[PATCH v5 0/3] Add a few tracepoints to panthor

[PATCH v5 0/3] Add a few tracepoints to panthor

[Nicolas Frattaroli]

This series adds two tracepoints to panthor.

The first tracepoint allows for inspecting the power status of the
hardware subdivisions, e.g. how many shader cores are powered on. This
is done by reading three hardware registers when a certain IRQ fires.

The second tracepoint instruments panthor's job IRQ handler. This is
more useful than the generic interrupt tracing functionality, as the
tracepoint has the events bit mask included, which indicates which
command stream group interfaces triggered the interrupt.

To test the tracepoints, the following can be used:

  :~# echo 1 > /sys/kernel/tracing/events/panthor/gpu_power_status/enable
  :~# echo 1 > /sys/kernel/tracing/events/panthor/gpu_job_irq/enable
  :~# echo 1 > /sys/kernel/tracing/tracing_on
  :~# cat /sys/kernel/tracing/trace_pipe

Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
---
Changes in v5:
- Change the panthor IRQ helpers to guard the mask member and register
  with a spinlock. The rationale behind using a spinlock, rather than
  some constellation of atomics, is that we have to guarantee mutual
  exclusion for state beyond just a single value, namely both the register
  write, and writes to/reads from the mask member, including
  reads-from-member-writes-to-register. Making the mask atomic does not do
  anything to avoid concurrency issues in such a case.
- Change the IRQ mask member to not get zeroed when suspended. It's
  possible something outside of the IRQ helpers depends on this
  behaviour, but I'd argue the code should not access the mask outside
  of the IRQ helpers, as it'll do so with no lock taken.
- Drop the mask_set function, but add mask_enable/mask_disable helpers
  to enable/disable individual parts of the IRQ mask.
- Add a resume_restore IRQ helper that does the same thing as resume,
  but does not overwrite the mask member. This avoids me having to
  refactor whatever panthor_mmu.c is doing with that poor mask member.
- Link to v4: https://lore.kernel.org/r/20251217-panthor-tracepoints-v4-0-916186cb8d03@collabora.com

Changes in v4:
- Include "panthor_hw.h" in panthor_trace.h instead of duplicating the
  reg/unreg function prototypes.
- Link to v3: https://lore.kernel.org/r/20251211-panthor-tracepoints-v3-0-924c9d356a5c@collabora.com

Changes in v3:
- Drop PWRFEATURES patch, as this register is no longer needed by this
  series.
- Eliminate the rt_on field from the gpu_power_status register, as per
  Steven Price's feedback.
- Make gpu_power_status tracepoint reg/unreg functions generic across
  hardware generations by wrapping a hw op in panthor_hw.c.
- Reimplement the <= v13 IRQ mask modification functions as the new hw
  ops functions. v14 can add its own ops in due time.
- Link to v2: https://lore.kernel.org/r/20251210-panthor-tracepoints-v2-0-ace2e29bad0f@collabora.com

Changes in v2:
- Only enable the GPU_IRQ_POWER_CHANGED_* IRQ mask bits when the
  tracepoint is enabled. Necessitates the new irq helper patch.
- Only enable the GPU_IRQ_POWER_CHANGED_* IRQ mask bits if the hardware
  architecture is <= v13, as v14 changes things.
- Use _READY instead of _PWRACTIVE registers, and rename the tracepoint
  accordingly.
- Also read the status of the ray tracing unit's power. This is a global
  flag for all shader cores, it seems. Necessitates the new register
  definition patch.
- Move the POWER_CHANGED_* check to earlier in the interrupt handler.
- Also listen to POWER_CHANGED, not just POWER_CHANGED_ALL, as this
  provides useful information with the _READY registers.
- Print the device name in both tracepoints, to disambiguate things on
  systems with multiple Mali GPUs.
- Document the gpu_power_status tracepoint, so the meaning of the fields
  is made clear.
- Link to v1: https://lore.kernel.org/r/20251203-panthor-tracepoints-v1-0-871c8917e084@collabora.com

---
Nicolas Frattaroli (3):
      drm/panthor: Extend IRQ helpers for mask modification/restoration
      drm/panthor: Add tracepoint for hardware utilisation changes
      drm/panthor: Add gpu_job_irq tracepoint

 drivers/gpu/drm/panthor/panthor_device.h | 55 +++++++++++++++++---
 drivers/gpu/drm/panthor/panthor_fw.c     | 13 +++++
 drivers/gpu/drm/panthor/panthor_gpu.c    | 30 ++++++++++-
 drivers/gpu/drm/panthor/panthor_gpu.h    |  2 +
 drivers/gpu/drm/panthor/panthor_hw.c     | 62 +++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_hw.h     |  8 +++
 drivers/gpu/drm/panthor/panthor_trace.h  | 86 ++++++++++++++++++++++++++++++++
 7 files changed, 247 insertions(+), 9 deletions(-)
---
base-commit: 7af813455943b4e998394368b4076ad056e77183
change-id: 20251203-panthor-tracepoints-488af09d46e7

Best regards,
-- 
Nicolas Frattaroli <nicolas.frattaroli@collabora.com>

[PATCH v5 1/3] drm/panthor: Extend IRQ helpers for mask modification/restoration

[Nicolas Frattaroli]

The current IRQ helpers do not guarantee mutual exclusion that covers
the entire transaction from accessing the mask member and modifying the
mask register.

