[PATCH v2 1/3] ARC: mcip: halt GFRC together with ARC cores

[PATCH v2 1/3] ARC: mcip: halt GFRC together with ARC cores

[Eugeniy Paltsev]

Currently GFRC is running regardless state of ARC cores in the SMP cluster.
That means even if ARC cores are halted when doing JTAG debugging GFRC
[our source of wall-time] continues to run giving us unexpected warnings
once we allow ARC cores to run due to some tasks being stuck for too
long.

Starting from ARC HS v3.0 it's possible to tie GFRC to state of up-to 4
ARC cores with help of GFRC's CORE register where we set a mask for
cores which state we need to rely on.

We update cpu mask every time new cpu came online instead of using
hardcoded one or using mask generated from "possible_cpus" as we
want it set correctly even if we run kernel on HW which has fewer cores
than expected (or we launch kernel via debugger and kick fever cores
than HW has)

Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---

Changes v1->v2:
 * wrap access to __mcip_cmd and __mcip_cmd_data into mcip_lock
   spinlock.
 * add GFRC_BUILD BCR check.
 * move MCIP_BCR read and check to mcip_setup_per_cpu function.

 arch/arc/kernel/mcip.c | 37 +++++++++++++++++++++++++++++++++++++
 include/soc/arc/mcip.h |  3 +++
 2 files changed, 40 insertions(+)

diff --git a/arch/arc/kernel/mcip.c b/arch/arc/kernel/mcip.c
index f61a52b..1119029 100644
--- a/arch/arc/kernel/mcip.c
+++ b/arch/arc/kernel/mcip.c
@@ -22,10 +22,47 @@ static DEFINE_RAW_SPINLOCK(mcip_lock);
 
 static char smp_cpuinfo_buf[128];
 
+/*
+ * Set mask to halt GFRC if any online core in SMP cluster is halted.
+ * Only works for ARC HS v3.0+, on earlier versions has no effect.
+ */
+static void mcip_update_gfrc_halt_mask(int cpu)
+{
+	struct bcr_generic gfrc;
+	unsigned long flags;
+	u32 gfrc_halt_mask;
+
+	READ_BCR(ARC_REG_GFRC_BUILD, gfrc);
+
+	/*
+	 * CMD_GFRC_SET_CORE and CMD_GFRC_READ_CORE commands were added in
+	 * GFRC 0x3 version.
+	 */
+	if (gfrc.ver < 0x3)
+		return;
+
+	raw_spin_lock_irqsave(&mcip_lock, flags);
+
+	__mcip_cmd(CMD_GFRC_READ_CORE, 0);
+	gfrc_halt_mask = read_aux_reg(ARC_REG_MCIP_READBACK);
+	gfrc_halt_mask |= BIT(cpu);
+	__mcip_cmd_data(CMD_GFRC_SET_CORE, 0, gfrc_halt_mask);
+
+	raw_spin_unlock_irqrestore(&mcip_lock, flags);
+}
+
 static void mcip_setup_per_cpu(int cpu)
 {
+	struct mcip_bcr mp;
+
+	READ_BCR(ARC_REG_MCIP_BCR, mp);
+
 	smp_ipi_irq_setup(cpu, IPI_IRQ);
 	smp_ipi_irq_setup(cpu, SOFTIRQ_IRQ);
+
+	/* Update GFRC halt mask as new CPU came online */
+	if (mp.gfrc)
+		mcip_update_gfrc_halt_mask(cpu);
 }
 
 static void mcip_ipi_send(int cpu)
diff --git a/include/soc/arc/mcip.h b/include/soc/arc/mcip.h
index c2d1b15..1138da5 100644
--- a/include/soc/arc/mcip.h
+++ b/include/soc/arc/mcip.h
@@ -15,6 +15,7 @@
 
 #define ARC_REG_MCIP_BCR	0x0d0
 #define ARC_REG_MCIP_IDU_BCR	0x0D5
+#define ARC_REG_GFRC_BUILD	0x0D6
 #define ARC_REG_MCIP_CMD	0x600
 #define ARC_REG_MCIP_WDATA	0x601
 #define ARC_REG_MCIP_READBACK	0x602
@@ -40,6 +41,8 @@ struct mcip_cmd {
 
 #define CMD_GFRC_READ_LO		0x42
 #define CMD_GFRC_READ_HI		0x43
+#define CMD_GFRC_SET_CORE		0x47
+#define CMD_GFRC_READ_CORE		0x48
 
 #define CMD_IDU_ENABLE			0x71
 #define CMD_IDU_DISABLE			0x72
-- 
2.9.3

[PATCH v2 2/3] ARC: mcip: update MCIP debug mask when the new cpu came online

[Eugeniy Paltsev]

As of today we use hardcoded MCIP debug mask, so if we launch
kernel via debugger and kick fever cores than HW has all cpus
hang at the momemt of setup MCIP debug mask.

