Skip to contentSkip to conversation

accel: rocket: Add RK3568 NPU support

RFC PATCH Superseded Midgy BALON 7 participants 38 messages Jun 4 – Jul 9

Subsystems (5)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover accel: rocket: Add RK3568 NPU support

RFC, not for merge. End-to-end inference does not produce correct output yet (see Status), so per the v2 discussion this is a request for design feedback. It now probes, attaches, and submits cleanly on a stock v7.1-rc6 tree; what remains is one hardware-internal issue.

The RK3568 has a single NVDLA-derived NPU core, the same IP family as the RK3588 NPU the driver already supports; the register layout matches. The RK3568 differences are a 32-bit NPU AXI/IOMMU (vs 40-bit) and explicit PVTPLL/PMU bring-up to power and de-idle the NPU before it is reachable.

Patches:
1-2 rocket: per-SoC data struct, then derive DMA width and core count
from match data (refactors, no functional change).
3 rocket: RK3568 SoC data + PVTPLL/PMU/NOC bring-up.
4 rocket: reset the NPU before detaching the IOMMU on a job timeout
(the detach otherwise stalls a wedged AXI master and WARNs).
5 rocket: keep the IOMMU domain attached across jobs instead of
re-attaching per job (the per-job rk_iommu handshake on the idle NPU MMU is slow and noisy).
6 iommu/rockchip: clear AUTO_GATING bit 1 on the RK356x v1 IOMMU so
the page-walker keeps its clock (else a TLB-miss walk never completes).
7 dt-bindings: add the RK3568 NPU compatible.
8-9 arm64 dts: add the NPU and its IOMMU, and enable them on ROCK 3B.

Dependency. The NPU MMU is rockchip-iommu v1 (32-bit) while the rest of the RK3568 uses v2 (40-bit). They cannot coexist until the driver carries per-device ops; this series is developed on top of Simon Xue's "iommu/rockchip: Drop global rk_ops in favor of per-device ops" [1]. Without it the NPU IOMMU fails to probe on a full RK3568 boot.

Power bring-up. The NPU is brought up through the power-domain layer (no driver hack): the NPU power-domain keeps its clocks but drops the pm_qos phandle (qos_npu sits behind the gated NPU NoC, so genpd's power-off QoS save faults reading it), and vdd_npu is marked always-on so the rail is up before genpd de-idles the NoC at power-on. The PMU de-idle then ACKs without PVTPLL running; PVTPLL is only needed for compute.

Status. On v7.1-rc6 the driver probes, creates /dev/accel/accel0, attaches an IOMMU domain, and submits jobs; the program controller fetches and broadcasts the command list. Inference output is still wrong, and the cause is split across three layers:
- kernel (this series): the RK3568 differences appear handled;
- mesa/Teflon userspace: still emits RK3588-tuned config, wrong for

    RK3568 (to be filed separately on mesa-dev);

- hardware: with corrected config the NPU's DMA reads the full input and weight tensors (confirmed via its DMA bandwidth counters), but the MAC/output stage never completes, the job times out, and the output stays at the buffer's zero-point. I have not found the missing step; it is not in the command list (replaying the vendor's byte-exact command list behaves the same). Pointers welcome, especially from anyone with RK3568 NPU experience.

Known residual. On the first IOMMU attach the NPU MMU is idle with paging already enabled; the rk_iommu stall/reset handshake does not complete in that state and logs one burst of timeouts before the (kept) domain settles. It is harmless here because the job times out regardless, but it points at an idle-MMU reconfiguration corner the rk_iommu code does not handle on this block.

[1] https://lore.kernel.org/linux-rockchip/[email protected]/

Changes since v2:
- Tagged RFC; now tested on a stock v7.1-rc6 tree.
- Bring-up moved into the power-domain/DT layer (no initcall hack).
- Added the IOMMU detach-on-timeout and attach-once driver fixes.
- Split the driver patch (Heiko): soc_data / match-data / RK3568.
- Derive DMA width and core count from match data; drop the DT rescans.
- Binding describes the hardware; added the missing $ref on rockchip,pmu.
- Disclosed the per-device-ops IOMMU dependency.

Midgy BALON (9):
accel: rocket: Introduce per-SoC rocket_soc_data
accel: rocket: Derive DMA width and core count from match data
accel: rocket: Add RK3568 SoC support
accel: rocket: Reset the NPU before detaching the IOMMU on timeout
accel: rocket: Keep the IOMMU domain attached across jobs iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU
dt-bindings: npu: rockchip,rk3588-rknn-core: Add RK3568
arm64: dts: rockchip: rk356x: Add the NPU and its IOMMU
arm64: dts: rockchip: rk3568-rock-3b: Enable the NPU

 .../npu/rockchip,rk3588-rknn-core.yaml        | 18 ++++-
 .../boot/dts/rockchip/rk3568-rock-3b.dts      | 14 +++-
 arch/arm64/boot/dts/rockchip/rk356x-base.dtsi | 38 +++++++++++
 drivers/accel/rocket/rocket_core.c            | 22 ++++++-
 drivers/accel/rocket/rocket_core.h            | 19 ++++++
 drivers/accel/rocket/rocket_device.c          | 15 ++---
 drivers/accel/rocket/rocket_device.h          |  3 +-
 drivers/accel/rocket/rocket_drv.c             | 66 ++++++++++++++++++-
 drivers/accel/rocket/rocket_job.c             | 35 ++++++++--
 drivers/iommu/rockchip-iommu.c                | 12 ++++
 10 files changed, 219 insertions(+), 23 deletions(-)

base-commit: 52c800fdcf11888ebeb50c3d707f782cc15b66eb

Signature · 1 line
2.39.5
#1 lore raw

RFC PATCH v3 1/9 accel: rocket: Introduce per-SoC rocket_soc_data

Add a per-SoC data structure carried in the OF match table, currently holding only the NPU AXI address width, and use it for the per-core DMA mask instead of a hardcoded 40-bit value. No functional change: the RK3588 AXI master is 40-bit. This prepares for SoCs with a narrower address width.

---
 drivers/accel/rocket/rocket_core.c |  7 ++++++-
 drivers/accel/rocket/rocket_core.h | 11 +++++++++++
 drivers/accel/rocket/rocket_drv.c  |  6 +++++-
 3 files changed, 22 insertions(+), 2 deletions(-)
drivers/accel/rocket/rocket_core.c +6 -1
index b3b2fa9ba645a..09c445af7de73 100644
@@ -7,6 +7,7 @@
 #include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/iommu.h>
+#include <linux/of.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
 #include <linux/reset.h>
@@ -21,6 +22,10 @@ int rocket_core_init(struct rocket_core *core)
 	u32 version;
 	int err = 0;
 
+	core->soc_data = of_device_get_match_data(dev);
+	if (!core->soc_data)
+		return dev_err_probe(dev, -EINVAL, "missing SoC match data\n");
+
 	core->resets[0].id = "srst_a";
 	core->resets[1].id = "srst_h";
 	err = devm_reset_control_bulk_get_exclusive(&pdev->dev, ARRAY_SIZE(core->resets),
@@ -52,7 +57,7 @@ int rocket_core_init(struct rocket_core *core)
 
 	dma_set_max_seg_size(dev, UINT_MAX);
 
-	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(40));
+	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(core->soc_data->dma_bits));
 	if (err)
 		return err;
 
drivers/accel/rocket/rocket_core.h +11 -0
index f6d7382854ca9..8ee105a0be40e 100644
@@ -12,6 +12,16 @@
 
 #include "rocket_registers.h"
 
+struct rocket_core;
+
+/**
+ * struct rocket_soc_data - per-SoC configuration data
+ * @dma_bits: Physical address width reachable by the NPU's AXI master.
+ */
+struct rocket_soc_data {
+	unsigned int dma_bits;
+};
+
 #define rocket_pc_readl(core, reg) \
 	readl((core)->pc_iomem + (REG_PC_##reg))
 #define rocket_pc_writel(core, reg, value) \
@@ -31,6 +41,7 @@ struct rocket_core {
 	struct device *dev;
 	struct rocket_device *rdev;
 	unsigned int index;
+	const struct rocket_soc_data *soc_data;
 
 	int irq;
 	void __iomem *pc_iomem;
drivers/accel/rocket/rocket_drv.c +5 -1
index 8bbbce594883e..384c38e13acce 100644
@@ -213,8 +213,12 @@ static void rocket_remove(struct platform_device *pdev)
 	}
 }
 
+static const struct rocket_soc_data rk3588_soc_data = {
+	.dma_bits = 40,
+};
+
 static const struct of_device_id dt_match[] = {
-	{ .compatible = "rockchip,rk3588-rknn-core" },
+	{ .compatible = "rockchip,rk3588-rknn-core", .data = &rk3588_soc_data },
 	{}
 };
 MODULE_DEVICE_TABLE(of, dt_match);
Signature · 1 line
2.39.5
#2 lore raw