This makes it hard, if not impossible, to implement mask modification
helpers that may change one of these outside the normal
suspend/resume/isr code paths.

Add a spinlock to struct panthor_irq that protects both the mask member
and register. Acquire it in all code paths that access these. Then, add
the aforementioned new helpers: mask_enable, mask_disable, and
resume_restore. The first two work by ORing and NANDing the mask bits,
and the latter relies on the new behaviour that panthor_irq::mask is not
set to 0 on suspend.

panthor_irq::suspended remains an atomic, as it's necessarily written to
outside the mask_lock in the suspend path.

Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
---
 drivers/gpu/drm/panthor/panthor_device.h | 55 +++++++++++++++++++++++++++-----
 1 file changed, 47 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
index f35e52b9546a..eb75c83e2db3 100644
--- a/drivers/gpu/drm/panthor/panthor_device.h
+++ b/drivers/gpu/drm/panthor/panthor_device.h
@@ -73,11 +73,14 @@ struct panthor_irq {
 	/** @irq: IRQ number. */
 	int irq;
 
-	/** @mask: Current mask being applied to xxx_INT_MASK. */
+	/** @mask: Values to write to xxx_INT_MASK if active. */
 	u32 mask;
 
 	/** @suspended: Set to true when the IRQ is suspended. */
 	atomic_t suspended;
+
+	/** @mask_lock: protects modifications to _INT_MASK and @mask */
+	spinlock_t mask_lock;
 };
 
 /**
@@ -410,6 +413,8 @@ static irqreturn_t panthor_ ## __name ## _irq_raw_handler(int irq, void *data)
 	struct panthor_irq *pirq = data;							\
 	struct panthor_device *ptdev = pirq->ptdev;						\
 												\
+	guard(spinlock_irqsave)(&pirq->mask_lock);						\
+												\
 	if (atomic_read(&pirq->suspended))							\
 		return IRQ_NONE;								\
 	if (!gpu_read(ptdev, __reg_prefix ## _INT_STAT))					\
@@ -425,8 +430,10 @@ static irqreturn_t panthor_ ## __name ## _irq_threaded_handler(int irq, void *da
 	struct panthor_device *ptdev = pirq->ptdev;						\
 	irqreturn_t ret = IRQ_NONE;								\
 												\
+	guard(spinlock_irqsave)(&pirq->mask_lock);						\
+												\
 	while (true) {										\
-		u32 status = gpu_read(ptdev, __reg_prefix ## _INT_RAWSTAT) & pirq->mask;	\
+		u32 status = (gpu_read(ptdev, __reg_prefix ## _INT_RAWSTAT) & pirq->mask);	\
 												\
 		if (!status)									\
 			break;									\
@@ -443,18 +450,30 @@ static irqreturn_t panthor_ ## __name ## _irq_threaded_handler(int irq, void *da
 												\
 static inline void panthor_ ## __name ## _irq_suspend(struct panthor_irq *pirq)			\
 {												\
-	pirq->mask = 0;										\
-	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, 0);					\
+	scoped_guard(spinlock_irqsave, &pirq->mask_lock) {					\
+		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, 0);				\
+	}											\
 	synchronize_irq(pirq->irq);								\
 	atomic_set(&pirq->suspended, true);							\
 }												\
 												\
 static inline void panthor_ ## __name ## _irq_resume(struct panthor_irq *pirq, u32 mask)	\
 {												\
-	atomic_set(&pirq->suspended, false);							\
+	guard(spinlock_irqsave)(&pirq->mask_lock);						\
+												\
 	pirq->mask = mask;									\
-	gpu_write(pirq->ptdev, __reg_prefix ## _INT_CLEAR, mask);				\
-	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, mask);				\
+	atomic_set(&pirq->suspended, false);							\
+	gpu_write(pirq->ptdev, __reg_prefix ## _INT_CLEAR, pirq->mask);				\
+	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);				\
+}												\
+												\
+static inline void panthor_ ## __name ## _irq_resume_restore(struct panthor_irq *pirq)		\
+{												\
+	guard(spinlock_irqsave)(&pirq->mask_lock);						\
+												\
+	atomic_set(&pirq->suspended, false);							\
+	gpu_write(pirq->ptdev, __reg_prefix ## _INT_CLEAR, pirq->mask);				\
+	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);				\
 }												\
 												\
 static int panthor_request_ ## __name ## _irq(struct panthor_device *ptdev,			\
@@ -463,13 +482,33 @@ static int panthor_request_ ## __name ## _irq(struct panthor_device *ptdev,			\
 {												\
 	pirq->ptdev = ptdev;									\
 	pirq->irq = irq;									\
-	panthor_ ## __name ## _irq_resume(pirq, mask);						\
+	pirq->mask = mask;									\
+	spin_lock_init(&pirq->mask_lock);							\
+	panthor_ ## __name ## _irq_resume_restore(pirq);					\
 												\
 	return devm_request_threaded_irq(ptdev->base.dev, irq,					\
 					 panthor_ ## __name ## _irq_raw_handler,		\
 					 panthor_ ## __name ## _irq_threaded_handler,		\
 					 IRQF_SHARED, KBUILD_MODNAME "-" # __name,		\
 					 pirq);							\
+}												\
+												\
+static inline void panthor_ ## __name ## _irq_mask_enable(struct panthor_irq *pirq, u32 mask)	\
+{												\
+	guard(spinlock_irqsave)(&pirq->mask_lock);						\
+												\
+	pirq->mask |= mask;									\
+	if (!atomic_read(&pirq->suspended))							\
+		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);			\
+}												\
+												\
+static inline void panthor_ ## __name ## _irq_mask_disable(struct panthor_irq *pirq, u32 mask)	\
+{												\
+	guard(spinlock_irqsave)(&pirq->mask_lock);						\
+												\
+	pirq->mask &= ~mask;									\
+	if (!atomic_read(&pirq->suspended))							\
+		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);			\
 }
 
