[PATCH v1 0/2] coresight: enable debug module

[PATCH v1 0/2] coresight: enable debug module

[Leo Yan]

ARMv8 architecture reference manual (ARM DDI 0487A.k) Chapter H7 "The
Sample-based Profiling Extension" has description for sampling
registers, we can utilize these registers to check program counter
value with combined CPU exception level, secure state, etc. So this is
helpful for CPU lockup bugs, e.g. if one CPU has run into infinite loop
with IRQ disabled; the 'hang' CPU cannot switch context and handle any
interrupt, so it cannot handle SMP call for stack dump, etc.

This patch series is to enable coresight debug module with sample-based
registers and register call back notifier for PCSR register dumping
when panic happens, so we can see below dumping info for panic; and
this patch series has considered the conditions for access permission
for debug registers self, so this can avoid access debug registers when
CPU power domain is off; the driver also try to figure out the CPU is
in secure or non-secure state.

ARM external debug module:
CPU[0]:
 EDPRSR:  0000000b (Power:On DLK:Unlock)
 EDPCSR:  [<ffff00000808eb54>] handle_IPI+0xe4/0x150
 EDCIDSR: 00000000
 EDVIDSR: 90000000 (State:Non-secure Mode:EL1/0 Width:64bits VMID:0)
CPU[1]:
 EDPRSR:  0000000b (Power:On DLK:Unlock)
 EDPCSR:  [<ffff0000087a64c0>] debug_notifier_call+0x108/0x288
 EDCIDSR: 00000000
 EDVIDSR: 90000000 (State:Non-secure Mode:EL1/0 Width:64bits VMID:0)

[...]

This patch series has been verified on 96boards Hikey.

Changes from RFC:
* According to Mike Leach suggestion, added check for EDPRSR to avoid
  lockup; added supporting EDVIDSR and EDCIDSR registers;
* According to Mark Rutland and Mathieu Poirier suggestion, rewrote
  the documentation for DT binding;
* According to Mark and Mathieu suggestion, refined debug driver;

Leo Yan (2):
  coresight: bindings for debug module
  coresight: add support for debug module

 .../devicetree/bindings/arm/coresight-debug.txt    |  39 ++
 drivers/hwtracing/coresight/Kconfig                |  10 +
 drivers/hwtracing/coresight/Makefile               |   1 +
 drivers/hwtracing/coresight/coresight-debug.c      | 396 +++++++++++++++++++++
 4 files changed, 446 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/arm/coresight-debug.txt
 create mode 100644 drivers/hwtracing/coresight/coresight-debug.c

-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH v1 1/2] coresight: bindings for debug module

[Leo Yan]

According to ARMv8 architecture reference manual (ARM DDI 0487A.k)
Chapter 'Part H: External debug', the CPU can integrate debug module
and it can support self-hosted debug and external debug. Especially
for supporting self-hosted debug, this means the program can access
the debug module from mmio region; and usually the mmio region is
integrated with coresight.

So add document for binding debug component, includes binding to two
clocks, one is apb clock for bus and another is debug clock for debug
module self; and also need specify the CPU node which the debug module
is dedicated to specific CPU.

Signed-off-by: Leo Yan <leo.yan-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>
---
 .../devicetree/bindings/arm/coresight-debug.txt    | 39 ++++++++++++++++++++++
 1 file changed, 39 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/arm/coresight-debug.txt

diff --git a/Documentation/devicetree/bindings/arm/coresight-debug.txt b/Documentation/devicetree/bindings/arm/coresight-debug.txt
new file mode 100644
index 0000000..6e03e9b
--- /dev/null
+++ b/Documentation/devicetree/bindings/arm/coresight-debug.txt
@@ -0,0 +1,39 @@
+* CoreSight Debug Component:
+
+CoreSight debug component are compliant with the ARMv8 architecture reference
+manual (ARM DDI 0487A.k) Chapter 'Part H: External debug'. The external debug
+module is mainly used for two modes: self-hosted debug and external debug, and
+it can be accessed from mmio region from Coresight and eventually the debug
+module connects with CPU for debugging. And the debug module provides
+sample-based profiling extension, which can be used to sample CPU program
+counter, secure state and exception level, etc; usually every CPU has one
+dedicated debug module to be connected.
+
+Required properties:
+
+- compatible : should be
+	     * "arm,coresight-debug", "arm,primecell"; supplemented with
+	       "arm,primecell" as driver is using the AMBA bus interface.
+
+- reg : physical base address and length of the register set.
+
+- clocks : the clocks associated to this component.
+
+- clock-names : the name of the clocks referenced by the code. Since we are
+                using the AMBA framework, the name of the clock providing
+		the interconnect should be "apb_pclk", and the debug module
+		has an additional clock "dbg_clk", which is used to provide
+		clock for debug module itself. Both clocks are mandatory.
+
+- cpu : the cpu phandle the debug module is affined to. When omitted
+	the source is considered to belong to CPU0.
+
+Example:
+
+	debug@f6590000 {
+		compatible = "arm,coresight-debug","arm,primecell";
+		reg = <0 0xf6590000 0 0x1000>;
+		clocks = <&sys_ctrl HI6220_CS_ATB>, <&acpu_ctrl HI6220_ACPU_DBG_CLK0>;
+		clock-names = "apb_pclk", "dbg_clk";
+		cpu = <&cpu0>;
+	};
-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 1/2] coresight: bindings for debug module

[Mike Leach]

Hi Leo

On 23 February 2017 at 01:57, Leo Yan <leo.yan@linaro.org> wrote:
> According to ARMv8 architecture reference manual (ARM DDI 0487A.k)
> Chapter 'Part H: External debug', the CPU can integrate debug module
> and it can support self-hosted debug and external debug. Especially
> for supporting self-hosted debug, this means the program can access
> the debug module from mmio region; and usually the mmio region is
> integrated with coresight.
>
> So add document for binding debug component, includes binding to two
> clocks, one is apb clock for bus and another is debug clock for debug
> module self; and also need specify the CPU node which the debug module
> is dedicated to specific CPU.
>
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> ---
>  .../devicetree/bindings/arm/coresight-debug.txt    | 39 ++++++++++++++++++++++
>  1 file changed, 39 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/arm/coresight-debug.txt
>
> diff --git a/Documentation/devicetree/bindings/arm/coresight-debug.txt b/Documentation/devicetree/bindings/arm/coresight-debug.txt
> new file mode 100644
> index 0000000..6e03e9b
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/arm/coresight-debug.txt
> @@ -0,0 +1,39 @@
> +* CoreSight Debug Component:
> +
> +CoreSight debug component are compliant with the ARMv8 architecture reference
> +manual (ARM DDI 0487A.k) Chapter 'Part H: External debug'. The external debug
> +module is mainly used for two modes: self-hosted debug and external debug, and
> +it can be accessed from mmio region from Coresight and eventually the debug
> +module connects with CPU for debugging. And the debug module provides
> +sample-based profiling extension, which can be used to sample CPU program
> +counter, secure state and exception level, etc; usually every CPU has one
> +dedicated debug module to be connected.
> +
> +Required properties:
> +
> +- compatible : should be
> +            * "arm,coresight-debug", "arm,primecell"; supplemented with
> +              "arm,primecell" as driver is using the AMBA bus interface.
> +
> +- reg : physical base address and length of the register set.
> +
> +- clocks : the clocks associated to this component.
> +
> +- clock-names : the name of the clocks referenced by the code. Since we are
> +                using the AMBA framework, the name of the clock providing
> +               the interconnect should be "apb_pclk", and the debug module
> +               has an additional clock "dbg_clk", which is used to provide
> +               clock for debug module itself. Both clocks are mandatory.
> +

Perhaps I am misunderstanding the nature of the .dts files, but I'm
puzzled about the dbg_clk. I cannot see anything in the architecture
or normal A53 implementation that mentions this.
To access external debug from the core/external debug agent then we do
need the APB clock, but the interface between the debug logic and the
processor core is clocked by the internal CPU clocks.


> +- cpu : the cpu phandle the debug module is affined to. When omitted
> +       the source is considered to belong to CPU0.
> +
> +Example:
> +
> +       debug@f6590000 {
> +               compatible = "arm,coresight-debug","arm,primecell";
> +               reg = <0 0xf6590000 0 0x1000>;
> +               clocks = <&sys_ctrl HI6220_CS_ATB>, <&acpu_ctrl HI6220_ACPU_DBG_CLK0>;
> +               clock-names = "apb_pclk", "dbg_clk";
> +               cpu = <&cpu0>;
> +       };
> --
> 2.7.4
>

When I was looking at clocks for trace on the HiKey board I noted:
HI6220_CS_DAPB -- which I assumed was the debug APB clock.
HI6220_CS_ATB - which I assumed was the ATB (trace bus) clock.


