[PATCH v2 0/8] drm/panthor: Add support for Mali-G1 GPUs

[PATCH v2 0/8] drm/panthor: Add support for Mali-G1 GPUs

[Karunika Choo]

This patch series extends the Panthor driver with basic support for
Mali-G1 GPUs.

The v14 architecture introduces several hardware and register-level
changes compared to prior GPUs. This series adds the necessary
architecture-specific support infrastructure, power control and reset
handling for Mali-G1 GPUs.

Patch Breakdown:
[Patch 1-2]:  Refactor panthor_hw to introduce architecture-specific
              hooks and abstractions to support the v14 architecture.
              These patches introduce architecture-specific HW binding
              for function pointers.
[Patch 3-5]:  Adds basic L2 power on/off and soft reset support for the
              PWR_CONTROL block introduced in v14.
[Patch 6]:    Update MCU halt and warm boot operations to reflect the
              GLB_REQ.STATE changes in v14. This ensures that the MCU is
              properly halted and the correct operations are performed
              on warm boot depending on the FW version.
[Patch 7]:    Align endpoint_req with changes introduced in v14, where
              the register is widened to 64-bit and shifed down by
              4-bytes. This patch adds the necessary infrastructure to
              discern the correct endpoint_req register to use.
[Patch 8]:    Enables Mali-G1 support on Panthor by adding HW bindings
              for v14 architecture, product names and path to FW binary.

v2:
 * Merged GPU_ID refactoring patch with the arch-specific panthor_hw
   binding patch (formerly PATCH 01/10 and PATCH 02/10).
 * Dropped panthor_hw feature bitmap patch in favor of functions that
   performs the relevant architecture version checks.
 * Fixed kernel test bot warnings.
 * Replaced function pointer accessor MACROs with static inline
   functions.
 * Refined power control logic, removed unnecessary checks and redundant
   stubs.
 * Replaced explicit CSG_IFACE_VERSION checks with functions describing
   the feature being checked for.
 * General readability improvements, more consistent error handling,
   behaviour clarifications, and formatting fixes.
 * Link to v1: https://lore.kernel.org/all/20251014094337.1009601-1-karunika.choo@arm.com/

Karunika Choo (8):
  drm/panthor: Add arch-specific panthor_hw binding
  drm/panthor: Add architecture-specific function operations
  drm/panthor: Introduce panthor_pwr API and power control framework
  drm/panthor: Implement L2 power on/off via PWR_CONTROL
  drm/panthor: Implement soft reset via PWR_CONTROL
  drm/panthor: Support GLB_REQ.STATE field for Mali-G1 GPUs
  drm/panthor: Support 64-bit endpoint_req register for Mali-G1
  drm/panthor: Add support for Mali-G1 GPUs

 drivers/gpu/drm/panthor/Makefile         |   1 +
 drivers/gpu/drm/panthor/panthor_device.c |  19 +-
 drivers/gpu/drm/panthor/panthor_device.h |  24 +
 drivers/gpu/drm/panthor/panthor_fw.c     | 130 +++++-
 drivers/gpu/drm/panthor/panthor_fw.h     |  32 +-
 drivers/gpu/drm/panthor/panthor_gpu.c    |  11 +-
 drivers/gpu/drm/panthor/panthor_gpu.h    |   1 +
 drivers/gpu/drm/panthor/panthor_hw.c     | 113 ++++-
 drivers/gpu/drm/panthor/panthor_hw.h     |  26 ++
 drivers/gpu/drm/panthor/panthor_pwr.c    | 548 +++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_pwr.h    |  23 +
 drivers/gpu/drm/panthor/panthor_regs.h   |  79 ++++
 drivers/gpu/drm/panthor/panthor_sched.c  |  21 +-
 13 files changed, 987 insertions(+), 41 deletions(-)
 create mode 100644 drivers/gpu/drm/panthor/panthor_pwr.c
 create mode 100644 drivers/gpu/drm/panthor/panthor_pwr.h

--
2.49.0

[PATCH v2 1/8] drm/panthor: Add arch-specific panthor_hw binding

[Karunika Choo]

This patch adds the framework for binding to a specific panthor_hw
structure based on the architecture major value parsed from the GPU_ID
register. This is in preparation of enabling architecture-specific
behaviours based on GPU_ID. As such, it also splits the GPU_ID register
read operation into its own helper function.

This framework allows a single panthor_hw structure to be shared across
multiple architectures should there be minimal changes between them via
the arch_min and arch_max field of the panthor_hw_entry structure,
instead of duplicating the structure across multiple architectures.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * merged GPU_ID refactoring patch with the arch-specific panthor_hw
   binding patch (PATCH 01/10 and PATCH 02/10 in v1).
---
 drivers/gpu/drm/panthor/panthor_device.h |  4 ++
 drivers/gpu/drm/panthor/panthor_hw.c     | 65 +++++++++++++++++++++++-
 drivers/gpu/drm/panthor/panthor_hw.h     |  6 +++
 3 files changed, 74 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
index a764111359d2..1457c1255f1f 100644
--- a/drivers/gpu/drm/panthor/panthor_device.h
+++ b/drivers/gpu/drm/panthor/panthor_device.h
@@ -26,6 +26,7 @@ struct panthor_device;
 struct panthor_gpu;
 struct panthor_group_pool;
 struct panthor_heap_pool;
+struct panthor_hw;
 struct panthor_job;
 struct panthor_mmu;
 struct panthor_fw;
@@ -122,6 +123,9 @@ struct panthor_device {
 	/** @csif_info: Command stream interface information. */
 	struct drm_panthor_csif_info csif_info;

+	/** @hw: GPU-specific data. */
+	struct panthor_hw *hw;
+
 	/** @gpu: GPU management data. */
 	struct panthor_gpu *gpu;

diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
index 4f2858114e5e..b6e7401327c3 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.c
+++ b/drivers/gpu/drm/panthor/panthor_hw.c
@@ -8,6 +8,28 @@
 #define GPU_PROD_ID_MAKE(arch_major, prod_major) \
 	(((arch_major) << 24) | (prod_major))

+/** struct panthor_hw_entry - HW arch major to panthor_hw binding entry */
+struct panthor_hw_entry {
+	/** @arch_min: Minimum supported architecture major value (inclusive) */
+	u8 arch_min;
+
+	/** @arch_max: Maximum supported architecture major value (inclusive) */
+	u8 arch_max;
+
+	/** @hwdev: Pointer to panthor_hw structure */
+	struct panthor_hw *hwdev;
+};
+
+static struct panthor_hw panthor_hw_arch_v10 = {};
+
+static struct panthor_hw_entry panthor_hw_match[] = {
+	{
+		.arch_min = 10,
+		.arch_max = 13,
+		.hwdev = &panthor_hw_arch_v10,
+	},
+};
+
 static char *get_gpu_model_name(struct panthor_device *ptdev)
 {
 	const u32 gpu_id = ptdev->gpu_info.gpu_id;
@@ -62,7 +84,6 @@ static void panthor_gpu_info_init(struct panthor_device *ptdev)
 {
 	unsigned int i;

-	ptdev->gpu_info.gpu_id = gpu_read(ptdev, GPU_ID);
 	ptdev->gpu_info.csf_id = gpu_read(ptdev, GPU_CSF_ID);
 	ptdev->gpu_info.gpu_rev = gpu_read(ptdev, GPU_REVID);
 	ptdev->gpu_info.core_features = gpu_read(ptdev, GPU_CORE_FEATURES);
@@ -117,8 +138,50 @@ static void panthor_hw_info_init(struct panthor_device *ptdev)
 		 ptdev->gpu_info.tiler_present);
 }

+static int panthor_hw_bind_device(struct panthor_device *ptdev)
+{
+	struct panthor_hw *hdev = NULL;
+	const u32 arch_major = GPU_ARCH_MAJOR(ptdev->gpu_info.gpu_id);
+	int i = 0;
+
+	for (i = 0; i < ARRAY_SIZE(panthor_hw_match); i++) {
+		struct panthor_hw_entry *entry = &panthor_hw_match[i];
+
+		if (arch_major >= entry->arch_min && arch_major <= entry->arch_max) {
+			hdev = entry->hwdev;
+			break;
+		}
+	}
+
+	if (!hdev)
+		return -EOPNOTSUPP;
+
+	ptdev->hw = hdev;
+
+	return 0;
+}
+
+static int panthor_hw_gpu_id_init(struct panthor_device *ptdev)
+{
+	ptdev->gpu_info.gpu_id = gpu_read(ptdev, GPU_ID);
+	if (!ptdev->gpu_info.gpu_id)
+		return -ENXIO;
+
+	return 0;
+}
+
 int panthor_hw_init(struct panthor_device *ptdev)
 {
+	int ret = 0;
+
+	ret = panthor_hw_gpu_id_init(ptdev);
+	if (ret)
+		return ret;
+
+	ret = panthor_hw_bind_device(ptdev);
+	if (ret)
+		return ret;
+
 	panthor_hw_info_init(ptdev);

 	return 0;
diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
index 0af6acc6aa6a..39752de3e7ad 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.h
+++ b/drivers/gpu/drm/panthor/panthor_hw.h
@@ -6,6 +6,12 @@

 struct panthor_device;

+/**
+ * struct panthor_hw - GPU specific register mapping and functions
+ */
+struct panthor_hw {
+};
+
 int panthor_hw_init(struct panthor_device *ptdev);

 #endif /* __PANTHOR_HW_H__ */
--
2.49.0

[PATCH v2 2/8] drm/panthor: Add architecture-specific function operations

[Karunika Choo]

Introduce architecture-specific function pointers to support
architecture-dependent behaviours. This patch adds the following
function pointers and updates their usage accordingly:

- soft_reset
- l2_power_on
- l2_power_off

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Updated includes for panthor_hw.h to allow static inline function
   pointer accessor functions instead of MACROs.
 * updated l2_power_off function signature to void instead of returning
   int as we have no way of handling a failure in this case.
---
 drivers/gpu/drm/panthor/panthor_device.c |  5 ++---
 drivers/gpu/drm/panthor/panthor_device.h | 18 +++++++++++++++++-
 drivers/gpu/drm/panthor/panthor_fw.c     |  4 ++--
 drivers/gpu/drm/panthor/panthor_gpu.c    | 11 ++++++++---
 drivers/gpu/drm/panthor/panthor_gpu.h    |  1 +
 drivers/gpu/drm/panthor/panthor_hw.c     | 10 ++++++++--
 drivers/gpu/drm/panthor/panthor_hw.h     | 18 ++++++++++++++++++
 7 files changed, 56 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/panthor/panthor_device.c b/drivers/gpu/drm/panthor/panthor_device.c
index 81df49880bd8..224a9237a2cc 100644
--- a/drivers/gpu/drm/panthor/panthor_device.c
+++ b/drivers/gpu/drm/panthor/panthor_device.c
@@ -18,7 +18,6 @@
 #include "panthor_device.h"
 #include "panthor_fw.h"
 #include "panthor_gpu.h"
-#include "panthor_hw.h"
 #include "panthor_mmu.h"
 #include "panthor_regs.h"
 #include "panthor_sched.h"
@@ -141,8 +140,8 @@ static void panthor_device_reset_work(struct work_struct *work)
 	panthor_sched_pre_reset(ptdev);
 	panthor_fw_pre_reset(ptdev, true);
 	panthor_mmu_pre_reset(ptdev);
-	panthor_gpu_soft_reset(ptdev);
-	panthor_gpu_l2_power_on(ptdev);
+	panthor_device_soft_reset(ptdev);
+	panthor_device_l2_power_on(ptdev);
 	panthor_mmu_post_reset(ptdev);
 	ret = panthor_fw_post_reset(ptdev);
 	atomic_set(&ptdev->reset.pending, 0);
diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
index 1457c1255f1f..2026cc6532ce 100644
--- a/drivers/gpu/drm/panthor/panthor_device.h
+++ b/drivers/gpu/drm/panthor/panthor_device.h
@@ -20,13 +20,14 @@
 #include <drm/gpu_scheduler.h>
 #include <drm/panthor_drm.h>

+#include "panthor_hw.h"
+
 struct panthor_csf;
 struct panthor_csf_ctx;
 struct panthor_device;
 struct panthor_gpu;
 struct panthor_group_pool;
 struct panthor_heap_pool;
-struct panthor_hw;
 struct panthor_job;
 struct panthor_mmu;
 struct panthor_fw;
@@ -532,4 +533,19 @@ static inline u64 gpu_read64_counter(struct panthor_device *ptdev, u32 reg)
 	read_poll_timeout(gpu_read64_relaxed, val, cond, delay_us, timeout_us,	\
 			  false, dev, reg)

+static inline int panthor_device_soft_reset(struct panthor_device *ptdev)
+{
+	return ptdev->hw->ops.soft_reset(ptdev);
+}
+
+static inline int panthor_device_l2_power_on(struct panthor_device *ptdev)
+{
+	return ptdev->hw->ops.l2_power_on(ptdev);
+}
+
+static inline void panthor_device_l2_power_off(struct panthor_device *ptdev)
+{
+	ptdev->hw->ops.l2_power_off(ptdev);
+}
+
 #endif
diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
index 9bf06e55eaee..6b91c3cb6678 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.c
+++ b/drivers/gpu/drm/panthor/panthor_fw.c
@@ -1184,7 +1184,7 @@ void panthor_fw_unplug(struct panthor_device *ptdev)
 	ptdev->fw->vm = NULL;

 	if (!IS_ENABLED(CONFIG_PM) || pm_runtime_active(ptdev->base.dev))
-		panthor_gpu_power_off(ptdev, L2, ptdev->gpu_info.l2_present, 20000);
+		panthor_device_l2_power_off(ptdev);
 }

 /**
@@ -1363,7 +1363,7 @@ int panthor_fw_init(struct panthor_device *ptdev)
 		return ret;
 	}

-	ret = panthor_gpu_l2_power_on(ptdev);
+	ret = panthor_device_l2_power_on(ptdev);
 	if (ret)
 		return ret;

diff --git a/drivers/gpu/drm/panthor/panthor_gpu.c b/drivers/gpu/drm/panthor/panthor_gpu.c
index db69449a5be0..f6181462047f 100644
--- a/drivers/gpu/drm/panthor/panthor_gpu.c
+++ b/drivers/gpu/drm/panthor/panthor_gpu.c
@@ -218,6 +218,11 @@ int panthor_gpu_block_power_on(struct panthor_device *ptdev,
 	return 0;
 }