 extern struct workqueue_struct *panthor_cleanup_wq;

-- 
2.52.0

[PATCH v5 2/3] drm/panthor: Add tracepoint for hardware utilisation changes

[Nicolas Frattaroli]

Mali GPUs have three registers that indicate which parts of the hardware
are powered at any moment. These take the form of bitmaps. In the case
of SHADER_READY for example, a high bit indicates that the shader core
corresponding to that bit index is powered on. These bitmaps aren't
solely contiguous bits, as it's common to have holes in the sequence of
shader core indices, and the actual set of which cores are present is
defined by the "shader present" register.

When the GPU finishes a power state transition, it fires a
GPU_IRQ_POWER_CHANGED_ALL interrupt. After such an interrupt is
received, the _READY registers will contain new interesting data. During
power transitions, the GPU_IRQ_POWER_CHANGED interrupt will fire, and
the registers will likewise contain potentially changed data.

This is not to be confused with the PWR_IRQ_POWER_CHANGED_ALL interrupt,
which is something related to Mali v14+'s power control logic. The
_READY registers and corresponding interrupts are already available in
v9 and onwards.

Expose the data as a tracepoint to userspace. This allows users to debug
various scenarios and gather interesting information, such as: knowing
how much hardware is lit up at any given time, correlating graphics
corruption with a specific powered shader core, measuring when hardware
is allowed to go to a powered off state again, and so on.

The registration/unregistration functions for the tracepoint go through
a wrapper in panthor_hw.c, so that v14+ can implement the same
tracepoint by adding its hardware specific IRQ on/off callbacks to the
panthor_hw.ops member.

Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
---
 drivers/gpu/drm/panthor/panthor_gpu.c   | 30 +++++++++++++++-
 drivers/gpu/drm/panthor/panthor_gpu.h   |  2 ++
 drivers/gpu/drm/panthor/panthor_hw.c    | 62 +++++++++++++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_hw.h    |  8 +++++
 drivers/gpu/drm/panthor/panthor_trace.h | 58 ++++++++++++++++++++++++++++++
 5 files changed, 159 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/panthor/panthor_gpu.c b/drivers/gpu/drm/panthor/panthor_gpu.c
index 057e167468d0..fc0a8c2aa7c7 100644
--- a/drivers/gpu/drm/panthor/panthor_gpu.c
+++ b/drivers/gpu/drm/panthor/panthor_gpu.c
@@ -22,6 +22,9 @@
 #include "panthor_hw.h"
 #include "panthor_regs.h"
 
+#define CREATE_TRACE_POINTS
+#include "panthor_trace.h"
+
 /**
  * struct panthor_gpu - GPU block management data.
  */
@@ -48,6 +51,9 @@ struct panthor_gpu {
 	 GPU_IRQ_RESET_COMPLETED | \
 	 GPU_IRQ_CLEAN_CACHES_COMPLETED)
 
+#define GPU_POWER_INTERRUPTS_MASK	\
+	(GPU_IRQ_POWER_CHANGED | GPU_IRQ_POWER_CHANGED_ALL)
+
 static void panthor_gpu_coherency_set(struct panthor_device *ptdev)
 {
 	gpu_write(ptdev, GPU_COHERENCY_PROTOCOL,
@@ -80,6 +86,12 @@ static void panthor_gpu_irq_handler(struct panthor_device *ptdev, u32 status)
 {
 	gpu_write(ptdev, GPU_INT_CLEAR, status);
 
+	if (tracepoint_enabled(gpu_power_status) && (status & GPU_POWER_INTERRUPTS_MASK))
+		trace_gpu_power_status(ptdev->base.dev,
+				       gpu_read64(ptdev, SHADER_READY),
+				       gpu_read64(ptdev, TILER_READY),
+				       gpu_read64(ptdev, L2_READY));
+
 	if (status & GPU_IRQ_FAULT) {
 		u32 fault_status = gpu_read(ptdev, GPU_FAULT_STATUS);
 		u64 address = gpu_read64(ptdev, GPU_FAULT_ADDR);
@@ -157,6 +169,22 @@ int panthor_gpu_init(struct panthor_device *ptdev)
 	return 0;
 }
 