Regards

Mike

-- 
Mike Leach
Principal Engineer, ARM Ltd.
Blackburn Design Centre. UK

Re: [PATCH v1 1/2] coresight: bindings for debug module

[Leo Yan]

Hi Mike,

On Mon, Feb 27, 2017 at 12:37:45PM +0000, Mike Leach wrote:
> Hi Leo
> 
> On 23 February 2017 at 01:57, Leo Yan <leo.yan@linaro.org> wrote:
> > According to ARMv8 architecture reference manual (ARM DDI 0487A.k)
> > Chapter 'Part H: External debug', the CPU can integrate debug module
> > and it can support self-hosted debug and external debug. Especially
> > for supporting self-hosted debug, this means the program can access
> > the debug module from mmio region; and usually the mmio region is
> > integrated with coresight.
> >
> > So add document for binding debug component, includes binding to two
> > clocks, one is apb clock for bus and another is debug clock for debug
> > module self; and also need specify the CPU node which the debug module
> > is dedicated to specific CPU.
> >
> > Signed-off-by: Leo Yan <leo.yan@linaro.org>
> > ---
> >  .../devicetree/bindings/arm/coresight-debug.txt    | 39 ++++++++++++++++++++++
> >  1 file changed, 39 insertions(+)
> >  create mode 100644 Documentation/devicetree/bindings/arm/coresight-debug.txt
> >
> > diff --git a/Documentation/devicetree/bindings/arm/coresight-debug.txt b/Documentation/devicetree/bindings/arm/coresight-debug.txt
> > new file mode 100644
> > index 0000000..6e03e9b
> > --- /dev/null
> > +++ b/Documentation/devicetree/bindings/arm/coresight-debug.txt
> > @@ -0,0 +1,39 @@
> > +* CoreSight Debug Component:
> > +
> > +CoreSight debug component are compliant with the ARMv8 architecture reference
> > +manual (ARM DDI 0487A.k) Chapter 'Part H: External debug'. The external debug
> > +module is mainly used for two modes: self-hosted debug and external debug, and
> > +it can be accessed from mmio region from Coresight and eventually the debug
> > +module connects with CPU for debugging. And the debug module provides
> > +sample-based profiling extension, which can be used to sample CPU program
> > +counter, secure state and exception level, etc; usually every CPU has one
> > +dedicated debug module to be connected.
> > +
> > +Required properties:
> > +
> > +- compatible : should be
> > +            * "arm,coresight-debug", "arm,primecell"; supplemented with
> > +              "arm,primecell" as driver is using the AMBA bus interface.
> > +
> > +- reg : physical base address and length of the register set.
> > +
> > +- clocks : the clocks associated to this component.
> > +
> > +- clock-names : the name of the clocks referenced by the code. Since we are
> > +                using the AMBA framework, the name of the clock providing
> > +               the interconnect should be "apb_pclk", and the debug module
> > +               has an additional clock "dbg_clk", which is used to provide
> > +               clock for debug module itself. Both clocks are mandatory.
> > +
> 
> Perhaps I am misunderstanding the nature of the .dts files, but I'm
> puzzled about the dbg_clk. I cannot see anything in the architecture
> or normal A53 implementation that mentions this.
> To access external debug from the core/external debug agent then we do
> need the APB clock, but the interface between the debug logic and the
> processor core is clocked by the internal CPU clocks.

You are right and sorry I may introduce confusion at here.

When I wrote this doc for binding, I assumed every debug logic has
its own clock source. This is because there have one register
ACPU_SC_PDBGUP_MBIST with eight bits, every bit is used to control
every CPU's DBGPWRDUP signal. I wrongly understood this bit is used
for debug logic clock gating.

I read CA53 TRM, the DBGPWRDUP signal actually is used between power
controller and CPU so can ensure CPU can be safely enter and exit low
power states; and "The EDPRSR.PU bit reflects the value of this
DBGPWRDUP signal". So I made mistake here and the bits in
ACPU_SC_PDBGUP_MBIST is no matter with debug logic clocks.

I will remove the "dbg_clk" from binding doc. Please correct me if my
understanding is still wrong or blur.

> > +- cpu : the cpu phandle the debug module is affined to. When omitted
> > +       the source is considered to belong to CPU0.
> > +
> > +Example:
> > +
> > +       debug@f6590000 {
> > +               compatible = "arm,coresight-debug","arm,primecell";
> > +               reg = <0 0xf6590000 0 0x1000>;
> > +               clocks = <&sys_ctrl HI6220_CS_ATB>, <&acpu_ctrl HI6220_ACPU_DBG_CLK0>;
> > +               clock-names = "apb_pclk", "dbg_clk";
> > +               cpu = <&cpu0>;
> > +       };
> > --
> > 2.7.4
> >
> 
> When I was looking at clocks for trace on the HiKey board I noted:
> HI6220_CS_DAPB -- which I assumed was the debug APB clock.
> HI6220_CS_ATB - which I assumed was the ATB (trace bus) clock.

The clock HI6220_CS_DAPB is a mux; and after went through the spec for
register definition, the clock driver misses the dapb gate clock in
its system controller 'sysctrl'; And I checked this bit is luckily
enabled by bootloaders (I have not checked it's enabled by ARM-TF or
UEFI) so the driver can access debug module registers.

diff --git a/drivers/clk/hisilicon/clk-hi6220.c b/drivers/clk/hisilicon/clk-hi6220.c
index 553d083..a4ac75e 100644
--- a/drivers/clk/hisilicon/clk-hi6220.c
+++ b/drivers/clk/hisilicon/clk-hi6220.c
@@ -134,6 +134,7 @@ static struct hisi_gate_clock hi6220_separated_gate_clks_sys[] __initdata = {
        { HI6220_UART4_PCLK,    "uart4_pclk",    "uart4_src",      CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x230, 8,  0, },
        { HI6220_SPI_CLK,       "spi_clk",       "clk_150m",       CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x230, 9,  0, },
        { HI6220_TSENSOR_CLK,   "tsensor_clk",   "clk_bus",        CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x230, 12, 0, },
+       { HI6220_DAPB_CLK,      "dapb_clk",      "cs_dapb",        CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x230, 18, 0, },
        { HI6220_MMU_CLK,       "mmu_clk",       "ddrc_axi1",      CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x240, 11, 0, },
        { HI6220_HIFI_SEL,      "hifi_sel",      "hifi_src",       CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x270, 0,  0, },
        { HI6220_MMC0_SYSPLL,   "mmc0_syspll",   "syspll",         CLK_SET_RATE_PARENT|CLK_IGNORE_UNUSED, 0x270, 1,  0, },

Thanks,
Leo Yan

[PATCH v1 2/2] coresight: add support for debug module

[Leo Yan]

Coresight includes debug module and usually the module connects with CPU
debug logic. ARMv8 architecture reference manual (ARM DDI 0487A.k) has
description for related info in "Part H: External Debug".

Chapter H7 "The Sample-based Profiling Extension" introduces several
sampling registers, e.g. we can check program counter value with
combined CPU exception level, secure state, etc. So this is helpful for
analysis CPU lockup scenarios, e.g. if one CPU has run into infinite
loop with IRQ disabled. In this case the CPU cannot switch context and
handle any interrupt (including IPIs), as the result it cannot handle
SMP call for stack dump.

This patch is to enable coresight debug module, so firstly this driver
is to enable clocks for debug module and this is to ensure the debug
module can be accessed from program or external debugger. And the driver
uses sample-based registers for debug purpose, e.g. when system detects
the CPU lockup and trigger panic, the driver will dump program counter
and combined context registers (EDCIDSR, EDVIDSR); by parsing context
registers so can quickly get to know CPU secure state, exception level,
etc.

Some of the debug module registers are located in CPU power domain, so
in the driver it has checked the power state for CPU before accessing
registers within CPU power domain. For most safe way to use this driver,
it's suggested to disable CPU low power states, this can simply set
"nohlt" in kernel command line.