+void panthor_gpu_l2_power_off(struct panthor_device *ptdev)
+{
+	panthor_gpu_power_off(ptdev, L2, ptdev->gpu_info.l2_present, 20000);
+}
+
 /**
  * panthor_gpu_l2_power_on() - Power-on the L2-cache
  * @ptdev: Device.
@@ -344,9 +349,9 @@ void panthor_gpu_suspend(struct panthor_device *ptdev)
 {
 	/* On a fast reset, simply power down the L2. */
 	if (!ptdev->reset.fast)
-		panthor_gpu_soft_reset(ptdev);
+		panthor_device_soft_reset(ptdev);
 	else
-		panthor_gpu_power_off(ptdev, L2, 1, 20000);
+		panthor_device_l2_power_off(ptdev);

 	panthor_gpu_irq_suspend(&ptdev->gpu->irq);
 }
@@ -361,6 +366,6 @@ 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_l2_power_on(ptdev);
+	panthor_device_l2_power_on(ptdev);
 }

diff --git a/drivers/gpu/drm/panthor/panthor_gpu.h b/drivers/gpu/drm/panthor/panthor_gpu.h
index 7c17a8c06858..12e66f48ced1 100644
--- a/drivers/gpu/drm/panthor/panthor_gpu.h
+++ b/drivers/gpu/drm/panthor/panthor_gpu.h
@@ -46,6 +46,7 @@ int panthor_gpu_block_power_off(struct panthor_device *ptdev,
 				   type ## _PWRTRANS, \
 				   mask, timeout_us)

+void panthor_gpu_l2_power_off(struct panthor_device *ptdev);
 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);
diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
index b6e7401327c3..092962db5ccd 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.c
+++ b/drivers/gpu/drm/panthor/panthor_hw.c
@@ -2,7 +2,7 @@
 /* Copyright 2025 ARM Limited. All rights reserved. */

 #include "panthor_device.h"
-#include "panthor_hw.h"
+#include "panthor_gpu.h"
 #include "panthor_regs.h"

 #define GPU_PROD_ID_MAKE(arch_major, prod_major) \
@@ -20,7 +20,13 @@ struct panthor_hw_entry {
 	struct panthor_hw *hwdev;
 };

-static struct panthor_hw panthor_hw_arch_v10 = {};
+static struct panthor_hw panthor_hw_arch_v10 = {
+	.ops = {
+		.soft_reset = panthor_gpu_soft_reset,
+		.l2_power_off = panthor_gpu_l2_power_off,
+		.l2_power_on = panthor_gpu_l2_power_on,
+	},
+};

 static struct panthor_hw_entry panthor_hw_match[] = {
 	{
diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
index 39752de3e7ad..2665d6dde2e3 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.h
+++ b/drivers/gpu/drm/panthor/panthor_hw.h
@@ -6,10 +6,28 @@

 struct panthor_device;

+/**
+ * struct panthor_hw_ops - HW operations that are specific to a GPU
+ */
+struct panthor_hw_ops {
+	/** @soft_reset: Soft reset function pointer */
+	int (*soft_reset)(struct panthor_device *ptdev);
+
+	/** @l2_power_off: L2 power off function pointer */
+	void (*l2_power_off)(struct panthor_device *ptdev);
+
+	/** @l2_power_on: L2 power on function pointer */
+	int (*l2_power_on)(struct panthor_device *ptdev);
+};
+
 /**
  * struct panthor_hw - GPU specific register mapping and functions
  */
 struct panthor_hw {
+	/** @features: Bitmap containing panthor_hw_feature */
+
+	/** @ops: Panthor HW specific operations */
+	struct panthor_hw_ops ops;
 };

 int panthor_hw_init(struct panthor_device *ptdev);
--
2.49.0

[PATCH v2 3/8] drm/panthor: Introduce panthor_pwr API and power control framework

[Karunika Choo]

Add the new panthor_pwr module, which provides basic power control
management for Mali-G1 GPUs. The initial implementation includes
infrastructure for initializing the PWR_CONTROL block, requesting and
handling its IRQ, and checking for PWR_CONTROL support based on GPU
architecture.

The patch also integrates panthor_pwr with the device lifecycle (init,
suspend, resume, and unplug) through the new API functions. It also
registers the IRQ handler under the 'gpu' IRQ as the PWR_CONTROL block
is located within the GPU_CONTROL block.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Removed stub functions.
 * Updated BIT() definitions for 64-bit fields to use BIT_U64() to
   address kernel test robot warnings for 32-bit systems.
 * Moved GPU_FEATURES_RAY_TRAVERSAL definition to the next patch where
   it is being used.
 * Drop the use of feature bits in favour of a function that performs a
   GPU_ARCH_MAJOR check instead.
---
 drivers/gpu/drm/panthor/Makefile         |   1 +
 drivers/gpu/drm/panthor/panthor_device.c |  14 ++-
 drivers/gpu/drm/panthor/panthor_device.h |   4 +
 drivers/gpu/drm/panthor/panthor_hw.c     |   5 +
 drivers/gpu/drm/panthor/panthor_hw.h     |   2 +
 drivers/gpu/drm/panthor/panthor_pwr.c    | 120 +++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_pwr.h    |  17 ++++
 drivers/gpu/drm/panthor/panthor_regs.h   |  78 +++++++++++++++
 8 files changed, 240 insertions(+), 1 deletion(-)
 create mode 100644 drivers/gpu/drm/panthor/panthor_pwr.c
 create mode 100644 drivers/gpu/drm/panthor/panthor_pwr.h

diff --git a/drivers/gpu/drm/panthor/Makefile b/drivers/gpu/drm/panthor/Makefile
index 02db21748c12..753a32c446df 100644
--- a/drivers/gpu/drm/panthor/Makefile
+++ b/drivers/gpu/drm/panthor/Makefile
@@ -10,6 +10,7 @@ panthor-y := \
 	panthor_heap.o \
 	panthor_hw.o \
 	panthor_mmu.o \
+	panthor_pwr.o \
 	panthor_sched.o

 obj-$(CONFIG_DRM_PANTHOR) += panthor.o
diff --git a/drivers/gpu/drm/panthor/panthor_device.c b/drivers/gpu/drm/panthor/panthor_device.c
index 224a9237a2cc..9d2adcb5431d 100644
--- a/drivers/gpu/drm/panthor/panthor_device.c
+++ b/drivers/gpu/drm/panthor/panthor_device.c
@@ -19,6 +19,7 @@
 #include "panthor_fw.h"
 #include "panthor_gpu.h"
 #include "panthor_mmu.h"
+#include "panthor_pwr.h"
 #include "panthor_regs.h"
 #include "panthor_sched.h"

@@ -101,6 +102,7 @@ void panthor_device_unplug(struct panthor_device *ptdev)
 	panthor_fw_unplug(ptdev);
 	panthor_mmu_unplug(ptdev);
 	panthor_gpu_unplug(ptdev);
+	panthor_pwr_unplug(ptdev);

 	pm_runtime_dont_use_autosuspend(ptdev->base.dev);
 	pm_runtime_put_sync_suspend(ptdev->base.dev);
@@ -248,10 +250,14 @@ int panthor_device_init(struct panthor_device *ptdev)
 	if (ret)
 		goto err_rpm_put;

-	ret = panthor_gpu_init(ptdev);
+	ret = panthor_pwr_init(ptdev);
 	if (ret)
 		goto err_rpm_put;

+	ret = panthor_gpu_init(ptdev);
+	if (ret)
+		goto err_unplug_pwr;
+
 	ret = panthor_gpu_coherency_init(ptdev);
 	if (ret)
 		goto err_unplug_gpu;
@@ -292,6 +298,9 @@ int panthor_device_init(struct panthor_device *ptdev)
 err_unplug_gpu:
 	panthor_gpu_unplug(ptdev);

+err_unplug_pwr:
+	panthor_pwr_unplug(ptdev);
+
 err_rpm_put:
 	pm_runtime_put_sync_suspend(ptdev->base.dev);
 	return ret;
@@ -445,6 +454,7 @@ static int panthor_device_resume_hw_components(struct panthor_device *ptdev)
 {
 	int ret;

+	panthor_pwr_resume(ptdev);
 	panthor_gpu_resume(ptdev);
 	panthor_mmu_resume(ptdev);

@@ -454,6 +464,7 @@ static int panthor_device_resume_hw_components(struct panthor_device *ptdev)

 	panthor_mmu_suspend(ptdev);
 	panthor_gpu_suspend(ptdev);
+	panthor_pwr_suspend(ptdev);
 	return ret;
 }

@@ -567,6 +578,7 @@ int panthor_device_suspend(struct device *dev)
 		panthor_fw_suspend(ptdev);
 		panthor_mmu_suspend(ptdev);
 		panthor_gpu_suspend(ptdev);
+		panthor_pwr_suspend(ptdev);
 		drm_dev_exit(cookie);
 	}

diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
index 2026cc6532ce..3231c8154e60 100644
--- a/drivers/gpu/drm/panthor/panthor_device.h
+++ b/drivers/gpu/drm/panthor/panthor_device.h
@@ -32,6 +32,7 @@ struct panthor_job;
 struct panthor_mmu;
 struct panthor_fw;
 struct panthor_perfcnt;
+struct panthor_pwr;
 struct panthor_vm;
 struct panthor_vm_pool;

@@ -127,6 +128,9 @@ struct panthor_device {
 	/** @hw: GPU-specific data. */
 	struct panthor_hw *hw;

+	/** @pwr: Power control management data. */
+	struct panthor_pwr *pwr;
+
 	/** @gpu: GPU management data. */
 	struct panthor_gpu *gpu;

diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
index 092962db5ccd..09aef34a6ce7 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.c
+++ b/drivers/gpu/drm/panthor/panthor_hw.c
@@ -192,3 +192,8 @@ int panthor_hw_init(struct panthor_device *ptdev)

 	return 0;
 }
+
+bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev)
+{
+	return GPU_ARCH_MAJOR(ptdev->gpu_info.gpu_id) >= 14;
+}
diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
index 2665d6dde2e3..4c71f27d1c0b 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.h
+++ b/drivers/gpu/drm/panthor/panthor_hw.h
@@ -32,4 +32,6 @@ struct panthor_hw {

 int panthor_hw_init(struct panthor_device *ptdev);

+bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev);
+
 #endif /* __PANTHOR_HW_H__ */
diff --git a/drivers/gpu/drm/panthor/panthor_pwr.c b/drivers/gpu/drm/panthor/panthor_pwr.c
new file mode 100644
index 000000000000..da64fe006a8b
--- /dev/null
+++ b/drivers/gpu/drm/panthor/panthor_pwr.c
@@ -0,0 +1,120 @@
+// SPDX-License-Identifier: GPL-2.0 or MIT
+/* Copyright 2025 ARM Limited. All rights reserved. */
+
+#include <linux/platform_device.h>
+#include <linux/interrupt.h>
+#include <linux/iopoll.h>
+#include <linux/wait.h>
+
+#include <drm/drm_managed.h>
+
+#include "panthor_device.h"
+#include "panthor_hw.h"
+#include "panthor_pwr.h"
+#include "panthor_regs.h"
+
+#define PWR_INTERRUPTS_MASK \
+	(PWR_IRQ_POWER_CHANGED_SINGLE | \
+	 PWR_IRQ_POWER_CHANGED_ALL | \
+	 PWR_IRQ_DELEGATION_CHANGED | \
+	 PWR_IRQ_RESET_COMPLETED | \
+	 PWR_IRQ_RETRACT_COMPLETED | \
+	 PWR_IRQ_INSPECT_COMPLETED | \
+	 PWR_IRQ_COMMAND_NOT_ALLOWED | \
+	 PWR_IRQ_COMMAND_INVALID)
+
+/**
+ * struct panthor_pwr - PWR_CONTROL block management data.
+ */
+struct panthor_pwr {
+	/** @irq: PWR irq. */
+	struct panthor_irq irq;
+
+	/** @reqs_lock: Lock protecting access to pending_reqs. */
+	spinlock_t reqs_lock;
+
+	/** @pending_reqs: Pending PWR requests. */
+	u32 pending_reqs;
+
+	/** @reqs_acked: PWR request wait queue. */
+	wait_queue_head_t reqs_acked;
+};
+
+static void panthor_pwr_irq_handler(struct panthor_device *ptdev, u32 status)
+{
+	spin_lock(&ptdev->pwr->reqs_lock);
+	gpu_write(ptdev, PWR_INT_CLEAR, status);
+
+	if (unlikely(status & PWR_IRQ_COMMAND_NOT_ALLOWED))
+		drm_err(&ptdev->base, "PWR_IRQ: COMMAND_NOT_ALLOWED");
+
+	if (unlikely(status & PWR_IRQ_COMMAND_INVALID))
+		drm_err(&ptdev->base, "PWR_IRQ: COMMAND_INVALID");
+
+	if (status & ptdev->pwr->pending_reqs) {
+		ptdev->pwr->pending_reqs &= ~status;
+		wake_up_all(&ptdev->pwr->reqs_acked);
+	}
+	spin_unlock(&ptdev->pwr->reqs_lock);
+}
+PANTHOR_IRQ_HANDLER(pwr, PWR, panthor_pwr_irq_handler);
+
+void panthor_pwr_unplug(struct panthor_device *ptdev)
+{
+	unsigned long flags;
+
+	if (!ptdev->pwr)
+		return;
+
+	/* Make sure the IRQ handler is not running after that point. */
+	panthor_pwr_irq_suspend(&ptdev->pwr->irq);
+
+	/* Wake-up all waiters. */
+	spin_lock_irqsave(&ptdev->pwr->reqs_lock, flags);
+	ptdev->pwr->pending_reqs = 0;
+	wake_up_all(&ptdev->pwr->reqs_acked);
+	spin_unlock_irqrestore(&ptdev->pwr->reqs_lock, flags);
+}
+
+int panthor_pwr_init(struct panthor_device *ptdev)
+{
+	struct panthor_pwr *pwr;
+	int err, irq;
+
+	if (!panthor_hw_has_pwr_ctrl(ptdev))
+		return 0;
+
+	pwr = drmm_kzalloc(&ptdev->base, sizeof(*pwr), GFP_KERNEL);
+	if (!pwr)
+		return -ENOMEM;
+
+	spin_lock_init(&pwr->reqs_lock);
+	init_waitqueue_head(&pwr->reqs_acked);
+	ptdev->pwr = pwr;
+
+	irq = platform_get_irq_byname(to_platform_device(ptdev->base.dev), "gpu");
+	if (irq < 0)
+		return irq;
+
+	err = panthor_request_pwr_irq(ptdev, &pwr->irq, irq, PWR_INTERRUPTS_MASK);
+	if (err)
+		return err;
+
+	return 0;
+}
+
+void panthor_pwr_suspend(struct panthor_device *ptdev)
+{
+	if (!ptdev->pwr)
+		return;
+
+	panthor_pwr_irq_suspend(&ptdev->pwr->irq);
+}
+
+void panthor_pwr_resume(struct panthor_device *ptdev)
+{
+	if (!ptdev->pwr)
+		return;
+
+	panthor_pwr_irq_resume(&ptdev->pwr->irq, PWR_INTERRUPTS_MASK);
+}
diff --git a/drivers/gpu/drm/panthor/panthor_pwr.h b/drivers/gpu/drm/panthor/panthor_pwr.h
new file mode 100644
index 000000000000..b325e5b7eba3
--- /dev/null
+++ b/drivers/gpu/drm/panthor/panthor_pwr.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: GPL-2.0 or MIT */
+/* Copyright 2025 ARM Limited. All rights reserved. */
+
+#ifndef __PANTHOR_PWR_H__
+#define __PANTHOR_PWR_H__
+
+struct panthor_device;
+
+void panthor_pwr_unplug(struct panthor_device *ptdev);
+
+int panthor_pwr_init(struct panthor_device *ptdev);
+
+void panthor_pwr_suspend(struct panthor_device *ptdev);
+
+void panthor_pwr_resume(struct panthor_device *ptdev);
+
+#endif /* __PANTHOR_PWR_H__ */
diff --git a/drivers/gpu/drm/panthor/panthor_regs.h b/drivers/gpu/drm/panthor/panthor_regs.h
index 8bee76d01bf8..5469eec02178 100644
--- a/drivers/gpu/drm/panthor/panthor_regs.h
+++ b/drivers/gpu/drm/panthor/panthor_regs.h
@@ -205,4 +205,82 @@
 #define CSF_DOORBELL(i)					(0x80000 + ((i) * 0x10000))
 #define CSF_GLB_DOORBELL_ID				0