+int panthor_gpu_power_changed_on(struct panthor_device *ptdev)
+{
+	guard(pm_runtime_active)(ptdev->base.dev);
+
+	panthor_gpu_irq_mask_enable(&ptdev->gpu->irq, GPU_POWER_INTERRUPTS_MASK);
+
+	return 0;
+}
+
+void panthor_gpu_power_changed_off(struct panthor_device *ptdev)
+{
+	guard(pm_runtime_active)(ptdev->base.dev);
+
+	panthor_gpu_irq_mask_disable(&ptdev->gpu->irq, GPU_POWER_INTERRUPTS_MASK);
+}
+
 /**
  * panthor_gpu_block_power_off() - Power-off a specific block of the GPU
  * @ptdev: Device.
@@ -395,7 +423,7 @@ void panthor_gpu_suspend(struct panthor_device *ptdev)
  */
 void panthor_gpu_resume(struct panthor_device *ptdev)
 {
-	panthor_gpu_irq_resume(&ptdev->gpu->irq, GPU_INTERRUPTS_MASK);
+	panthor_gpu_irq_resume_restore(&ptdev->gpu->irq);
 	panthor_hw_l2_power_on(ptdev);
 }
 
diff --git a/drivers/gpu/drm/panthor/panthor_gpu.h b/drivers/gpu/drm/panthor/panthor_gpu.h
index 12e66f48ced1..12c263a39928 100644
--- a/drivers/gpu/drm/panthor/panthor_gpu.h
+++ b/drivers/gpu/drm/panthor/panthor_gpu.h
@@ -51,5 +51,7 @@ int panthor_gpu_l2_power_on(struct panthor_device *ptdev);
 int panthor_gpu_flush_caches(struct panthor_device *ptdev,
 			     u32 l2, u32 lsc, u32 other);
 int panthor_gpu_soft_reset(struct panthor_device *ptdev);
+void panthor_gpu_power_changed_off(struct panthor_device *ptdev);
+int panthor_gpu_power_changed_on(struct panthor_device *ptdev);
 
 #endif
diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
index 87ebb7ae42c4..ae3320d0e251 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.c
+++ b/drivers/gpu/drm/panthor/panthor_hw.c
@@ -1,6 +1,8 @@
 // SPDX-License-Identifier: GPL-2.0 or MIT
 /* Copyright 2025 ARM Limited. All rights reserved. */
 
+#include <linux/platform_device.h>
+
 #include <drm/drm_print.h>
 
 #include "panthor_device.h"
@@ -29,6 +31,8 @@ static struct panthor_hw panthor_hw_arch_v10 = {
 		.soft_reset = panthor_gpu_soft_reset,
 		.l2_power_off = panthor_gpu_l2_power_off,
 		.l2_power_on = panthor_gpu_l2_power_on,
+		.power_changed_off = panthor_gpu_power_changed_off,
+		.power_changed_on = panthor_gpu_power_changed_on,
 	},
 };
 
@@ -53,6 +57,64 @@ static struct panthor_hw_entry panthor_hw_match[] = {
 	},
 };
 
+static int panthor_hw_set_power_tracing(struct device *dev, void *data)
+{
+	struct panthor_device *ptdev = dev_get_drvdata(dev);
+
+	if (!ptdev)
+		return -ENODEV;
+
+	if (!ptdev->hw)
+		return 0;
+
+	if (data) {
+		if (ptdev->hw->ops.power_changed_on)
+			return ptdev->hw->ops.power_changed_on(ptdev);
+	} else {
+		if (ptdev->hw->ops.power_changed_off)
+			ptdev->hw->ops.power_changed_off(ptdev);
+	}
+
+	return 0;
+}
+
+int panthor_hw_power_status_register(void)
+{
+	struct device_driver *drv;
+	int ret;
+
+	drv = driver_find("panthor", &platform_bus_type);
+	if (!drv)
+		return -ENODEV;
+
+	ret = driver_for_each_device(drv, NULL, (void *)true,
+				     panthor_hw_set_power_tracing);
+
+	return ret;
+}
+
+void panthor_hw_power_status_unregister(void)
+{
+	struct device_driver *drv;
+	int ret;
+
+	drv = driver_find("panthor", &platform_bus_type);
+	if (!drv)
+		return;
+
+	ret = driver_for_each_device(drv, NULL, NULL, panthor_hw_set_power_tracing);
+
+	/*
+	 * Ideally, it'd be possible to ask driver_for_each_device to hand us
+	 * another "start" to keep going after the failing device, but it
+	 * doesn't do that. Minor inconvenience in what is probably a bad day
+	 * on the computer already though.
+	 */
+	if (ret)
+		pr_warn("Couldn't mask power IRQ for at least one device: %pe\n",
+			ERR_PTR(ret));
+}
+
 static char *get_gpu_model_name(struct panthor_device *ptdev)
 {
 	const u32 gpu_id = ptdev->gpu_info.gpu_id;
diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
index 56c68c1e9c26..2c28aea82841 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.h
+++ b/drivers/gpu/drm/panthor/panthor_hw.h
@@ -19,6 +19,12 @@ struct panthor_hw_ops {
 