Signed-off-by: Leo Yan <leo.yan-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>
---
 drivers/hwtracing/coresight/Kconfig           |  10 +
 drivers/hwtracing/coresight/Makefile          |   1 +
 drivers/hwtracing/coresight/coresight-debug.c | 396 ++++++++++++++++++++++++++
 3 files changed, 407 insertions(+)
 create mode 100644 drivers/hwtracing/coresight/coresight-debug.c

diff --git a/drivers/hwtracing/coresight/Kconfig b/drivers/hwtracing/coresight/Kconfig
index 130cb21..2c8883b 100644
--- a/drivers/hwtracing/coresight/Kconfig
+++ b/drivers/hwtracing/coresight/Kconfig
@@ -89,4 +89,14 @@ config CORESIGHT_STM
 	  logging useful software events or data coming from various entities
 	  in the system, possibly running different OSs
 
+config CORESIGHT_DEBUG
+	bool "CoreSight debug driver"
+	depends on ARM || ARM64
+	help
+	  This driver provides support for coresight debugging module. This
+	  is primarily used to dump sample-based profiling registers for
+	  panic. By disable CPU low power states (like "nohlt" in kernel
+	  command line), this is more safe and avoid lockup issue when
+	  access debug module registers.
+
 endif
diff --git a/drivers/hwtracing/coresight/Makefile b/drivers/hwtracing/coresight/Makefile
index af480d9..d540d45 100644
--- a/drivers/hwtracing/coresight/Makefile
+++ b/drivers/hwtracing/coresight/Makefile
@@ -16,3 +16,4 @@ obj-$(CONFIG_CORESIGHT_SOURCE_ETM4X) += coresight-etm4x.o \
 					coresight-etm4x-sysfs.o
 obj-$(CONFIG_CORESIGHT_QCOM_REPLICATOR) += coresight-replicator-qcom.o
 obj-$(CONFIG_CORESIGHT_STM) += coresight-stm.o
+obj-$(CONFIG_CORESIGHT_DEBUG) += coresight-debug.o
diff --git a/drivers/hwtracing/coresight/coresight-debug.c b/drivers/hwtracing/coresight/coresight-debug.c
new file mode 100644
index 0000000..a109fef
--- /dev/null
+++ b/drivers/hwtracing/coresight/coresight-debug.c
@@ -0,0 +1,396 @@
+/*
+ * Copyright (c) 2017 Linaro Limited. All rights reserved.
+ *
+ * Author: Leo Yan <leo.yan-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+#include <linux/amba/bus.h>
+#include <linux/coresight.h>
+#include <linux/cpu.h>
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/moduleparam.h>
+#include <linux/slab.h>
+#include <linux/smp.h>
+#include <linux/types.h>
+#include <linux/uaccess.h>
+
+#include "coresight-priv.h"
+
+#define EDPCSR				0x0A0
+#define EDCIDSR				0x0A4
+#define EDVIDSR				0x0A8
+#define EDPCSR_HI			0x0AC
+#define EDOSLAR				0x300
+#define EDPRSR				0x314
+#define EDDEVID1			0xFC4
+#define EDDEVID				0xFC8
+
+#define EDPCSR_PROHIBITED		0xFFFFFFFF
+
+/* bits definition for EDPCSR */
+#ifndef CONFIG_64BIT
+#define EDPCSR_THUMB			BIT(0)
+#define EDPCSR_ARM_INST_MASK		GENMASK(31, 2)
+#define EDPCSR_THUMB_INST_MASK		GENMASK(31, 1)
+#endif
+
+/* bits definition for EDPRSR */
+#define EDPRSR_DLK			BIT(6)
+#define EDPRSR_PU			BIT(0)
+
+/* bits definition for EDVIDSR */
+#define EDVIDSR_NS			BIT(31)
+#define EDVIDSR_E2			BIT(30)
+#define EDVIDSR_E3			BIT(29)
+#define EDVIDSR_HV			BIT(28)
+#define EDVIDSR_VMID			GENMASK(7, 0)
+
+/* bits definition for EDDEVID1 */
+#define EDDEVID1_PCSR_OFFSET_MASK	GENMASK(3, 0)
+#define EDDEVID1_PCSR_OFFSET_INS_SET	(0x0)
+#define EDDEVID1_PCSR_NO_OFFSET		(0x1)
+
+/* bits definition for EDDEVID */
+#define EDDEVID_PCSAMPLE_MODE		GENMASK(3, 0)
+#define EDDEVID_IMPL_NONE		(0x0)
+#define EDDEVID_IMPL_EDPCSR_EDCIDSR	(0x2)
+#define EDDEVID_IMPL_FULL		(0x3)
+
+struct debug_drvdata {
+	void __iomem	*base;
+	struct device	*dev;
+	int		cpu;
+	struct clk	*dbg_clk;
+
+	bool		edpcsr_present;
+	bool		edvidsr_present;
+	bool		pc_has_offset;
+
+	u32		eddevid;
+	u32		eddevid1;
+
+	u32		edpcsr;
+	u32		edpcsr_hi;
+	u32		edprsr;
+	u32		edvidsr;
+	u32		edcidsr;
+};
+
+static int debug_count;
+static DEFINE_PER_CPU(struct debug_drvdata *, debug_drvdata);
+
+static void debug_os_unlock(struct debug_drvdata *drvdata)
+{
+	/* Unlocks the debug registers */
+	writel_relaxed(0x0, drvdata->base + EDOSLAR);
+	wmb();
+}
+
+/*
+ * According to ARM DDI 0487A.k, before access external debug
+ * registers should firstly check the access permission; if any
+ * below condition has been met then cannot access debug
+ * registers to avoid lockup issue:
+ *
+ * - CPU power domain is powered off;
+ * - The OS Double Lock is locked;
+ *
+ * By checking EDPRSR can get to know if meet these conditions.
+ */
+static bool debug_access_permit(struct debug_drvdata *drvdata)
+{
+	/* CPU is powered off */
+	if (!(drvdata->edprsr & EDPRSR_PU))
+		return false;
+
+	/* The OS Double Lock is locked */
+	if (drvdata->edprsr & EDPRSR_DLK)
+		return false;
+
+	return true;
+}
+
+static void debug_read_regs(struct debug_drvdata *drvdata)
+{
+	drvdata->edprsr = readl_relaxed(drvdata->base + EDPRSR);
+
+	if (!debug_access_permit(drvdata))
+		return;
+
+	if (!drvdata->edpcsr_present)
+		return;
+
+	CS_UNLOCK(drvdata->base);
+
+	debug_os_unlock(drvdata);
+
+	drvdata->edpcsr = readl_relaxed(drvdata->base + EDPCSR);
+
+	/*
+	 * As described in ARM DDI 0487A.k, if the processing
+	 * element (PE) is in debug state, or sample-based
+	 * profiling is prohibited, EDPCSR reads as 0xFFFFFFFF;
+	 * EDCIDSR, EDVIDSR and EDPCSR_HI registers also become
+	 * UNKNOWN state. So directly bail out for this case.
+	 */
+	if (drvdata->edpcsr == EDPCSR_PROHIBITED) {
+		CS_LOCK(drvdata->base);
+		return;
+	}
+
+	/*
+	 * A read of the EDPCSR normally has the side-effect of
+	 * indirectly writing to EDCIDSR, EDVIDSR and EDPCSR_HI;
+	 * at this point it's safe to read value from them.
+	 */
+	drvdata->edcidsr = readl_relaxed(drvdata->base + EDCIDSR);
+#ifdef CONFIG_64BIT
+	drvdata->edpcsr_hi = readl_relaxed(drvdata->base + EDPCSR_HI);
+#endif
+
+	if (drvdata->edvidsr_present)
+		drvdata->edvidsr = readl_relaxed(drvdata->base + EDVIDSR);
+