+/* PWR Control registers */
+
+#define PWR_CONTROL_BASE				0x800
+#define PWR_CTRL_REG(x)					(PWR_CONTROL_BASE + (x))
+
+#define PWR_INT_RAWSTAT					PWR_CTRL_REG(0x0)
+#define PWR_INT_CLEAR					PWR_CTRL_REG(0x4)
+#define PWR_INT_MASK					PWR_CTRL_REG(0x8)
+#define PWR_INT_STAT					PWR_CTRL_REG(0xc)
+#define   PWR_IRQ_POWER_CHANGED_SINGLE			BIT(0)
+#define   PWR_IRQ_POWER_CHANGED_ALL			BIT(1)
+#define   PWR_IRQ_DELEGATION_CHANGED			BIT(2)
+#define   PWR_IRQ_RESET_COMPLETED			BIT(3)
+#define   PWR_IRQ_RETRACT_COMPLETED			BIT(4)
+#define   PWR_IRQ_INSPECT_COMPLETED			BIT(5)
+#define   PWR_IRQ_COMMAND_NOT_ALLOWED			BIT(30)
+#define   PWR_IRQ_COMMAND_INVALID			BIT(31)
+
+#define PWR_STATUS					PWR_CTRL_REG(0x20)
+#define   PWR_STATUS_ALLOW_L2				BIT_U64(0)
+#define   PWR_STATUS_ALLOW_TILER			BIT_U64(1)
+#define   PWR_STATUS_ALLOW_SHADER			BIT_U64(8)
+#define   PWR_STATUS_ALLOW_BASE				BIT_U64(14)
+#define   PWR_STATUS_ALLOW_STACK			BIT_U64(15)
+#define   PWR_STATUS_DOMAIN_ALLOWED(x)			BIT_U64(x)
+#define   PWR_STATUS_DELEGATED_L2			BIT_U64(16)
+#define   PWR_STATUS_DELEGATED_TILER			BIT_U64(17)
+#define   PWR_STATUS_DELEGATED_SHADER			BIT_U64(24)
+#define   PWR_STATUS_DELEGATED_BASE			BIT_U64(30)
+#define   PWR_STATUS_DELEGATED_STACK			BIT_U64(31)
+#define   PWR_STATUS_DELEGATED_SHIFT			16
+#define   PWR_STATUS_DOMAIN_DELEGATED(x)		BIT_U64((x) + PWR_STATUS_DELEGATED_SHIFT)
+#define   PWR_STATUS_ALLOW_SOFT_RESET			BIT_U64(33)
+#define   PWR_STATUS_ALLOW_FAST_RESET			BIT_U64(34)
+#define   PWR_STATUS_POWER_PENDING			BIT_U64(41)
+#define   PWR_STATUS_RESET_PENDING			BIT_U64(42)
+#define   PWR_STATUS_RETRACT_PENDING			BIT_U64(43)
+#define   PWR_STATUS_INSPECT_PENDING			BIT_U64(44)
+
+#define PWR_COMMAND					PWR_CTRL_REG(0x28)
+#define   PWR_COMMAND_POWER_UP				0x10
+#define   PWR_COMMAND_POWER_DOWN			0x11
+#define   PWR_COMMAND_DELEGATE				0x20
+#define   PWR_COMMAND_RETRACT				0x21
+#define   PWR_COMMAND_RESET_SOFT			0x31
+#define   PWR_COMMAND_RESET_FAST			0x32
+#define   PWR_COMMAND_INSPECT				0xF0
+#define   PWR_COMMAND_DOMAIN_L2				0
+#define   PWR_COMMAND_DOMAIN_TILER			1
+#define   PWR_COMMAND_DOMAIN_SHADER			8
+#define   PWR_COMMAND_DOMAIN_BASE			14
+#define   PWR_COMMAND_DOMAIN_STACK			15
+#define   PWR_COMMAND_SUBDOMAIN_RTU			BIT(0)
+#define   PWR_COMMAND_DEF(cmd, domain, subdomain)	\
+	(((subdomain) << 16) | ((domain) << 8) | (cmd))
+
+#define PWR_CMDARG					PWR_CTRL_REG(0x30)
+
+#define PWR_L2_PRESENT					PWR_CTRL_REG(0x100)
+#define PWR_L2_READY					PWR_CTRL_REG(0x108)
+#define PWR_L2_PWRTRANS					PWR_CTRL_REG(0x110)
+#define PWR_L2_PWRACTIVE				PWR_CTRL_REG(0x118)
+#define PWR_TILER_PRESENT				PWR_CTRL_REG(0x140)
+#define PWR_TILER_READY					PWR_CTRL_REG(0x148)
+#define PWR_TILER_PWRTRANS				PWR_CTRL_REG(0x150)
+#define PWR_TILER_PWRACTIVE				PWR_CTRL_REG(0x158)
+#define PWR_SHADER_PRESENT				PWR_CTRL_REG(0x200)
+#define PWR_SHADER_READY				PWR_CTRL_REG(0x208)
+#define PWR_SHADER_PWRTRANS				PWR_CTRL_REG(0x210)
+#define PWR_SHADER_PWRACTIVE				PWR_CTRL_REG(0x218)
+#define PWR_BASE_PRESENT				PWR_CTRL_REG(0x380)
+#define PWR_BASE_READY					PWR_CTRL_REG(0x388)
+#define PWR_BASE_PWRTRANS				PWR_CTRL_REG(0x390)
+#define PWR_BASE_PWRACTIVE				PWR_CTRL_REG(0x398)
+#define PWR_STACK_PRESENT				PWR_CTRL_REG(0x3c0)
+#define PWR_STACK_READY					PWR_CTRL_REG(0x3c8)
+#define PWR_STACK_PWRTRANS				PWR_CTRL_REG(0x3d0)
+
 #endif
--
2.49.0

Re: [PATCH v2 3/8] drm/panthor: Introduce panthor_pwr API and power control framework

[Boris Brezillon]

On Fri, 24 Oct 2025 21:21:12 +0100
Karunika Choo <karunika.choo@arm.com> wrote:
> diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
> index 092962db5ccd..09aef34a6ce7 100644
> --- a/drivers/gpu/drm/panthor/panthor_hw.c
> +++ b/drivers/gpu/drm/panthor/panthor_hw.c
> @@ -192,3 +192,8 @@ int panthor_hw_init(struct panthor_device *ptdev)
> 
>  	return 0;
>  }
> +
> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev)
> +{
> +	return GPU_ARCH_MAJOR(ptdev->gpu_info.gpu_id) >= 14;
> +}
> diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
> index 2665d6dde2e3..4c71f27d1c0b 100644
> --- a/drivers/gpu/drm/panthor/panthor_hw.h
> +++ b/drivers/gpu/drm/panthor/panthor_hw.h
> @@ -32,4 +32,6 @@ struct panthor_hw {
> 
>  int panthor_hw_init(struct panthor_device *ptdev);
> 
> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev);

Let's make this a static inline function so the compiler can inline its
content at compile time.

Re: [PATCH v2 3/8] drm/panthor: Introduce panthor_pwr API and power control framework

[Karunika Choo]

On 26/10/2025 08:13, Boris Brezillon wrote:
> On Fri, 24 Oct 2025 21:21:12 +0100
> Karunika Choo <karunika.choo@arm.com> wrote:
>> diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
>> index 092962db5ccd..09aef34a6ce7 100644
>> --- a/drivers/gpu/drm/panthor/panthor_hw.c
>> +++ b/drivers/gpu/drm/panthor/panthor_hw.c
>> @@ -192,3 +192,8 @@ int panthor_hw_init(struct panthor_device *ptdev)
>>
>>  	return 0;
>>  }
>> +
>> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev)
>> +{
>> +	return GPU_ARCH_MAJOR(ptdev->gpu_info.gpu_id) >= 14;
>> +}
>> diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
>> index 2665d6dde2e3..4c71f27d1c0b 100644
>> --- a/drivers/gpu/drm/panthor/panthor_hw.h
>> +++ b/drivers/gpu/drm/panthor/panthor_hw.h
>> @@ -32,4 +32,6 @@ struct panthor_hw {
>>
>>  int panthor_hw_init(struct panthor_device *ptdev);
>>
>> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev);
> 
> Let's make this a static inline function so the compiler can inline its
> content at compile time.
> 

I wonder if making it static inline is a good idea. We will need to move
this function into panthor_device.h to be able to do so as we need to
access the panthor_device structure. Would this still be desirable?

Re: [PATCH v2 3/8] drm/panthor: Introduce panthor_pwr API and power control framework

[Boris Brezillon]

On Mon, 27 Oct 2025 10:36:13 +0000
Karunika Choo <karunika.choo@arm.com> wrote:

> On 26/10/2025 08:13, Boris Brezillon wrote:
> > On Fri, 24 Oct 2025 21:21:12 +0100
> > Karunika Choo <karunika.choo@arm.com> wrote:  
> >> diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
> >> index 092962db5ccd..09aef34a6ce7 100644
> >> --- a/drivers/gpu/drm/panthor/panthor_hw.c
> >> +++ b/drivers/gpu/drm/panthor/panthor_hw.c
> >> @@ -192,3 +192,8 @@ int panthor_hw_init(struct panthor_device *ptdev)
> >>
> >>  	return 0;
> >>  }
> >> +
> >> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev)
> >> +{
> >> +	return GPU_ARCH_MAJOR(ptdev->gpu_info.gpu_id) >= 14;
> >> +}
> >> diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
> >> index 2665d6dde2e3..4c71f27d1c0b 100644
> >> --- a/drivers/gpu/drm/panthor/panthor_hw.h
> >> +++ b/drivers/gpu/drm/panthor/panthor_hw.h
> >> @@ -32,4 +32,6 @@ struct panthor_hw {
> >>
> >>  int panthor_hw_init(struct panthor_device *ptdev);
> >>
> >> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev);  
> > 
> > Let's make this a static inline function so the compiler can inline its
> > content at compile time.
> >   
> 
> I wonder if making it static inline is a good idea. We will need to move
> this function into panthor_device.h to be able to do so as we need to
> access the panthor_device structure. Would this still be desirable?

Can't we just include panthor_device.h from panthor_hw.h instead? As
far as I can tell, panthor_device.h doesn't include panthor_hw.h, so we
should be fine.

Re: [PATCH v2 3/8] drm/panthor: Introduce panthor_pwr API and power control framework

[Karunika Choo]

On 27/10/2025 11:36, Boris Brezillon wrote:
> On Mon, 27 Oct 2025 10:36:13 +0000
> Karunika Choo <karunika.choo@arm.com> wrote:
> 
>> On 26/10/2025 08:13, Boris Brezillon wrote:
>>> On Fri, 24 Oct 2025 21:21:12 +0100
>>> Karunika Choo <karunika.choo@arm.com> wrote:  
>>>> diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
>>>> index 092962db5ccd..09aef34a6ce7 100644
>>>> --- a/drivers/gpu/drm/panthor/panthor_hw.c
>>>> +++ b/drivers/gpu/drm/panthor/panthor_hw.c
>>>> @@ -192,3 +192,8 @@ int panthor_hw_init(struct panthor_device *ptdev)
>>>>
>>>>  	return 0;
>>>>  }
>>>> +
>>>> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev)
>>>> +{
>>>> +	return GPU_ARCH_MAJOR(ptdev->gpu_info.gpu_id) >= 14;
>>>> +}
>>>> diff --git a/drivers/gpu/drm/panthor/panthor_hw.h b/drivers/gpu/drm/panthor/panthor_hw.h
>>>> index 2665d6dde2e3..4c71f27d1c0b 100644
>>>> --- a/drivers/gpu/drm/panthor/panthor_hw.h
>>>> +++ b/drivers/gpu/drm/panthor/panthor_hw.h
>>>> @@ -32,4 +32,6 @@ struct panthor_hw {
>>>>
>>>>  int panthor_hw_init(struct panthor_device *ptdev);
>>>>
>>>> +bool panthor_hw_has_pwr_ctrl(struct panthor_device *ptdev);  
>>>
>>> Let's make this a static inline function so the compiler can inline its
>>> content at compile time.
>>>   
>>
>> I wonder if making it static inline is a good idea. We will need to move
>> this function into panthor_device.h to be able to do so as we need to
>> access the panthor_device structure. Would this still be desirable?
> 
> Can't we just include panthor_device.h from panthor_hw.h instead? As
> far as I can tell, panthor_device.h doesn't include panthor_hw.h, so we
> should be fine.

v2 changed this, but let me see if I can rework these to enable both
your's and Steve's suggestions.

Kind regards,
Karunika Choo

[PATCH v2 4/8] drm/panthor: Implement L2 power on/off via PWR_CONTROL

[Karunika Choo]

This patch adds common helpers to issue power commands, poll
transitions, and validate domain state, then wires them into the L2
on/off paths.

The L2 power-on sequence now delegates control of the SHADER and TILER
domains to the MCU when allowed, while the L2 itself is never delegated.
On power-off, dependent domains beneath the L2 are checked, and if
necessary, retracted and powered down to maintain proper domain
ordering.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Updated GENMASK to GENMASK_U64 to address kernel test robot warnings
   for 32-bit systems.
 * Removed panthor_pwr_read_status() in favour of a simple gpu_read64()
   operation on the PWR_STATUS register.
 * Renamed panthor_pwr_info_show() to panthor_pwr_debug_info_show() for
   more clarity.
 * Added additional WARN_ON for an invalid domain when requesting power
   domain transition.
 * Made panthor_pwr_domain_transition()'s expected val logic more
   readable and clearer.
 * Wait on domain power transition instead of failing the operation.
 * Fixed inconsistent error return value vs kerneldoc.
 * Removed confusing drm_dbg in delegate_domain() in favor of a comment.
 * Add unwind to panthor_pwr_delegate_domains().
 * Moved child domain handling logic from panthor_pwr_l2_power_off()
   into panthor_pwr_domain_force_off().
 * Added additional clarification regarding delegation and retraction of
   power domains.
 * Minor formatting and readability changes and remove unnecessary
   checks.
---
 drivers/gpu/drm/panthor/panthor_pwr.c  | 378 +++++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_pwr.h  |   4 +
 drivers/gpu/drm/panthor/panthor_regs.h |   1 +
 3 files changed, 383 insertions(+)

diff --git a/drivers/gpu/drm/panthor/panthor_pwr.c b/drivers/gpu/drm/panthor/panthor_pwr.c
index da64fe006a8b..cd529660a276 100644
--- a/drivers/gpu/drm/panthor/panthor_pwr.c
+++ b/drivers/gpu/drm/panthor/panthor_pwr.c
@@ -23,6 +23,14 @@
 	 PWR_IRQ_COMMAND_NOT_ALLOWED | \
 	 PWR_IRQ_COMMAND_INVALID)

+#define PWR_ALL_CORES_MASK		GENMASK_U64(63, 0)
+
+#define PWR_DOMAIN_MAX_BITS		16
+
+#define PWR_TRANSITION_TIMEOUT_US	(2ULL * USEC_PER_SEC)
+
+#define PWR_RETRACT_TIMEOUT_US		(2ULL * USEC_PER_MSEC)
+
 /**
  * struct panthor_pwr - PWR_CONTROL block management data.
  */
@@ -59,6 +67,323 @@ static void panthor_pwr_irq_handler(struct panthor_device *ptdev, u32 status)
 }
 PANTHOR_IRQ_HANDLER(pwr, PWR, panthor_pwr_irq_handler);

+static void panthor_pwr_write_command(struct panthor_device *ptdev, u32 command, u64 args)
+{
+	if (args)
+		gpu_write64(ptdev, PWR_CMDARG, args);
+
+	gpu_write(ptdev, PWR_COMMAND, command);
+}
+
+static const char *get_domain_name(u8 domain)
+{
+	switch (domain) {
+	case PWR_COMMAND_DOMAIN_L2:
+		return "L2";
+	case PWR_COMMAND_DOMAIN_TILER:
+		return "Tiler";
+	case PWR_COMMAND_DOMAIN_SHADER:
+		return "Shader";
+	case PWR_COMMAND_DOMAIN_BASE:
+		return "Base";
+	case PWR_COMMAND_DOMAIN_STACK:
+		return "Stack";
+	}
+	return "Unknown";
+}
+
+static u32 get_domain_base(u8 domain)
+{
+	switch (domain) {
+	case PWR_COMMAND_DOMAIN_L2:
+		return PWR_L2_PRESENT;
+	case PWR_COMMAND_DOMAIN_TILER:
+		return PWR_TILER_PRESENT;
+	case PWR_COMMAND_DOMAIN_SHADER:
+		return PWR_SHADER_PRESENT;
+	case PWR_COMMAND_DOMAIN_BASE:
+		return PWR_BASE_PRESENT;
+	case PWR_COMMAND_DOMAIN_STACK:
+		return PWR_STACK_PRESENT;
+	}
+	return 0;
+}
+
+static u32 get_domain_ready_reg(u32 domain)
+{
+	return get_domain_base(domain) + (PWR_L2_READY - PWR_L2_PRESENT);
+}
+
+static u32 get_domain_pwrtrans_reg(u32 domain)
+{
+	return get_domain_base(domain) + (PWR_L2_PWRTRANS - PWR_L2_PRESENT);
+}
+
+static bool is_valid_domain(u32 domain)
+{
+	return get_domain_base(domain) != 0;
+}
+
+static bool has_rtu(struct panthor_device *ptdev)
+{
+	return ptdev->gpu_info.gpu_features & GPU_FEATURES_RAY_TRAVERSAL;
+}
+
+static u8 get_domain_subdomain(struct panthor_device *ptdev, u32 domain)
+{
+	if ((domain == PWR_COMMAND_DOMAIN_SHADER) && has_rtu(ptdev))
+		return PWR_COMMAND_SUBDOMAIN_RTU;
+
+	return 0;
+}
+
+static int panthor_pwr_domain_wait_transition(struct panthor_device *ptdev, u32 domain,
+					      u32 timeout_us)
+{
+	u32 pwrtrans_reg = get_domain_pwrtrans_reg(domain);
+	u64 val;
+	int ret = 0;
+
+	ret = gpu_read64_poll_timeout(ptdev, pwrtrans_reg, val, !(PWR_ALL_CORES_MASK & val), 100,
+				      timeout_us);
+	if (ret) {
+		drm_err(&ptdev->base, "%s domain power in transition, pwrtrans(0x%llx)",
+			get_domain_name(domain), val);
+		return ret;
+	}
+
+	return 0;
+}
+
+static void panthor_pwr_debug_info_show(struct panthor_device *ptdev)
+{
+	drm_info(&ptdev->base, "GPU_FEATURES:    0x%016llx", gpu_read64(ptdev, GPU_FEATURES));
+	drm_info(&ptdev->base, "PWR_STATUS:      0x%016llx", gpu_read64(ptdev, PWR_STATUS));
+	drm_info(&ptdev->base, "L2_PRESENT:      0x%016llx", gpu_read64(ptdev, PWR_L2_PRESENT));
+	drm_info(&ptdev->base, "L2_PWRTRANS:     0x%016llx", gpu_read64(ptdev, PWR_L2_PWRTRANS));
+	drm_info(&ptdev->base, "L2_READY:        0x%016llx", gpu_read64(ptdev, PWR_L2_READY));
+	drm_info(&ptdev->base, "TILER_PRESENT:   0x%016llx", gpu_read64(ptdev, PWR_TILER_PRESENT));
+	drm_info(&ptdev->base, "TILER_PWRTRANS:  0x%016llx", gpu_read64(ptdev, PWR_TILER_PWRTRANS));
+	drm_info(&ptdev->base, "TILER_READY:     0x%016llx", gpu_read64(ptdev, PWR_TILER_READY));
+	drm_info(&ptdev->base, "SHADER_PRESENT:  0x%016llx", gpu_read64(ptdev, PWR_SHADER_PRESENT));
+	drm_info(&ptdev->base, "SHADER_PWRTRANS: 0x%016llx", gpu_read64(ptdev, PWR_SHADER_PWRTRANS));
+	drm_info(&ptdev->base, "SHADER_READY:    0x%016llx", gpu_read64(ptdev, PWR_SHADER_READY));
+}
+
+static int panthor_pwr_domain_transition(struct panthor_device *ptdev, u32 cmd, u32 domain,
+					 u64 mask, u32 timeout_us)
+{
+	u32 ready_reg = get_domain_ready_reg(domain);
+	u32 pwr_cmd = PWR_COMMAND_DEF(cmd, domain, get_domain_subdomain(ptdev, domain));
+	u64 expected_val = 0;
+	u64 val;
+	int ret = 0;
+
+	if (drm_WARN_ON(&ptdev->base, !is_valid_domain(domain)))
+		return -EINVAL;
+
+	switch (cmd) {
+	case PWR_COMMAND_POWER_DOWN:
+		expected_val = 0;
+		break;
+	case PWR_COMMAND_POWER_UP:
+		expected_val = mask;
+		break;
+	default:
+		drm_err(&ptdev->base, "Invalid power domain transition command (0x%x)", cmd);
+		return -EINVAL;
+	}
+
+	ret = panthor_pwr_domain_wait_transition(ptdev, domain, timeout_us);
+	if (ret)
+		return ret;
+
+	/* domain already in target state, return early */
+	if ((gpu_read64(ptdev, ready_reg) & mask) == expected_val)
+		return 0;
+
+	panthor_pwr_write_command(ptdev, pwr_cmd, mask);
+
+	ret = gpu_read64_poll_timeout(ptdev, ready_reg, val, (mask & val) == expected_val, 100,
+				      timeout_us);
+	if (ret) {
+		drm_err(&ptdev->base,
+			"timeout waiting on %s power domain transition, cmd(0x%x), arg(0x%llx)",
+			get_domain_name(domain), pwr_cmd, mask);
+		panthor_pwr_debug_info_show(ptdev);
+		return ret;
+	}
+
+	return 0;
+}
+
+#define panthor_pwr_domain_power_off(__ptdev, __domain, __mask, __timeout_us)            \
+	panthor_pwr_domain_transition(__ptdev, PWR_COMMAND_POWER_DOWN, __domain, __mask, \
+				      __timeout_us)
+
+#define panthor_pwr_domain_power_on(__ptdev, __domain, __mask, __timeout_us) \
+	panthor_pwr_domain_transition(__ptdev, PWR_COMMAND_POWER_UP, __domain, __mask, __timeout_us)
+
+/**
+ * retract_domain() - Retract control of a domain from MCU
+ * @ptdev: Device.
+ * @domain: Domain to retract the control
+ *
+ * Retracting L2 domain is not expected since it won't be delegated.
+ *
+ * Return: 0 on success or retracted already.
+ *         -EPERM if domain is L2.
+ *         A negative error code otherwise.
+ */
+static int retract_domain(struct panthor_device *ptdev, u32 domain)
+{
+	const u32 pwr_cmd = PWR_COMMAND_DEF(PWR_COMMAND_RETRACT, domain, 0);
+	const u64 pwr_status = gpu_read64(ptdev, PWR_STATUS);
+	const u64 delegated_mask = PWR_STATUS_DOMAIN_DELEGATED(domain);
+	const u64 allow_mask = PWR_STATUS_DOMAIN_ALLOWED(domain);
+	u64 val;
+	int ret;
+
+	if (drm_WARN_ON(&ptdev->base, domain == PWR_COMMAND_DOMAIN_L2))
+		return -EPERM;
+
+	ret = gpu_read64_poll_timeout(ptdev, PWR_STATUS, val, !(PWR_STATUS_RETRACT_PENDING & val),
+				      0, PWR_RETRACT_TIMEOUT_US);
+	if (ret) {
+		drm_err(&ptdev->base, "%s domain retract pending", get_domain_name(domain));
+		return ret;
+	}
+
+	if (!(pwr_status & delegated_mask)) {
+		drm_dbg(&ptdev->base, "%s domain already retracted", get_domain_name(domain));
+		return 0;
+	}
+
+	panthor_pwr_write_command(ptdev, pwr_cmd, 0);
+
+	/*
+	 * On successful retraction
+	 * allow-flag will be set with delegated-flag being cleared.
+	 */
+	ret = gpu_read64_poll_timeout(ptdev, PWR_STATUS, val,
+				      ((delegated_mask | allow_mask) & val) == allow_mask, 10,
+				      PWR_TRANSITION_TIMEOUT_US);
+	if (ret) {
+		drm_err(&ptdev->base, "Retracting %s domain timeout, cmd(0x%x)",
+			get_domain_name(domain), pwr_cmd);
+		return ret;
+	}
+
+	return 0;
+}
+
+/**
+ * delegate_domain() - Delegate control of a domain to MCU
+ * @ptdev: Device.
+ * @domain: Domain to delegate the control
+ *
+ * Delegating L2 domain is prohibited.
+ *
+ * Return:
+ * *       0 on success or delegated already.
+ * *       -EPERM if domain is L2.
+ * *       A negative error code otherwise.
+ */
+static int delegate_domain(struct panthor_device *ptdev, u32 domain)
+{
+	const u32 pwr_cmd = PWR_COMMAND_DEF(PWR_COMMAND_DELEGATE, domain, 0);
+	const u64 pwr_status = gpu_read64(ptdev, PWR_STATUS);
+	const u64 allow_mask = PWR_STATUS_DOMAIN_ALLOWED(domain);
+	const u64 delegated_mask = PWR_STATUS_DOMAIN_DELEGATED(domain);
+	u64 val;
+	int ret;
+
+	if (drm_WARN_ON(&ptdev->base, domain == PWR_COMMAND_DOMAIN_L2))
+		return -EPERM;
+
+	/* Already delegated, exit early */
+	if (pwr_status & delegated_mask)
+		return 0;
+
+	/* Check if the command is allowed before delegating. */
+	if (!(pwr_status & allow_mask)) {
+		drm_warn(&ptdev->base, "Delegating %s domain not allowed", get_domain_name(domain));
+		return -EPERM;
+	}
+
+	ret = panthor_pwr_domain_wait_transition(ptdev, domain, PWR_TRANSITION_TIMEOUT_US);
+	if (ret)
+		return ret;
+
+	panthor_pwr_write_command(ptdev, pwr_cmd, 0);
+
+	/*
+	 * On successful delegation
+	 * allow-flag will be cleared with delegated-flag being set.
+	 */
+	ret = gpu_read64_poll_timeout(ptdev, PWR_STATUS, val,
+				      ((delegated_mask | allow_mask) & val) == delegated_mask,
+				      10, PWR_TRANSITION_TIMEOUT_US);
+	if (ret) {
+		drm_err(&ptdev->base, "Delegating %s domain timeout, cmd(0x%x)",
+			get_domain_name(domain), pwr_cmd);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int panthor_pwr_delegate_domains(struct panthor_device *ptdev)
+{
+	int ret;
+
+	if (!ptdev->pwr)
+		return 0;
+
+	ret = delegate_domain(ptdev, PWR_COMMAND_DOMAIN_SHADER);
+	if (ret)
+		return ret;
+
+	ret = delegate_domain(ptdev, PWR_COMMAND_DOMAIN_TILER);
+	if (ret)
+		goto err_retract_shader;
+
+	return 0;
+
+err_retract_shader:
+	retract_domain(ptdev, PWR_COMMAND_DOMAIN_SHADER);
+
+	return ret;
+}
+
+/**
+ * panthor_pwr_domain_force_off - Forcefully power down a domain.
+ * @ptdev: Device.
+ * @domain: Domain to forcefully power down.
+ *
+ * This function will attempt to retract and power off the requested power
+ * domain. However, if retraction fails, the operation is aborted. If power off
+ * fails, the domain will remain retracted and under the host control.
+ *
+ * Return: 0 on success or a negative error code on failure.
+ */
+static int panthor_pwr_domain_force_off(struct panthor_device *ptdev, u32 domain)
+{
+	const u64 domain_ready = gpu_read64(ptdev, get_domain_ready_reg(domain));
+	int ret;
+
+	/* Domain already powered down, early exit. */
+	if (!domain_ready)
+		return 0;
+
+	/* Domain has to be in host control to issue power off command. */
+	ret = retract_domain(ptdev, domain);
+	if (ret)
+		return ret;
+
+	return panthor_pwr_domain_power_off(ptdev, domain, domain_ready, PWR_TRANSITION_TIMEOUT_US);
+}
+
 void panthor_pwr_unplug(struct panthor_device *ptdev)
 {
 	unsigned long flags;
@@ -103,6 +428,59 @@ int panthor_pwr_init(struct panthor_device *ptdev)
 	return 0;
 }

+void panthor_pwr_l2_power_off(struct panthor_device *ptdev)
+{
+	const u64 l2_allow_mask = PWR_STATUS_DOMAIN_ALLOWED(PWR_COMMAND_DOMAIN_L2);
+	const u64 pwr_status = gpu_read64(ptdev, PWR_STATUS);
+
+	/* Abort if L2 power off constraints are not satisfied */
+	if (!(pwr_status & l2_allow_mask)) {
+		drm_warn(&ptdev->base, "Power off L2 domain not allowed");
+		return;
+	}
+
+	/* It is expected that when halting the MCU, it would power down its
+	 * delegated domains. However, an unresponsive or hung MCU may not do
+	 * so, which is why we need to check and retract the domains back into
+	 * host control to be powered down in the right order before powering
+	 * down the L2.
+	 */
+	if (panthor_pwr_domain_force_off(ptdev, PWR_COMMAND_DOMAIN_TILER))
+		return;
+
+	if (panthor_pwr_domain_force_off(ptdev, PWR_COMMAND_DOMAIN_SHADER))
+		return;
+
+	panthor_pwr_domain_power_off(ptdev, PWR_COMMAND_DOMAIN_L2, ptdev->gpu_info.l2_present,
+				     PWR_TRANSITION_TIMEOUT_US);
+}
+
+int panthor_pwr_l2_power_on(struct panthor_device *ptdev)
+{
+	const u32 pwr_status = gpu_read64(ptdev, PWR_STATUS);
+	const u32 l2_allow_mask = PWR_STATUS_DOMAIN_ALLOWED(PWR_COMMAND_DOMAIN_L2);
+	int ret;
+
+	if ((pwr_status & l2_allow_mask) == 0) {
+		drm_warn(&ptdev->base, "Power on L2 domain not allowed");
+		return -EPERM;
+	}
+
+	ret = panthor_pwr_domain_power_on(ptdev, PWR_COMMAND_DOMAIN_L2, ptdev->gpu_info.l2_present,
+					  PWR_TRANSITION_TIMEOUT_US);
+	if (ret)
+		return ret;
+
+	/* Delegate control of the shader and tiler power domains to the MCU as
+	 * it can better manage which shader/tiler cores need to be powered up
+	 * or can be powered down based on currently running jobs.
+	 *
+	 * If the shader and tiler domains are already delegated to the MCU,
+	 * this call would just return early.
+	 */
+	return panthor_pwr_delegate_domains(ptdev);
+}
+
 void panthor_pwr_suspend(struct panthor_device *ptdev)
 {
 	if (!ptdev->pwr)
diff --git a/drivers/gpu/drm/panthor/panthor_pwr.h b/drivers/gpu/drm/panthor/panthor_pwr.h
index b325e5b7eba3..3c834059a860 100644
--- a/drivers/gpu/drm/panthor/panthor_pwr.h
+++ b/drivers/gpu/drm/panthor/panthor_pwr.h
@@ -10,6 +10,10 @@ void panthor_pwr_unplug(struct panthor_device *ptdev);

 int panthor_pwr_init(struct panthor_device *ptdev);

+void panthor_pwr_l2_power_off(struct panthor_device *ptdev);
+
+int panthor_pwr_l2_power_on(struct panthor_device *ptdev);
+
 void panthor_pwr_suspend(struct panthor_device *ptdev);

 void panthor_pwr_resume(struct panthor_device *ptdev);
diff --git a/drivers/gpu/drm/panthor/panthor_regs.h b/drivers/gpu/drm/panthor/panthor_regs.h
index 5469eec02178..18702d7001e2 100644
--- a/drivers/gpu/drm/panthor/panthor_regs.h
+++ b/drivers/gpu/drm/panthor/panthor_regs.h
@@ -72,6 +72,7 @@

 #define GPU_FEATURES					0x60
 #define   GPU_FEATURES_RAY_INTERSECTION			BIT(2)
+#define   GPU_FEATURES_RAY_TRAVERSAL			BIT(5)

 #define GPU_TIMESTAMP_OFFSET				0x88
 #define GPU_CYCLE_COUNT					0x90
--
2.49.0

Re: [PATCH v2 4/8] drm/panthor: Implement L2 power on/off via PWR_CONTROL

[Erik Faye-Lund]

On Fri, 2025-10-24 at 21:21 +0100, Karunika Choo wrote:
> This patch adds common helpers to issue power commands, poll
> transitions, and validate domain state, then wires them into the L2
> on/off paths.
> 
> The L2 power-on sequence now delegates control of the SHADER and
> TILER
> domains to the MCU when allowed, while the L2 itself is never
> delegated.
> On power-off, dependent domains beneath the L2 are checked, and if
> necessary, retracted and powered down to maintain proper domain
> ordering.
> 
> Signed-off-by: Karunika Choo <karunika.choo@arm.com>
> ---
> v2:
>  * Updated GENMASK to GENMASK_U64 to address kernel test robot
> warnings
>    for 32-bit systems.
>  * Removed panthor_pwr_read_status() in favour of a simple
> gpu_read64()
>    operation on the PWR_STATUS register.
>  * Renamed panthor_pwr_info_show() to panthor_pwr_debug_info_show()
> for
>    more clarity.
>  * Added additional WARN_ON for an invalid domain when requesting
> power
>    domain transition.
>  * Made panthor_pwr_domain_transition()'s expected val logic more
>    readable and clearer.
>  * Wait on domain power transition instead of failing the operation.
>  * Fixed inconsistent error return value vs kerneldoc.
>  * Removed confusing drm_dbg in delegate_domain() in favor of a
> comment.
>  * Add unwind to panthor_pwr_delegate_domains().
>  * Moved child domain handling logic from panthor_pwr_l2_power_off()
>    into panthor_pwr_domain_force_off().
>  * Added additional clarification regarding delegation and retraction
> of
>    power domains.
>  * Minor formatting and readability changes and remove unnecessary
>    checks.
> ---
>  drivers/gpu/drm/panthor/panthor_pwr.c  | 378
> +++++++++++++++++++++++++
>  drivers/gpu/drm/panthor/panthor_pwr.h  |   4 +
>  drivers/gpu/drm/panthor/panthor_regs.h |   1 +
>  3 files changed, 383 insertions(+)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_pwr.c
> b/drivers/gpu/drm/panthor/panthor_pwr.c
> index da64fe006a8b..cd529660a276 100644
> --- a/drivers/gpu/drm/panthor/panthor_pwr.c
> +++ b/drivers/gpu/drm/panthor/panthor_pwr.c
> @@ -23,6 +23,14 @@
>  	 PWR_IRQ_COMMAND_NOT_ALLOWED | \
>  	 PWR_IRQ_COMMAND_INVALID)
> 
> +#define PWR_ALL_CORES_MASK		GENMASK_U64(63, 0)
> +
> +#define PWR_DOMAIN_MAX_BITS		16
> +
> +#define PWR_TRANSITION_TIMEOUT_US	(2ULL * USEC_PER_SEC)
> +
> +#define PWR_RETRACT_TIMEOUT_US		(2ULL * USEC_PER_MSEC)
> +
>  /**
>   * struct panthor_pwr - PWR_CONTROL block management data.
>   */
> @@ -59,6 +67,323 @@ static void panthor_pwr_irq_handler(struct
> panthor_device *ptdev, u32 status)
>  }
>  PANTHOR_IRQ_HANDLER(pwr, PWR, panthor_pwr_irq_handler);
> 
> +static void panthor_pwr_write_command(struct panthor_device *ptdev,
> u32 command, u64 args)
> +{
> +	if (args)
> +		gpu_write64(ptdev, PWR_CMDARG, args);
> +
> +	gpu_write(ptdev, PWR_COMMAND, command);
> +}
> +
> +static const char *get_domain_name(u8 domain)
> +{
> +	switch (domain) {
> +	case PWR_COMMAND_DOMAIN_L2:
> +		return "L2";
> +	case PWR_COMMAND_DOMAIN_TILER:
> +		return "Tiler";
> +	case PWR_COMMAND_DOMAIN_SHADER:
> +		return "Shader";
> +	case PWR_COMMAND_DOMAIN_BASE:
> +		return "Base";
> +	case PWR_COMMAND_DOMAIN_STACK:
> +		return "Stack";
> +	}
> +	return "Unknown";
> +}
> +
> +static u32 get_domain_base(u8 domain)
> +{
> +	switch (domain) {
> +	case PWR_COMMAND_DOMAIN_L2:
> +		return PWR_L2_PRESENT;
> +	case PWR_COMMAND_DOMAIN_TILER:
> +		return PWR_TILER_PRESENT;
> +	case PWR_COMMAND_DOMAIN_SHADER:
> +		return PWR_SHADER_PRESENT;
> +	case PWR_COMMAND_DOMAIN_BASE:
> +		return PWR_BASE_PRESENT;
> +	case PWR_COMMAND_DOMAIN_STACK:
> +		return PWR_STACK_PRESENT;
> +	}
> +	return 0;
> +}
> +
> +static u32 get_domain_ready_reg(u32 domain)
> +{
> +	return get_domain_base(domain) + (PWR_L2_READY -
> PWR_L2_PRESENT);
> +}
> +
> +static u32 get_domain_pwrtrans_reg(u32 domain)
> +{
> +	return get_domain_base(domain) + (PWR_L2_PWRTRANS -
> PWR_L2_PRESENT);
> +}
> +
> +static bool is_valid_domain(u32 domain)
> +{
> +	return get_domain_base(domain) != 0;
> +}
> +
> +static bool has_rtu(struct panthor_device *ptdev)
> +{
> +	return ptdev->gpu_info.gpu_features &
> GPU_FEATURES_RAY_TRAVERSAL;
> +}
> +
> +static u8 get_domain_subdomain(struct panthor_device *ptdev, u32
> domain)
> +{
> +	if ((domain == PWR_COMMAND_DOMAIN_SHADER) && has_rtu(ptdev))
> +		return PWR_COMMAND_SUBDOMAIN_RTU;
> +
> +	return 0;
> +}
> +
> +static int panthor_pwr_domain_wait_transition(struct panthor_device
> *ptdev, u32 domain,
> +					      u32 timeout_us)
> +{
> +	u32 pwrtrans_reg = get_domain_pwrtrans_reg(domain);
> +	u64 val;
> +	int ret = 0;
> +
> +	ret = gpu_read64_poll_timeout(ptdev, pwrtrans_reg, val,
> !(PWR_ALL_CORES_MASK & val), 100,
> +				      timeout_us);
> +	if (ret) {
> +		drm_err(&ptdev->base, "%s domain power in
> transition, pwrtrans(0x%llx)",
> +			get_domain_name(domain), val);
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +static void panthor_pwr_debug_info_show(struct panthor_device
> *ptdev)
> +{
> +	drm_info(&ptdev->base, "GPU_FEATURES:    0x%016llx",
> gpu_read64(ptdev, GPU_FEATURES));
> +	drm_info(&ptdev->base, "PWR_STATUS:      0x%016llx",
> gpu_read64(ptdev, PWR_STATUS));
> +	drm_info(&ptdev->base, "L2_PRESENT:      0x%016llx",
> gpu_read64(ptdev, PWR_L2_PRESENT));
> +	drm_info(&ptdev->base, "L2_PWRTRANS:     0x%016llx",
> gpu_read64(ptdev, PWR_L2_PWRTRANS));
> +	drm_info(&ptdev->base, "L2_READY:        0x%016llx",
> gpu_read64(ptdev, PWR_L2_READY));
> +	drm_info(&ptdev->base, "TILER_PRESENT:   0x%016llx",
> gpu_read64(ptdev, PWR_TILER_PRESENT));
> +	drm_info(&ptdev->base, "TILER_PWRTRANS:  0x%016llx",
> gpu_read64(ptdev, PWR_TILER_PWRTRANS));
> +	drm_info(&ptdev->base, "TILER_READY:     0x%016llx",
> gpu_read64(ptdev, PWR_TILER_READY));
> +	drm_info(&ptdev->base, "SHADER_PRESENT:  0x%016llx",
> gpu_read64(ptdev, PWR_SHADER_PRESENT));
> +	drm_info(&ptdev->base, "SHADER_PWRTRANS: 0x%016llx",
> gpu_read64(ptdev, PWR_SHADER_PWRTRANS));
> +	drm_info(&ptdev->base, "SHADER_READY:    0x%016llx",
> gpu_read64(ptdev, PWR_SHADER_READY));
> +}
> +
> +static int panthor_pwr_domain_transition(struct panthor_device
> *ptdev, u32 cmd, u32 domain,
> +					 u64 mask, u32 timeout_us)
> +{
> +	u32 ready_reg = get_domain_ready_reg(domain);
> +	u32 pwr_cmd = PWR_COMMAND_DEF(cmd, domain,
> get_domain_subdomain(ptdev, domain));
> +	u64 expected_val = 0;
> +	u64 val;
> +	int ret = 0;
> +
> +	if (drm_WARN_ON(&ptdev->base, !is_valid_domain(domain)))
> +		return -EINVAL;
> +
> +	switch (cmd) {
> +	case PWR_COMMAND_POWER_DOWN:
> +		expected_val = 0;
> +		break;
> +	case PWR_COMMAND_POWER_UP:
> +		expected_val = mask;
> +		break;
> +	default:
> +		drm_err(&ptdev->base, "Invalid power domain
> transition command (0x%x)", cmd);
> +		return -EINVAL;
> +	}
> +
> +	ret = panthor_pwr_domain_wait_transition(ptdev, domain,
> timeout_us);
> +	if (ret)
> +		return ret;
> +
> +	/* domain already in target state, return early */
> +	if ((gpu_read64(ptdev, ready_reg) & mask) == expected_val)
> +		return 0;
> +
> +	panthor_pwr_write_command(ptdev, pwr_cmd, mask);
> +
> +	ret = gpu_read64_poll_timeout(ptdev, ready_reg, val, (mask &
> val) == expected_val, 100,
> +				      timeout_us);
> +	if (ret) {
> +		drm_err(&ptdev->base,
> +			"timeout waiting on %s power domain
> transition, cmd(0x%x), arg(0x%llx)",
> +			get_domain_name(domain), pwr_cmd, mask);
> +		panthor_pwr_debug_info_show(ptdev);
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +#define panthor_pwr_domain_power_off(__ptdev, __domain, __mask,
> __timeout_us)            \
> +	panthor_pwr_domain_transition(__ptdev,
> PWR_COMMAND_POWER_DOWN, __domain, __mask, \
> +				      __timeout_us)
> +
> +#define panthor_pwr_domain_power_on(__ptdev, __domain, __mask,
> __timeout_us) \
> +	panthor_pwr_domain_transition(__ptdev, PWR_COMMAND_POWER_UP,
> __domain, __mask, __timeout_us)
> +
> +/**
> + * retract_domain() - Retract control of a domain from MCU
> + * @ptdev: Device.
> + * @domain: Domain to retract the control
> + *
> + * Retracting L2 domain is not expected since it won't be delegated.
> + *
> + * Return: 0 on success or retracted already.
> + *         -EPERM if domain is L2.
> + *         A negative error code otherwise.
> + */
> +static int retract_domain(struct panthor_device *ptdev, u32 domain)
> +{
> +	const u32 pwr_cmd = PWR_COMMAND_DEF(PWR_COMMAND_RETRACT,
> domain, 0);
> +	const u64 pwr_status = gpu_read64(ptdev, PWR_STATUS);
> +	const u64 delegated_mask =
> PWR_STATUS_DOMAIN_DELEGATED(domain);
> +	const u64 allow_mask = PWR_STATUS_DOMAIN_ALLOWED(domain);
> +	u64 val;
> +	int ret;
> +
> +	if (drm_WARN_ON(&ptdev->base, domain ==
> PWR_COMMAND_DOMAIN_L2))
> +		return -EPERM;
> +
> +	ret = gpu_read64_poll_timeout(ptdev, PWR_STATUS, val,
> !(PWR_STATUS_RETRACT_PENDING & val),
> +				      0, PWR_RETRACT_TIMEOUT_US);
> +	if (ret) {
> +		drm_err(&ptdev->base, "%s domain retract pending",
> get_domain_name(domain));
> +		return ret;
> +	}
> +
> +	if (!(pwr_status & delegated_mask)) {
> +		drm_dbg(&ptdev->base, "%s domain already retracted",
> get_domain_name(domain));
> +		return 0;
> +	}
> +
> +	panthor_pwr_write_command(ptdev, pwr_cmd, 0);
> +
> +	/*
> +	 * On successful retraction
> +	 * allow-flag will be set with delegated-flag being cleared.
> +	 */
> +	ret = gpu_read64_poll_timeout(ptdev, PWR_STATUS, val,
> +				      ((delegated_mask | allow_mask)
> & val) == allow_mask, 10,
> +				      PWR_TRANSITION_TIMEOUT_US);
> +	if (ret) {
> +		drm_err(&ptdev->base, "Retracting %s domain timeout,
> cmd(0x%x)",
> +			get_domain_name(domain), pwr_cmd);
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +/**
> + * delegate_domain() - Delegate control of a domain to MCU
> + * @ptdev: Device.
> + * @domain: Domain to delegate the control
> + *
> + * Delegating L2 domain is prohibited.
> + *
> + * Return:
> + * *       0 on success or delegated already.
> + * *       -EPERM if domain is L2.
> + * *       A negative error code otherwise.
> + */
> +static int delegate_domain(struct panthor_device *ptdev, u32 domain)
> +{
> +	const u32 pwr_cmd = PWR_COMMAND_DEF(PWR_COMMAND_DELEGATE,
> domain, 0);
> +	const u64 pwr_status = gpu_read64(ptdev, PWR_STATUS);
> +	const u64 allow_mask = PWR_STATUS_DOMAIN_ALLOWED(domain);
> +	const u64 delegated_mask =
> PWR_STATUS_DOMAIN_DELEGATED(domain);
> +	u64 val;
> +	int ret;
> +
> +	if (drm_WARN_ON(&ptdev->base, domain ==
> PWR_COMMAND_DOMAIN_L2))
> +		return -EPERM;
> +
> +	/* Already delegated, exit early */
> +	if (pwr_status & delegated_mask)
> +		return 0;
> +
> +	/* Check if the command is allowed before delegating. */
> +	if (!(pwr_status & allow_mask)) {
> +		drm_warn(&ptdev->base, "Delegating %s domain not
> allowed", get_domain_name(domain));
> +		return -EPERM;
> +	}
> +
> +	ret = panthor_pwr_domain_wait_transition(ptdev, domain,
> PWR_TRANSITION_TIMEOUT_US);
> +	if (ret)
> +		return ret;
> +
> +	panthor_pwr_write_command(ptdev, pwr_cmd, 0);
> +
> +	/*
> +	 * On successful delegation
> +	 * allow-flag will be cleared with delegated-flag being set.
> +	 */
> +	ret = gpu_read64_poll_timeout(ptdev, PWR_STATUS, val,
> +				      ((delegated_mask | allow_mask)
> & val) == delegated_mask,
> +				      10,
> PWR_TRANSITION_TIMEOUT_US);
> +	if (ret) {
> +		drm_err(&ptdev->base, "Delegating %s domain timeout,
> cmd(0x%x)",
> +			get_domain_name(domain), pwr_cmd);
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +static int panthor_pwr_delegate_domains(struct panthor_device
> *ptdev)
> +{
> +	int ret;
> +
> +	if (!ptdev->pwr)
> +		return 0;
> +
> +	ret = delegate_domain(ptdev, PWR_COMMAND_DOMAIN_SHADER);
> +	if (ret)
> +		return ret;
> +
> +	ret = delegate_domain(ptdev, PWR_COMMAND_DOMAIN_TILER);
> +	if (ret)
> +		goto err_retract_shader;
> +
> +	return 0;
> +
> +err_retract_shader:
> +	retract_domain(ptdev, PWR_COMMAND_DOMAIN_SHADER);
> +
> +	return ret;
> +}
> +
> +/**
> + * panthor_pwr_domain_force_off - Forcefully power down a domain.
> + * @ptdev: Device.
> + * @domain: Domain to forcefully power down.
> + *
> + * This function will attempt to retract and power off the requested
> power
> + * domain. However, if retraction fails, the operation is aborted.
> If power off
> + * fails, the domain will remain retracted and under the host
> control.
> + *
> + * Return: 0 on success or a negative error code on failure.
> + */
> +static int panthor_pwr_domain_force_off(struct panthor_device
> *ptdev, u32 domain)
> +{
> +	const u64 domain_ready = gpu_read64(ptdev,
> get_domain_ready_reg(domain));
> +	int ret;
> +
> +	/* Domain already powered down, early exit. */
> +	if (!domain_ready)
> +		return 0;
> +
> +	/* Domain has to be in host control to issue power off
> command. */
> +	ret = retract_domain(ptdev, domain);
> +	if (ret)
> +		return ret;
> +
> +	return panthor_pwr_domain_power_off(ptdev, domain,
> domain_ready, PWR_TRANSITION_TIMEOUT_US);
> +}
> +
>  void panthor_pwr_unplug(struct panthor_device *ptdev)
>  {
>  	unsigned long flags;
> @@ -103,6 +428,59 @@ int panthor_pwr_init(struct panthor_device
> *ptdev)
>  	return 0;
>  }
> 
> +void panthor_pwr_l2_power_off(struct panthor_device *ptdev)
> +{
> +	const u64 l2_allow_mask =
> PWR_STATUS_DOMAIN_ALLOWED(PWR_COMMAND_DOMAIN_L2);
> +	const u64 pwr_status = gpu_read64(ptdev, PWR_STATUS);
> +
> +	/* Abort if L2 power off constraints are not satisfied */
> +	if (!(pwr_status & l2_allow_mask)) {
> +		drm_warn(&ptdev->base, "Power off L2 domain not
> allowed");
> +		return;
> +	}
> +
> +	/* It is expected that when halting the MCU, it would power
> down its
> +	 * delegated domains. However, an unresponsive or hung MCU
> may not do
> +	 * so, which is why we need to check and retract the domains
> back into
> +	 * host control to be powered down in the right order before
> powering
> +	 * down the L2.
> +	 */
> +	if (panthor_pwr_domain_force_off(ptdev,
> PWR_COMMAND_DOMAIN_TILER))
> +		return;
> +
> +	if (panthor_pwr_domain_force_off(ptdev,
> PWR_COMMAND_DOMAIN_SHADER))
> +		return;
> +
> +	panthor_pwr_domain_power_off(ptdev, PWR_COMMAND_DOMAIN_L2,
> ptdev->gpu_info.l2_present,
> +				     PWR_TRANSITION_TIMEOUT_US);
> +}
> +
> +int panthor_pwr_l2_power_on(struct panthor_device *ptdev)
> +{
> +	const u32 pwr_status = gpu_read64(ptdev, PWR_STATUS);
> +	const u32 l2_allow_mask =
> PWR_STATUS_DOMAIN_ALLOWED(PWR_COMMAND_DOMAIN_L2);
> +	int ret;
> +
> +	if ((pwr_status & l2_allow_mask) == 0) {
> +		drm_warn(&ptdev->base, "Power on L2 domain not
> allowed");
> +		return -EPERM;
> +	}
> +
> +	ret = panthor_pwr_domain_power_on(ptdev,
> PWR_COMMAND_DOMAIN_L2, ptdev->gpu_info.l2_present,
> +					 
> PWR_TRANSITION_TIMEOUT_US);
> +	if (ret)
> +		return ret;
> +
> +	/* Delegate control of the shader and tiler power domains to
> the MCU as
> +	 * it can better manage which shader/tiler cores need to be
> powered up
> +	 * or can be powered down based on currently running jobs.
> +	 *
> +	 * If the shader and tiler domains are already delegated to
> the MCU,
> +	 * this call would just return early.
> +	 */
> +	return panthor_pwr_delegate_domains(ptdev);
> +}
> +
>  void panthor_pwr_suspend(struct panthor_device *ptdev)
>  {
>  	if (!ptdev->pwr)
> diff --git a/drivers/gpu/drm/panthor/panthor_pwr.h
> b/drivers/gpu/drm/panthor/panthor_pwr.h
> index b325e5b7eba3..3c834059a860 100644
> --- a/drivers/gpu/drm/panthor/panthor_pwr.h
> +++ b/drivers/gpu/drm/panthor/panthor_pwr.h
> @@ -10,6 +10,10 @@ void panthor_pwr_unplug(struct panthor_device
> *ptdev);
> 
>  int panthor_pwr_init(struct panthor_device *ptdev);
> 
> +void panthor_pwr_l2_power_off(struct panthor_device *ptdev);
> +
> +int panthor_pwr_l2_power_on(struct panthor_device *ptdev);
> +
>  void panthor_pwr_suspend(struct panthor_device *ptdev);
> 
>  void panthor_pwr_resume(struct panthor_device *ptdev);
> diff --git a/drivers/gpu/drm/panthor/panthor_regs.h
> b/drivers/gpu/drm/panthor/panthor_regs.h
> index 5469eec02178..18702d7001e2 100644
> --- a/drivers/gpu/drm/panthor/panthor_regs.h
> +++ b/drivers/gpu/drm/panthor/panthor_regs.h
> @@ -72,6 +72,7 @@
> 
>  #define GPU_FEATURES					0x60
>  #define  
> GPU_FEATURES_RAY_INTERSECTION			BIT(2)
> +#define   GPU_FEATURES_RAY_TRAVERSAL			BIT(5)

This seems out of place in this commit, which is about power stuff...

[PATCH v2 5/8] drm/panthor: Implement soft reset via PWR_CONTROL

[Karunika Choo]

Add helpers to issue reset commands through the PWR_CONTROL interface
and wait for reset completion using IRQ signaling. This enables support
for RESET_SOFT operations with timeout handling and status verification.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Dropped RESET_FAST implementation as it is not currently being used.
 * Renamed reset_completed to reset_pending to align with underlying
   logic and fixed the logic of its callers accordingly.
 * Improved readability of panthor_pwr_reset() and removed inline
   ternary expressions.
---
 drivers/gpu/drm/panthor/panthor_pwr.c | 50 +++++++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_pwr.h |  2 ++
 2 files changed, 52 insertions(+)

diff --git a/drivers/gpu/drm/panthor/panthor_pwr.c b/drivers/gpu/drm/panthor/panthor_pwr.c
index cd529660a276..4edb818c7ac4 100644
--- a/drivers/gpu/drm/panthor/panthor_pwr.c
+++ b/drivers/gpu/drm/panthor/panthor_pwr.c
@@ -3,6 +3,7 @@

 #include <linux/platform_device.h>
 #include <linux/interrupt.h>
+#include <linux/cleanup.h>
 #include <linux/iopoll.h>
 #include <linux/wait.h>

@@ -31,6 +32,8 @@

 #define PWR_RETRACT_TIMEOUT_US		(2ULL * USEC_PER_MSEC)

+#define PWR_RESET_TIMEOUT_MS		500
+
 /**
  * struct panthor_pwr - PWR_CONTROL block management data.
  */
@@ -75,6 +78,43 @@ static void panthor_pwr_write_command(struct panthor_device *ptdev, u32 command,
 	gpu_write(ptdev, PWR_COMMAND, command);
 }

+static bool reset_irq_raised(struct panthor_device *ptdev)
+{
+	return gpu_read(ptdev, PWR_INT_RAWSTAT) & PWR_IRQ_RESET_COMPLETED;
+}
+
+static bool reset_pending(struct panthor_device *ptdev)
+{
+	return (ptdev->pwr->pending_reqs & PWR_IRQ_RESET_COMPLETED);
+}
+
+static int panthor_pwr_reset(struct panthor_device *ptdev, u32 reset_cmd)
+{
+	scoped_guard(spinlock_irqsave, &ptdev->pwr->reqs_lock) {
+		if (reset_pending(ptdev)) {
+			drm_WARN(&ptdev->base, 1, "Reset already pending");
+		} else {
+			ptdev->pwr->pending_reqs |= PWR_IRQ_RESET_COMPLETED;
+			gpu_write(ptdev, PWR_INT_CLEAR, PWR_IRQ_RESET_COMPLETED);
+			panthor_pwr_write_command(ptdev, reset_cmd, 0);
+		}
+	}
+
+	if (!wait_event_timeout(ptdev->pwr->reqs_acked, !reset_pending(ptdev),
+				msecs_to_jiffies(PWR_RESET_TIMEOUT_MS))) {
+		guard(spinlock_irqsave)(&ptdev->pwr->reqs_lock);
+
+		if (reset_pending(ptdev) && !reset_irq_raised(ptdev)) {
+			drm_err(&ptdev->base, "RESET timed out (0x%x)", reset_cmd);
+			return -ETIMEDOUT;
+		}
+
+		ptdev->pwr->pending_reqs &= ~PWR_IRQ_RESET_COMPLETED;
+	}
+
+	return 0;
+}
+
 static const char *get_domain_name(u8 domain)
 {
 	switch (domain) {
@@ -428,6 +468,16 @@ int panthor_pwr_init(struct panthor_device *ptdev)
 	return 0;
 }

+int panthor_pwr_reset_soft(struct panthor_device *ptdev)
+{
+	if (!(gpu_read64(ptdev, PWR_STATUS) & PWR_STATUS_ALLOW_SOFT_RESET)) {
+		drm_err(&ptdev->base, "RESET_SOFT not allowed");
+		return -EOPNOTSUPP;
+	}
+
+	return panthor_pwr_reset(ptdev, PWR_COMMAND_RESET_SOFT);
+}
+
 void panthor_pwr_l2_power_off(struct panthor_device *ptdev)
 {
 	const u64 l2_allow_mask = PWR_STATUS_DOMAIN_ALLOWED(PWR_COMMAND_DOMAIN_L2);
diff --git a/drivers/gpu/drm/panthor/panthor_pwr.h b/drivers/gpu/drm/panthor/panthor_pwr.h
index 3c834059a860..adf1f6136abc 100644
--- a/drivers/gpu/drm/panthor/panthor_pwr.h
+++ b/drivers/gpu/drm/panthor/panthor_pwr.h
@@ -10,6 +10,8 @@ void panthor_pwr_unplug(struct panthor_device *ptdev);

 int panthor_pwr_init(struct panthor_device *ptdev);

+int panthor_pwr_reset_soft(struct panthor_device *ptdev);
+
 void panthor_pwr_l2_power_off(struct panthor_device *ptdev);

 int panthor_pwr_l2_power_on(struct panthor_device *ptdev);
--
2.49.0

[PATCH v2 6/8] drm/panthor: Support GLB_REQ.STATE field for Mali-G1 GPUs

[Karunika Choo]

Add support for the GLB_REQ.STATE field introduced in CSF v4.1+, which
replaces the HALT bit to provide finer control over the MCU state. This
change implements basic handling for transitioning the MCU between
ACTIVE and HALT states on Mali-G1 GPUs.