 	/** @l2_power_on: L2 power on function pointer */
 	int (*l2_power_on)(struct panthor_device *ptdev);
+
+	/** @power_changed_on: Start listening to power change IRQs */
+	int (*power_changed_on)(struct panthor_device *ptdev);
+
+	/** @power_changed_off: Stop listening to power change IRQs */
+	void (*power_changed_off)(struct panthor_device *ptdev);
 };
 
 /**
@@ -32,6 +38,8 @@ struct panthor_hw {
 };
 
 int panthor_hw_init(struct panthor_device *ptdev);
+int panthor_hw_power_status_register(void);
+void panthor_hw_power_status_unregister(void);
 
 static inline int panthor_hw_soft_reset(struct panthor_device *ptdev)
 {
diff --git a/drivers/gpu/drm/panthor/panthor_trace.h b/drivers/gpu/drm/panthor/panthor_trace.h
new file mode 100644
index 000000000000..5bd420894745
--- /dev/null
+++ b/drivers/gpu/drm/panthor/panthor_trace.h
@@ -0,0 +1,58 @@
+/* SPDX-License-Identifier: GPL-2.0 or MIT */
+/* Copyright 2025 Collabora ltd. */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM panthor
+
+#if !defined(__PANTHOR_TRACE_H__) || defined(TRACE_HEADER_MULTI_READ)
+#define __PANTHOR_TRACE_H__
+
+#include <linux/tracepoint.h>
+#include <linux/types.h>
+
+#include "panthor_hw.h"
+
+/**
+ * gpu_power_status - called whenever parts of GPU hardware are turned on or off
+ * @dev: pointer to the &struct device, for printing the device name
+ * @shader_bitmap: bitmap where a high bit indicates the shader core at a given
+ *                 bit index is on, and a low bit indicates a shader core is
+ *                 either powered off or absent
+ * @tiler_bitmap: bitmap where a high bit indicates the tiler unit at a given
+ *                bit index is on, and a low bit indicates a tiler unit is
+ *                either powered off or absent
+ * @l2_bitmap: bitmap where a high bit indicates the L2 cache at a given bit
+ *             index is on, and a low bit indicates the L2 cache is either
+ *             powered off or absent
+ */
+TRACE_EVENT_FN(gpu_power_status,
+	TP_PROTO(const struct device *dev, u64 shader_bitmap, u64 tiler_bitmap,
+		 u64 l2_bitmap),
+	TP_ARGS(dev, shader_bitmap, tiler_bitmap, l2_bitmap),
+	TP_STRUCT__entry(
+		__string(dev_name, dev_name(dev))
+		__field(u64, shader_bitmap)
+		__field(u64, tiler_bitmap)
+		__field(u64, l2_bitmap)
+	),
+	TP_fast_assign(
+		__assign_str(dev_name);
+		__entry->shader_bitmap	= shader_bitmap;
+		__entry->tiler_bitmap	= tiler_bitmap;
+		__entry->l2_bitmap	= l2_bitmap;
+	),
+	TP_printk("%s: shader_bitmap=0x%llx tiler_bitmap=0x%llx l2_bitmap=0x%llx",
+		  __get_str(dev_name), __entry->shader_bitmap, __entry->tiler_bitmap,
+		  __entry->l2_bitmap
+	),
+	panthor_hw_power_status_register, panthor_hw_power_status_unregister
+);
+
+#endif /* __PANTHOR_TRACE_H__ */
+
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_FILE panthor_trace
+
+#include <trace/define_trace.h>

-- 
2.52.0

[PATCH v5 3/3] drm/panthor: Add gpu_job_irq tracepoint

[Nicolas Frattaroli]

Mali's CSF firmware triggers the job IRQ whenever there's new firmware
events for processing. While this can be a global event (BIT(31) of the
status register), it's usually an event relating to a command stream
group (the other bit indices).

Panthor throws these events onto a workqueue for processing outside the
IRQ handler. It's therefore useful to have an instrumented tracepoint
that goes beyond the generic IRQ tracepoint for this specific case, as
it can be augmented with additional data, namely the events bit mask.

This can then be used to debug problems relating to GPU jobs events not
being processed quickly enough. The duration_ns field can be used to
work backwards from when the tracepoint fires (at the end of the IRQ
handler) to figure out when the interrupt itself landed, providing not
just information on how long the work queueing took, but also when the
actual interrupt itself arrived.

With this information in hand, the IRQ handler itself being slow can be
excluded as a possible source of problems, and attention can be directed
to the workqueue processing instead.

Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
---
 drivers/gpu/drm/panthor/panthor_fw.c    | 13 +++++++++++++
 drivers/gpu/drm/panthor/panthor_trace.h | 28 ++++++++++++++++++++++++++++
 2 files changed, 41 insertions(+)

diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
index a64ec8756bed..513ed42f7b41 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.c
+++ b/drivers/gpu/drm/panthor/panthor_fw.c
@@ -26,6 +26,7 @@
 #include "panthor_mmu.h"
 #include "panthor_regs.h"
 #include "panthor_sched.h"
+#include "panthor_trace.h"
 
 #define CSF_FW_NAME "mali_csffw.bin"
 
@@ -1060,6 +1061,12 @@ static void panthor_fw_init_global_iface(struct panthor_device *ptdev)
 
 static void panthor_job_irq_handler(struct panthor_device *ptdev, u32 status)
 {
+	u32 duration;
+	u64 start;
+
+	if (tracepoint_enabled(gpu_job_irq))
+		start = ktime_get_ns();
+
 	gpu_write(ptdev, JOB_INT_CLEAR, status);
 
 	if (!ptdev->fw->booted && (status & JOB_INT_GLOBAL_IF))
@@ -1072,6 +1079,12 @@ static void panthor_job_irq_handler(struct panthor_device *ptdev, u32 status)
 		return;
 
 	panthor_sched_report_fw_events(ptdev, status);
+
+	if (tracepoint_enabled(gpu_job_irq)) {
+		if (check_sub_overflow(ktime_get_ns(), start, &duration))
+			duration = U32_MAX;
+		trace_gpu_job_irq(ptdev->base.dev, status, duration);
+	}
 }
 PANTHOR_IRQ_HANDLER(job, JOB, panthor_job_irq_handler);
 
diff --git a/drivers/gpu/drm/panthor/panthor_trace.h b/drivers/gpu/drm/panthor/panthor_trace.h
index 5bd420894745..6ffeb4fe6599 100644
--- a/drivers/gpu/drm/panthor/panthor_trace.h
+++ b/drivers/gpu/drm/panthor/panthor_trace.h
@@ -48,6 +48,34 @@ TRACE_EVENT_FN(gpu_power_status,
 	panthor_hw_power_status_register, panthor_hw_power_status_unregister
 );
 
+/**
+ * gpu_job_irq - called after a job interrupt from firmware completes
+ * @dev: pointer to the &struct device, for printing the device name
+ * @events: bitmask of BIT(CSG id) | BIT(31) for a global event
+ * @duration_ns: Nanoseconds between job IRQ handler entry and exit
+ *
+ * The panthor_job_irq_handler() function instrumented by this tracepoint exits
+ * once it has queued the firmware interrupts for processing, not when the
+ * firmware interrupts are fully processed. This tracepoint allows for debugging
+ * issues with delays in the workqueue's processing of events.
+ */
+TRACE_EVENT(gpu_job_irq,
+	TP_PROTO(const struct device *dev, u32 events, u32 duration_ns),
+	TP_ARGS(dev, events, duration_ns),
+	TP_STRUCT__entry(
+		__string(dev_name, dev_name(dev))
+		__field(u32, events)
+		__field(u32, duration_ns)
+	),
+	TP_fast_assign(
+		__assign_str(dev_name);
+		__entry->events		= events;
+		__entry->duration_ns	= duration_ns;
+	),
+	TP_printk("%s: events=0x%x duration_ns=%d", __get_str(dev_name),
+		  __entry->events, __entry->duration_ns)
+);
+
 #endif /* __PANTHOR_TRACE_H__ */
 
 #undef TRACE_INCLUDE_PATH

-- 
2.52.0

Re: [PATCH v5 1/3] drm/panthor: Extend IRQ helpers for mask modification/restoration

[Steven Price]

On 21/12/2025 17:10, Nicolas Frattaroli wrote:
> The current IRQ helpers do not guarantee mutual exclusion that covers
> the entire transaction from accessing the mask member and modifying the
> mask register.
> 
> This makes it hard, if not impossible, to implement mask modification
> helpers that may change one of these outside the normal
> suspend/resume/isr code paths.
> 
> Add a spinlock to struct panthor_irq that protects both the mask member
> and register. Acquire it in all code paths that access these. Then, add
> the aforementioned new helpers: mask_enable, mask_disable, and
> resume_restore. The first two work by ORing and NANDing the mask bits,
> and the latter relies on the new behaviour that panthor_irq::mask is not
> set to 0 on suspend.
> 
> panthor_irq::suspended remains an atomic, as it's necessarily written to
> outside the mask_lock in the suspend path.
> 
> Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
> ---
>  drivers/gpu/drm/panthor/panthor_device.h | 55 +++++++++++++++++++++++++++-----
>  1 file changed, 47 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
> index f35e52b9546a..eb75c83e2db3 100644
> --- a/drivers/gpu/drm/panthor/panthor_device.h
> +++ b/drivers/gpu/drm/panthor/panthor_device.h
> @@ -73,11 +73,14 @@ struct panthor_irq {
>  	/** @irq: IRQ number. */
>  	int irq;
>  
> -	/** @mask: Current mask being applied to xxx_INT_MASK. */
> +	/** @mask: Values to write to xxx_INT_MASK if active. */
>  	u32 mask;
>  
>  	/** @suspended: Set to true when the IRQ is suspended. */
>  	atomic_t suspended;
> +
> +	/** @mask_lock: protects modifications to _INT_MASK and @mask */
> +	spinlock_t mask_lock;
>  };
>  
>  /**
> @@ -410,6 +413,8 @@ static irqreturn_t panthor_ ## __name ## _irq_raw_handler(int irq, void *data)
>  	struct panthor_irq *pirq = data;							\
>  	struct panthor_device *ptdev = pirq->ptdev;						\
>  												\
> +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> +												\
>  	if (atomic_read(&pirq->suspended))							\
>  		return IRQ_NONE;								\
>  	if (!gpu_read(ptdev, __reg_prefix ## _INT_STAT))					\
> @@ -425,8 +430,10 @@ static irqreturn_t panthor_ ## __name ## _irq_threaded_handler(int irq, void *da
>  	struct panthor_device *ptdev = pirq->ptdev;						\
>  	irqreturn_t ret = IRQ_NONE;								\
>  												\
> +	guard(spinlock_irqsave)(&pirq->mask_lock);						\