+	CS_LOCK(drvdata->base);
+}
+
+#ifndef CONFIG_64BIT
+static bool debug_pc_has_offset(struct debug_drvdata *drvdata)
+{
+	u32 pcsr_offset;
+
+	pcsr_offset = drvdata->eddevid1 & EDDEVID1_PCSR_OFFSET_MASK;
+
+	return (pcsr_offset == EDDEVID1_PCSR_OFFSET_INS_SET);
+}
+
+static unsigned long debug_adjust_pc(struct debug_drvdata *drvdata,
+				     unsigned long pc)
+{
+	unsigned long arm_inst_offset = 0, thumb_inst_offset = 0;
+
+	if (debug_pc_has_offset(drvdata)) {
+		arm_inst_offset = 8;
+		thumb_inst_offset = 4;
+	}
+
+	/* Handle thumb instruction */
+	if (pc & EDPCSR_THUMB) {
+		pc = (pc & EDPCSR_THUMB_INST_MASK) - thumb_inst_offset;
+		return pc;
+	}
+
+	/*
+	 * Handle arm instruction offset, if the arm instruction
+	 * is not 4 byte alignment then it's possible the case
+	 * for implementation defined; keep original value for this
+	 * case and print info for notice.
+	 */
+	if (pc & BIT(1))
+		pr_emerg("Instruction offset is implementation defined\n");
+	else
+		pc = (pc & EDPCSR_ARM_INST_MASK) - arm_inst_offset;
+
+	return pc;
+}
+#endif
+
+static void debug_dump_regs(struct debug_drvdata *drvdata)
+{
+	unsigned long pc;
+
+	pr_emerg("\tEDPRSR:  %08x (Power:%s DLK:%s)\n", drvdata->edprsr,
+		 drvdata->edprsr & EDPRSR_PU ? "On" : "Off",
+		 drvdata->edprsr & EDPRSR_DLK ? "Lock" : "Unlock");
+
+	if (!debug_access_permit(drvdata) || !drvdata->edpcsr_present) {
+		pr_emerg("No permission to access debug registers!\n");
+		return;
+	}
+
+	if (drvdata->edpcsr == EDPCSR_PROHIBITED) {
+		pr_emerg("CPU is in Debug state or profiling is prohibited!\n");
+		return;
+	}
+
+#ifdef CONFIG_64BIT
+	pc = (unsigned long)drvdata->edpcsr_hi << 32 |
+	     (unsigned long)drvdata->edpcsr;
+#else
+	pc = debug_adjust_pc(drvdata, (unsigned long)drvdata->edpcsr);
+#endif
+
+	pr_emerg("\tEDPCSR:  [<%p>] %pS\n", (void *)pc, (void *)pc);
+	pr_emerg("\tEDCIDSR: %08x\n", drvdata->edcidsr);
+
+	if (!drvdata->edvidsr_present)
+		return;
+
+	pr_emerg("\tEDVIDSR: %08x (State:%s Mode:%s Width:%s VMID:%x)\n",
+		 drvdata->edvidsr,
+		 drvdata->edvidsr & EDVIDSR_NS ? "Non-secure" : "Secure",
+		 drvdata->edvidsr & EDVIDSR_E3 ? "EL3" :
+			(drvdata->edvidsr & EDVIDSR_E2 ? "EL2" : "EL1/0"),
+		 drvdata->edvidsr & EDVIDSR_HV ? "64bits" : "32bits",
+		 drvdata->edvidsr & (u32)EDVIDSR_VMID);
+}
+
+/*
+ * Dump out information on panic.
+ */
+static int debug_notifier_call(struct notifier_block *self,
+			       unsigned long v, void *p)
+{
+	int cpu;
+
+	pr_emerg("ARM external debug module:\n");
+
+	for_each_possible_cpu(cpu) {
+
+		if (!per_cpu(debug_drvdata, cpu))
+			continue;
+
+		pr_emerg("CPU[%d]:\n", per_cpu(debug_drvdata, cpu)->cpu);
+
+		debug_read_regs(per_cpu(debug_drvdata, cpu));
+		debug_dump_regs(per_cpu(debug_drvdata, cpu));
+	}
+
+	return 0;
+}
+
+static struct notifier_block debug_notifier = {
+	.notifier_call = debug_notifier_call,
+};
+
+static void debug_init_arch_data(void *info)
+{
+	struct debug_drvdata *drvdata = info;
+	u32 mode;
+
+	CS_UNLOCK(drvdata->base);
+
+	debug_os_unlock(drvdata);
+
+	/* Read device info */
+	drvdata->eddevid  = readl_relaxed(drvdata->base + EDDEVID);
+	drvdata->eddevid1 = readl_relaxed(drvdata->base + EDDEVID1);
+
+	/* Parse implementation feature */
+	mode = drvdata->eddevid & EDDEVID_PCSAMPLE_MODE;
+	if (mode == EDDEVID_IMPL_FULL) {
+		drvdata->edpcsr_present  = true;
+		drvdata->edvidsr_present = true;
+	} else if (mode == EDDEVID_IMPL_EDPCSR_EDCIDSR) {
+		drvdata->edpcsr_present  = true;
+		drvdata->edvidsr_present = false;
+	} else {
+		drvdata->edpcsr_present  = false;
+		drvdata->edvidsr_present = false;
+	}
+
+	CS_LOCK(drvdata->base);
+}
+
+static int debug_probe(struct amba_device *adev, const struct amba_id *id)
+{
+	int ret;
+	void __iomem *base;
+	struct device *dev = &adev->dev;
+	struct coresight_platform_data *pdata = NULL;
+	struct debug_drvdata *drvdata;
+	struct resource *res = &adev->res;
+	struct device_node *np = adev->dev.of_node;
+
+	drvdata = devm_kzalloc(dev, sizeof(*drvdata), GFP_KERNEL);
+	if (!drvdata)
+		return -ENOMEM;
+
+	if (np) {
+		pdata = of_get_coresight_platform_data(dev, np);
+		if (IS_ERR(pdata))
+			return PTR_ERR(pdata);
+		adev->dev.platform_data = pdata;
+	}
+
+	drvdata->dev = &adev->dev;
+	drvdata->dbg_clk = devm_clk_get(&adev->dev, "dbg_clk");
+	if (IS_ERR(drvdata->dbg_clk)) {
+		dev_err(dev, "debug clock initialization failed.\n");
+		return PTR_ERR(drvdata->dbg_clk);
+	}
+
+	ret = clk_prepare_enable(drvdata->dbg_clk);
+	if (ret)
+		return ret;
+
+	dev_set_drvdata(dev, drvdata);
+
+	/* Validity for the resource is already checked by the AMBA core */
+	base = devm_ioremap_resource(dev, res);
+	if (IS_ERR(base))
+		return PTR_ERR(base);
+
+	drvdata->base = base;
+	drvdata->cpu = pdata ? pdata->cpu : 0;
+
+	get_online_cpus();
+	per_cpu(debug_drvdata, drvdata->cpu) = drvdata;
+
+	if (smp_call_function_single(drvdata->cpu,
+				debug_init_arch_data, drvdata, 1))
+		dev_err(dev, "Debug arch init failed\n");
+
+	if (!debug_count++)
+		atomic_notifier_chain_register(&panic_notifier_list,
+					       &debug_notifier);
+	put_online_cpus();
+
+	dev_info(dev, "%s initialized\n", (char *)id->data);
+	return 0;
+}
+
+static struct amba_id debug_ids[] = {
+	{       /* Debug for Cortex-A53 */
+		.id	= 0x000bbd03,
+		.mask	= 0x000fffff,
+		.data	= "debug",
+	},
+	{       /* Debug for Cortex-A57 */
+		.id	= 0x000bbd07,
+		.mask	= 0x000fffff,
+		.data	= "debug",
+	},
+	{       /* Debug for Cortex-A72 */
+		.id	= 0x000bbd08,
+		.mask	= 0x000fffff,
+		.data	= "debug",
+	},
+	{ 0, 0},
+};
+
+static struct amba_driver debug_driver = {
+	.drv = {
+		.name   = "coresight-debug",
+		.suppress_bind_attrs = true,
+	},
+	.probe		= debug_probe,
+	.id_table	= debug_ids,
+};
+builtin_amba_driver(debug_driver);
-- 
2.7.4

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 2/2] coresight: add support for debug module

[Mathieu Poirier]

On Thu, Feb 23, 2017 at 09:57:47AM +0800, Leo Yan wrote:
> Coresight includes debug module and usually the module connects with CPU
> debug logic. ARMv8 architecture reference manual (ARM DDI 0487A.k) has
> description for related info in "Part H: External Debug".
> 
> Chapter H7 "The Sample-based Profiling Extension" introduces several
> sampling registers, e.g. we can check program counter value with
> combined CPU exception level, secure state, etc. So this is helpful for
> analysis CPU lockup scenarios, e.g. if one CPU has run into infinite
> loop with IRQ disabled. In this case the CPU cannot switch context and
> handle any interrupt (including IPIs), as the result it cannot handle
> SMP call for stack dump.
> 
> This patch is to enable coresight debug module, so firstly this driver
> is to enable clocks for debug module and this is to ensure the debug
> module can be accessed from program or external debugger. And the driver
> uses sample-based registers for debug purpose, e.g. when system detects
> the CPU lockup and trigger panic, the driver will dump program counter
> and combined context registers (EDCIDSR, EDVIDSR); by parsing context
> registers so can quickly get to know CPU secure state, exception level,
> etc.
> 
> Some of the debug module registers are located in CPU power domain, so
> in the driver it has checked the power state for CPU before accessing
> registers within CPU power domain. For most safe way to use this driver,
> it's suggested to disable CPU low power states, this can simply set
> "nohlt" in kernel command line.
> 
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> ---
>  drivers/hwtracing/coresight/Kconfig           |  10 +
>  drivers/hwtracing/coresight/Makefile          |   1 +
>  drivers/hwtracing/coresight/coresight-debug.c | 396 ++++++++++++++++++++++++++
>  3 files changed, 407 insertions(+)
>  create mode 100644 drivers/hwtracing/coresight/coresight-debug.c
> 
> diff --git a/drivers/hwtracing/coresight/Kconfig b/drivers/hwtracing/coresight/Kconfig
> index 130cb21..2c8883b 100644
> --- a/drivers/hwtracing/coresight/Kconfig
> +++ b/drivers/hwtracing/coresight/Kconfig
> @@ -89,4 +89,14 @@ config CORESIGHT_STM
>  	  logging useful software events or data coming from various entities
>  	  in the system, possibly running different OSs
>  
> +config CORESIGHT_DEBUG
> +	bool "CoreSight debug driver"
> +	depends on ARM || ARM64
> +	help
> +	  This driver provides support for coresight debugging module. This
> +	  is primarily used to dump sample-based profiling registers for
> +	  panic. By disable CPU low power states (like "nohlt" in kernel
> +	  command line), this is more safe and avoid lockup issue when
> +	  access debug module registers.
> +
>  endif
> diff --git a/drivers/hwtracing/coresight/Makefile b/drivers/hwtracing/coresight/Makefile
> index af480d9..d540d45 100644
> --- a/drivers/hwtracing/coresight/Makefile
> +++ b/drivers/hwtracing/coresight/Makefile
> @@ -16,3 +16,4 @@ obj-$(CONFIG_CORESIGHT_SOURCE_ETM4X) += coresight-etm4x.o \
>  					coresight-etm4x-sysfs.o
>  obj-$(CONFIG_CORESIGHT_QCOM_REPLICATOR) += coresight-replicator-qcom.o
>  obj-$(CONFIG_CORESIGHT_STM) += coresight-stm.o
> +obj-$(CONFIG_CORESIGHT_DEBUG) += coresight-debug.o
> diff --git a/drivers/hwtracing/coresight/coresight-debug.c b/drivers/hwtracing/coresight/coresight-debug.c
> new file mode 100644
> index 0000000..a109fef
> --- /dev/null
> +++ b/drivers/hwtracing/coresight/coresight-debug.c
> @@ -0,0 +1,396 @@
> +/*
> + * Copyright (c) 2017 Linaro Limited. All rights reserved.
> + *
> + * Author: Leo Yan <leo.yan@linaro.org>
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms of the GNU General Public License version 2 as published by
> + * the Free Software Foundation.
> + *
> + * This program is distributed in the hope that it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License along with
> + * this program.  If not, see <http://www.gnu.org/licenses/>.
> + *
> + */
> +#include <linux/amba/bus.h>
> +#include <linux/coresight.h>
> +#include <linux/cpu.h>
> +#include <linux/device.h>
> +#include <linux/err.h>
> +#include <linux/init.h>
> +#include <linux/io.h>
> +#include <linux/kernel.h>
> +#include <linux/moduleparam.h>
> +#include <linux/slab.h>
> +#include <linux/smp.h>
> +#include <linux/types.h>
> +#include <linux/uaccess.h>
> +
> +#include "coresight-priv.h"
> +
> +#define EDPCSR				0x0A0
> +#define EDCIDSR				0x0A4
> +#define EDVIDSR				0x0A8
> +#define EDPCSR_HI			0x0AC
> +#define EDOSLAR				0x300
> +#define EDPRSR				0x314
> +#define EDDEVID1			0xFC4
> +#define EDDEVID				0xFC8
> +
> +#define EDPCSR_PROHIBITED		0xFFFFFFFF
> +
> +/* bits definition for EDPCSR */
> +#ifndef CONFIG_64BIT
> +#define EDPCSR_THUMB			BIT(0)
> +#define EDPCSR_ARM_INST_MASK		GENMASK(31, 2)
> +#define EDPCSR_THUMB_INST_MASK		GENMASK(31, 1)
> +#endif
> +
> +/* bits definition for EDPRSR */
> +#define EDPRSR_DLK			BIT(6)
> +#define EDPRSR_PU			BIT(0)
> +
> +/* bits definition for EDVIDSR */
> +#define EDVIDSR_NS			BIT(31)
> +#define EDVIDSR_E2			BIT(30)
> +#define EDVIDSR_E3			BIT(29)
> +#define EDVIDSR_HV			BIT(28)
> +#define EDVIDSR_VMID			GENMASK(7, 0)
> +
> +/* bits definition for EDDEVID1 */
> +#define EDDEVID1_PCSR_OFFSET_MASK	GENMASK(3, 0)
> +#define EDDEVID1_PCSR_OFFSET_INS_SET	(0x0)
> +#define EDDEVID1_PCSR_NO_OFFSET		(0x1)

Hi Leo,

The above #define isn't used in the code - please remove.

> +
> +/* bits definition for EDDEVID */
> +#define EDDEVID_PCSAMPLE_MODE		GENMASK(3, 0)
> +#define EDDEVID_IMPL_NONE		(0x0)
> +#define EDDEVID_IMPL_EDPCSR_EDCIDSR	(0x2)
> +#define EDDEVID_IMPL_FULL		(0x3)
> +
> +struct debug_drvdata {
> +	void __iomem	*base;
> +	struct device	*dev;
> +	int		cpu;
> +	struct clk	*dbg_clk;
> +
> +	bool		edpcsr_present;
> +	bool		edvidsr_present;
> +	bool		pc_has_offset;
> +
> +	u32		eddevid;
> +	u32		eddevid1;
> +
> +	u32		edpcsr;
> +	u32		edpcsr_hi;
> +	u32		edprsr;
> +	u32		edvidsr;
> +	u32		edcidsr;
> +};
> +
> +static int debug_count;

No need to make this global - please move this to debug_probe() 

> +static DEFINE_PER_CPU(struct debug_drvdata *, debug_drvdata);
> +
> +static void debug_os_unlock(struct debug_drvdata *drvdata)
> +{
> +	/* Unlocks the debug registers */
> +	writel_relaxed(0x0, drvdata->base + EDOSLAR);
> +	wmb();
> +}
> +
> +/*
> + * According to ARM DDI 0487A.k, before access external debug
> + * registers should firstly check the access permission; if any
> + * below condition has been met then cannot access debug
> + * registers to avoid lockup issue:
> + *
> + * - CPU power domain is powered off;
> + * - The OS Double Lock is locked;
> + *
> + * By checking EDPRSR can get to know if meet these conditions.
> + */
> +static bool debug_access_permit(struct debug_drvdata *drvdata)

s/debug_access_permit/debug_access_permitted

> +{
> +	/* CPU is powered off */
> +	if (!(drvdata->edprsr & EDPRSR_PU))
> +		return false;
> +
> +	/* The OS Double Lock is locked */
> +	if (drvdata->edprsr & EDPRSR_DLK)
> +		return false;
> +
> +	return true;
> +}
> +
> +static void debug_read_regs(struct debug_drvdata *drvdata)
> +{
> +	drvdata->edprsr = readl_relaxed(drvdata->base + EDPRSR);
> +
> +	if (!debug_access_permit(drvdata))
> +		return;
> +
> +	if (!drvdata->edpcsr_present)
> +		return;
> +
> +	CS_UNLOCK(drvdata->base);
> +
> +	debug_os_unlock(drvdata);
> +
> +	drvdata->edpcsr = readl_relaxed(drvdata->base + EDPCSR);
> +
> +	/*
> +	 * As described in ARM DDI 0487A.k, if the processing
> +	 * element (PE) is in debug state, or sample-based
> +	 * profiling is prohibited, EDPCSR reads as 0xFFFFFFFF;
> +	 * EDCIDSR, EDVIDSR and EDPCSR_HI registers also become
> +	 * UNKNOWN state. So directly bail out for this case.
> +	 */
> +	if (drvdata->edpcsr == EDPCSR_PROHIBITED) {
> +		CS_LOCK(drvdata->base);
> +		return;
> +	}
> +
> +	/*
> +	 * A read of the EDPCSR normally has the side-effect of
> +	 * indirectly writing to EDCIDSR, EDVIDSR and EDPCSR_HI;
> +	 * at this point it's safe to read value from them.
> +	 */
> +	drvdata->edcidsr = readl_relaxed(drvdata->base + EDCIDSR);
> +#ifdef CONFIG_64BIT
> +	drvdata->edpcsr_hi = readl_relaxed(drvdata->base + EDPCSR_HI);
> +#endif
> +
> +	if (drvdata->edvidsr_present)
> +		drvdata->edvidsr = readl_relaxed(drvdata->base + EDVIDSR);
> +
> +	CS_LOCK(drvdata->base);
> +}
> +
> +#ifndef CONFIG_64BIT
> +static bool debug_pc_has_offset(struct debug_drvdata *drvdata)
> +{
> +	u32 pcsr_offset;
> +
> +	pcsr_offset = drvdata->eddevid1 & EDDEVID1_PCSR_OFFSET_MASK;
> +
> +	return (pcsr_offset == EDDEVID1_PCSR_OFFSET_INS_SET);
> +}
> +
> +static unsigned long debug_adjust_pc(struct debug_drvdata *drvdata,
> +				     unsigned long pc)
> +{
> +	unsigned long arm_inst_offset = 0, thumb_inst_offset = 0;
> +
> +	if (debug_pc_has_offset(drvdata)) {
> +		arm_inst_offset = 8;
> +		thumb_inst_offset = 4;
> +	}
> +
> +	/* Handle thumb instruction */
> +	if (pc & EDPCSR_THUMB) {
> +		pc = (pc & EDPCSR_THUMB_INST_MASK) - thumb_inst_offset;
> +		return pc;
> +	}
> +
> +	/*
> +	 * Handle arm instruction offset, if the arm instruction
> +	 * is not 4 byte alignment then it's possible the case
> +	 * for implementation defined; keep original value for this
> +	 * case and print info for notice.
> +	 */
> +	if (pc & BIT(1))
> +		pr_emerg("Instruction offset is implementation defined\n");
> +	else
> +		pc = (pc & EDPCSR_ARM_INST_MASK) - arm_inst_offset;
> +
> +	return pc;
> +}
> +#endif
> +
> +static void debug_dump_regs(struct debug_drvdata *drvdata)
> +{
> +	unsigned long pc;
> +
> +	pr_emerg("\tEDPRSR:  %08x (Power:%s DLK:%s)\n", drvdata->edprsr,
> +		 drvdata->edprsr & EDPRSR_PU ? "On" : "Off",
> +		 drvdata->edprsr & EDPRSR_DLK ? "Lock" : "Unlock");
> +
> +	if (!debug_access_permit(drvdata) || !drvdata->edpcsr_present) {
> +		pr_emerg("No permission to access debug registers!\n");
> +		return;
> +	}
> +
> +	if (drvdata->edpcsr == EDPCSR_PROHIBITED) {
> +		pr_emerg("CPU is in Debug state or profiling is prohibited!\n");
> +		return;
> +	}
> +
> +#ifdef CONFIG_64BIT
> +	pc = (unsigned long)drvdata->edpcsr_hi << 32 |
> +	     (unsigned long)drvdata->edpcsr;
> +#else
> +	pc = debug_adjust_pc(drvdata, (unsigned long)drvdata->edpcsr);
> +#endif
> +
> +	pr_emerg("\tEDPCSR:  [<%p>] %pS\n", (void *)pc, (void *)pc);
> +	pr_emerg("\tEDCIDSR: %08x\n", drvdata->edcidsr);
> +
> +	if (!drvdata->edvidsr_present)
> +		return;
> +
> +	pr_emerg("\tEDVIDSR: %08x (State:%s Mode:%s Width:%s VMID:%x)\n",
> +		 drvdata->edvidsr,
> +		 drvdata->edvidsr & EDVIDSR_NS ? "Non-secure" : "Secure",
> +		 drvdata->edvidsr & EDVIDSR_E3 ? "EL3" :
> +			(drvdata->edvidsr & EDVIDSR_E2 ? "EL2" : "EL1/0"),
> +		 drvdata->edvidsr & EDVIDSR_HV ? "64bits" : "32bits",
> +		 drvdata->edvidsr & (u32)EDVIDSR_VMID);
> +}
> +
> +/*
> + * Dump out information on panic.
> + */
> +static int debug_notifier_call(struct notifier_block *self,
> +			       unsigned long v, void *p)
> +{
> +	int cpu;
> +
> +	pr_emerg("ARM external debug module:\n");
> +
> +	for_each_possible_cpu(cpu) {
> +
> +		if (!per_cpu(debug_drvdata, cpu))
> +			continue;
> +
> +		pr_emerg("CPU[%d]:\n", per_cpu(debug_drvdata, cpu)->cpu);
> +
> +		debug_read_regs(per_cpu(debug_drvdata, cpu));
> +		debug_dump_regs(per_cpu(debug_drvdata, cpu));
> +	}
> +
> +	return 0;
> +}
> +
> +static struct notifier_block debug_notifier = {
> +	.notifier_call = debug_notifier_call,
> +};
> +
> +static void debug_init_arch_data(void *info)
> +{
> +	struct debug_drvdata *drvdata = info;
> +	u32 mode;
> +
> +	CS_UNLOCK(drvdata->base);
> +
> +	debug_os_unlock(drvdata);
> +
> +	/* Read device info */
> +	drvdata->eddevid  = readl_relaxed(drvdata->base + EDDEVID);
> +	drvdata->eddevid1 = readl_relaxed(drvdata->base + EDDEVID1);
> +
> +	/* Parse implementation feature */
> +	mode = drvdata->eddevid & EDDEVID_PCSAMPLE_MODE;
> +	if (mode == EDDEVID_IMPL_FULL) {
> +		drvdata->edpcsr_present  = true;
> +		drvdata->edvidsr_present = true;
> +	} else if (mode == EDDEVID_IMPL_EDPCSR_EDCIDSR) {
> +		drvdata->edpcsr_present  = true;
> +		drvdata->edvidsr_present = false;
> +	} else {
> +		drvdata->edpcsr_present  = false;
> +		drvdata->edvidsr_present = false;
> +	}
> +
> +	CS_LOCK(drvdata->base);
> +}
> +
> +static int debug_probe(struct amba_device *adev, const struct amba_id *id)
> +{
> +	int ret;
> +	void __iomem *base;
> +	struct device *dev = &adev->dev;
> +	struct coresight_platform_data *pdata = NULL;
> +	struct debug_drvdata *drvdata;
> +	struct resource *res = &adev->res;
> +	struct device_node *np = adev->dev.of_node;
> +
> +	drvdata = devm_kzalloc(dev, sizeof(*drvdata), GFP_KERNEL);
> +	if (!drvdata)
> +		return -ENOMEM;
> +
> +	if (np) {
> +		pdata = of_get_coresight_platform_data(dev, np);
> +		if (IS_ERR(pdata))
> +			return PTR_ERR(pdata);
> +		adev->dev.platform_data = pdata;
> +	}

I wonder if dragging a coresight_platform_data is worth it.  From what I see the
only thing it is used for is to retrieve the CPU.  As such I think it is better
get the portion of code in of_get_coresight_platform_data() that deals with the
CPU and create a new function (of_coresight_get_cpu()).  That way you can reuse
it here and in of_get_coresight_platform_data(). 

> +
> +	drvdata->dev = &adev->dev;
> +	drvdata->dbg_clk = devm_clk_get(&adev->dev, "dbg_clk");
> +	if (IS_ERR(drvdata->dbg_clk)) {
> +		dev_err(dev, "debug clock initialization failed.\n");
> +		return PTR_ERR(drvdata->dbg_clk);
> +	}
> +
> +	ret = clk_prepare_enable(drvdata->dbg_clk);
> +	if (ret)
> +		return ret;
> +
> +	dev_set_drvdata(dev, drvdata);
> +
> +	/* Validity for the resource is already checked by the AMBA core */
> +	base = devm_ioremap_resource(dev, res);
> +	if (IS_ERR(base))
> +		return PTR_ERR(base);
> +
> +	drvdata->base = base;
> +	drvdata->cpu = pdata ? pdata->cpu : 0;
> +
> +	get_online_cpus();
> +	per_cpu(debug_drvdata, drvdata->cpu) = drvdata;
> +
> +	if (smp_call_function_single(drvdata->cpu,
> +				debug_init_arch_data, drvdata, 1))
> +		dev_err(dev, "Debug arch init failed\n");
> +
> +	if (!debug_count++)
> +		atomic_notifier_chain_register(&panic_notifier_list,
> +					       &debug_notifier);
> +	put_online_cpus();

There is no need to hold the CPUs in place through the chain registration.  This
should go before the if().