The update introduces new helpers to issue the state change requests,
poll for MCU halt completion, and restore the MCU to an active state
after halting.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Reduced MCU_HALT_TIMEOUT_US to 1 second.
 * Wrap the CSG_IFACE_VERSION checks for v4.1.0 with
   panthor_fw_has_glb_state().
 * Removed use of undefined panthor_fw_csf_version() MACRO.
---
 drivers/gpu/drm/panthor/panthor_fw.c | 89 +++++++++++++++++++++++-----
 drivers/gpu/drm/panthor/panthor_fw.h |  7 +++
 2 files changed, 80 insertions(+), 16 deletions(-)

diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
index 6b91c3cb6678..78f884ccb8b0 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.c
+++ b/drivers/gpu/drm/panthor/panthor_fw.c
@@ -32,6 +32,7 @@
 #define PROGRESS_TIMEOUT_SCALE_SHIFT		10
 #define IDLE_HYSTERESIS_US			800
 #define PWROFF_HYSTERESIS_US			10000
+#define MCU_HALT_TIMEOUT_US			(1ULL * USEC_PER_SEC)

 /**
  * struct panthor_fw_binary_hdr - Firmware binary header.
@@ -316,6 +317,13 @@ panthor_fw_get_cs_iface(struct panthor_device *ptdev, u32 csg_slot, u32 cs_slot)
 	return &ptdev->fw->iface.streams[csg_slot][cs_slot];
 }

+static bool panthor_fw_has_glb_state(struct panthor_device *ptdev)
+{
+	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
+
+	return glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0);
+}
+
 /**
  * panthor_fw_conv_timeout() - Convert a timeout into a cycle-count
  * @ptdev: Device.
@@ -995,6 +1003,9 @@ static void panthor_fw_init_global_iface(struct panthor_device *ptdev)
 					 GLB_IDLE_EN |
 					 GLB_IDLE;

+	if (glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0))
+		glb_iface->input->ack_irq_mask |= GLB_STATE_MASK;
+
 	panthor_fw_update_reqs(glb_iface, req, GLB_IDLE_EN, GLB_IDLE_EN);
 	panthor_fw_toggle_reqs(glb_iface, req, ack,
 			       GLB_CFG_ALLOC_EN |
@@ -1068,6 +1079,54 @@ static void panthor_fw_stop(struct panthor_device *ptdev)
 		drm_err(&ptdev->base, "Failed to stop MCU");
 }

+static bool panthor_fw_mcu_halted(struct panthor_device *ptdev)
+{
+	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
+	bool halted;
+
+	halted = gpu_read(ptdev, MCU_STATUS) == MCU_STATUS_HALT;
+
+	if (panthor_fw_has_glb_state(ptdev))
+		halted &= (GLB_STATE_GET(glb_iface->output->ack) == GLB_STATE_HALT);
+
+	return halted;
+}
+
+static void panthor_fw_halt_mcu(struct panthor_device *ptdev)
+{
+	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
+
+	if (panthor_fw_has_glb_state(ptdev))
+		panthor_fw_update_reqs(glb_iface, req, GLB_STATE(GLB_STATE_HALT), GLB_STATE_MASK);
+	else
+		panthor_fw_update_reqs(glb_iface, req, GLB_HALT, GLB_HALT);
+
+	gpu_write(ptdev, CSF_DOORBELL(CSF_GLB_DOORBELL_ID), 1);
+}
+
+static bool panthor_fw_wait_mcu_halted(struct panthor_device *ptdev)
+{
+	bool halted = false;
+
+	if (read_poll_timeout_atomic(panthor_fw_mcu_halted, halted, halted, 10,
+				     MCU_HALT_TIMEOUT_US, 0, ptdev)) {
+		drm_warn(&ptdev->base, "Timed out waiting for MCU to halt");
+		return false;
+	}
+
+	return true;
+}
+
+static void panthor_fw_mcu_set_active(struct panthor_device *ptdev)
+{
+	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
+
+	if (panthor_fw_has_glb_state(ptdev))
+		panthor_fw_update_reqs(glb_iface, req, GLB_STATE(GLB_STATE_ACTIVE), GLB_STATE_MASK);
+	else
+		panthor_fw_update_reqs(glb_iface, req, 0, GLB_HALT);
+}
+
 /**
  * panthor_fw_pre_reset() - Call before a reset.
  * @ptdev: Device.
@@ -1084,19 +1143,13 @@ void panthor_fw_pre_reset(struct panthor_device *ptdev, bool on_hang)
 	ptdev->reset.fast = false;

 	if (!on_hang) {
-		struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
-		u32 status;
-
-		panthor_fw_update_reqs(glb_iface, req, GLB_HALT, GLB_HALT);
-		gpu_write(ptdev, CSF_DOORBELL(CSF_GLB_DOORBELL_ID), 1);
-		if (!gpu_read_poll_timeout(ptdev, MCU_STATUS, status,
-					   status == MCU_STATUS_HALT, 10,
-					   100000)) {
-			ptdev->reset.fast = true;
-		} else {
+		panthor_fw_halt_mcu(ptdev);
+		if (!panthor_fw_wait_mcu_halted(ptdev))
 			drm_warn(&ptdev->base, "Failed to cleanly suspend MCU");
-		}
+		else
+			ptdev->reset.fast = true;
 	}
+	panthor_fw_stop(ptdev);

 	panthor_job_irq_suspend(&ptdev->fw->irq);
 }
@@ -1124,14 +1177,14 @@ int panthor_fw_post_reset(struct panthor_device *ptdev)
 		 */
 		panthor_reload_fw_sections(ptdev, true);
 	} else {
-		/* The FW detects 0 -> 1 transitions. Make sure we reset
-		 * the HALT bit before the FW is rebooted.
+		/*
+		 * If the FW was previously successfully halted in the pre-reset
+		 * operation, we need to transition it to active again before
+		 * the FW is rebooted.
 		 * This is not needed on a slow reset because FW sections are
 		 * re-initialized.
 		 */
-		struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
-
-		panthor_fw_update_reqs(glb_iface, req, 0, GLB_HALT);
+		panthor_fw_mcu_set_active(ptdev);
 	}

 	ret = panthor_fw_start(ptdev);
@@ -1169,6 +1222,10 @@ void panthor_fw_unplug(struct panthor_device *ptdev)
 		if (ptdev->fw->irq.irq)
 			panthor_job_irq_suspend(&ptdev->fw->irq);

+		panthor_fw_halt_mcu(ptdev);
+		if (!panthor_fw_wait_mcu_halted(ptdev))
+			drm_warn(&ptdev->base, "Failed to halt MCU on unplug");
+
 		panthor_fw_stop(ptdev);
 	}

diff --git a/drivers/gpu/drm/panthor/panthor_fw.h b/drivers/gpu/drm/panthor/panthor_fw.h
index 6598d96c6d2a..a19ed48b2d0b 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.h
+++ b/drivers/gpu/drm/panthor/panthor_fw.h
@@ -214,6 +214,13 @@ struct panthor_fw_global_input_iface {
 #define GLB_FWCFG_UPDATE			BIT(9)
 #define GLB_IDLE_EN				BIT(10)
 #define GLB_SLEEP				BIT(12)
+#define GLB_STATE_MASK				GENMASK(14, 12)
+#define   GLB_STATE_ACTIVE			0
+#define   GLB_STATE_HALT			1
+#define   GLB_STATE_SLEEP			2
+#define   GLB_STATE_SUSPEND			3
+#define   GLB_STATE(x)				(((x) << 12) & GLB_STATE_MASK)
+#define   GLB_STATE_GET(x)			(((x) & GLB_STATE_MASK) >> 12)
 #define GLB_INACTIVE_COMPUTE			BIT(20)
 #define GLB_INACTIVE_FRAGMENT			BIT(21)
 #define GLB_INACTIVE_TILER			BIT(22)
--
2.49.0

Re: [PATCH v2 6/8] drm/panthor: Support GLB_REQ.STATE field for Mali-G1 GPUs

[Boris Brezillon]

On Fri, 24 Oct 2025 21:21:15 +0100
Karunika Choo <karunika.choo@arm.com> wrote:

> +static bool panthor_fw_has_glb_state(struct panthor_device *ptdev)
> +{
> +	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
> +
> +	return glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0);
> +}

It's okay for now, but if we start having more of these, we probably
want to automate the generation of those has_<featurex>() helpers with
something like:

#define FW_FEATURE(feat_name, major, minor)						\
static bool panthor_fw_has_ ## feat_name(struct panthor_device *ptdev) 			\
{											\
	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev); 	\
											\
	return glb_iface->control->version >= CSF_IFACE_VERSION(major, minor, 0);	\
}

Same goes for the HW features BTW.

> +
>  /**
>   * panthor_fw_conv_timeout() - Convert a timeout into a cycle-count
>   * @ptdev: Device.
> @@ -995,6 +1003,9 @@ static void panthor_fw_init_global_iface(struct panthor_device *ptdev)
>  					 GLB_IDLE_EN |
>  					 GLB_IDLE;
> 
> +	if (glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0))

	if (panthor_fw_has_glb_state(ptdev))

> +		glb_iface->input->ack_irq_mask |= GLB_STATE_MASK;
> +
>  	panthor_fw_update_reqs(glb_iface, req, GLB_IDLE_EN, GLB_IDLE_EN);
>  	panthor_fw_toggle_reqs(glb_iface, req, ack,
>  			       GLB_CFG_ALLOC_EN |
> @@ -1068,6 +1079,54 @@ static void panthor_fw_stop(struct panthor_device *ptdev)
>  		drm_err(&ptdev->base, "Failed to stop MCU");
>  }

Re: [PATCH v2 6/8] drm/panthor: Support GLB_REQ.STATE field for Mali-G1 GPUs

[Karunika Choo]

On 26/10/2025 08:27, Boris Brezillon wrote:
> On Fri, 24 Oct 2025 21:21:15 +0100
> Karunika Choo <karunika.choo@arm.com> wrote:
> 
>> +static bool panthor_fw_has_glb_state(struct panthor_device *ptdev)
>> +{
>> +	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
>> +
>> +	return glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0);
>> +}
> 
> It's okay for now, but if we start having more of these, we probably
> want to automate the generation of those has_<featurex>() helpers with
> something like:
> 
> #define FW_FEATURE(feat_name, major, minor)						\
> static bool panthor_fw_has_ ## feat_name(struct panthor_device *ptdev) 			\
> {											\
> 	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev); 	\
> 											\
> 	return glb_iface->control->version >= CSF_IFACE_VERSION(major, minor, 0);	\
> }
> 
> Same goes for the HW features BTW.
> 

I wonder if at that point, would a bitmask as previously proposed be a
cleaner solution? I have a minor bone to pick with MACROs that generate
functions as they make finding its definition unnecessarily complicated
and obtuse. Not to mention if the conditions for a feature changes and
is disjoint from all the others, updating the macro to handle these
would cause additional churn that I would really hope to avoid.

>> +
>>  /**
>>   * panthor_fw_conv_timeout() - Convert a timeout into a cycle-count
>>   * @ptdev: Device.
>> @@ -995,6 +1003,9 @@ static void panthor_fw_init_global_iface(struct panthor_device *ptdev)
>>  					 GLB_IDLE_EN |
>>  					 GLB_IDLE;
>>
>> +	if (glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0))
> 
> 	if (panthor_fw_has_glb_state(ptdev))
> 

Thanks for catching this. Will fix in v3.

Kind regards,
Karunika Choo

>> +		glb_iface->input->ack_irq_mask |= GLB_STATE_MASK;
>> +
>>  	panthor_fw_update_reqs(glb_iface, req, GLB_IDLE_EN, GLB_IDLE_EN);
>>  	panthor_fw_toggle_reqs(glb_iface, req, ack,
>>  			       GLB_CFG_ALLOC_EN |
>> @@ -1068,6 +1079,54 @@ static void panthor_fw_stop(struct panthor_device *ptdev)
>>  		drm_err(&ptdev->base, "Failed to stop MCU");
>>  }
> 

Re: [PATCH v2 6/8] drm/panthor: Support GLB_REQ.STATE field for Mali-G1 GPUs

[Boris Brezillon]

On Mon, 27 Oct 2025 10:43:42 +0000
Karunika Choo <karunika.choo@arm.com> wrote:

> On 26/10/2025 08:27, Boris Brezillon wrote:
> > On Fri, 24 Oct 2025 21:21:15 +0100
> > Karunika Choo <karunika.choo@arm.com> wrote:
> >   
> >> +static bool panthor_fw_has_glb_state(struct panthor_device *ptdev)
> >> +{
> >> +	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
> >> +
> >> +	return glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0);
> >> +}  
> > 
> > It's okay for now, but if we start having more of these, we probably
> > want to automate the generation of those has_<featurex>() helpers with
> > something like:
> > 
> > #define FW_FEATURE(feat_name, major, minor)						\
> > static bool panthor_fw_has_ ## feat_name(struct panthor_device *ptdev) 			\
> > {											\
> > 	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev); 	\
> > 											\
> > 	return glb_iface->control->version >= CSF_IFACE_VERSION(major, minor, 0);	\
> > }
> > 
> > Same goes for the HW features BTW.
> >   
> 
> I wonder if at that point, would a bitmask as previously proposed be a
> cleaner solution? I have a minor bone to pick with MACROs that generate
> functions as they make finding its definition unnecessarily complicated
> and obtuse. Not to mention if the conditions for a feature changes and
> is disjoint from all the others, updating the macro to handle these
> would cause additional churn that I would really hope to avoid.

Let's do without the macros for now.