Re: [RFC PATCH v3 1/9] accel: rocket: Introduce per-SoC rocket_soc_data

Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 11 potential issue(s) to consider:

Pre-existing issues:
- [Critical] kref_put() dereferences a NULL pointer when job initialization fails early or a BO is imported.
- [High] Error path in BO creation frees an object that is already owned by the handle table.
- [High] Device is permanently left in an active power state upon unbind.
- [High] Array mismanagement during core removal prevents multi-core unbind.
- [High] Memory and DRM device leak on probe failure or deferral.
- [High] Asynchronous reset work is canceled after the DRM scheduler is destroyed.
- [High] The DRM device is registered and exposed to userspace before the NPU cores are initialized.
- [High] The global rdev pointer and rdev->num_cores are accessed and modified locklessly during probe and removal.
- [High] The shared interrupt handler accesses hardware registers without checking the PM state, and suspend executes without synchronizing IRQs.
- [High] A failure in rocket_job_run leaks the PM usage counter and returns an unsignaled hardware fence, hanging the DRM scheduler.
- [High] Registers are accessed during initialization before resets are deasserted.
--

commit e6cc2e46cba8fcba41d70a8b1e9422607b444db8
Author: Midgy BALON <[email protected]>

accel: rocket: Introduce per-SoC rocket_soc_data

Add a per-SoC data structure carried in the OF match table, currently holding only the NPU AXI address width, and use it for the per-core DMA mask instead of a hardcoded 40-bit value.

drivers/accel/rocket/rocket_core.c int rocket_core_init(struct rocket_core *core) · line 28

Quoted 12 lines
drivers/accel/rocket/rocket_core.c +4 -0
index b3b2fa9ba645a..09c445af7de73 100644
@@ -21,6 +22,10 @@ int rocket_core_init(struct rocket_core *core)
 	u32 version;
 	int err = 0;
 
+	core->soc_data = of_device_get_match_data(dev);
+	if (!core->soc_data)
+		return dev_err_probe(dev, -EINVAL, "missing SoC match data\n");
+

[Severity: High]
This is a pre-existing issue, but I noticed that later in this function, registers are accessed before resets are deasserted:

drivers/accel/rocket/rocket_core.c:rocket_core_init() {
	...
	core->resets[0].id = "srst_a";
	core->resets[1].id = "srst_h";
	err = devm_reset_control_bulk_get_exclusive(&pdev->dev, ARRAY_SIZE(core->resets),
						    core->resets);
	...
	version = rocket_pc_readl(core, VERSION);
	version += rocket_pc_readl(core, VERSION_NUM) & 0xffff;
	...
}

Will this cause a bus stall and trigger an asynchronous external abort if the hardware or bootloader leaves these resets asserted?

[Severity: High]
This is a pre-existing issue, but does rocket_core_fini() leave the device permanently powered on with its clocks running after the driver has been detached?

drivers/accel/rocket/rocket_core.c:rocket_core_fini() {
	pm_runtime_dont_use_autosuspend(core->dev);
	pm_runtime_disable(core->dev);
	...
}

Since rocket_core_init() drops its Runtime PM reference asynchronously via pm_runtime_put_autosuspend(dev), if the driver is unbound before the autosuspend delay elapses, pm_runtime_disable() forcefully cancels the pending asynchronous suspend work without first synchronously powering down the hardware (e.g., via pm_runtime_put_sync()).

Quoted 3 lines
drivers/accel/rocket/rocket_core.c +0 -0
 	core->resets[0].id = "srst_a";
 	core->resets[1].id = "srst_h";
 	err = devm_reset_control_bulk_get_exclusive(&pdev->dev, ARRAY_SIZE(core->resets),

[ ... ]

Quoted 14 lines
drivers/accel/rocket/rocket_core.c +1 -1
@@ -52,7 +57,7 @@ int rocket_core_init(struct rocket_core *core)
 
 	dma_set_max_seg_size(dev, UINT_MAX);
 
-	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(40));
+	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(core->soc_data->dma_bits));
 	if (err)
 		return err;
 
drivers/accel/rocket/rocket_drv.c +0 -0
index 8bbbce594883e..384c38e13acce 100644

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 215

Quoted 3 lines
drivers/accel/rocket/rocket_drv.c +0 -0
@@ -213,8 +213,12 @@ static void rocket_remove(struct platform_device *pdev)
 	}
 }
 

[Severity: High]
This is a pre-existing issue, but can this array mismanagement during core removal prevent multi-core unbind?

drivers/accel/rocket/rocket_drv.c:rocket_remove() {
	...
	rocket_core_fini(&rdev->cores[core]);
	rdev->num_cores--;
	...
}
drivers/accel/rocket/rocket_drv.c:find_core_for_dev() {
	for (unsigned int core = 0; core < rdev->num_cores; core++) {
		if (dev == rdev->cores[core].dev)
			return core;
	}
	...
}

When a core is unbound, rdev->num_cores is decremented, but the unbound core's slot in the static rdev->cores array is left unmodified, and the remaining elements are not shifted.

Because find_core_for_dev() unconditionally loops only rdev->num_cores times, any core residing at an array index greater than or equal to the decremented num_cores becomes invisible.

Will subsequent attempts to unbind those remaining cores fail, silently exiting early and completely leaking the cores' resources and power state?

[Severity: High]
This is a pre-existing issue, but could rocket_probe() leak memory and the DRM device on probe failure or deferral?

drivers/accel/rocket/rocket_drv.c:rocket_probe() {
	if (rdev == NULL) {
		/* First core probing, initialize DRM device. */
		rdev = rocket_device_init(drm_dev, &rocket_drm_driver, soc_data);
	...
}
drivers/accel/rocket/rocket_device.c:rocket_device_init() {
	rdev = devm_drm_dev_alloc(dev, rocket_drm_driver, struct rocket_device, ddev);
	...
	rdev->cores = devm_kcalloc(dev, soc_data->num_cores, sizeof(*rdev->cores), GFP_KERNEL);
	...
}

The global facade device rdev is dynamically initialized via rocket_device_init() using the global, virtually registered drm_dev platform device. Allocations like devm_drm_dev_alloc() and devm_kcalloc() are bound to the drm_dev lifecycle rather than the physical pdev currently being probed.

If rocket_core_init() subsequently fails (e.g., due to -EPROBE_DEFER), rocket_probe() cleans up by calling rocket_device_fini() and resetting rdev = NULL. However, rocket_device_fini() only unregisters the DRM device.

Because drm_dev remains globally alive until module unload, none of the devm_ allocations are freed. On the next probe attempt, a new rdev is allocated on drm_dev. Can this create a permanent memory leak?

[Severity: High]
This is a pre-existing issue, but might the DRM device be registered and exposed to userspace before the NPU cores are initialized?

drivers/accel/rocket/rocket_device.c:rocket_device_init() {
	...
	err = drm_dev_register(ddev, 0);
	if (err)
		return ERR_PTR(err);

	return rdev;
}
drivers/accel/rocket/rocket_drv.c:rocket_probe() {
	...
		rdev = rocket_device_init(drm_dev, &rocket_drm_driver, soc_data);
	...
	unsigned int core = rdev->num_cores;
	dev_set_drvdata(&pdev->dev, rdev);

	rdev->cores[core].rdev = rdev;
	rdev->cores[core].dev = &pdev->dev;
	...
}

When the first core is probed, it calls rocket_device_init(), which ends with drm_dev_register(). At this moment, the DRM character device becomes visible and openable by userspace.

However, the core hardware, IRQs, and core pointers (e.g. rdev->cores[core].dev) are not assigned or initialized until immediately afterward, when rocket_core_init() runs.

If userspace opens the device in this narrow window, will rocket_open() evaluate the uninitialized rdev->cores[0].dev (which is NULL) and pass it to rocket_iommu_domain_create(), causing an immediate NULL pointer dereference?

[Severity: High]
This is a pre-existing issue, but is it possible that the global rdev pointer and rdev->num_cores are accessed and modified locklessly during probe and removal?

drivers/accel/rocket/rocket_drv.c:rocket_probe() {
	if (rdev == NULL) {
		/* First core probing, initialize DRM device. */
		rdev = rocket_device_init(drm_dev, &rocket_drm_driver, soc_data);
	...
	rdev->num_cores++;
	...
}

If multiple cores are probed concurrently (e.g., via sysfs asynchronous binding), multiple threads will evaluate if (rdev == NULL) simultaneously without a lock.

Can this create a data race where concurrent threads allocate redundant DRM devices, overwrite the rdev singleton (causing memory leaks), and corrupt the shared rdev->num_cores counter and rdev->cores array?

