h>
 #include <linux/seq_file.h>
-#include <asm/smp.h>
+#include <asm/sbi.h>
 
 static struct fwnode_handle *(*__get_intc_node)(void);
 
@@ -39,4 +39,5 @@ void __init init_IRQ(void)
 	irqchip_init();
 	if (!handle_arch_irq)
 		panic("No interrupt controller found.");
+	sbi_ipi_init();
 }
arch/riscv/kernel/sbi-ipi.c new file +86 -0

86 lines not on this page · raw message

arch/riscv/kernel/sbi.c +0 -17
index fc614650a2e3..e9f04eba0e09 100644
@@ -638,21 +638,6 @@ long sbi_get_mimpid(void)
 	return __sbi_base_ecall(SBI_EXT_BASE_GET_MIMPID);
 }
 
-static void sbi_send_cpumask_ipi(const struct cpumask *target)
-{
-	sbi_send_ipi(target);
-}
-
-static void sbi_ipi_clear(void)
-{
-	csr_clear(CSR_IP, IE_SIE);
-}
-
-static const struct riscv_ipi_ops sbi_ipi_ops = {
-	.ipi_inject = sbi_send_cpumask_ipi,
-	.ipi_clear = sbi_ipi_clear
-};
-
 void __init sbi_init(void)
 {
 	int ret;
@@ -699,6 +684,4 @@ void __init sbi_init(void)
 		__sbi_send_ipi	= __sbi_send_ipi_v01;
 		__sbi_rfence	= __sbi_rfence_v01;
 	}
-
-	riscv_set_ipi_ops(&sbi_ipi_ops);
 }
arch/riscv/kernel/smp.c +78 -77

211 lines not on this page · raw message

arch/riscv/kernel/smpboot.c +2 -3
index 3373df413c88..9cbdb960515b 100644
@@ -30,7 +30,6 @@
 #include <asm/numa.h>
 #include <asm/tlbflush.h>
 #include <asm/sections.h>
-#include <asm/sbi.h>
 #include <asm/smp.h>
 
 #include "head.h"
@@ -157,12 +156,12 @@ asmlinkage __visible void smp_callin(void)
 	struct mm_struct *mm = &init_mm;
 	unsigned int curr_cpuid = smp_processor_id();
 
-	riscv_clear_ipi();
-
 	/* All kernel threads share the same mm context.  */
 	mmgrab(mm);
 	current->active_mm = mm;
 
+	riscv_ipi_enable();
+
 	store_cpu_topology(curr_cpuid);
 	notify_cpu_starting(curr_cpuid);
 	numa_add_cpu(curr_cpuid);
drivers/clocksource/timer-clint.c +47 -11

121 lines not on this page · raw message

drivers/irqchip/Kconfig +1 -0
index 7ef9f5e696d3..131379aa8424 100644
@@ -540,6 +540,7 @@ config TI_PRUSS_INTC
 config RISCV_INTC
 	bool "RISC-V Local Interrupt Controller"
 	depends on RISCV
+	select IRQ_DOMAIN_HIERARCHY
 	default y
 	help
 	   This enables support for the per-HART local interrupt controller
drivers/irqchip/irq-riscv-intc.c +24 -31

82 lines not on this page · raw message

Signature · 7 lines
2.34.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author
#9 lore raw

PATCH v12 5/7 RISC-V: Allow marking IPIs as suitable for remote FENCEs

In reply to Anup Patel

To do remote FENCEs (i.e. remote TLB flushes) using IPI calls on the RISC-V kernel, we need hardware mechanism to directly inject IPI from the supervisor mode (i.e. RISC-V kernel) instead of using SBI calls.

The upcoming AIA IMSIC devices allow direct IPI injection from the supervisor mode (i.e. RISC-V kernel). To support this, we extend the riscv_ipi_set_virq_range() function so that IPI provider (i.e. irqchip drivers can mark IPIs as suitable for remote FENCEs.