[PATCH v2 7/8] drm/panthor: Support 64-bit endpoint_req register for Mali-G1

[Karunika Choo]

Add support for the 64-bit endpoint_req register introduced in CSF v4.0+
GPUs. Unlike a simple register widening, the 64-bit variant occupies the
next 64 bits after the original 32-bit field, requiring
version-dependent access.

This change introduces helper functions to read, write, and update the
endpoint_req register, ensuring correct handling on both pre-v4.0 and
v4.0+ firmwares.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Wrap the CSG_IFACE_VERSION checks for v4.0.0 with
   panthor_fw_has_64bit_ep_req().
 * Removed wrongly included code from previous patch.
 * Reordered CSG_EP_REQ_PRIORITY_GET() and CSG_EP_REQ_PRIORITY() to
   reuse CSG_EP_REQ_PRIORITY_MASK definition.
 * Updated panthor_fw_csg_endpoint_req_*() functions to accept CSG iface
   structure instead of a CSG id.
 * Update endpoint_req variables to u64.
 * Minor readability and code quality fixes.
---
 drivers/gpu/drm/panthor/panthor_fw.c    | 36 +++++++++++++++++++++++++
 drivers/gpu/drm/panthor/panthor_fw.h    | 25 +++++++++++++++--
 drivers/gpu/drm/panthor/panthor_sched.c | 21 +++++++++------
 3 files changed, 72 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
index 78f884ccb8b0..9ba10ab1d7c0 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.c
+++ b/drivers/gpu/drm/panthor/panthor_fw.c
@@ -324,6 +324,42 @@ static bool panthor_fw_has_glb_state(struct panthor_device *ptdev)
 	return glb_iface->control->version >= CSF_IFACE_VERSION(4, 1, 0);
 }

+static bool panthor_fw_has_64bit_ep_req(struct panthor_device *ptdev)
+{
+	struct panthor_fw_global_iface *glb_iface = panthor_fw_get_glb_iface(ptdev);
+
+	return glb_iface->control->version >= CSF_IFACE_VERSION(4, 0, 0);
+}
+
+u64 panthor_fw_csg_endpoint_req_get(struct panthor_device *ptdev,
+				    struct panthor_fw_csg_iface *csg_iface)
+{
+	if (panthor_fw_has_64bit_ep_req(ptdev))
+		return csg_iface->input->endpoint_req2;
+	else
+		return csg_iface->input->endpoint_req;
+}
+
+void panthor_fw_csg_endpoint_req_set(struct panthor_device *ptdev,
+				     struct panthor_fw_csg_iface *csg_iface, u64 value)
+{
+	if (panthor_fw_has_64bit_ep_req(ptdev))
+		csg_iface->input->endpoint_req2 = value;
+	else
+		csg_iface->input->endpoint_req = lower_32_bits(value);
+}
+
+void panthor_fw_csg_endpoint_req_update(struct panthor_device *ptdev,
+					struct panthor_fw_csg_iface *csg_iface, u64 value,
+					u64 mask)
+{
+	if (panthor_fw_has_64bit_ep_req(ptdev))
+		panthor_fw_update_reqs64(csg_iface, endpoint_req2, value, mask);
+	else
+		panthor_fw_update_reqs(csg_iface, endpoint_req, lower_32_bits(value),
+				       lower_32_bits(mask));
+}
+
 /**
  * panthor_fw_conv_timeout() - Convert a timeout into a cycle-count
  * @ptdev: Device.
diff --git a/drivers/gpu/drm/panthor/panthor_fw.h b/drivers/gpu/drm/panthor/panthor_fw.h
index a19ed48b2d0b..fbdc21469ba3 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.h
+++ b/drivers/gpu/drm/panthor/panthor_fw.h
@@ -167,10 +167,11 @@ struct panthor_fw_csg_input_iface {
 #define CSG_EP_REQ_TILER(x)			(((x) << 16) & GENMASK(19, 16))
 #define CSG_EP_REQ_EXCL_COMPUTE			BIT(20)
 #define CSG_EP_REQ_EXCL_FRAGMENT		BIT(21)
-#define CSG_EP_REQ_PRIORITY(x)			(((x) << 28) & GENMASK(31, 28))
 #define CSG_EP_REQ_PRIORITY_MASK		GENMASK(31, 28)
+#define CSG_EP_REQ_PRIORITY(x)			(((x) << 28) & CSG_EP_REQ_PRIORITY_MASK)
+#define CSG_EP_REQ_PRIORITY_GET(x)		(((x) & CSG_EP_REQ_PRIORITY_MASK) >> 28)
 	u32 endpoint_req;
-	u32 reserved2[2];
+	u64 endpoint_req2;
 	u64 suspend_buf;
 	u64 protm_suspend_buf;
 	u32 config;
@@ -464,6 +465,16 @@ struct panthor_fw_global_iface {
 		spin_unlock(&(__iface)->lock); \
 	} while (0)

+#define panthor_fw_update_reqs64(__iface, __in_reg, __val, __mask) \
+	do { \
+		u64 __cur_val, __new_val; \
+		spin_lock(&(__iface)->lock); \
+		__cur_val = READ_ONCE((__iface)->input->__in_reg); \
+		__new_val = (__cur_val & ~(__mask)) | ((__val) & (__mask)); \
+		WRITE_ONCE((__iface)->input->__in_reg, __new_val); \
+		spin_unlock(&(__iface)->lock); \
+	} while (0)
+
 struct panthor_fw_global_iface *
 panthor_fw_get_glb_iface(struct panthor_device *ptdev);

@@ -473,6 +484,16 @@ panthor_fw_get_csg_iface(struct panthor_device *ptdev, u32 csg_slot);
 struct panthor_fw_cs_iface *
 panthor_fw_get_cs_iface(struct panthor_device *ptdev, u32 csg_slot, u32 cs_slot);

+u64 panthor_fw_csg_endpoint_req_get(struct panthor_device *ptdev,
+				    struct panthor_fw_csg_iface *csg_iface);
+
+void panthor_fw_csg_endpoint_req_set(struct panthor_device *ptdev,
+				     struct panthor_fw_csg_iface *csg_iface, u64 value);
+
+void panthor_fw_csg_endpoint_req_update(struct panthor_device *ptdev,
+					struct panthor_fw_csg_iface *csg_iface, u64 value,
+					u64 mask);
+
 int panthor_fw_csg_wait_acks(struct panthor_device *ptdev, u32 csg_id, u32 req_mask,
 			     u32 *acked, u32 timeout_ms);

diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
index 0cc9055f4ee5..d6f5efc10312 100644
--- a/drivers/gpu/drm/panthor/panthor_sched.c
+++ b/drivers/gpu/drm/panthor/panthor_sched.c
@@ -1139,11 +1139,13 @@ csg_slot_sync_priority_locked(struct panthor_device *ptdev, u32 csg_id)
 {
 	struct panthor_csg_slot *csg_slot = &ptdev->scheduler->csg_slots[csg_id];
 	struct panthor_fw_csg_iface *csg_iface;
+	u64 endpoint_req;

 	lockdep_assert_held(&ptdev->scheduler->lock);

 	csg_iface = panthor_fw_get_csg_iface(ptdev, csg_id);
-	csg_slot->priority = (csg_iface->input->endpoint_req & CSG_EP_REQ_PRIORITY_MASK) >> 28;
+	endpoint_req = panthor_fw_csg_endpoint_req_get(ptdev, csg_iface);
+	csg_slot->priority = CSG_EP_REQ_PRIORITY_GET(endpoint_req);
 }

 /**
@@ -1303,6 +1305,7 @@ csg_slot_prog_locked(struct panthor_device *ptdev, u32 csg_id, u32 priority)
 	struct panthor_csg_slot *csg_slot;
 	struct panthor_group *group;
 	u32 queue_mask = 0, i;
+	u64 endpoint_req;

 	lockdep_assert_held(&ptdev->scheduler->lock);

@@ -1329,10 +1332,12 @@ csg_slot_prog_locked(struct panthor_device *ptdev, u32 csg_id, u32 priority)
 	csg_iface->input->allow_compute = group->compute_core_mask;
 	csg_iface->input->allow_fragment = group->fragment_core_mask;
 	csg_iface->input->allow_other = group->tiler_core_mask;
-	csg_iface->input->endpoint_req = CSG_EP_REQ_COMPUTE(group->max_compute_cores) |
-					 CSG_EP_REQ_FRAGMENT(group->max_fragment_cores) |
-					 CSG_EP_REQ_TILER(group->max_tiler_cores) |
-					 CSG_EP_REQ_PRIORITY(priority);
+	endpoint_req = CSG_EP_REQ_COMPUTE(group->max_compute_cores) |
+		       CSG_EP_REQ_FRAGMENT(group->max_fragment_cores) |
+		       CSG_EP_REQ_TILER(group->max_tiler_cores) |
+		       CSG_EP_REQ_PRIORITY(priority);
+	panthor_fw_csg_endpoint_req_set(ptdev, csg_iface, endpoint_req);
+
 	csg_iface->input->config = panthor_vm_as(group->vm);

 	if (group->suspend_buf)
@@ -2230,9 +2235,9 @@ tick_ctx_apply(struct panthor_scheduler *sched, struct panthor_sched_tick_ctx *c
 				continue;
 			}

-			panthor_fw_update_reqs(csg_iface, endpoint_req,
-					       CSG_EP_REQ_PRIORITY(new_csg_prio),
-					       CSG_EP_REQ_PRIORITY_MASK);
+			panthor_fw_csg_endpoint_req_update(ptdev, csg_iface,
+							   CSG_EP_REQ_PRIORITY(new_csg_prio),
+							   CSG_EP_REQ_PRIORITY_MASK);
 			csgs_upd_ctx_queue_reqs(ptdev, &upd_ctx, csg_id,
 						csg_iface->output->ack ^ CSG_ENDPOINT_CONFIG,
 						CSG_ENDPOINT_CONFIG);
--
2.49.0

[PATCH v2 8/8] drm/panthor: Add support for Mali-G1 GPUs

[Karunika Choo]

Add support for Mali-G1 GPUs (CSF architecture v14), introducing a new
panthor_hw_arch_v14 entry with reset and L2 power management operations
via the PWR_CONTROL block.

Mali-G1 introduces a dedicated PWR_CONTROL block for managing resets and
power domains. panthor_gpu_info_init() is updated to use this block for
L2, tiler, and shader domain present register reads.

Signed-off-by: Karunika Choo <karunika.choo@arm.com>
---
v2:
 * Removed feature bits usage.
 * Check panthor_hw_has_pwr_ctrl() to read the correct *_PRESENT
   registers instead of reading reserved registers on newer GPUs.
---
 drivers/gpu/drm/panthor/panthor_fw.c |  1 +
 drivers/gpu/drm/panthor/panthor_hw.c | 35 ++++++++++++++++++++++++----
 2 files changed, 32 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
index 9ba10ab1d7c0..031d23166c18 100644
--- a/drivers/gpu/drm/panthor/panthor_fw.c
+++ b/drivers/gpu/drm/panthor/panthor_fw.c
@@ -1500,3 +1500,4 @@ MODULE_FIRMWARE("arm/mali/arch10.12/mali_csffw.bin");
 MODULE_FIRMWARE("arm/mali/arch11.8/mali_csffw.bin");
 MODULE_FIRMWARE("arm/mali/arch12.8/mali_csffw.bin");
 MODULE_FIRMWARE("arm/mali/arch13.8/mali_csffw.bin");
+MODULE_FIRMWARE("arm/mali/arch14.8/mali_csffw.bin");
diff --git a/drivers/gpu/drm/panthor/panthor_hw.c b/drivers/gpu/drm/panthor/panthor_hw.c
index 09aef34a6ce7..29c7a6e60300 100644
--- a/drivers/gpu/drm/panthor/panthor_hw.c
+++ b/drivers/gpu/drm/panthor/panthor_hw.c
@@ -3,6 +3,7 @@

 #include "panthor_device.h"
 #include "panthor_gpu.h"
+#include "panthor_pwr.h"
 #include "panthor_regs.h"

 #define GPU_PROD_ID_MAKE(arch_major, prod_major) \
@@ -28,12 +29,25 @@ static struct panthor_hw panthor_hw_arch_v10 = {
 	},
 };

+static struct panthor_hw panthor_hw_arch_v14 = {
+	.ops = {
+		.soft_reset = panthor_pwr_reset_soft,
+		.l2_power_off = panthor_pwr_l2_power_off,
+		.l2_power_on = panthor_pwr_l2_power_on,
+	},
+};
+
 static struct panthor_hw_entry panthor_hw_match[] = {
 	{
 		.arch_min = 10,
 		.arch_max = 13,
 		.hwdev = &panthor_hw_arch_v10,
 	},
+	{
+		.arch_min = 14,
+		.arch_max = 14,
+		.hwdev = &panthor_hw_arch_v14,
+	},
 };

 static char *get_gpu_model_name(struct panthor_device *ptdev)
@@ -81,6 +95,12 @@ static char *get_gpu_model_name(struct panthor_device *ptdev)
 		fallthrough;
 	case GPU_PROD_ID_MAKE(13, 1):
 		return "Mali-G625";
+	case GPU_PROD_ID_MAKE(14, 0):
+		return "Mali-G1-Ultra";
+	case GPU_PROD_ID_MAKE(14, 1):
+		return "Mali-G1-Premium";
+	case GPU_PROD_ID_MAKE(14, 3):
+		return "Mali-G1-Pro";
 	}

 	return "(Unknown Mali GPU)";
@@ -107,12 +127,19 @@ static void panthor_gpu_info_init(struct panthor_device *ptdev)

 	ptdev->gpu_info.as_present = gpu_read(ptdev, GPU_AS_PRESENT);

-	ptdev->gpu_info.shader_present = gpu_read64(ptdev, GPU_SHADER_PRESENT);
-	ptdev->gpu_info.tiler_present = gpu_read64(ptdev, GPU_TILER_PRESENT);
-	ptdev->gpu_info.l2_present = gpu_read64(ptdev, GPU_L2_PRESENT);
-
 	/* Introduced in arch 11.x */
 	ptdev->gpu_info.gpu_features = gpu_read64(ptdev, GPU_FEATURES);
+
+	if (panthor_hw_has_pwr_ctrl(ptdev)) {
+		/* Introduced in arch 14.x */
+		ptdev->gpu_info.l2_present = gpu_read64(ptdev, PWR_L2_PRESENT);
+		ptdev->gpu_info.tiler_present = gpu_read64(ptdev, PWR_TILER_PRESENT);
+		ptdev->gpu_info.shader_present = gpu_read64(ptdev, PWR_SHADER_PRESENT);
+	} else {
+		ptdev->gpu_info.shader_present = gpu_read64(ptdev, GPU_SHADER_PRESENT);
+		ptdev->gpu_info.tiler_present = gpu_read64(ptdev, GPU_TILER_PRESENT);
+		ptdev->gpu_info.l2_present = gpu_read64(ptdev, GPU_L2_PRESENT);
+	}
 }

 static void panthor_hw_info_init(struct panthor_device *ptdev)
--
2.49.0