[Severity: Critical]
This is a pre-existing issue, but will kref_put() dereference a NULL pointer when job initialization fails early or a BO is imported?

drivers/accel/rocket/rocket_job.c:rocket_ioctl_submit_job() {
	...
	goto out_cleanup_job;
	...
out_cleanup_job:
	if (ret)
		drm_sched_job_cleanup(&rjob->base);
out_put_job:
	rocket_job_put(rjob);
}
drivers/accel/rocket/rocket_drv.c:rocket_iommu_domain_put() {
	kref_put(&domain->kref, rocket_iommu_domain_destroy);
}

If an early failure occurs in rocket_ioctl_submit_job() (e.g., in rocket_copy_tasks() or drm_gem_objects_lookup()), the error path calls rocket_job_put(), which invokes rocket_job_cleanup(). At this point, job->domain is still NULL.

rocket_job_cleanup() unconditionally passes this to rocket_iommu_domain_put(), which immediately evaluates &domain->kref and causes a NULL pointer dereference.

The same panic appears to occur when a BO is imported via PRIME (rocket_gem_create_object() leaves domain NULL), and rocket_gem_bo_free() subsequently accesses bo->domain->domain and bo->domain.

[Severity: High]
This is a pre-existing issue, but could the error path in BO creation free an object that is already owned by the handle table?

drivers/accel/rocket/rocket_gem.c:rocket_ioctl_create_bo() {
	...
	ret = drm_gem_handle_create(file, gem_obj, &args->handle);
	drm_gem_object_put(gem_obj);
	if (ret)
		goto err;
	...
err:
	drm_gem_shmem_object_free(gem_obj);
}

drm_gem_handle_create() successfully exposes gem_obj to userspace and increments its handle refcount, after which the local reference is correctly dropped via drm_gem_object_put().

If a subsequent step fails (such as drm_gem_shmem_get_pages_sgt()), the error path jumps to err: and manually calls drm_gem_shmem_object_free(gem_obj).

This explicitly frees the backing memory while the object is still actively referenced by the handle table. When userspace later closes the handle, will the GEM core double-free the object?

[Severity: High]
This is a pre-existing issue, but should the asynchronous reset work be canceled before the DRM scheduler is destroyed?

drivers/accel/rocket/rocket_job.c:rocket_job_fini() {
	drm_sched_fini(&core->sched);

	cancel_work_sync(&core->reset.work);
	destroy_workqueue(core->reset.wq);
}

drm_sched_fini(&core->sched) is called before cancel_work_sync(&core->reset.work).

If the background reset task rocket_reset_work() is running concurrently, it will execute rocket_reset(), which attempts to call drm_sched_stop() and drm_sched_start() on the core->sched instance that has already been torn down or freed.

[Severity: High]
This is a pre-existing issue, but could the shared interrupt handler access hardware registers without checking the PM state, and does suspend execute without synchronizing IRQs?

drivers/accel/rocket/rocket_job.c:rocket_job_irq_handler() {
	struct rocket_core *core = data;
	u32 raw_status = rocket_pc_readl(core, INTERRUPT_RAW_STATUS);
	...
}
drivers/accel/rocket/rocket_drv.c:rocket_device_runtime_suspend() {
	...
	clk_bulk_disable_unprepare(ARRAY_SIZE(rdev->cores[core].clks), rdev->cores[core].clks);

	return 0;
}

The driver requests its interrupt using IRQF_SHARED, meaning rocket_job_irq_handler() can execute at any time if another device on the same line interrupts the CPU.

When called, the handler immediately reads hardware registers (INTERRUPT_RAW_STATUS) without using pm_runtime_get_if_active() to verify the device is powered. If the device was suspended, will this read return garbage or cause a bus fault?

Additionally, the rocket_device_runtime_suspend() callback disables the device's clocks without invoking synchronize_irq(), which might allow hardware to lose power while an interrupt handler is executing mid-flight.

[Severity: High]
This is a pre-existing issue, but can a failure in rocket_job_run() leak the PM usage counter and return an unsignaled hardware fence, hanging the DRM scheduler?

drivers/accel/rocket/rocket_job.c:rocket_job_run() {
	...
	ret = pm_runtime_get_sync(core->dev);
	if (ret < 0)
		return fence;
	...
		ret = iommu_attach_group(job->domain->domain, core->iommu_group);
		if (ret < 0)
			return fence;
	...
}
drivers/accel/rocket/rocket_job.c:rocket_reset() {
	scoped_guard(mutex, &core->job_lock) {
		if (core->in_flight_job)
			pm_runtime_put_noidle(core->dev);

		core->in_flight_job = NULL;
	}
}

If pm_runtime_get_sync() returns an error, it returns the created fence directly. Since pm_runtime_get_sync() increments the usage counter even upon failure, this leaks the PM reference (missing pm_runtime_put_noidle()).

Similarly, if iommu_attach_group() fails shortly after, the function again returns the fence without dropping the successfully acquired PM reference.

Because the hardware submission is skipped in both error paths, the returned fence is never signaled. The DRM scheduler will block indefinitely waiting for it, triggering a job timeout.

Will the timeout reset logic (rocket_reset()) fail to clean up these leaked PM references because core->in_flight_job is evaluated as NULL (it was never assigned before the early returns)?

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +5 -1
+static const struct rocket_soc_data rk3588_soc_data = {
+	.dma_bits = 40,
+};
+
 static const struct of_device_id dt_match[] = {
-	{ .compatible = "rockchip,rk3588-rknn-core" },
+	{ .compatible = "rockchip,rk3588-rknn-core", .data = &rk3588_soc_data },
 	{}
 };
 MODULE_DEVICE_TABLE(of, dt_match);
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#3 lore raw

RFC PATCH v3 2/9 accel: rocket: Derive DMA width and core count from match data

In reply to Midgy BALON

The probe already has the per-SoC match data, which now records the core count and DMA width. Use it for the cores array allocation and the device DMA mask instead of re-scanning the device tree for available core nodes.

---
 drivers/accel/rocket/rocket_core.h   |  2 ++
 drivers/accel/rocket/rocket_device.c | 15 +++++----------
 drivers/accel/rocket/rocket_device.h |  3 ++-
 drivers/accel/rocket/rocket_drv.c    |  7 ++++++-
 4 files changed, 15 insertions(+), 12 deletions(-)
drivers/accel/rocket/rocket_core.h +2 -0
index 8ee105a0be40e..d6421251670dc 100644
@@ -16,9 +16,11 @@ struct rocket_core;
 
 /**
  * struct rocket_soc_data - per-SoC configuration data
+ * @num_cores: Number of NPU cores in this SoC.
  * @dma_bits: Physical address width reachable by the NPU's AXI master.
  */
 struct rocket_soc_data {
+	unsigned int num_cores;
 	unsigned int dma_bits;
 };
 
drivers/accel/rocket/rocket_device.c +5 -10
index 46e6ee1e72c5f..6186f4faa3a2a 100644
@@ -6,18 +6,16 @@
 #include <linux/clk.h>
 #include <linux/dma-mapping.h>
 #include <linux/platform_device.h>
-#include <linux/of.h>
 
 #include "rocket_device.h"
 
 struct rocket_device *rocket_device_init(struct platform_device *pdev,
-					 const struct drm_driver *rocket_drm_driver)
+					 const struct drm_driver *rocket_drm_driver,
+					 const struct rocket_soc_data *soc_data)
 {
 	struct device *dev = &pdev->dev;
-	struct device_node *core_node;
 	struct rocket_device *rdev;
 	struct drm_device *ddev;
-	unsigned int num_cores = 0;
 	int err;
 
 	rdev = devm_drm_dev_alloc(dev, rocket_drm_driver, struct rocket_device, ddev);
@@ -27,17 +25,14 @@ struct rocket_device *rocket_device_init(struct platform_device *pdev,
 	ddev = &rdev->ddev;
 	dev_set_drvdata(dev, rdev);
 
-	for_each_compatible_node(core_node, NULL, "rockchip,rk3588-rknn-core")
-		if (of_device_is_available(core_node))
-			num_cores++;
-
-	rdev->cores = devm_kcalloc(dev, num_cores, sizeof(*rdev->cores), GFP_KERNEL);
+	rdev->cores = devm_kcalloc(dev, soc_data->num_cores, sizeof(*rdev->cores),
+				   GFP_KERNEL);
 	if (!rdev->cores)
 		return ERR_PTR(-ENOMEM);
 
 	dma_set_max_seg_size(dev, UINT_MAX);
 
-	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(40));
+	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(soc_data->dma_bits));
 	if (err)
 		return ERR_PTR(err);
 
drivers/accel/rocket/rocket_device.h +2 -1
index ce662abc01d3d..2f74e078974e3 100644
@@ -22,7 +22,8 @@ struct rocket_device {
 };
 
 struct rocket_device *rocket_device_init(struct platform_device *pdev,
-					 const struct drm_driver *rocket_drm_driver);
+					 const struct drm_driver *rocket_drm_driver,
+					 const struct rocket_soc_data *soc_data);
 void rocket_device_fini(struct rocket_device *rdev);
 #define to_rocket_device(drm_dev) \
 	((struct rocket_device *)(container_of((drm_dev), struct rocket_device, ddev)))
drivers/accel/rocket/rocket_drv.c +6 -1
index 384c38e13acce..c18840e5aff76 100644
@@ -159,11 +159,15 @@ static const struct drm_driver rocket_drm_driver = {
 
 static int rocket_probe(struct platform_device *pdev)
 {
+	const struct rocket_soc_data *soc_data = of_device_get_match_data(&pdev->dev);
 	int ret;
 