---
 arch/riscv/include/asm/smp.h      | 18 ++++++++++++++++--
 arch/riscv/kernel/sbi-ipi.c       |  2 +-
 arch/riscv/kernel/smp.c           | 11 ++++++++++-
 drivers/clocksource/timer-clint.c |  2 +-
 4 files changed, 28 insertions(+), 5 deletions(-)
arch/riscv/include/asm/smp.h +16 -2
index 79ed0b73cd4e..56976e41a21e 100644
@@ -16,6 +16,9 @@ struct seq_file;
 extern unsigned long boot_cpu_hartid;
 
 #ifdef CONFIG_SMP
+
+#include <linux/jump_label.h>
+
 /*
  * Mapping between linux logical cpu index and hartid.
  */
@@ -46,7 +49,12 @@ void riscv_ipi_disable(void);
 bool riscv_ipi_have_virq_range(void);
 
 /* Set the IPI interrupt numbers for arch (called by irqchip drivers) */
-void riscv_ipi_set_virq_range(int virq, int nr);
+void riscv_ipi_set_virq_range(int virq, int nr, bool use_for_rfence);
+
+/* Check if we can use IPIs for remote FENCEs */
+DECLARE_STATIC_KEY_FALSE(riscv_ipi_for_rfence);
+#define riscv_use_ipi_for_rfence() \
+	static_branch_unlikely(&riscv_ipi_for_rfence)
 
 /* Secondary hart entry */
 asmlinkage void smp_callin(void);
@@ -93,10 +101,16 @@ static inline bool riscv_ipi_have_virq_range(void)
 	return false;
 }
 
-static inline void riscv_ipi_set_virq_range(int virq, int nr)
+static inline void riscv_ipi_set_virq_range(int virq, int nr,
+					    bool use_for_rfence)
 {
 }
 
+static inline bool riscv_use_ipi_for_rfence(void)
+{
+	return false;
+}
+
 #endif /* CONFIG_SMP */
 
 #if defined(CONFIG_HOTPLUG_CPU) && (CONFIG_SMP)
arch/riscv/kernel/sbi-ipi.c +1 -1
index b1f654b861fc..d4cd4d96ab49 100644
@@ -81,6 +81,6 @@ void __init sbi_ipi_init(void)
 			  "irqchip/sbi-ipi:starting",
 			  sbi_ipi_starting_cpu, sbi_ipi_dying_cpu);
 
-	riscv_ipi_set_virq_range(virq, BITS_PER_BYTE);
+	riscv_ipi_set_virq_range(virq, BITS_PER_BYTE, false);
 	pr_info("providing IPIs using SBI IPI extension\n");
 }
arch/riscv/kernel/smp.c +10 -1
index e8a20454d65b..74b8cb1a89ab 100644
@@ -145,7 +145,10 @@ bool riscv_ipi_have_virq_range(void)
 	return (ipi_virq_base) ? true : false;
 }
 
-void riscv_ipi_set_virq_range(int virq, int nr)
+DEFINE_STATIC_KEY_FALSE(riscv_ipi_for_rfence);
+EXPORT_SYMBOL_GPL(riscv_ipi_for_rfence);
+
+void riscv_ipi_set_virq_range(int virq, int nr, bool use_for_rfence)
 {
 	int i, err;
 
@@ -168,6 +171,12 @@ void riscv_ipi_set_virq_range(int virq, int nr)
 
 	/* Enabled IPIs for boot CPU immediately */
 	riscv_ipi_enable();
+
+	/* Update RFENCE static key */
+	if (use_for_rfence)
+		static_branch_enable(&riscv_ipi_for_rfence);
+	else
+		static_branch_disable(&riscv_ipi_for_rfence);
 }
 EXPORT_SYMBOL_GPL(riscv_ipi_set_virq_range);
 
drivers/clocksource/timer-clint.c +1 -1
index d5dbe8883e40..1146753a373a 100644
@@ -251,7 +251,7 @@ static int __init clint_timer_init_dt(struct device_node *np)
 		goto fail_free_irq;
 	}
 
