Received-SPF: pass (google.com: domain of linux-kernel+bounces-176864-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Feedback-ID: ifd894703:Fastmail
From: Jiaxun Yang <jiaxun.yang@flygoat.com>
Date: Sun, 12 May 2024 13:01:50 +0100
Subject: [PATCH] clocksource: Add node counter timer driver for
 MIPS/Loongson64
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
Message-Id: <20240512-loongson_nodecnt-v1-1-2157b92ef8f8@flygoat.com>
To: Huacai Chen <chenhuacai@kernel.org>, 
 Thomas Bogendoerfer <tsbogend@alpha.franken.de>, 
 Daniel Lezcano <daniel.lezcano@linaro.org>, 
 Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org, linux-mips@vger.kernel.org, 
 Jiaxun Yang <jiaxun.yang@flygoat.com>

Node counter is a timer presents on many Loongson-3 series CPUs.
It is maintained on every node in system. To avoid synchronisation
complexity we only access the copy from first node in system.

It also has many ways to be accessed, on latest Loongson-3 CPU with
IOCSR instruction support it should be accessed with a IOCSR request,
while on earlier Loongson-3 CPUs it is attached to a 32 bits MMIO bus.
For QEMU's Loongson-3 virt system it is mapped to a 64 bit MMIO location.

On some rare case the counter is disabled by firmware or not present
on chip, so we need to perform a lightweight test to ensure it is
running before actually use it.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
 MAINTAINERS                                      |   1 +
 arch/mips/include/asm/mach-loongson64/loongson.h |   3 +
 arch/mips/loongson64/time.c                      |   1 +
 drivers/clocksource/Kconfig                      |   8 ++
 drivers/clocksource/loongson-nodecnt.c           | 112 +++++++++++++++++++++++
 5 files changed, 125 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index c675fc296b19..b36bff5b9803 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -15117,6 +15117,7 @@ L:	linux-mips@vger.kernel.org
 S:	Maintained
 F:	arch/mips/include/asm/mach-loongson64/
 F:	arch/mips/loongson64/
+F:	drivers/clocksource/loongson-nodecnt.c
 F:	drivers/irqchip/irq-loongson*
 F:	drivers/platform/mips/cpu_hwmon.c
 
diff --git a/arch/mips/include/asm/mach-loongson64/loongson.h b/arch/mips/include/asm/mach-loongson64/loongson.h
index f7c3ab6d724e..d07f4be06595 100644
--- a/arch/mips/include/asm/mach-loongson64/loongson.h
+++ b/arch/mips/include/asm/mach-loongson64/loongson.h
@@ -56,6 +56,9 @@ extern void *loongson_fdt_blob;
 extern void mach_irq_dispatch(unsigned int pending);
 extern int mach_i8259_irq(void);
 
+/* Time functions */
+extern int __init nodecnt_clocksource_init(void);
+
 /* We need this in some places... */
 #define delay() ({		\
 	int x;				\
diff --git a/arch/mips/loongson64/time.c b/arch/mips/loongson64/time.c
index f6d2c1e30570..95d4110c1ee6 100644
--- a/arch/mips/loongson64/time.c
+++ b/arch/mips/loongson64/time.c
@@ -41,6 +41,7 @@ void __init plat_time_init(void)
 	/* setup mips r4k timer */
 	mips_hpt_frequency = cpu_clock_freq / 2;
 
+	nodecnt_clocksource_init();
 #ifdef CONFIG_RS780_HPET
 	setup_hpet_timer();
 #endif
diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig
index 34faa0320ece..1c068f604333 100644
--- a/drivers/clocksource/Kconfig
+++ b/drivers/clocksource/Kconfig
@@ -743,4 +743,12 @@ config EP93XX_TIMER
 	  Enables support for the Cirrus Logic timer block
 	  EP93XX.
 