> +
> +	dev_info(dev, "%s initialized\n", (char *)id->data);
        
        fprintf(buffer, (char *)id->data, drvdata->cpu);
        dev_info(dev, "%s initialized\n", buffer);

> +	return 0;
> +}
> +
> +static struct amba_id debug_ids[] = {
> +	{       /* Debug for Cortex-A53 */
> +		.id	= 0x000bbd03,
> +		.mask	= 0x000fffff,
> +		.data	= "debug",
       
                .data   = "Coresight debug-CPU%d",

> +	},
> +	{       /* Debug for Cortex-A57 */
> +		.id	= 0x000bbd07,
> +		.mask	= 0x000fffff,
> +		.data	= "debug",
> +	},
> +	{       /* Debug for Cortex-A72 */
> +		.id	= 0x000bbd08,
> +		.mask	= 0x000fffff,
> +		.data	= "debug",
> +	},
> +	{ 0, 0},
> +};
> +
> +static struct amba_driver debug_driver = {
> +	.drv = {
> +		.name   = "coresight-debug",
> +		.suppress_bind_attrs = true,
> +	},
> +	.probe		= debug_probe,
> +	.id_table	= debug_ids,
> +};
> +builtin_amba_driver(debug_driver);
> -- 
> 2.7.4
> 

Re: [PATCH v1 2/2] coresight: add support for debug module

[Leo Yan]

On Fri, Feb 24, 2017 at 12:07:11PM -0700, Mathieu Poirier wrote:
> On Thu, Feb 23, 2017 at 09:57:47AM +0800, Leo Yan wrote:

[...]

> > +/* bits definition for EDDEVID1 */
> > +#define EDDEVID1_PCSR_OFFSET_MASK	GENMASK(3, 0)
> > +#define EDDEVID1_PCSR_OFFSET_INS_SET	(0x0)
> > +#define EDDEVID1_PCSR_NO_OFFSET		(0x1)
> 
> Hi Leo,
> 
> The above #define isn't used in the code - please remove.

Thanks a lot for quite detailed reivew and suggestion for below
refactors. I will follow up them and send out new version for
reviewing.