-	riscv_ipi_set_virq_range(rc, BITS_PER_BYTE);
+	riscv_ipi_set_virq_range(rc, BITS_PER_BYTE, true);
 	clint_clear_ipi();
 #endif
 
Signature · 7 lines
2.34.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author
#10 lore raw

PATCH v12 6/7 RISC-V: Use IPIs for remote TLB flush when possible

In reply to Anup Patel

If we have specialized interrupt controller (such as AIA IMSIC) which allows supervisor mode to directly inject IPIs without any assistance from M-mode or HS-mode then using such specialized interrupt controller, we can do remote TLB flushes directly from supervisor mode instead of using the SBI RFENCE calls.

This patch extends remote TLB flush functions to use supervisor mode IPIs whenever direct supervisor mode IPIs.are supported by interrupt controller.

---
 arch/riscv/mm/tlbflush.c | 93 +++++++++++++++++++++++++++++++++-------
 1 file changed, 78 insertions(+), 15 deletions(-)
arch/riscv/mm/tlbflush.c +78 -15
index 37ed760d007c..27a7db8eb2c4 100644
@@ -23,14 +23,62 @@ static inline void local_flush_tlb_page_asid(unsigned long addr,
 			: "memory");
 }
 
+static inline void local_flush_tlb_range(unsigned long start,
+		unsigned long size, unsigned long stride)
+{
+	if (size <= stride)
+		local_flush_tlb_page(start);
+	else
+		local_flush_tlb_all();
+}
+
+static inline void local_flush_tlb_range_asid(unsigned long start,
+		unsigned long size, unsigned long stride, unsigned long asid)
+{
+	if (size <= stride)
+		local_flush_tlb_page_asid(start, asid);
+	else
+		local_flush_tlb_all_asid(asid);
+}
+
+static void __ipi_flush_tlb_all(void *info)
+{
+	local_flush_tlb_all();
+}
+
 void flush_tlb_all(void)
 {
-	sbi_remote_sfence_vma(NULL, 0, -1);
+	if (riscv_use_ipi_for_rfence())
+		on_each_cpu(__ipi_flush_tlb_all, NULL, 1);
+	else
+		sbi_remote_sfence_vma(NULL, 0, -1);
+}
+
+struct flush_tlb_range_data {
+	unsigned long asid;
+	unsigned long start;
+	unsigned long size;
+	unsigned long stride;
+};
+
+static void __ipi_flush_tlb_range_asid(void *info)
+{
+	struct flush_tlb_range_data *d = info;
+
+	local_flush_tlb_range_asid(d->start, d->size, d->stride, d->asid);
+}
+
+static void __ipi_flush_tlb_range(void *info)
+{
+	struct flush_tlb_range_data *d = info;
+
+	local_flush_tlb_range(d->start, d->size, d->stride);
 }
 
-static void __sbi_tlb_flush_range(struct mm_struct *mm, unsigned long start,
-				  unsigned long size, unsigned long stride)
+static void __flush_tlb_range(struct mm_struct *mm, unsigned long start,
+			      unsigned long size, unsigned long stride)
 {
+	struct flush_tlb_range_data ftd;
 	struct cpumask *cmask = mm_cpumask(mm);
 	unsigned int cpuid;
 	bool broadcast;
@@ -45,19 +93,34 @@ static void __sbi_tlb_flush_range(struct mm_struct *mm, unsigned long start,
 		unsigned long asid = atomic_long_read(&mm->context.id);
 