So update MCIP debug mask when the new cpu came online, instead of
use hardcoded MCIP debug mask.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---

Changes v1->v2:
 * update MCIP debug mask when the new cpu came online instead of
   using MCIP debug mask generated from "possible_cpus" mask.

 arch/arc/kernel/mcip.c | 37 ++++++++++++++++++++++++++++++++-----
 include/soc/arc/mcip.h |  2 ++
 2 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/arch/arc/kernel/mcip.c b/arch/arc/kernel/mcip.c
index 1119029..5fe84e4 100644
--- a/arch/arc/kernel/mcip.c
+++ b/arch/arc/kernel/mcip.c
@@ -51,6 +51,34 @@ static void mcip_update_gfrc_halt_mask(int cpu)
 	raw_spin_unlock_irqrestore(&mcip_lock, flags);
 }
 
+static void mcip_update_debug_halt_mask(int cpu)
+{
+	u32 mcip_mask = 0;
+	unsigned long flags;
+
+	raw_spin_lock_irqsave(&mcip_lock, flags);
+
+	/*
+	 * mcip_mask is same for CMD_DEBUG_SET_SELECT and CMD_DEBUG_SET_MASK
+	 * commands. So read it once instead of reading both CMD_DEBUG_READ_MASK
+	 * and CMD_DEBUG_READ_SELECT.
+	 */
+	__mcip_cmd(CMD_DEBUG_READ_SELECT, 0);
+	mcip_mask = read_aux_reg(ARC_REG_MCIP_READBACK);
+
+	mcip_mask |= BIT(cpu);
+
+	__mcip_cmd_data(CMD_DEBUG_SET_SELECT, 0, mcip_mask);
+	/*
+	 * Parameter specified halt cause:
+	 * STATUS32[H]/actionpoint/breakpoint/self-halt
+	 * We choose all of them (0xF).
+	 */
+	__mcip_cmd_data(CMD_DEBUG_SET_MASK, 0xF, mcip_mask);
+
+	raw_spin_unlock_irqrestore(&mcip_lock, flags);
+}
+
 static void mcip_setup_per_cpu(int cpu)
 {
 	struct mcip_bcr mp;
@@ -63,6 +91,10 @@ static void mcip_setup_per_cpu(int cpu)
 	/* Update GFRC halt mask as new CPU came online */
 	if (mp.gfrc)
 		mcip_update_gfrc_halt_mask(cpu);
+
+	/* Update MCIP debug mask as new CPU came online */
+	if (mp.dbg)
+		mcip_update_debug_halt_mask(cpu);
 }
 
 static void mcip_ipi_send(int cpu)
@@ -138,11 +170,6 @@ static void mcip_probe_n_setup(void)
 		IS_AVAIL1(mp.gfrc, "GFRC"));
 
 	cpuinfo_arc700[0].extn.gfrc = mp.gfrc;
-
-	if (mp.dbg) {
-		__mcip_cmd_data(CMD_DEBUG_SET_SELECT, 0, 0xf);
-		__mcip_cmd_data(CMD_DEBUG_SET_MASK, 0xf, 0xf);
-	}
 }
 
 struct plat_smp_ops plat_smp_ops = {
diff --git a/include/soc/arc/mcip.h b/include/soc/arc/mcip.h
index 1138da5..a91f251 100644
--- a/include/soc/arc/mcip.h
+++ b/include/soc/arc/mcip.h
@@ -37,7 +37,9 @@ struct mcip_cmd {
 #define CMD_SEMA_RELEASE		0x12
 
 #define CMD_DEBUG_SET_MASK		0x34
+#define CMD_DEBUG_READ_MASK		0x35
 #define CMD_DEBUG_SET_SELECT		0x36
+#define CMD_DEBUG_READ_SELECT		0x37
 
 #define CMD_GFRC_READ_LO		0x42
 #define CMD_GFRC_READ_HI		0x43
-- 
2.9.3

[PATCH v2 3/3] ARC: setup cpu possible mask according to possible-cpus dts property

[Eugeniy Paltsev]

As we have option in u-boot to set CPU mask for running linux,
we want to pass information to kernel about CPU cores should
be brought up. So we patch kernel dtb in u-boot to set
possible-cpus property.

This also allows us to have correctly setuped MCIP debug mask.