Thanks,
Leo Yan

> > +
> > +/* bits definition for EDDEVID */
> > +#define EDDEVID_PCSAMPLE_MODE		GENMASK(3, 0)
> > +#define EDDEVID_IMPL_NONE		(0x0)
> > +#define EDDEVID_IMPL_EDPCSR_EDCIDSR	(0x2)
> > +#define EDDEVID_IMPL_FULL		(0x3)
> > +
> > +struct debug_drvdata {
> > +	void __iomem	*base;
> > +	struct device	*dev;
> > +	int		cpu;
> > +	struct clk	*dbg_clk;
> > +
> > +	bool		edpcsr_present;
> > +	bool		edvidsr_present;
> > +	bool		pc_has_offset;
> > +
> > +	u32		eddevid;
> > +	u32		eddevid1;
> > +
> > +	u32		edpcsr;
> > +	u32		edpcsr_hi;
> > +	u32		edprsr;
> > +	u32		edvidsr;
> > +	u32		edcidsr;
> > +};
> > +
> > +static int debug_count;
> 
> No need to make this global - please move this to debug_probe() 
> 
> > +static DEFINE_PER_CPU(struct debug_drvdata *, debug_drvdata);
> > +
> > +static void debug_os_unlock(struct debug_drvdata *drvdata)
> > +{
> > +	/* Unlocks the debug registers */
> > +	writel_relaxed(0x0, drvdata->base + EDOSLAR);
> > +	wmb();
> > +}
> > +
> > +/*
> > + * According to ARM DDI 0487A.k, before access external debug
> > + * registers should firstly check the access permission; if any
> > + * below condition has been met then cannot access debug
> > + * registers to avoid lockup issue:
> > + *
> > + * - CPU power domain is powered off;
> > + * - The OS Double Lock is locked;
> > + *
> > + * By checking EDPRSR can get to know if meet these conditions.
> > + */
> > +static bool debug_access_permit(struct debug_drvdata *drvdata)
> 
> s/debug_access_permit/debug_access_permitted
> 
> > +{
> > +	/* CPU is powered off */
> > +	if (!(drvdata->edprsr & EDPRSR_PU))
> > +		return false;
> > +
> > +	/* The OS Double Lock is locked */
> > +	if (drvdata->edprsr & EDPRSR_DLK)
> > +		return false;
> > +
> > +	return true;
> > +}
> > +
> > +static void debug_read_regs(struct debug_drvdata *drvdata)
> > +{
> > +	drvdata->edprsr = readl_relaxed(drvdata->base + EDPRSR);
> > +
> > +	if (!debug_access_permit(drvdata))
> > +		return;
> > +
> > +	if (!drvdata->edpcsr_present)
> > +		return;
> > +
> > +	CS_UNLOCK(drvdata->base);
> > +
> > +	debug_os_unlock(drvdata);
> > +
> > +	drvdata->edpcsr = readl_relaxed(drvdata->base + EDPCSR);
> > +
> > +	/*
> > +	 * As described in ARM DDI 0487A.k, if the processing
> > +	 * element (PE) is in debug state, or sample-based
> > +	 * profiling is prohibited, EDPCSR reads as 0xFFFFFFFF;
> > +	 * EDCIDSR, EDVIDSR and EDPCSR_HI registers also become
> > +	 * UNKNOWN state. So directly bail out for this case.
> > +	 */
> > +	if (drvdata->edpcsr == EDPCSR_PROHIBITED) {
> > +		CS_LOCK(drvdata->base);
> > +		return;
> > +	}
> > +
> > +	/*
> > +	 * A read of the EDPCSR normally has the side-effect of
> > +	 * indirectly writing to EDCIDSR, EDVIDSR and EDPCSR_HI;
> > +	 * at this point it's safe to read value from them.
> > +	 */
> > +	drvdata->edcidsr = readl_relaxed(drvdata->base + EDCIDSR);
> > +#ifdef CONFIG_64BIT
> > +	drvdata->edpcsr_hi = readl_relaxed(drvdata->base + EDPCSR_HI);
> > +#endif
> > +
> > +	if (drvdata->edvidsr_present)
> > +		drvdata->edvidsr = readl_relaxed(drvdata->base + EDVIDSR);
> > +
> > +	CS_LOCK(drvdata->base);
> > +}
> > +
> > +#ifndef CONFIG_64BIT
> > +static bool debug_pc_has_offset(struct debug_drvdata *drvdata)
> > +{
> > +	u32 pcsr_offset;
> > +
> > +	pcsr_offset = drvdata->eddevid1 & EDDEVID1_PCSR_OFFSET_MASK;
> > +
> > +	return (pcsr_offset == EDDEVID1_PCSR_OFFSET_INS_SET);
> > +}
> > +
> > +static unsigned long debug_adjust_pc(struct debug_drvdata *drvdata,
> > +				     unsigned long pc)
> > +{
> > +	unsigned long arm_inst_offset = 0, thumb_inst_offset = 0;
> > +
> > +	if (debug_pc_has_offset(drvdata)) {
> > +		arm_inst_offset = 8;
> > +		thumb_inst_offset = 4;
> > +	}
> > +
> > +	/* Handle thumb instruction */
> > +	if (pc & EDPCSR_THUMB) {
> > +		pc = (pc & EDPCSR_THUMB_INST_MASK) - thumb_inst_offset;
> > +		return pc;
> > +	}
> > +
> > +	/*
> > +	 * Handle arm instruction offset, if the arm instruction
> > +	 * is not 4 byte alignment then it's possible the case
> > +	 * for implementation defined; keep original value for this
> > +	 * case and print info for notice.
> > +	 */
> > +	if (pc & BIT(1))
> > +		pr_emerg("Instruction offset is implementation defined\n");
> > +	else
> > +		pc = (pc & EDPCSR_ARM_INST_MASK) - arm_inst_offset;
> > +
> > +	return pc;
> > +}
> > +#endif
> > +
> > +static void debug_dump_regs(struct debug_drvdata *drvdata)
> > +{
> > +	unsigned long pc;
> > +
> > +	pr_emerg("\tEDPRSR:  %08x (Power:%s DLK:%s)\n", drvdata->edprsr,
> > +		 drvdata->edprsr & EDPRSR_PU ? "On" : "Off",
> > +		 drvdata->edprsr & EDPRSR_DLK ? "Lock" : "Unlock");
> > +
> > +	if (!debug_access_permit(drvdata) || !drvdata->edpcsr_present) {
> > +		pr_emerg("No permission to access debug registers!\n");
> > +		return;
> > +	}
> > +
> > +	if (drvdata->edpcsr == EDPCSR_PROHIBITED) {
> > +		pr_emerg("CPU is in Debug state or profiling is prohibited!\n");
> > +		return;
> > +	}
> > +
> > +#ifdef CONFIG_64BIT
> > +	pc = (unsigned long)drvdata->edpcsr_hi << 32 |
> > +	     (unsigned long)drvdata->edpcsr;
> > +#else
> > +	pc = debug_adjust_pc(drvdata, (unsigned long)drvdata->edpcsr);
> > +#endif
> > +
> > +	pr_emerg("\tEDPCSR:  [<%p>] %pS\n", (void *)pc, (void *)pc);
> > +	pr_emerg("\tEDCIDSR: %08x\n", drvdata->edcidsr);
> > +
> > +	if (!drvdata->edvidsr_present)
> > +		return;
> > +
> > +	pr_emerg("\tEDVIDSR: %08x (State:%s Mode:%s Width:%s VMID:%x)\n",
> > +		 drvdata->edvidsr,
> > +		 drvdata->edvidsr & EDVIDSR_NS ? "Non-secure" : "Secure",
> > +		 drvdata->edvidsr & EDVIDSR_E3 ? "EL3" :
> > +			(drvdata->edvidsr & EDVIDSR_E2 ? "EL2" : "EL1/0"),
> > +		 drvdata->edvidsr & EDVIDSR_HV ? "64bits" : "32bits",
> > +		 drvdata->edvidsr & (u32)EDVIDSR_VMID);
> > +}
> > +
> > +/*
> > + * Dump out information on panic.
> > + */
> > +static int debug_notifier_call(struct notifier_block *self,
> > +			       unsigned long v, void *p)
> > +{
> > +	int cpu;
> > +
> > +	pr_emerg("ARM external debug module:\n");
> > +
> > +	for_each_possible_cpu(cpu) {
> > +
> > +		if (!per_cpu(debug_drvdata, cpu))
> > +			continue;
> > +
> > +		pr_emerg("CPU[%d]:\n", per_cpu(debug_drvdata, cpu)->cpu);
> > +
> > +		debug_read_regs(per_cpu(debug_drvdata, cpu));
> > +		debug_dump_regs(per_cpu(debug_drvdata, cpu));
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static struct notifier_block debug_notifier = {
> > +	.notifier_call = debug_notifier_call,
> > +};
> > +
> > +static void debug_init_arch_data(void *info)
> > +{
> > +	struct debug_drvdata *drvdata = info;
> > +	u32 mode;
> > +
> > +	CS_UNLOCK(drvdata->base);
> > +
> > +	debug_os_unlock(drvdata);
> > +
> > +	/* Read device info */
> > +	drvdata->eddevid  = readl_relaxed(drvdata->base + EDDEVID);
> > +	drvdata->eddevid1 = readl_relaxed(drvdata->base + EDDEVID1);
> > +
> > +	/* Parse implementation feature */
> > +	mode = drvdata->eddevid & EDDEVID_PCSAMPLE_MODE;
> > +	if (mode == EDDEVID_IMPL_FULL) {
> > +		drvdata->edpcsr_present  = true;
> > +		drvdata->edvidsr_present = true;
> > +	} else if (mode == EDDEVID_IMPL_EDPCSR_EDCIDSR) {
> > +		drvdata->edpcsr_present  = true;
> > +		drvdata->edvidsr_present = false;
> > +	} else {
> > +		drvdata->edpcsr_present  = false;
> > +		drvdata->edvidsr_present = false;
> > +	}
> > +
> > +	CS_LOCK(drvdata->base);
> > +}
> > +
> > +static int debug_probe(struct amba_device *adev, const struct amba_id *id)
> > +{
> > +	int ret;
> > +	void __iomem *base;
> > +	struct device *dev = &adev->dev;
> > +	struct coresight_platform_data *pdata = NULL;
> > +	struct debug_drvdata *drvdata;
> > +	struct resource *res = &adev->res;
> > +	struct device_node *np = adev->dev.of_node;
> > +
> > +	drvdata = devm_kzalloc(dev, sizeof(*drvdata), GFP_KERNEL);
> > +	if (!drvdata)
> > +		return -ENOMEM;
> > +
> > +	if (np) {
> > +		pdata = of_get_coresight_platform_data(dev, np);
> > +		if (IS_ERR(pdata))
> > +			return PTR_ERR(pdata);
> > +		adev->dev.platform_data = pdata;
> > +	}
> 
> I wonder if dragging a coresight_platform_data is worth it.  From what I see the
> only thing it is used for is to retrieve the CPU.  As such I think it is better
> get the portion of code in of_get_coresight_platform_data() that deals with the
> CPU and create a new function (of_coresight_get_cpu()).  That way you can reuse
> it here and in of_get_coresight_platform_data(). 
> 
> > +
> > +	drvdata->dev = &adev->dev;
> > +	drvdata->dbg_clk = devm_clk_get(&adev->dev, "dbg_clk");
> > +	if (IS_ERR(drvdata->dbg_clk)) {
> > +		dev_err(dev, "debug clock initialization failed.\n");
> > +		return PTR_ERR(drvdata->dbg_clk);
> > +	}
> > +
> > +	ret = clk_prepare_enable(drvdata->dbg_clk);
> > +	if (ret)
> > +		return ret;
> > +
> > +	dev_set_drvdata(dev, drvdata);
> > +
> > +	/* Validity for the resource is already checked by the AMBA core */
> > +	base = devm_ioremap_resource(dev, res);
> > +	if (IS_ERR(base))
> > +		return PTR_ERR(base);
> > +
> > +	drvdata->base = base;
> > +	drvdata->cpu = pdata ? pdata->cpu : 0;
> > +
> > +	get_online_cpus();
> > +	per_cpu(debug_drvdata, drvdata->cpu) = drvdata;
> > +
> > +	if (smp_call_function_single(drvdata->cpu,
> > +				debug_init_arch_data, drvdata, 1))
> > +		dev_err(dev, "Debug arch init failed\n");
> > +
> > +	if (!debug_count++)
> > +		atomic_notifier_chain_register(&panic_notifier_list,
> > +					       &debug_notifier);
> > +	put_online_cpus();
> 
> There is no need to hold the CPUs in place through the chain registration.  This
> should go before the if().
> 
> > +
> > +	dev_info(dev, "%s initialized\n", (char *)id->data);
>         
>         fprintf(buffer, (char *)id->data, drvdata->cpu);
>         dev_info(dev, "%s initialized\n", buffer);
> 
> > +	return 0;
> > +}
> > +
> > +static struct amba_id debug_ids[] = {
> > +	{       /* Debug for Cortex-A53 */
> > +		.id	= 0x000bbd03,
> > +		.mask	= 0x000fffff,
> > +		.data	= "debug",
>        
>                 .data   = "Coresight debug-CPU%d",
> 
> > +	},
> > +	{       /* Debug for Cortex-A57 */
> > +		.id	= 0x000bbd07,
> > +		.mask	= 0x000fffff,
> > +		.data	= "debug",
> > +	},
> > +	{       /* Debug for Cortex-A72 */
> > +		.id	= 0x000bbd08,
> > +		.mask	= 0x000fffff,
> > +		.data	= "debug",
> > +	},
> > +	{ 0, 0},
> > +};
> > +
> > +static struct amba_driver debug_driver = {
> > +	.drv = {
> > +		.name   = "coresight-debug",
> > +		.suppress_bind_attrs = true,
> > +	},
> > +	.probe		= debug_probe,
> > +	.id_table	= debug_ids,
> > +};
> > +builtin_amba_driver(debug_driver);
> > -- 
> > 2.7.4
> >