 		if (broadcast) {
-			sbi_remote_sfence_vma_asid(cmask, start, size, asid);
-		} else if (size <= stride) {
-			local_flush_tlb_page_asid(start, asid);
+			if (riscv_use_ipi_for_rfence()) {
+				ftd.asid = asid;
+				ftd.start = start;
+				ftd.size = size;
+				ftd.stride = stride;
+				on_each_cpu_mask(cmask,
+						 __ipi_flush_tlb_range_asid,
+						 &ftd, 1);
+			} else
+				sbi_remote_sfence_vma_asid(cmask,
+							   start, size, asid);
 		} else {
-			local_flush_tlb_all_asid(asid);
+			local_flush_tlb_range_asid(start, size, stride, asid);
 		}
 	} else {
 		if (broadcast) {
-			sbi_remote_sfence_vma(cmask, start, size);
-		} else if (size <= stride) {
-			local_flush_tlb_page(start);
+			if (riscv_use_ipi_for_rfence()) {
+				ftd.asid = 0;
+				ftd.start = start;
+				ftd.size = size;
+				ftd.stride = stride;
+				on_each_cpu_mask(cmask,
+						 __ipi_flush_tlb_range,
+						 &ftd, 1);
+			} else
+				sbi_remote_sfence_vma(cmask, start, size);
 		} else {
-			local_flush_tlb_all();
+			local_flush_tlb_range(start, size, stride);
 		}
 	}
 
@@ -66,23 +129,23 @@ static void __sbi_tlb_flush_range(struct mm_struct *mm, unsigned long start,
 
 void flush_tlb_mm(struct mm_struct *mm)
 {
-	__sbi_tlb_flush_range(mm, 0, -1, PAGE_SIZE);
+	__flush_tlb_range(mm, 0, -1, PAGE_SIZE);
 }
 
 void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
 {
-	__sbi_tlb_flush_range(vma->vm_mm, addr, PAGE_SIZE, PAGE_SIZE);
+	__flush_tlb_range(vma->vm_mm, addr, PAGE_SIZE, PAGE_SIZE);
 }
 
 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 		     unsigned long end)
 {
-	__sbi_tlb_flush_range(vma->vm_mm, start, end - start, PAGE_SIZE);
+	__flush_tlb_range(vma->vm_mm, start, end - start, PAGE_SIZE);
 }
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start,
 			unsigned long end)
 {
-	__sbi_tlb_flush_range(vma->vm_mm, start, end - start, PMD_SIZE);
+	__flush_tlb_range(vma->vm_mm, start, end - start, PMD_SIZE);
 }
 #endif
Signature · 7 lines
2.34.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author
#11 lore raw

PATCH v12 7/7 RISC-V: Use IPIs for remote icache flush when possible

In reply to Anup Patel

If we have specialized interrupt controller (such as AIA IMSIC) which allows supervisor mode to directly inject IPIs without any assistance from M-mode or HS-mode then using such specialized interrupt controller, we can do remote icache flushe directly from supervisor mode instead of using the SBI RFENCE calls.

This patch extends remote icache flush functions to use supervisor mode IPIs whenever direct supervisor mode IPIs.are supported by interrupt controller.

---
 arch/riscv/mm/cacheflush.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
arch/riscv/mm/cacheflush.c +3 -2
index 57b40a350420..f10cb47eac3a 100644
@@ -19,7 +19,7 @@ void flush_icache_all(void)
 {
 	local_flush_icache_all();
 
-	if (IS_ENABLED(CONFIG_RISCV_SBI))
+	if (IS_ENABLED(CONFIG_RISCV_SBI) && !riscv_use_ipi_for_rfence())
 		sbi_remote_fence_i(NULL);
 	else
 		on_each_cpu(ipi_remote_fence_i, NULL, 1);
@@ -67,7 +67,8 @@ void flush_icache_mm(struct mm_struct *mm, bool local)
 		 * with flush_icache_deferred().
 		 */
 		smp_mb();
-	} else if (IS_ENABLED(CONFIG_RISCV_SBI)) {
+	} else if (IS_ENABLED(CONFIG_RISCV_SBI) &&
+		   !riscv_use_ipi_for_rfence()) {
 		sbi_remote_fence_i(&others);
 	} else {
 		on_each_cpu_mask(&others, ipi_remote_fence_i, NULL, 1);
Signature · 7 lines
2.34.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv