Skip to contentSkip to conversation

ARM: socfpga: Soft reset, hotplug and device tree clocks

PATCH Superseded dinguyen at altera.com 5 participants 13 messages Mar 13, 2013 – Mar 18, 2013

Subsystems (6)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

From: Dinh Nguyen <[email protected]>

Hi Arnd/Olof,

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

0002: ARM: socfpga: Add clock entries into device tree
- Correctly use the clock framework with device tree entries.
- Based from clk-highbank.c

Based on arm-soc/for-next.

Thanks,
Dinh

Dinh Nguyen (2):
ARM: socfpga: Enable hotplug and soft reset
ARM: socfpga: Add clock entries into device tree

 .../bindings/arm/altera/socfpga-clk-manager.txt    |   11 ++
 .../devicetree/bindings/clock/altr_socfpga.txt     |   15 ++
 arch/arm/boot/dts/socfpga.dtsi                     |  154 +++++++++++++++++
 arch/arm/boot/dts/socfpga_cyclone5.dts             |    8 +
 arch/arm/boot/dts/socfpga_vt.dts                   |    8 +
 arch/arm/mach-socfpga/core.h                       |   17 ++
 arch/arm/mach-socfpga/platsmp.c                    |    9 +-
 arch/arm/mach-socfpga/socfpga.c                    |   16 +-
 drivers/clk/socfpga/clk.c                          |  174 +++++++++++++++++---
 9 files changed, 387 insertions(+), 25 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt
 create mode 100644 Documentation/devicetree/bindings/clock/altr_socfpga.txt
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 1/2 ARM: socfpga: Enable hotplug and soft reset

From: Dinh Nguyen <[email protected]>

Put CPU1 into reset when it is hotplugged. Enable a cold or warm reset to the HW from userspace.

Also fix a few sparse errors:

warning: symbol 'sys_manager_base_addr' was not declared. Should it be static?
warning: symbol 'rst_manager_base_addr' was not declared. Should it be static?

---
 arch/arm/mach-socfpga/core.h    |   17 +++++++++++++++++
 arch/arm/mach-socfpga/platsmp.c |    9 ++++++---
 arch/arm/mach-socfpga/socfpga.c |   10 +++++++++-
 3 files changed, 32 insertions(+), 4 deletions(-)
arch/arm/mach-socfpga/core.h +17 -0
index 315edff..d2a251f 100644
@@ -20,12 +20,29 @@
 #ifndef __MACH_CORE_H
 #define __MACH_CORE_H
 
+#define SOCFPGA_RSTMGR_CTRL	0x04
+#define SOCFPGA_RSTMGR_MODPERRST	0x14
+#define SOCFPGA_RSTMGR_BRGMODRST	0x1c
+
+/* System Manager bits */
+#define RSTMGR_CTRL_SWCOLDRSTREQ	0x1	/* Cold Reset */
+#define RSTMGR_CTRL_SWWARMRSTREQ	0x2	/* Warm Reset */
+/*MPU Module Reset Register */
+#define RSTMGR_MPUMODRST_CPU0	0x1	/*CPU0 Reset*/
+#define RSTMGR_MPUMODRST_CPU1	0x2	/*CPU1 Reset*/
+#define RSTMGR_MPUMODRST_WDS		0x4	/*Watchdog Reset*/
+#define RSTMGR_MPUMODRST_SCUPER	0x8	/*SCU and periphs reset*/
+#define RSTMGR_MPUMODRST_L2		0x10	/*L2 Cache reset*/
+
 extern void socfpga_secondary_startup(void);
 extern void __iomem *socfpga_scu_base_addr;
 
 extern void socfpga_init_clocks(void);
 extern void socfpga_sysmgr_init(void);
 
+extern void __iomem *sys_manager_base_addr;
+extern void __iomem *rst_manager_base_addr;
+
 extern struct smp_operations socfpga_smp_ops;
 extern char secondary_trampoline, secondary_trampoline_end;
 
arch/arm/mach-socfpga/platsmp.c +6 -3
index 84c60fa..c75c33d 100644
@@ -30,9 +30,6 @@
 
 #include "core.h"
 
-extern void __iomem *sys_manager_base_addr;
-extern void __iomem *rst_manager_base_addr;
-
 static void __cpuinit socfpga_secondary_init(unsigned int cpu)
 {
 	/*
@@ -100,6 +97,12 @@ static void __init socfpga_smp_prepare_cpus(unsigned int max_cpus)
  */
 static void socfpga_cpu_die(unsigned int cpu)
 {
+	/* Flush the L1 data cache. */
+	flush_cache_all();
+
+	/* This will put CPU1 into reset.*/
+	__raw_writel(RSTMGR_MPUMODRST_CPU1, rst_manager_base_addr + 0x10);
+
 	cpu_do_idle();
 
 	/* We should have never returned from idle */
arch/arm/mach-socfpga/socfpga.c +9 -1
index 1042c02..b41a945 100644
@@ -87,7 +87,15 @@ static void __init socfpga_init_irq(void)
 
 static void socfpga_cyclone5_restart(char mode, const char *cmd)
 {
-	/* TODO: */
+	u32 temp;
+
+	temp = __raw_readl(rst_manager_base_addr + SOCFPGA_RSTMGR_CTRL);
+
+	if (mode == 'h')
+		temp |= RSTMGR_CTRL_SWCOLDRSTREQ;
+	else
+		temp |= RSTMGR_CTRL_SWWARMRSTREQ;
+	__raw_writel(temp, rst_manager_base_addr + SOCFPGA_RSTMGR_CTRL);
 }
 
 static void __init socfpga_cyclone5_init(void)
Signature · 1 line
1.7.9.5
#2 lore raw

PATCH 1/2 ARM: socfpga: Enable hotplug and soft reset

Hi!

Quoted 9 lines

Put CPU1 into reset when it is hotplugged. Enable a cold or warm reset to the HW from userspace.

Also fix a few sparse errors:

warning: symbol 'sys_manager_base_addr' was not declared. Should it be static?
warning: symbol 'rst_manager_base_addr' was not declared. Should it be static?

It seems pretty good.

arch/arm/mach-socfpga/core.h line 36

Quoted 9 lines
arch/arm/mach-socfpga/core.h +7 -0
+/*MPU Module Reset Register */
+#define RSTMGR_MPUMODRST_CPU0	0x1	/*CPU0 Reset*/
+#define RSTMGR_MPUMODRST_CPU1	0x2	/*CPU1 Reset*/
+#define RSTMGR_MPUMODRST_WDS		0x4	/*Watchdog Reset*/
+#define RSTMGR_MPUMODRST_SCUPER	0x8	/*SCU and periphs reset*/
+#define RSTMGR_MPUMODRST_L2		0x10	/*L2 Cache reset*/
+

I'd include space after /* and before */.

arch/arm/mach-socfpga/core.h static void __init socfpga_smp_prepare_cpus(unsigned int max_cpus) · line 104

Quoted 9 lines
arch/arm/mach-socfpga/core.h +5 -0
@@ -100,6 +97,12 @@ static void __init socfpga_smp_prepare_cpus(unsigned int max_cpus)
  */
 static void socfpga_cpu_die(unsigned int cpu)
 {
+	/* Flush the L1 data cache. */
+	flush_cache_all();
+
+	/* This will put CPU1 into reset.*/
+	__raw_writel(RSTMGR_MPUMODRST_CPU1, rst_manager_base_addr + 0x10);

Is there reason to do __raw_* operations? I believe we want the memory barriers etc, so we should use plain writel(). Same for actual reset code.

Thanks,
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#3 lore raw

PATCH 2/2 ARM: socfpga: Add clock entries into device tree

In reply to dinguyen at altera.com

From: Dinh Nguyen <[email protected]>

Adds the main PLL clock groups for SOCFPGA into device tree file so that the clock framework to query the clock and clock rates appropriately.

$cat /sys/kernel/debug/clk/clk_summary
clock enable_cnt prepare_cnt rate
---------------------------------------------------------------------

 osc1                           2           2            25000000
    sdram_pll                   0           0            400000000
       s2f_usr2_clk             0           0            66666666
       ddr_dq_clk               0           0            200000000
       ddr_2x_dqs_clk           0           0            400000000
       ddr_dqs_clk              0           0            200000000
    periph_pll                  2           2            500000000
       s2f_usr1_clk             0           0            50000000
       per_base_clk             4           4            100000000
       per_nand_mmc_clk         0           0            25000000
       per_qsi_clk              0           0            250000000
       emac1_clk                1           1            125000000
       emac0_clk                0           0            125000000
    main_pll                    1           1            1600000000
       cfg_s2f_usr0_clk         0           0            100000000
       main_nand_sdmmc_clk      0           0            100000000
       main_qspi_clk            0           0            400000000
       dbg_base_clk             0           0            400000000
       mainclk                  0           0            400000000
       mpuclk                   1           1            800000000
          smp_twd               1           1            200000000
---
 .../bindings/arm/altera/socfpga-clk-manager.txt    |   11 ++
 .../devicetree/bindings/clock/altr_socfpga.txt     |   15 ++
 arch/arm/boot/dts/socfpga.dtsi                     |  154 +++++++++++++++++
 arch/arm/boot/dts/socfpga_cyclone5.dts             |    8 +
 arch/arm/boot/dts/socfpga_vt.dts                   |    8 +
 arch/arm/mach-socfpga/socfpga.c                    |    6 +
 drivers/clk/socfpga/clk.c                          |  174 +++++++++++++++++---
 7 files changed, 355 insertions(+), 21 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt
 create mode 100644 Documentation/devicetree/bindings/clock/altr_socfpga.txt
Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt new file +11 -0
new file mode 100644
index 0000000..38cf62b
@@ -0,0 +1,11 @@
+Altera SOCFPGA Clock Manager
+
+Required properties:
+- compatible : "altr,clk-mgr"
+- reg : Should contain 1 register ranges(address and length)
+
+Example:
+	 clkmgr at ffd04000 {
+		compatible = "altr,clk-mgr";
+		reg = <0xffd04000 0x1000>;
+	};
Documentation/devicetree/bindings/clock/altr_socfpga.txt new file +15 -0
new file mode 100644
index 0000000..47b8b76
@@ -0,0 +1,15 @@
+Device Tree Clock bindings for Altera's SoCFPGA platform
+
+This binding uses the common clock binding[1].
+
+[1] Documentation/devicetree/bindings/clock/clock-bindings.txt
+
+Required properties:
+- compatible : shall be one of the following:
+	"altr,socfpga-pll-clock" - for a PLL clock
+	"altr,socfpga-perip-clock" - The peripheral clock divided from the
+		PLL clock.
+- reg : shall be the control register offset from CLOCK_MANAGER's base for the clock.
+- clocks : shall be the input parent clock phandle for the clock. This is
+	either an oscillator or a pll output.
+- #clock-cells : from common clock binding; shall be set to 0.
\ No newline at end of file
arch/arm/boot/dts/socfpga.dtsi +154 -0

160 lines not on this page · raw message

arch/arm/boot/dts/socfpga_cyclone5.dts +8 -0
index 3ae8a83..2495958 100644
@@ -33,6 +33,14 @@
 	};
 
 	soc {
+		clkmgr at ffd04000 {
+			clocks {
+				osc1 {
+					clock-frequency = <25000000>;
+				};
+			};
+		};
+
 		timer0 at ffc08000 {
 			clock-frequency = <100000000>;
 		};
arch/arm/boot/dts/socfpga_vt.dts +8 -0
index 1036eba..0bf035d 100644
@@ -33,6 +33,14 @@
 	};
 
 	soc {
+		clkmgr at ffd04000 {
+			clocks {
+				osc1 {
+					clock-frequency = <10000000>;
+				};
+			};
+		};
+
 		timer0 at ffc08000 {
 			clock-frequency = <7000000>;
 		};
arch/arm/mach-socfpga/socfpga.c +6 -0
index b41a945..856625a 100644
@@ -15,6 +15,7 @@
  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 #include <linux/dw_apb_timer.h>
+#include <linux/clk-provider.h>
 #include <linux/irqchip.h>
 #include <linux/of_address.h>
 #include <linux/of_irq.h>
@@ -29,6 +30,7 @@
 void __iomem *socfpga_scu_base_addr = ((void __iomem *)(SOCFPGA_SCU_VIRT_BASE));
 void __iomem *sys_manager_base_addr;
 void __iomem *rst_manager_base_addr;
+void __iomem *clk_mgr_base_addr;
 unsigned long cpu1start_addr;
 
 static struct map_desc scu_io_desc __initdata = {
@@ -77,6 +79,9 @@ void __init socfpga_sysmgr_init(void)
 
 	np = of_find_compatible_node(NULL, NULL, "altr,rst-mgr");
 	rst_manager_base_addr = of_iomap(np, 0);
+
+	np = of_find_compatible_node(NULL, NULL, "altr,clk-mgr");
+	clk_mgr_base_addr = of_iomap(np, 0);
 }
 
 static void __init socfpga_init_irq(void)
@@ -102,6 +107,7 @@ static void __init socfpga_cyclone5_init(void)
 {
 	l2x0_of_init(0, ~0UL);
 	of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
+	of_clk_init(NULL);
 	socfpga_init_clocks();
 }
 
drivers/clk/socfpga/clk.c +153 -21

199 lines not on this page · raw message

Signature · 1 line
1.7.9.5
#4 lore raw

PATCH 2/2 ARM: socfpga: Add clock entries into device tree

Hi!

Quoted 2 lines

Adds the main PLL clock groups for SOCFPGA into device tree file so that the clock framework to query the clock and clock rates

insert "works" here?

Quoted 9 lines

appropriately.

Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt new file +0 -0
new file mode 100644
index 0000000..38cf62b

Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt line 5

Quoted 5 lines
Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt new file +5 -0
@@ -0,0 +1,11 @@
+Altera SOCFPGA Clock Manager
+
+Required properties:
+- compatible : "altr,clk-mgr"
+- reg : Should contain 1 register ranges(address and length)

"1 register range (address"... ?

Documentation/devicetree/bindings/clock/altr_socfpga.txt line 4

Quoted 8 lines
Documentation/devicetree/bindings/clock/altr_socfpga.txt new file +4 -0
@@ -0,0 +1,15 @@
+- reg : shall be the control register offset from CLOCK_MANAGER's base for the clock.
+- clocks : shall be the input parent clock phandle for the clock. This is
+	either an oscillator or a pll output.
+- #clock-cells : from common clock binding; shall be set to 0.
\ No newline at end of file

Newline would be nice :-).

Quoted 12 lines
drivers/clk/socfpga/clk.c +1 -1
index 2c855a6..9379b1c 100644
@@ -1,5 +1,5 @@
 /*
- *  Copyright (C) 2012 Altera Corporation <www.altera.com>
+ *  Copyright (C) 2012-2013 Altera Corporation <www.altera.com>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by

drivers/clk/socfpga/clk.c line 14

Quoted 4 lines
drivers/clk/socfpga/clk.c +1 -0
@@ -11,41 +11,173 @@
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  * GNU General Public License for more details.
  *
+ * Based from clk-highbank.c

"Based on"? Does it make sense to add
* Copyright 2011-2012 Calxeda, Inc.
?

drivers/clk/socfpga/clk.c line 50

Quoted 7 lines
drivers/clk/socfpga/clk.c +7 -0
+extern void __iomem *clk_mgr_base_addr;
+
+struct socfpga_clk {
+	struct clk_hw hw;
+	void __iomem	*reg;
+	char *parent_name;
+};

Either align all fields using tabs, or none. This looks weird.

drivers/clk/socfpga/clk.c line 83

Quoted 9 lines
drivers/clk/socfpga/clk.c +9 -0
+static unsigned long clk_pll_recalc_rate(struct clk_hw *hwclk,
+					 unsigned long parent_rate)
+{
+	struct socfpga_clk *socfpgaclk = to_socfpga_clk(hwclk);
+	unsigned long divf, divq, vco_freq, reg;
+	unsigned long bypass;
+
+	reg = readl(socfpgaclk->reg);
+	bypass = readl(clk_mgr_base_addr + CLKMGR_BYPASS);

Would it make sense to create structure for clk_mgr_base_addr, too? At least code would be consistent that way.

Quoted 4 lines
drivers/clk/socfpga/clk.c +4 -0
+static unsigned long clk_periclk_recalc_rate(struct clk_hw *hwclk,
+					     unsigned long parent_rate)
+{
+	struct socfpga_clk *socfpgaclk = to_socfpga_clk(hwclk);

drivers/clk/socfpga/clk.c line 113

Quoted 10 lines
drivers/clk/socfpga/clk.c +10 -0
+	u32 div;
+
+	/* mpuclk, mainclk, and dbg_base_clk divisors are fixed.*/
+	if (strcmp(hwclk->init->name, "mpuclk") == 0)
+		div = 2;
+	else if ((strcmp(hwclk->init->name, "mainclk") == 0) ||
+			(strcmp(hwclk->init->name, "dbg_base_clk") == 0))
+		div = 4;
+	else
+		div = ((readl(socfpgaclk->reg) & 0x1ff) + 1);

Would it make sense to cache have div value in struct socfpga_clk ... so that it does not have to be recomputed based on strcmp?

Otherwise it looks good to me.

Thanks,
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#5 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

In reply to dinguyen at altera.com

Dinh,

Quoted 9 lines · dinguyen at altera.com

From: Dinh Nguyen <[email protected]>

Hi Arnd/Olof,

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

Rob

Quoted 26 lines

0002: ARM: socfpga: Add clock entries into device tree
- Correctly use the clock framework with device tree entries.
- Based from clk-highbank.c

Based on arm-soc/for-next.

Thanks,
Dinh

Dinh Nguyen (2):
ARM: socfpga: Enable hotplug and soft reset
ARM: socfpga: Add clock entries into device tree

 .../bindings/arm/altera/socfpga-clk-manager.txt    |   11 ++
 .../devicetree/bindings/clock/altr_socfpga.txt     |   15 ++
 arch/arm/boot/dts/socfpga.dtsi                     |  154 +++++++++++++++++
 arch/arm/boot/dts/socfpga_cyclone5.dts             |    8 +
 arch/arm/boot/dts/socfpga_vt.dts                   |    8 +
 arch/arm/mach-socfpga/core.h                       |   17 ++
 arch/arm/mach-socfpga/platsmp.c                    |    9 +-
 arch/arm/mach-socfpga/socfpga.c                    |   16 +-
 drivers/clk/socfpga/clk.c                          |  174 +++++++++++++++++---
 9 files changed, 387 insertions(+), 25 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt
 create mode 100644 Documentation/devicetree/bindings/clock/altr_socfpga.txt
#6 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

Hi!

Quoted 10 lines
Quoted 5 lines

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

I tried googling PSCI, but got:

Power State Coordination Interface
System Software on ARM Processors
This document is only available in a PDF version to registered ARM customers.

Do you have better pointer?
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#7 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

Quoted 14 lines · Pavel Machek

Hi!

Quoted 10 lines
Quoted 5 lines

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

Quoted 8 lines

I tried googling PSCI, but got:

Power State Coordination Interface
System Software on ARM Processors
This document is only available in a PDF version to registered ARM customers.

Do you have better pointer?

The easiest ist if you look at the kernel source at arch/arm/kernel/psci.c.

I think the registration at ARM is free, but I agree it's annoying to force people to go through that.

Arnd

#8 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

In reply to Rob Herring

Hi Rob,

Quoted 9 lines · Rob Herring

Dinh,

Quoted 6 lines · dinguyen at altera.com

From: Dinh Nguyen <[email protected]>

Hi Arnd/Olof,

Just 2 patches for mach-socfpga:

Quoted 7 lines
Quoted 3 lines

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

I just briefly skimmed the ARM doc and Linaro slidedeck. It looks like PSCI is already in the kernel, but I don't see any usage of it yet? I'll try to use PSCI for our hotplug. Do you know where those opcodes for cpu_on or cpu_off came from?

Currently we are running in non-secure mode, but may have a requirement to run in secure mode in other applications, so will PSCI still work for that case?

Thanks,
Dinh

Quoted 31 lines

Rob

Quoted 26 lines

0002: ARM: socfpga: Add clock entries into device tree
- Correctly use the clock framework with device tree entries.
- Based from clk-highbank.c

Based on arm-soc/for-next.

Thanks,
Dinh

Dinh Nguyen (2):
ARM: socfpga: Enable hotplug and soft reset
ARM: socfpga: Add clock entries into device tree

 .../bindings/arm/altera/socfpga-clk-manager.txt    |   11 ++
 .../devicetree/bindings/clock/altr_socfpga.txt     |   15 ++
 arch/arm/boot/dts/socfpga.dtsi                     |  154 +++++++++++++++++
 arch/arm/boot/dts/socfpga_cyclone5.dts             |    8 +
 arch/arm/boot/dts/socfpga_vt.dts                   |    8 +
 arch/arm/mach-socfpga/core.h                       |   17 ++
 arch/arm/mach-socfpga/platsmp.c                    |    9 +-
 arch/arm/mach-socfpga/socfpga.c                    |   16 +-
 drivers/clk/socfpga/clk.c                          |  174 +++++++++++++++++---
 9 files changed, 387 insertions(+), 25 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/arm/altera/socfpga-clk-manager.txt
 create mode 100644 Documentation/devicetree/bindings/clock/altr_socfpga.txt
#9 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

In reply to Rob Herring

Hi!

Quoted 6 lines
Quoted 6 lines

From: Dinh Nguyen <[email protected]>

Hi Arnd/Olof,

Just 2 patches for mach-socfpga:

Quoted 8 lines
Quoted 3 lines

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

What would be the advantage?

We do not have suitable hypervisor at the moment. Nor I see why it would be good to push power management into it.

(Plus, the implementation seems pretty incomplete for arm32.).

Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#10 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

Quoted 12 lines · Pavel Machek

Hi!