+	if (!soc_data)
+		return -EINVAL;
+
 	if (rdev == NULL) {
 		/* First core probing, initialize DRM device. */
-		rdev = rocket_device_init(drm_dev, &rocket_drm_driver);
+		rdev = rocket_device_init(drm_dev, &rocket_drm_driver, soc_data);
 		if (IS_ERR(rdev)) {
 			dev_err(&pdev->dev, "failed to initialize rocket device\n");
 			return PTR_ERR(rdev);
@@ -214,6 +218,7 @@ static void rocket_remove(struct platform_device *pdev)
 }
 
 static const struct rocket_soc_data rk3588_soc_data = {
+	.num_cores = 3,
 	.dma_bits = 40,
 };
 
Signature · 1 line
2.39.5
#4 lore raw

Re: [RFC PATCH v3 2/9] accel: rocket: Derive DMA width and core count from match data

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 5 potential issue(s) to consider:

New issues:
- [High] Missing bounds check on rdev->num_cores against soc_data->num_cores causes an out-of-bounds array write.

Pre-existing issues:
- [Critical] Early registration of the DRM device creates a race window leading to a NULL pointer dereference in rocket_open.
- [High] Array indexing mismatch on unbind/rebind causes state corruption and failed lookups.
- [High] Race condition during global device initialization and array modification due to lack of locking.
- [High] Failure in rocket_device_init() leaves an ERR_PTR in the global rdev variable, causing a panic on the next probe. --

commit d76917aa771a1a958c02a6121211647bc25a0160
Author: Midgy BALON <[email protected]>

accel: rocket: Derive DMA width and core count from match data

The probe already has the per-SoC match data, which now records the core count and DMA width. Use it for the cores array allocation and the device DMA mask instead of re-scanning the device tree for available core nodes.

Quoted 4 lines
drivers/accel/rocket/rocket_device.c +0 -0
index 46e6ee1e72c5f..6186f4faa3a2a 100644

[ ... ]

drivers/accel/rocket/rocket_device.c struct rocket_device *rocket_device_init(struct platform_device *pdev, · line 29

Quoted 11 lines
drivers/accel/rocket/rocket_device.c +2 -5
@@ -27,17 +25,14 @@ struct rocket_device *rocket_device_init(struct platform_device *pdev,
 	ddev = &rdev->ddev;
 	dev_set_drvdata(dev, rdev);
 
-	for_each_compatible_node(core_node, NULL, "rockchip,rk3588-rknn-core")
-		if (of_device_is_available(core_node))
-			num_cores++;
-
-	rdev->cores = devm_kcalloc(dev, num_cores, sizeof(*rdev->cores), GFP_KERNEL);
+	rdev->cores = devm_kcalloc(dev, soc_data->num_cores, sizeof(*rdev->cores),
+				   GFP_KERNEL);

[Severity: Critical]
This is a pre-existing issue, but does rocket_device_init() register the DRM device prematurely?

The first core to probe invokes rocket_device_init(), which immediately calls drm_dev_register() at the end of the function. This exposes the DRM device to userspace before the caller can initialize the core array elements.

If userspace quickly opens the DRM device node:

rocket_open()
rocket_iommu_domain_create(rdev->cores[0].dev)

Since the device pointer in the array has not yet been assigned in rocket_probe() and is still NULL, this results in a NULL pointer dereference.

Quoted 15 lines
drivers/accel/rocket/rocket_device.c +1 -1
 	if (!rdev->cores)
 		return ERR_PTR(-ENOMEM);
 
 	dma_set_max_seg_size(dev, UINT_MAX);
 
-	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(40));
+	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(soc_data->dma_bits));
 	if (err)
 		return ERR_PTR(err);
 
drivers/accel/rocket/rocket_drv.c +0 -0
index 384c38e13acce..c18840e5aff76 100644

drivers/accel/rocket/rocket_drv.c static const struct drm_driver rocket_drm_driver = { · line 168

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +4 -0
@@ -159,11 +159,15 @@ static const struct drm_driver rocket_drm_driver = {
 
 static int rocket_probe(struct platform_device *pdev)
 {
+	const struct rocket_soc_data *soc_data = of_device_get_match_data(&pdev->dev);
 	int ret;
 
+	if (!soc_data)
+		return -EINVAL;
+
 	if (rdev == NULL) {

[Severity: High]
This isn't a bug introduced by this patch, but does this lack locking?

The kernel driver core can probe multiple matching rockchip,rk3588-rknn-core devices in parallel. Concurrent probes could observe rdev == NULL simultaneously, instantiating the DRM device multiple times and leaking memory. Additionally, concurrent reads and increments of rdev->num_cores can cause multiple threads to overwrite the same array slot in rdev->cores[] while leaving others uninitialized.

drivers/accel/rocket/rocket_drv.c static const struct drm_driver rocket_drm_driver = { · line 170

Quoted 3 lines
drivers/accel/rocket/rocket_drv.c +1 -1
 		/* First core probing, initialize DRM device. */
-		rdev = rocket_device_init(drm_dev, &rocket_drm_driver);
+		rdev = rocket_device_init(drm_dev, &rocket_drm_driver, soc_data);

[Severity: High]
This is a pre-existing issue, but what happens if rocket_device_init() fails?

If rocket_device_init() returns an error pointer, it is directly assigned to the global rdev variable. The error path correctly returns the error but fails to reset the static global rdev back to NULL. When the driver core invokes rocket_probe() for the next matching device tree node, the rdev == NULL check evaluates to false because rdev contains an error pointer. The probe will then attempt to access rdev->num_cores, causing an invalid pointer dereference.

drivers/accel/rocket/rocket_drv.c

Quoted 5 lines
drivers/accel/rocket/rocket_drv.c +0 -0
 		if (IS_ERR(rdev)) {
 			dev_err(&pdev->dev, "failed to initialize rocket device\n");
 			return PTR_ERR(rdev);
drivers/accel/rocket/rocket_drv.c +0 -0
 		}
 	}

[Severity: High]
Can this cause an out-of-bounds array write?

The allocation of the rdev->cores array now uses a fixed size defined by soc_data->num_cores instead of dynamically counting the device tree nodes. However, rocket_probe() is called by the driver core for every matching DT node, and it unconditionally uses the auto-incrementing rdev->num_cores as the array index later in the function.

If a malformed or custom device tree contains more core nodes than soc_data->num_cores, rocket_probe() will be invoked more times than the array can hold, writing past the end of the rdev->cores buffer.

[ ... ]

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 218

Quoted 2 lines
drivers/accel/rocket/rocket_drv.c +0 -0
@@ -214,6 +218,7 @@ static void rocket_remove(struct platform_device *pdev)
 }

[Severity: High]
This isn't a bug introduced by this patch, but does the array indexing handle unbind and rebind correctly?

When a core is unbound, rocket_remove() decrements rdev->num_cores without compacting the remaining active cores in the rdev->cores array to fill the empty slot. Consequently, find_core_for_dev() (which relies on rdev->num_cores for its loop bound) fails to find any remaining cores located at indices >= the new rdev->num_cores, preventing their cleanup.

Furthermore, if a core is re-bound, rocket_probe() will use the decremented rdev->num_cores as the index and overwrite an existing active core, causing resource leaks and state corruption.

Quoted 5 lines
drivers/accel/rocket/rocket_drv.c +1 -0
 static const struct rocket_soc_data rk3588_soc_data = {
+	.num_cores = 3,
 	.dma_bits = 40,
 };
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
#5 lore raw

RFC PATCH v3 3/9 accel: rocket: Add RK3568 SoC support

In reply to Midgy BALON

The RK3568 has a single core of the same NVDLA-derived NPU IP as the RK3588, with a 32-bit AXI master. Unlike the RK3588 it must be powered on and de-idled through the PMU, and its PVTPLL clock started via SCMI, before the NPU is reachable. Add rk3568_soc_data with an noc_init callback performing this bring-up.