Woah! You can't do that. That means there's a spinlock held while
calling the __handler() function. So you'll get "sleeping while atomic"
bug reports. Specifically I can see that panthor_mmu_irq_handler() takes
a mutex.

But the whole point of a threaded handler is so that it can sleep, so we
definitely don't want a spinlock held.

Thanks,
Steve

> +												\
>  	while (true) {										\
> -		u32 status = gpu_read(ptdev, __reg_prefix ## _INT_RAWSTAT) & pirq->mask;	\
> +		u32 status = (gpu_read(ptdev, __reg_prefix ## _INT_RAWSTAT) & pirq->mask);	\
>  												\
>  		if (!status)									\
>  			break;									\
> @@ -443,18 +450,30 @@ static irqreturn_t panthor_ ## __name ## _irq_threaded_handler(int irq, void *da
>  												\
>  static inline void panthor_ ## __name ## _irq_suspend(struct panthor_irq *pirq)			\
>  {												\
> -	pirq->mask = 0;										\
> -	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, 0);					\
> +	scoped_guard(spinlock_irqsave, &pirq->mask_lock) {					\
> +		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, 0);				\
> +	}											\
>  	synchronize_irq(pirq->irq);								\
>  	atomic_set(&pirq->suspended, true);							\
>  }												\
>  												\
>  static inline void panthor_ ## __name ## _irq_resume(struct panthor_irq *pirq, u32 mask)	\
>  {												\
> -	atomic_set(&pirq->suspended, false);							\
> +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> +												\
>  	pirq->mask = mask;									\
> -	gpu_write(pirq->ptdev, __reg_prefix ## _INT_CLEAR, mask);				\
> -	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, mask);				\
> +	atomic_set(&pirq->suspended, false);							\
> +	gpu_write(pirq->ptdev, __reg_prefix ## _INT_CLEAR, pirq->mask);				\
> +	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);				\
> +}												\
> +												\
> +static inline void panthor_ ## __name ## _irq_resume_restore(struct panthor_irq *pirq)		\
> +{												\
> +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> +												\
> +	atomic_set(&pirq->suspended, false);							\
> +	gpu_write(pirq->ptdev, __reg_prefix ## _INT_CLEAR, pirq->mask);				\
> +	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);				\
>  }												\
>  												\
>  static int panthor_request_ ## __name ## _irq(struct panthor_device *ptdev,			\
> @@ -463,13 +482,33 @@ static int panthor_request_ ## __name ## _irq(struct panthor_device *ptdev,			\
>  {												\
>  	pirq->ptdev = ptdev;									\
>  	pirq->irq = irq;									\
> -	panthor_ ## __name ## _irq_resume(pirq, mask);						\
> +	pirq->mask = mask;									\
> +	spin_lock_init(&pirq->mask_lock);							\
> +	panthor_ ## __name ## _irq_resume_restore(pirq);					\
>  												\
>  	return devm_request_threaded_irq(ptdev->base.dev, irq,					\
>  					 panthor_ ## __name ## _irq_raw_handler,		\
>  					 panthor_ ## __name ## _irq_threaded_handler,		\
>  					 IRQF_SHARED, KBUILD_MODNAME "-" # __name,		\
>  					 pirq);							\
> +}												\
> +												\
> +static inline void panthor_ ## __name ## _irq_mask_enable(struct panthor_irq *pirq, u32 mask)	\
> +{												\
> +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> +												\
> +	pirq->mask |= mask;									\
> +	if (!atomic_read(&pirq->suspended))							\
> +		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);			\
> +}												\
> +												\
> +static inline void panthor_ ## __name ## _irq_mask_disable(struct panthor_irq *pirq, u32 mask)	\
> +{												\
> +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> +												\
> +	pirq->mask &= ~mask;									\
> +	if (!atomic_read(&pirq->suspended))							\
> +		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, pirq->mask);			\
>  }
>  
>  extern struct workqueue_struct *panthor_cleanup_wq;
> 

Re: [PATCH v5 1/3] drm/panthor: Extend IRQ helpers for mask modification/restoration

[Nicolas Frattaroli]