Quoted 10 lines
Quoted 9 lines

From: Dinh Nguyen <[email protected]>

Hi Arnd/Olof,

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Quoted 6 lines
Quoted 4 lines

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

What would be the advantage?

It gets rid of some platform code. If you do an A15 part, it abstracts out the differences between the cores on entering/exiting coherency. It should also save you from writing suspend/resume and cpuidle support for your platform.

Once you do go to an A15, you will have to run in non-secure world. Once you run in non-secure mode, you cannot do everything within the kernel. So either you create custom smc calls or you use PSCI. If you really support secure mode too, then the SOC design should prevent non-secure accesses to anything that would affect the secure world like powergating a core or cache.

Quoted 2 lines

We do not have suitable hypervisor at the moment. Nor I see why it would be good to push power management into it.

It's not a hypervisor. A minimal implementation is only about 100 lines of assembly.

Quoted 1 line

(Plus, the implementation seems pretty incomplete for arm32.).

That's the point. The linux side is simple. The secure monitor side is also simplified somewhat. When you enter secure monitor mode, you are automatically running with the cache and mmu off. This avoids the side effects of cache allocations while trying to flush the caches.

Rob

Quoted 3 lines

Pavel

#11 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

Hi!

Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 5 lines

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

Quoted 6 lines
Quoted 1 line

What would be the advantage?

It gets rid of some platform code. If you do an A15 part, it abstracts out the differences between the cores on entering/exiting coherency. It should also save you from writing suspend/resume and cpuidle support for your platform.

I don't see how PSCI would be useful at this point. Clock framework is needed for basic MMC and Ethernet support; PSCI would not help there.

Quoted 5 lines
Quoted 2 lines

We do not have suitable hypervisor at the moment. Nor I see why it would be good to push power management into it.

It's not a hypervisor. A minimal implementation is only about 100 lines of assembly.

So... we could reduce some complexity of the kernel by inventing another component, inventing API, and putting the complex code into the newly invented component... when no other arm32 architecture does that. (Where would you put the 100 lines? u-boot?)

Quoted 6 lines
Quoted 1 line

(Plus, the implementation seems pretty incomplete for arm32.).

That's the point. The linux side is simple. The secure monitor side is also simplified somewhat. When you enter secure monitor mode, you are automatically running with the cache and mmu off. This avoids the side effects of cache allocations while trying to flush the caches.

This is clock framework. It does not touch any caches, and does not have problems with MMU. Even if we would implement PSCI in future, this code would stay the same AFAICT.
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#12 lore raw

cover ARM: socfpga: Soft reset, hotplug and device tree clocks

Hi Pavel,

Quoted 26 lines · ZY - pavel

Hi!

Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 5 lines

Just 2 patches for mach-socfpga:

0001: ARM: socfpga: Enable hotplug and soft reset
- Able to hotplug CPU1 by putting it into reset and bringing back online.

Have you seen the discussion on PSCI? There's an ARM doc on it and Linaro session from last week. Is there a possibility you can use that? You would need to be able to run in non-secure mode and implement smc calls.

Quoted 6 lines
Quoted 1 line

What would be the advantage?

It gets rid of some platform code. If you do an A15 part, it abstracts out the differences between the cores on entering/exiting coherency. It should also save you from writing suspend/resume and cpuidle support for your platform.

I don't see how PSCI would be useful at this point. Clock framework is needed for basic MMC and Ethernet support; PSCI would not help there.

Quoted 3 lines
Quoted 2 lines

We do not have suitable hypervisor at the moment. Nor I see why it would be good to push power management into it.

Quoted 7 lines
Quoted 2 lines

It's not a hypervisor. A minimal implementation is only about 100 lines of assembly.

So... we could reduce some complexity of the kernel by inventing another component, inventing API, and putting the complex code into the newly invented component... when no other arm32 architecture does that. (Where would you put the 100 lines? u-boot?)

Let me just rework the patches to have just the clock framework and a soft reset support for SOCFPGA. I'll work on add PSCI for hotplug in a separate patch.

Dinh

Quoted 12 lines
Quoted 7 lines
Quoted 2 lines

(Plus, the implementation seems pretty incomplete for arm32.).

That's the point. The linux side is simple. The secure monitor side is also simplified somewhat. When you enter secure monitor mode, you are automatically running with the cache and mmu off. This avoids the side effects of cache allocations while trying to flush the caches.

This is clock framework. It does not touch any caches, and does not have problems with MMU. Even if we would implement PSCI in future, this code would stay the same AFAICT.
Pavel