---
 drivers/accel/rocket/rocket_core.c |  9 +++++
 drivers/accel/rocket/rocket_core.h |  3 ++
 drivers/accel/rocket/rocket_drv.c  | 53 ++++++++++++++++++++++++++++++
 3 files changed, 65 insertions(+)
drivers/accel/rocket/rocket_core.c +9 -0
index 09c445af7de73..a8de876365873 100644
@@ -88,6 +88,15 @@ int rocket_core_init(struct rocket_core *core)
 		return err;
 	}
 
+	if (core->soc_data->noc_init) {
+		err = core->soc_data->noc_init(core);
+		if (err) {
+			pm_runtime_put_sync(dev);
+			rocket_job_fini(core);
+			return err;
+		}
+	}
+
 	version = rocket_pc_readl(core, VERSION);
 	version += rocket_pc_readl(core, VERSION_NUM) & 0xffff;
 
drivers/accel/rocket/rocket_core.h +3 -0
index d6421251670dc..66d138a8ed773 100644
@@ -18,10 +18,13 @@ struct rocket_core;
  * struct rocket_soc_data - per-SoC configuration data
  * @num_cores: Number of NPU cores in this SoC.
  * @dma_bits: Physical address width reachable by the NPU's AXI master.
+ * @noc_init: Optional callback to power on and de-idle the NPU NOC bus.
+ *            Required on RK3568, where this is done through the PMU.
  */
 struct rocket_soc_data {
 	unsigned int num_cores;
 	unsigned int dma_bits;
+	int (*noc_init)(struct rocket_core *core);
 };
 
 #define rocket_pc_readl(core, reg) \
drivers/accel/rocket/rocket_drv.c +53 -0

74 lines not on this page · raw message

Signature · 1 line
2.39.5
#6 lore raw

Re: [RFC PATCH v3 3/9] accel: rocket: Add RK3568 SoC support

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

New issues:
- [High] The error path for noc_init failure in rocket_core_init calls rocket_job_fini() instead of rocket_core_fini(), leaking an IOMMU group reference and leaving PM runtime active.

Pre-existing issues:
- [High] The core->clks array lacks initialization of its id fields before being passed to devm_clk_bulk_get(), causing the driver to configure the rate on the wrong clock in rk3568_noc_init().
- [High] Failed probe leaks struct rocket_device and associated DRM structures because they are devm-allocated against a global persistent platform device.
--

commit 60f18710aa215157b0b6ac91ed70f00cc049eb2c
Author: Midgy BALON <[email protected]>

accel: rocket: Add RK3568 SoC support

The RK3568 has a single core of the same NVDLA-derived NPU IP as the RK3588, with a 32-bit AXI master. Unlike the RK3588 it must be powered on and de-idled through the PMU, and its PVTPLL clock started via SCMI, before the NPU is reachable. Add rk3568_soc_data with an noc_init callback performing this bring-up.

drivers/accel/rocket/rocket_core.c int rocket_core_init(struct rocket_core *core) · line 95

Quoted 13 lines
drivers/accel/rocket/rocket_core.c +5 -0
index 09c445af7de73..a8de876365873 100644
@@ -88,6 +88,15 @@ int rocket_core_init(struct rocket_core *core)
 		return err;
 	}
 
+	if (core->soc_data->noc_init) {
+		err = core->soc_data->noc_init(core);
+		if (err) {
+			pm_runtime_put_sync(dev);
+			rocket_job_fini(core);

[Severity: High]
If noc_init() fails, should this call rocket_core_fini() instead of rocket_job_fini()?

Calling rocket_job_fini() here bypasses the rest of the teardown that rocket_core_fini() handles. This leaves the PM runtime enabled and autosuspend incorrectly configured, and skips iommu_group_put() which leaks the IOMMU group reference.

Quoted 5 lines
drivers/accel/rocket/rocket_core.c +4 -0
+			return err;
+		}
+	}
+
 	version = rocket_pc_readl(core, VERSION);

[ ... ]

Quoted 4 lines
drivers/accel/rocket/rocket_drv.c +0 -0
index c18840e5aff76..5a72d0b5f4dff 100644

[ ... ]

Quoted 15 lines
drivers/accel/rocket/rocket_drv.c +11 -0
@@ -217,12 +219,63 @@ static void rocket_remove(struct platform_device *pdev)
 	}
 }
 
+/*
+ * On RK3568 the NPU NOC bus is gated and idle out of reset and must be
+ * powered on and de-idled through the PMU before the NPU is reachable.  PMU
+ * registers use a write-mask protocol: the upper 16 bits enable writes to the
+ * matching lower 16 bits.
+ *
+ * The NPU's high-speed clock is a PVTPLL managed by TF-A via SCMI and must be
+ * running before the NOC acknowledges the de-idle request.  Force a real SCMI
+ * rate change (an intermediate rate defeats the clock framework's
+ * unchanged-rate shortcut) now that the power domain is on and clocks enabled.
+ */

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 242

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +10 -0
+#define ROCKET_RK3568_SCMI_CLK	2
+
+static int rk3568_noc_init(struct rocket_core *core)
+{
+	struct regmap *pmu;
+	unsigned int val;
+	int ret;
+
+	clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 600000000UL);
+	clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 1000000000UL);

[Severity: High]
This is a pre-existing issue, but does core->clks actually contain the correct SCMI clock here?

Looking at rocket_core_init(), devm_clk_bulk_get() is called without first initializing the .id fields of the core->clks array:

drivers/accel/rocket/rocket_core.c:rocket_core_init() {
    ...
    err = devm_clk_bulk_get(dev, ARRAY_SIZE(core->clks), core->clks);
    ...
}

Because the IDs are uninitialized (NULL), clk_get() will repeatedly return the very first clock of the device. Does this mean these clk_set_rate() calls are erroneously forcing a rate change on the wrong clock while leaving the PVTPLL unconfigured?

[ ... ]

[Severity: High]
This is a pre-existing issue, but could probe failures (such as noc_init returning an error) cause an unbounded memory leak?

In rocket_probe(), the global drm_dev is passed to rocket_device_init():

drivers/accel/rocket/rocket_drv.c:rocket_probe() {
    ...
    rdev = rocket_device_init(drm_dev, &rocket_drm_driver, soc_data);
    ...
}

Inside rocket_device_init(), memory is allocated against this global device's devres list:

drivers/accel/rocket/rocket_device.c:rocket_device_init() {
    ...
    rdev = devm_drm_dev_alloc(dev, rocket_drm_driver, struct rocket_device, ddev);
    ...
}

If probe fails, rocket_probe() calls rocket_device_fini() to unregister the DRM device. However, it does not free the devm-allocated memory because the allocations are tied to the global drm_dev that persists until module unload. Does this mean every deferred probe or bind/unbind action allocates new memory on the global devres list?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=3
#7 lore raw

RFC PATCH v3 4/9 accel: rocket: Reset the NPU before detaching the IOMMU on timeout

In reply to Midgy BALON

On a job timeout the NPU AXI master can be left wedged with outstanding transactions. rocket_reset() detached the IOMMU group before resetting the hardware, so iommu_detach_group() -> __iommu_group_set_core_domain() asked the rk_iommu to stall and wait for the in-flight transactions to drain. They never did, the stall request timed out (-ETIMEDOUT) and the IOMMU core WARNed:

  WARNING: drivers/iommu/iommu.c:157 __iommu_group_set_core_domain
    iommu_detach_group
    rocket_reset
    rocket_job_timedout

Assert the core reset first: it quiesces the AXI master so the following IOMMU detach completes cleanly. Move the detach after rocket_core_reset() and out of the job_lock (it does not touch in_flight_job).

---
 drivers/accel/rocket/rocket_job.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
drivers/accel/rocket/rocket_job.c +9 -3
index ac51bff39833f..e25234261536b 100644
@@ -364,14 +364,20 @@ rocket_reset(struct rocket_core *core, struct drm_sched_job *bad)
 		if (core->in_flight_job)
 			pm_runtime_put_noidle(core->dev);
 
-		iommu_detach_group(NULL, core->iommu_group);
-
 		core->in_flight_job = NULL;
 	}
 
-	/* Proceed with reset now. */
+	/*
+	 * Reset the NPU hardware before detaching the IOMMU. A timed-out job
+	 * leaves the NPU AXI master wedged; detaching the IOMMU then issues a
+	 * stall request that never drains and times out (warning in the IOMMU
+	 * core). Asserting the core reset first quiesces the master so the
+	 * detach completes cleanly.
+	 */
 	rocket_core_reset(core);
 
+	iommu_detach_group(NULL, core->iommu_group);
+
 	/* NPU has been reset, we can clear the reset pending bit. */
 	atomic_set(&core->reset.pending, 0);
 
Signature · 1 line
2.39.5
#8 lore raw

Re: [RFC PATCH v3 4/9] accel: rocket: Reset the NPU before detaching the IOMMU on timeout

Thank you for your contribution! Sashiko AI review found 9 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 9 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Double free of rjob->tasks array when copy_from_user() fails.
- [High] The shared IRQ handler unconditionally accesses hardware registers without verifying the device's PM active state.
- [High] Missing PM reference drop and unsignaled fence return on error paths in rocket_job_run().
- [High] Concurrent hardware reset and IRQ handler execution can cause an external abort.
- [High] Unbalanced teardown order in rocket_core_fini() causes UAF and data races.
- [High] NULL pointer dereference of job->domain in rocket_job_cleanup() on early error paths.
- [High] Missing synchronize_irq() in runtime suspend callback.
- [Medium] The TDR reset path permanently leaks power by preventing the device from entering runtime suspend.
- [Medium] The hardware reset is performed after dropping the PM runtime reference, potentially targeting an unpowered device. --

commit 323afa3b038052ef6c02ee0548658a65126910a0
Author: Midgy BALON <[email protected]>

accel: rocket: Reset the NPU before detaching the IOMMU on timeout

On a job timeout the NPU AXI master can be left wedged with outstanding transactions. rocket_reset() detached the IOMMU group before resetting the hardware, causing the IOMMU core to stall. Assert the core reset first to quiesce the master so the detach completes cleanly.

drivers/accel/rocket/rocket_job.c rocket_reset(struct rocket_core *core, struct drm_sched_job *bad) · line 365

Quoted 6 lines
drivers/accel/rocket/rocket_job.c +0 -0
@@ -364,14 +364,20 @@ rocket_reset(struct rocket_core *core, struct drm_sched_job *bad)
 		if (core->in_flight_job)
 			pm_runtime_put_noidle(core->dev);

[Severity: Medium]
This is a pre-existing issue, but does this TDR reset path permanently leak power by preventing the device from entering runtime suspend?

It appears pm_runtime_put_noidle() drops the reference without triggering an idle power-down. If the timed-out job was the last active job, the usage count reaches zero but the device remains in the RPM_ACTIVE state indefinitely, staying powered on until another job is submitted and correctly triggers autosuspend.

Quoted 4 lines
drivers/accel/rocket/rocket_job.c +0 -2
-		iommu_detach_group(NULL, core->iommu_group);
-
 		core->in_flight_job = NULL;
 	}

drivers/accel/rocket/rocket_job.c rocket_reset(struct rocket_core *core, struct drm_sched_job *bad) · line 377

Quoted 10 lines
drivers/accel/rocket/rocket_job.c +7 -1
 
-	/* Proceed with reset now. */
+	/*
+	 * Reset the NPU hardware before detaching the IOMMU. A timed-out job
+	 * leaves the NPU AXI master wedged; detaching the IOMMU then issues a
+	 * stall request that never drains and times out (warning in the IOMMU
+	 * core). Asserting the core reset first quiesces the master so the
+	 * detach completes cleanly.
+	 */
 	rocket_core_reset(core);

[Severity: High]
This isn't a bug introduced by this patch, but can a concurrent hardware reset and IRQ handler execution cause an external abort?

rocket_core_reset() asserts a hardware reset without first disabling device interrupts. If an interrupt fires just before or during the reset, rocket_job_irq_handler() will execute and perform MMIO reads and writes to the bus while the hardware is actively in reset.

[Severity: Medium]
This is also a pre-existing issue, but is it safe to perform the hardware reset after dropping the PM runtime reference?

Dropping the PM reference via pm_runtime_put_noidle() earlier decrements the usage count. A concurrently scheduled autosuspend timer could expire precisely in this window and power down the device before or during the reset operations.

Asserting resets on a suspended device can lead to system bus faults. Should the PM reference be dropped after the hardware reset is complete?

drivers/accel/rocket/rocket_job.c rocket_reset(struct rocket_core *core, struct drm_sched_job *bad) · line 382

Quoted 5 lines
drivers/accel/rocket/rocket_job.c +3 -0
+
+	iommu_detach_group(NULL, core->iommu_group);
+
 	/* NPU has been reset, we can clear the reset pending bit. */
 	atomic_set(&core->reset.pending, 0);

[Severity: High]
This isn't a bug introduced by this patch, but does the shared IRQ handler unconditionally access hardware registers without verifying the device's PM active state?

drivers/accel/rocket/rocket_job.c:rocket_job_irq_handler() {
    ...
    u32 raw_status = rocket_pc_readl(core, INTERRUPT_RAW_STATUS);
    ...
}

The driver registers rocket_job_irq_handler() with the IRQF_SHARED flag. Without first calling pm_runtime_get_if_active(), if the NPU is runtime suspended, this read accesses an unpowered register domain. This typically returns 0xffffffff (causing spurious handling) or triggers a synchronous external abort.

[Severity: High]
This is a pre-existing issue, but does rocket_job_run() leak the PM reference and return an unsignaled fence on error paths?

drivers/accel/rocket/rocket_job.c:rocket_job_run() {
    ...
    ret = pm_runtime_get_sync(core->dev);
    if (ret < 0)
        return fence;
    ...
}

If pm_runtime_get_sync() or iommu_attach_group() fails, the function returns a newly created, un-signaled fence instead of an error pointer, and leaves the PM usage counter incremented. Because the hardware submission step is skipped, the hardware will never signal this fence, causing the DRM scheduler to hang indefinitely.

[Severity: High]
This is a pre-existing issue, but does the teardown order in rocket_core_fini() cause a use-after-free and data races?

drivers/accel/rocket/rocket_core.c:rocket_core_fini() {
    ...
    iommu_group_put(core->iommu_group);
    core->iommu_group = NULL;
    rocket_job_fini(core);
    ...
}

Resources are torn down in the wrong order. The iommu_group and attached_domain members are freed, and pm_runtime_disable() is called before rocket_job_fini().

If a scheduled job is processed by the DRM scheduler during this window, rocket_job_run() will concurrently access the now-NULL iommu_group without locking. Should rocket_job_fini() be called first to stop the scheduler and wait for jobs?

[Severity: Critical]
This isn't a bug introduced by this patch, but is there a double free of the rjob->tasks array when copy_from_user() fails?

drivers/accel/rocket/rocket_job.c:rocket_copy_tasks() {
    ...
fail:
    kvfree(rjob->tasks);
    return ret;
}

If copy_from_user() fails, it frees rjob->tasks but leaves the dangling pointer.

The function returns an error, prompting rocket_ioctl_submit_job() to jump to out_cleanup_job and call rocket_job_put(rjob). This invokes rocket_job_cleanup(), which unconditionally calls kvfree(job->tasks), resulting in a double-free.

[Severity: High]
This is a pre-existing issue, but does rocket_job_cleanup() dereference a NULL job->domain on early error paths?

drivers/accel/rocket/rocket_job.c:rocket_job_cleanup() {
    ...
    rocket_iommu_domain_put(job->domain);
    ...
}

The job->domain structure is initially NULL and only populated late in the function. If an earlier step fails (such as rocket_copy_tasks()), the cleanup unconditionally calls rocket_iommu_domain_put(job->domain).

This wrapper dereferences the domain struct to access its kref without a NULL check:

drivers/accel/rocket/rocket_drv.c:rocket_iommu_domain_put() {
    kref_put(&domain->kref, rocket_iommu_domain_destroy);
}

[Severity: High]
This isn't a bug introduced by this patch, but is there a missing synchronize_irq() in the runtime suspend callback?

drivers/accel/rocket/rocket_drv.c:rocket_device_runtime_suspend() {
    ...
    clk_bulk_disable_unprepare(ARRAY_SIZE(rdev->cores[core].clks),
                               rdev->cores[core].clks);
    ...
}

The driver registers a shared interrupt but fails to call synchronize_irq(). Disabling the clocks while the IRQ handler is executing concurrently on another CPU could cause an AXI bus fault or a hard system crash.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4
#9 lore raw

RFC PATCH v3 5/9 accel: rocket: Keep the IOMMU domain attached across jobs

In reply to Midgy BALON

rocket attached the job's IOMMU domain in rocket_job_run() and detached it again on every completion and reset. Each attach/detach toggles the rk_iommu stall/force-reset/paging handshake, and on RK3568 the NPU MMU is idle between jobs, so that handshake times out and logs a burst of "stall/paging request timed out" errors for every job.

Attach the per-context domain once and keep it: track the attached domain in the core, swap it only when a job from a different context runs, and detach it at core teardown. A reference on the attached domain is held so it outlives the job that first attached it and is released on swap/teardown.

---
 drivers/accel/rocket/rocket_core.c |  6 ++++++
 drivers/accel/rocket/rocket_core.h |  3 +++
 drivers/accel/rocket/rocket_job.c  | 27 +++++++++++++++++++++------
 3 files changed, 30 insertions(+), 6 deletions(-)
drivers/accel/rocket/rocket_core.c +6 -0
index a8de876365873..634f78dfe2887 100644
@@ -13,6 +13,7 @@
 #include <linux/reset.h>
 
 #include "rocket_core.h"
+#include "rocket_drv.h"
 #include "rocket_job.h"
 
 int rocket_core_init(struct rocket_core *core)
@@ -112,6 +113,11 @@ void rocket_core_fini(struct rocket_core *core)
 {
 	pm_runtime_dont_use_autosuspend(core->dev);
 	pm_runtime_disable(core->dev);
+	if (core->attached_domain) {
+		iommu_detach_group(NULL, core->iommu_group);
+		rocket_iommu_domain_put(core->attached_domain);
+		core->attached_domain = NULL;
+	}
 	iommu_group_put(core->iommu_group);
 	core->iommu_group = NULL;
 	rocket_job_fini(core);
drivers/accel/rocket/rocket_core.h +3 -0
index 66d138a8ed773..05a197a9c0113 100644
@@ -42,6 +42,8 @@ struct rocket_soc_data {
 #define rocket_core_writel(core, reg, value) \
 	writel(value, (core)->core_iomem + (REG_CORE_##reg) - REG_CORE_S_STATUS)
 
+struct rocket_iommu_domain;
+
 struct rocket_core {
 	struct device *dev;
 	struct rocket_device *rdev;
@@ -56,6 +58,7 @@ struct rocket_core {
 	struct reset_control_bulk_data resets[2];
 
 	struct iommu_group *iommu_group;
+	struct rocket_iommu_domain *attached_domain;
 
 	struct mutex job_lock;
 	struct rocket_job *in_flight_job;
drivers/accel/rocket/rocket_job.c +21 -6
index e25234261536b..b248371be8a1e 100644
@@ -9,6 +9,7 @@
 #include <drm/rocket_accel.h>
 #include <linux/interrupt.h>
 #include <linux/iommu.h>
+#include <linux/kref.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
 
@@ -314,9 +315,26 @@ static struct dma_fence *rocket_job_run(struct drm_sched_job *sched_job)
 	if (ret < 0)
 		return fence;
 
-	ret = iommu_attach_group(job->domain->domain, core->iommu_group);
-	if (ret < 0)
-		return fence;
+	/*
+	 * Attach the job's IOMMU domain only when it differs from the one
+	 * already attached. Re-attaching per job toggles the rk_iommu
+	 * stall/reset handshake on an idle NPU MMU, which is slow and
+	 * noisy; keep the domain attached across jobs instead.
+	 */
+	if (core->attached_domain != job->domain) {
+		if (core->attached_domain) {
+			iommu_detach_group(NULL, core->iommu_group);
+			rocket_iommu_domain_put(core->attached_domain);
+			core->attached_domain = NULL;
+		}
+
+		ret = iommu_attach_group(job->domain->domain, core->iommu_group);
+		if (ret < 0)
+			return fence;
+
+		kref_get(&job->domain->kref);
+		core->attached_domain = job->domain;
+	}
 
 	scoped_guard(mutex, &core->job_lock) {
 		core->in_flight_job = job;
@@ -340,7 +358,6 @@ static void rocket_job_handle_irq(struct rocket_core *core)
 				return;
 			}
 
-			iommu_detach_group(NULL, iommu_group_get(core->dev));
 			dma_fence_signal(core->in_flight_job->done_fence);
 			pm_runtime_put_autosuspend(core->dev);
 			core->in_flight_job = NULL;
@@ -376,8 +393,6 @@ rocket_reset(struct rocket_core *core, struct drm_sched_job *bad)
 	 */
 	rocket_core_reset(core);
 
-	iommu_detach_group(NULL, core->iommu_group);
-
 	/* NPU has been reset, we can clear the reset pending bit. */
 	atomic_set(&core->reset.pending, 0);
 
Signature · 1 line
2.39.5
#10 lore raw

Re: [RFC PATCH v3 5/9] accel: rocket: Keep the IOMMU domain attached across jobs

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

New issues:
- [High] The IOMMU hardware registers are left uninitialized after a job timeout and hardware reset, causing subsequent jobs to immediately fault.

Pre-existing issues:
- [High] Incorrect teardown ordering in rocket_core_fini() destroys IOMMU and PM resources before stopping the DRM scheduler, leading to data races, use-after-free, and potential hardware faults.
- [High] PM runtime usage counts and unsignaled fences are leaked on error paths in rocket_job_run().
--

commit 49c877f1ba7baae4a28268850b55e2ac3ab4a8cc
Author: Midgy BALON <[email protected]>

accel: rocket: Keep the IOMMU domain attached across jobs

rocket attached the job's IOMMU domain in rocket_job_run() and detached it again on every completion and reset. Each attach/detach toggles the rk_iommu stall/force-reset/paging handshake, and on RK3568 the NPU MMU is idle between jobs, so that handshake times out and logs a burst of "stall/paging request timed out" errors for every job.

Attach the per-context domain once and keep it: track the attached domain in the core, swap it only when a job from a different context runs, and detach it at core teardown. A reference on the attached domain is held so it outlives the job that first attached it and is released on swap/teardown.

Quoted 4 lines
drivers/accel/rocket/rocket_core.c +0 -0
index a8de876365873..634f78dfe2887 100644

[ ... ]

drivers/accel/rocket/rocket_core.c void rocket_core_fini(struct rocket_core *core) · line 122

Quoted 11 lines
drivers/accel/rocket/rocket_core.c +5 -0
@@ -112,6 +113,11 @@ void rocket_core_fini(struct rocket_core *core)
 {
 	pm_runtime_dont_use_autosuspend(core->dev);
 	pm_runtime_disable(core->dev);
+	if (core->attached_domain) {
+		iommu_detach_group(NULL, core->iommu_group);
+		rocket_iommu_domain_put(core->attached_domain);
+		core->attached_domain = NULL;
+	}
 	iommu_group_put(core->iommu_group);
 	core->iommu_group = NULL;

[Severity: High]
This is a pre-existing issue, but IOMMU resources are being torn down before the DRM scheduler is stopped in rocket_job_fini(core).

Because the scheduler might still be active, couldn't an in-flight job be executing concurrently? This could lead to a data race on core->attached_domain, the IOMMU domain being forcefully unmapped from a running job causing AXI bus faults, or a use-after-free when rocket_job_run() accesses core->iommu_group.

Should rocket_job_fini(core) be called before tearing down the IOMMU state to ensure the hardware is idle and the scheduler is stopped?

Quoted 2 lines
drivers/accel/rocket/rocket_core.c +0 -0
 	rocket_job_fini(core);

}

[ ... ]

Quoted 4 lines
drivers/accel/rocket/rocket_job.c +0 -0
index e25234261536b..b248371be8a1e 100644

[ ... ]

drivers/accel/rocket/rocket_job.c static struct dma_fence *rocket_job_run(struct drm_sched_job *sched_job) · line 316

Quoted 3 lines
drivers/accel/rocket/rocket_job.c +0 -0
@@ -314,9 +315,26 @@ static struct dma_fence *rocket_job_run(struct drm_sched_job *sched_job)
 	if (ret < 0)
 		return fence;

[Severity: High]
This isn't a bug introduced by this patch, but if pm_runtime_get_sync() fails, it returns a negative error code but still increments the PM usage counter.

The error path returns without calling pm_runtime_put_noidle() or signaling the fence, which leaks the PM counter and causes the scheduler to wait indefinitely.

Quoted 9 lines
drivers/accel/rocket/rocket_job.c +6 -3
-	ret = iommu_attach_group(job->domain->domain, core->iommu_group);
-	if (ret < 0)
-		return fence;
+	/*
+	 * Attach the job's IOMMU domain only when it differs from the one
+	 * already attached. Re-attaching per job toggles the rk_iommu
+	 * stall/reset handshake on an idle NPU MMU, which is slow and
+	 * noisy; keep the domain attached across jobs instead.
+	 */

drivers/accel/rocket/rocket_job.c static struct dma_fence *rocket_job_run(struct drm_sched_job *sched_job) · line 333

Quoted 10 lines
drivers/accel/rocket/rocket_job.c +10 -0
+	if (core->attached_domain != job->domain) {
+		if (core->attached_domain) {
+			iommu_detach_group(NULL, core->iommu_group);
+			rocket_iommu_domain_put(core->attached_domain);
+			core->attached_domain = NULL;
+		}
+
+		ret = iommu_attach_group(job->domain->domain, core->iommu_group);
+		if (ret < 0)
+			return fence;

[Severity: High]
Also a pre-existing issue, but if iommu_attach_group() fails, it returns the fence early without dropping the successfully acquired PM reference. Should this path include a pm_runtime_put_noidle() call to avoid leaking the reference?

Quoted 7 lines
drivers/accel/rocket/rocket_job.c +4 -0
+
+		kref_get(&job->domain->kref);
+		core->attached_domain = job->domain;
+	}
 
 	scoped_guard(mutex, &core->job_lock) {
 		core->in_flight_job = job;

[ ... ]

drivers/accel/rocket/rocket_job.c rocket_reset(struct rocket_core *core, struct drm_sched_job *bad) · line 397

Quoted 8 lines
drivers/accel/rocket/rocket_job.c +0 -2
@@ -376,8 +393,6 @@ rocket_reset(struct rocket_core *core, struct drm_sched_job *bad)
 	 */
 	rocket_core_reset(core);
 
-	iommu_detach_group(NULL, core->iommu_group);
-
 	/* NPU has been reset, we can clear the reset pending bit. */
 	atomic_set(&core->reset.pending, 0);

[Severity: High]
When the NPU is hard-reset by rocket_core_reset(), the IOMMU's hardware registers (like page table base addresses) are wiped.