Signed-off-by: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
---

Changes v1->v2:
 * no changes.

 arch/arc/kernel/smp.c | 50 ++++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 40 insertions(+), 10 deletions(-)

diff --git a/arch/arc/kernel/smp.c b/arch/arc/kernel/smp.c
index efe8b42..21d86c3 100644
--- a/arch/arc/kernel/smp.c
+++ b/arch/arc/kernel/smp.c
@@ -24,6 +24,7 @@
 #include <linux/reboot.h>
 #include <linux/irqdomain.h>
 #include <linux/export.h>
+#include <linux/of_fdt.h>
 
 #include <asm/processor.h>
 #include <asm/setup.h>
@@ -47,6 +48,42 @@ void __init smp_prepare_boot_cpu(void)
 {
 }
 
+static int __init arc_get_cpu_map(const char *name, struct cpumask *cpumask)
+{
+	unsigned long dt_root = of_get_flat_dt_root();
+	const char *buf;
+
+	buf = of_get_flat_dt_prop(dt_root, name, NULL);
+	if (!buf)
+		return -EINVAL;
+
+	if (cpulist_parse(buf, cpumask))
+		return -EINVAL;
+
+	return 0;
+}
+
+/*
+ * Read from DeviceTree and setup cpu possible mask. If there is no
+ * "possible-cpus" property in DeviceTree pretend all [0..NR_CPUS-1] exist.
+ */
+static void __init arc_init_cpu_possible(void)
+{
+	struct cpumask cpumask;
+
+	if (arc_get_cpu_map("possible-cpus", &cpumask)) {
+		pr_warn("Failed to get possible-cpus from dtb, pretending all %u cpus exist\n",
+			NR_CPUS);
+
+		cpumask_setall(&cpumask);
+	}
+
+	if (!cpumask_test_cpu(0, &cpumask))
+		panic("Master cpu (cpu[0]) is missed in cpu possible mask!");
+
+	init_cpu_possible(&cpumask);
+}
+
 /*
  * Called from setup_arch() before calling setup_processor()
  *
@@ -58,10 +95,7 @@ void __init smp_prepare_boot_cpu(void)
  */
 void __init smp_init_cpus(void)
 {
-	unsigned int i;
-
-	for (i = 0; i < NR_CPUS; i++)
-		set_cpu_possible(i, true);
+	arc_init_cpu_possible();
 
 	if (plat_smp_ops.init_early_smp)
 		plat_smp_ops.init_early_smp();
@@ -70,16 +104,12 @@ void __init smp_init_cpus(void)
 /* called from init ( ) =>  process 1 */
 void __init smp_prepare_cpus(unsigned int max_cpus)
 {
-	int i;
-
 	/*
 	 * if platform didn't set the present map already, do it now
 	 * boot cpu is set to present already by init/main.c
 	 */
-	if (num_present_cpus() <= 1) {
-		for (i = 0; i < max_cpus; i++)
-			set_cpu_present(i, true);
-	}
+	if (num_present_cpus() <= 1)
+		init_cpu_present(cpu_possible_mask);
 }
 
 void __init smp_cpus_done(unsigned int max_cpus)
-- 
2.9.3

Re: [PATCH v2 2/3] ARC: mcip: update MCIP debug mask when the new cpu came online

[Vineet Gupta]

On 02/23/2018 08:42 AM, Eugeniy Paltsev wrote:
> As of today we use hardcoded MCIP debug mask, so if we launch
> kernel via debugger and kick fever cores than HW has all cpus
> hang at the momemt of setup MCIP debug mask.
>
> So update MCIP debug mask when the new cpu came online, instead of
> use hardcoded MCIP debug mask.
>
> Signed-off-by: Eugeniy Paltsev<Eugeniy.Paltsev@synopsys.com>

Acked-by: Vineet Gupta <vgupta@ynopsys.com>

Re: [PATCH v2 1/3] ARC: mcip: halt GFRC together with ARC cores

[Vineet Gupta]

On 02/23/2018 08:41 AM, Eugeniy Paltsev wrote:
> Currently GFRC is running regardless state of ARC cores in the SMP cluster.
> That means even if ARC cores are halted when doing JTAG debugging GFRC
> [our source of wall-time] continues to run giving us unexpected warnings
> once we allow ARC cores to run due to some tasks being stuck for too
> long.
>
> Starting from ARC HS v3.0 it's possible to tie GFRC to state of up-to 4
> ARC cores with help of GFRC's CORE register where we set a mask for
> cores which state we need to rely on.
>
> We update cpu mask every time new cpu came online instead of using
> hardcoded one or using mask generated from "possible_cpus" as we
> want it set correctly even if we run kernel on HW which has fewer cores
> than expected (or we launch kernel via debugger and kick fever cores
> than HW has)
>
> Signed-off-by: Alexey Brodkin<abrodkin@synopsys.com>
> Signed-off-by: Eugeniy Paltsev<Eugeniy.Paltsev@synopsys.com>

Acked-by: vineet Gupta <vgupta@synopsys.com>