+config LOONGSON_NODECNT
+	bool "Loongson Node Conunter timer driver"
+	default y if MIPS && MACH_LOONGSON64
+	depends on (MIPS && MACH_LOONGSON64) || COMPILE_TEST
+	depends on GENERIC_SCHED_CLOCK
+	help
+	  Enables support for the Loongson Node Counter timer.
+
 endmenu
diff --git a/drivers/clocksource/loongson-nodecnt.c b/drivers/clocksource/loongson-nodecnt.c
new file mode 100644
index 000000000000..3cea4045ce75
--- /dev/null
+++ b/drivers/clocksource/loongson-nodecnt.c
@@ -0,0 +1,112 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ *  Copyright (C) 2024, Jiaxun Yang <jiaxun.yang@flygoat.com>
+ *  Loongson-3 Node Counter clocksource
+ */
+
+#include <linux/clocksource.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/sched_clock.h>
+
+#include <loongson.h>
+#include <loongson_regs.h>
+
+#define NODECNT_REGBASE		0x3ff00408
+
+static void __iomem *nodecnt_reg;
+static u64 (*nodecnt_read_fn)(void);
+
+static u64 notrace nodecnt_read_2x32(void)
+{
+	unsigned int hi, hi2, lo;
+
+	do {
+		hi = readl_relaxed(nodecnt_reg + 4);
+		lo = readl_relaxed(nodecnt_reg);
+		hi2 = readl_relaxed(nodecnt_reg + 4);
+	} while (hi2 != hi);
+
+	return (((u64) hi) << 32) + lo;
+}
+
+static u64 notrace nodecnt_read_64(void)
+{
+	return readq_relaxed(nodecnt_reg);
+}
+
+static u64 notrace nodecnt_read_csr(void)
+{
+	return csr_readq(LOONGSON_CSR_NODECNT);
+}
+
+static u64 nodecnt_clocksource_read(struct clocksource *cs)
+{
+	return nodecnt_read_fn();
+}
+
+static struct clocksource nodecnt_clocksource = {
+	.name	= "nodecnt",
+	.read	= nodecnt_clocksource_read,
+	.mask	= CLOCKSOURCE_MASK(64),
+	.flags	= CLOCK_SOURCE_IS_CONTINUOUS,
+};
+
+int __init nodecnt_clocksource_init(void)
+{
+	int err;
+	uint64_t delta;
+
+	if (!cpu_clock_freq)
+		return -ENODEV;
+
+	if (cpu_has_csr() && csr_readl(LOONGSON_CSR_FEATURES) & LOONGSON_CSRF_NODECNT) {
+		nodecnt_read_fn = nodecnt_read_csr;
+	} else if (loongson_sysconf.bridgetype == VIRTUAL) {
+		nodecnt_reg = ioremap(NODECNT_REGBASE, 8);
+		if (!nodecnt_reg)
+			return -ENOMEM;
+		nodecnt_read_fn = nodecnt_read_64;
+	} else {
+		switch (boot_cpu_data.processor_id & (PRID_IMP_MASK | PRID_REV_MASK)) {
+		case PRID_IMP_LOONGSON_64C | PRID_REV_LOONGSON3A_R2_0:
+		case PRID_IMP_LOONGSON_64C | PRID_REV_LOONGSON3A_R2_1:
+		case PRID_IMP_LOONGSON_64C | PRID_REV_LOONGSON3A_R3_0:
+		case PRID_IMP_LOONGSON_64C | PRID_REV_LOONGSON3A_R3_1:
+			break;
+		default:
+			return -ENODEV;
+		}
+		nodecnt_reg = ioremap(NODECNT_REGBASE, 8);
+		if (!nodecnt_reg)
+			return -ENOMEM;
+		nodecnt_read_fn = nodecnt_read_2x32;
+	}
+
+	/* Test if nodecnt is usable */
+	delta = nodecnt_read_fn();
+	udelay(10);
+	delta = nodecnt_read_fn() - delta;
+
+	if (!delta) {
+		pr_info("nodecnt: clocksource unusable\n");
+		err = -ENODEV;
+		goto out;
+	}
+
+	err = clocksource_register_hz(&nodecnt_clocksource, cpu_clock_freq);
+	if (err) {
+		pr_err("nodecnt: clocksource register failed\n");
+		goto out;
+	}
+
+	/* It fits for sched_clock if we don't suffer from cross node access */
+	if (loongson_sysconf.bridgetype == VIRTUAL || loongson_sysconf.nr_nodes <= 1)
+		sched_clock_register(nodecnt_read_fn, 64, cpu_clock_freq);
+
+out:
+	if (nodecnt_reg)
+		iounmap(nodecnt_reg);
+	return err;
+}

---
base-commit: 75fa778d74b786a1608d55d655d42b480a6fa8bd
change-id: 20240512-loongson_nodecnt-0704f76bc959

Best regards,
-- 
Jiaxun Yang <jiaxun.yang@flygoat.com>