On Monday, 22 December 2025 16:23:35 Central European Standard Time Steven Price wrote:
> On 21/12/2025 17:10, Nicolas Frattaroli wrote:
> > The current IRQ helpers do not guarantee mutual exclusion that covers
> > the entire transaction from accessing the mask member and modifying the
> > mask register.
> > 
> > This makes it hard, if not impossible, to implement mask modification
> > helpers that may change one of these outside the normal
> > suspend/resume/isr code paths.
> > 
> > Add a spinlock to struct panthor_irq that protects both the mask member
> > and register. Acquire it in all code paths that access these. Then, add
> > the aforementioned new helpers: mask_enable, mask_disable, and
> > resume_restore. The first two work by ORing and NANDing the mask bits,
> > and the latter relies on the new behaviour that panthor_irq::mask is not
> > set to 0 on suspend.
> > 
> > panthor_irq::suspended remains an atomic, as it's necessarily written to
> > outside the mask_lock in the suspend path.
> > 
> > Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
> > ---
> >  drivers/gpu/drm/panthor/panthor_device.h | 55 +++++++++++++++++++++++++++-----
> >  1 file changed, 47 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
> > index f35e52b9546a..eb75c83e2db3 100644
> > --- a/drivers/gpu/drm/panthor/panthor_device.h
> > +++ b/drivers/gpu/drm/panthor/panthor_device.h
> > @@ -73,11 +73,14 @@ struct panthor_irq {
> >  	/** @irq: IRQ number. */
> >  	int irq;
> >  
> > -	/** @mask: Current mask being applied to xxx_INT_MASK. */
> > +	/** @mask: Values to write to xxx_INT_MASK if active. */
> >  	u32 mask;
> >  
> >  	/** @suspended: Set to true when the IRQ is suspended. */
> >  	atomic_t suspended;
> > +
> > +	/** @mask_lock: protects modifications to _INT_MASK and @mask */
> > +	spinlock_t mask_lock;
> >  };
> >  
> >  /**
> > @@ -410,6 +413,8 @@ static irqreturn_t panthor_ ## __name ## _irq_raw_handler(int irq, void *data)
> >  	struct panthor_irq *pirq = data;							\
> >  	struct panthor_device *ptdev = pirq->ptdev;						\
> >  												\
> > +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> > +												\
> >  	if (atomic_read(&pirq->suspended))							\
> >  		return IRQ_NONE;								\
> >  	if (!gpu_read(ptdev, __reg_prefix ## _INT_STAT))					\
> > @@ -425,8 +430,10 @@ static irqreturn_t panthor_ ## __name ## _irq_threaded_handler(int irq, void *da
> >  	struct panthor_device *ptdev = pirq->ptdev;						\
> >  	irqreturn_t ret = IRQ_NONE;								\
> >  												\
> > +	guard(spinlock_irqsave)(&pirq->mask_lock);						\
> 
> Woah! You can't do that. That means there's a spinlock held while
> calling the __handler() function. So you'll get "sleeping while atomic"
> bug reports. Specifically I can see that panthor_mmu_irq_handler() takes
> a mutex.
> 
> But the whole point of a threaded handler is so that it can sleep, so we
> definitely don't want a spinlock held.

Hmm, yeah, I was worried about that. The core issue is that I'm
trying to make sure we don't restore the mask if we suspended
in the meantime, and don't mess anything up if the mask got
modified in the meantime as well.

I guess I can get us there by reading the mask member into a
local with the lock held for the &'ing, dropping it before
we enter the while, and at the mask restoration point we
take the lock and OR the mask register contents with the
mask we squirrelled away into the local, unless we're
suspended at this point.

I'm not sure if the behaviour when the mask is modified while
we're in the while and we're still using the old one is the
one we want, but going off vibes it seems that in this case,
it'd just conclude that all it has to process has been
processed, and the handler will presumably run again with
the new remaining interrupt status bits.

Kind regards,
Nicolas Frattaroli

> 
> Thanks,
> Steve
> 
> > +												\
> >  	while (true) {										\
> > -		u32 status = gpu_read(ptdev, __reg_prefix ## _INT_RAWSTAT) & pirq->mask;	\
> > +		u32 status = (gpu_read(ptdev, __reg_prefix ## _INT_RAWSTAT) & pirq->mask);	\
> >  												\
> >  		if (!status)									\
> >  			break;									\
> > @@ -443,18 +450,30 @@ static irqreturn_t panthor_ ## __name ## _irq_threaded_handler(int irq, void *da
> >  												\
> >  static inline void panthor_ ## __name ## _irq_suspend(struct panthor_irq *pirq)			\
> >  {												\
> > -	pirq->mask = 0;										\
> > -	gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, 0);					\
> > +	scoped_guard(spinlock_irqsave, &pirq->mask_lock) {					\
> > +		gpu_write(pirq->ptdev, __reg_prefix ## _INT_MASK, 0);				\
> > +	}											\
> >  	synchronize_irq(pirq->irq);								\
> >  	atomic_set(&pirq->suspended, true);							\
> >  }												\
> >  												\
> > [...]