Since core->attached_domain is not cleared here, when the scheduler restarts and executes the next job, rocket_job_run() will see core->attached_domain matches the job's domain and skip calling iommu_attach_group().

Will this leave the IOMMU hardware unconfigured and cause subsequent jobs to immediately throw DMA faults? Should core->attached_domain be cleared here to force a re-attach and reprogramming of the IOMMU?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=5
#11 lore raw

RFC PATCH v3 6/9 iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

In reply to Midgy BALON

On the RK356x v1 IOMMU, RK_MMU_AUTO_GATING resets to 0x3. Bit 1 enables auto clock-gating of the page-table walker, so the walker's AXI master loses its clock between transactions; a TLB-miss page walk then never completes and the IOMMU is left stuck (PAGING_ENABLED, never IDLE).

Clear bit 1 (keeping bit 0, the slave-port gate) once paging is enabled so the walker keeps its clock. This is required for the RK3568 NPU MMU.

---
 drivers/iommu/rockchip-iommu.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
drivers/iommu/rockchip-iommu.c +12 -0
index 4da80136933c4..e3d8b6e9ca12b 100644
@@ -953,6 +953,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
 	ret = rk_iommu_enable_paging(iommu);
 
+	if (!ret) {
+		/*
+		 * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+		 * auto clock-gating; the walker's AXI master then loses its clock
+		 * between transactions and a TLB-miss page walk never completes,
+		 * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+		 * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+		 */
+		for (i = 0; i < iommu->num_mmu; i++)
+			rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);
+	}
+
 out_disable_stall:
 	rk_iommu_disable_stall(iommu);
 out_disable_clocks:
Signature · 1 line
2.39.5
#12 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [High] The patch writes 0x2 to RK_MMU_AUTO_GATING to clear bit 1, but this value actually sets bit 1 and…

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] The patch writes 0x2 to RK_MMU_AUTO_GATING to clear bit 1, but this value actually sets bit 1 and clears bit 0, doing the exact opposite of the intended behavior.
- [High] An SoC-specific workaround is applied unconditionally to all Rockchip IOMMUs using a brittle hardcoded register overwrite.
--

commit 86bf3e3b9a68d419b8e6ba9d4d3cf8afb0c324bb
Author: Midgy BALON <[email protected]>

iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

This commit aims to fix a hardware lockup on the RK356x v1 IOMMU by clearing bit 1 of the RK_MMU_AUTO_GATING register. This change is intended to prevent auto clock-gating of the page-table walker.

Quoted 7 lines
drivers/iommu/rockchip-iommu.c +0 -0
index 0013cf196c573..7d40ecb26123f 100644
@@ -952,6 +952,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
 	ret = rk_iommu_enable_paging(iommu);

drivers/iommu/rockchip-iommu.c static int rk_iommu_enable(struct rk_iommu *iommu) · line 963

Quoted 10 lines
drivers/iommu/rockchip-iommu.c +9 -0
 
+	if (!ret) {
+		/*
+		 * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+		 * auto clock-gating; the walker's AXI master then loses its clock
+		 * between transactions and a TLB-miss page walk never completes,
+		 * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+		 * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+		 */
+		for (i = 0; i < iommu->num_mmu; i++)

[Severity: High]
Will this loop unconditionally apply the RK356x v1 workaround to all SoCs supported by the rockchip-iommu driver? It seems older variants like the RK3288 or RK3399 might unexpectedly receive this register write in rk_iommu_enable() without a check for the specific SoC version.

drivers/iommu/rockchip-iommu.c static int rk_iommu_enable(struct rk_iommu *iommu) · line 965

Quoted 1 line
drivers/iommu/rockchip-iommu.c +1 -0
+			rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);

[Severity: High]
Does writing 0x2 here achieve the intended result?

The comment mentions clearing bit 1 and keeping bit 0. Since the register resets to 0x3 (binary 11), writing 0x2 (binary 10) actually sets bit 1 and clears bit 0. This seems to do the exact opposite of what is described, keeping the buggy page-walker auto clock-gating enabled and inadvertently disabling the working slave-port gate. Should this be 0x1 instead?

Additionally, because rk_iommu_write() completely overwrites the register, could this clobber other bits in RK_MMU_AUTO_GATING? Would it be safer to use a read-modify-write pattern here?

Quoted 4 lines
drivers/iommu/rockchip-iommu.c +2 -0
+	}
+
 out_disable_stall:
 	rk_iommu_disable_stall(iommu);
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6
maintainer
#13 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

In reply to Midgy BALON
Quoted 8 lines · Midgy BALON

On the RK356x v1 IOMMU, RK_MMU_AUTO_GATING resets to 0x3. Bit 1 enables auto clock-gating of the page-table walker, so the walker's AXI master loses its clock between transactions; a TLB-miss page walk then never completes and the IOMMU is left stuck (PAGING_ENABLED, never IDLE).

Clear bit 1 (keeping bit 0, the slave-port gate) once paging is enabled so the walker keeps its clock. This is required for the RK3568 NPU MMU.

Hi,

I'm not able to review this patch myself, but maybe it can be submitted separately while we work on the NPU bits?

Regards,

Tomeu

Quoted 31 lines
---
 drivers/iommu/rockchip-iommu.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
drivers/iommu/rockchip-iommu.c +12 -1
index 4da80136933c4..e3d8b6e9ca12b 100644
@@ -953,6 +953,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
        ret = rk_iommu_enable_paging(iommu);
 
+       if (!ret) {
+               /*
+                * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+                * auto clock-gating; the walker's AXI master then loses its clock
+                * between transactions and a TLB-miss page walk never completes,
+                * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+                * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+                */
+               for (i = 0; i < iommu->num_mmu; i++)
+                       rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);
+       }
+
 out_disable_stall:
        rk_iommu_disable_stall(iommu);
 out_disable_clocks:
--

2.39.5

#14 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

In reply to Midgy BALON

Hello Midgy,

Quoted 26 lines · Midgy BALON

On the RK356x v1 IOMMU, RK_MMU_AUTO_GATING resets to 0x3. Bit 1 enables auto clock-gating of the page-table walker, so the walker's AXI master loses its clock between transactions; a TLB-miss page walk then never completes and the IOMMU is left stuck (PAGING_ENABLED, never IDLE).

Clear bit 1 (keeping bit 0, the slave-port gate) once paging is enabled so the walker keeps its clock. This is required for the RK3568 NPU MMU.

---
 drivers/iommu/rockchip-iommu.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
drivers/iommu/rockchip-iommu.c +5 -0
index 4da80136933c4..e3d8b6e9ca12b 100644
@@ -953,6 +953,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
 	ret = rk_iommu_enable_paging(iommu);
 
+	if (!ret) {
+		/*
+		 * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+		 * auto clock-gating; the walker's AXI master then loses its clock
+		 * between transactions and a TLB-miss page walk never completes,

drivers/iommu/rockchip-iommu.c static int rk_iommu_enable(struct rk_iommu *iommu) · line 970

Quoted 10 lines
drivers/iommu/rockchip-iommu.c +7 -0
+		 * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+		 * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+		 */
+		for (i = 0; i < iommu->num_mmu; i++)
+			rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);
+	}
+
 out_disable_stall:
 	rk_iommu_disable_stall(iommu);
 out_disable_clocks:

As I said, it is v2. Could you please try using the code below instead and see if it works? Thank you.

drivers/iommu/rockchip-iommu.c +5 -0
index 0013cf196c57..89e3a83a0251 100644
@@ -930,6 +930,7 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
        struct iommu_domain *domain = iommu->domain;
        struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
        int ret, i;
+       u32 auto_gate;
 
        ret = clk_bulk_enable(iommu->num_clocks, iommu->clocks);
        if (ret)
@@ -948,6 +949,10 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
                               rk_ops->mk_dtentries(rk_domain->dt_dma));
                rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_ZAP_CACHE);
                rk_iommu_write(iommu->bases[i], RK_MMU_INT_MASK, RK_MMU_IRQ_MASK);
+
+               auto_gate = rk_iommu_read(iommu->bases[i], RK_MMU_AUTO_GATING);
+               auto_gate |= BIT(31);
+               rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, auto_gate);
        }
 
        ret = rk_iommu_enable_paging(iommu);
Signature · 2 lines
Best, 
Chaoyi
#15 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

Hi Chaoyi,

Quoted 3 lines

As I said, it is v2. Could you please try using the code below instead and see if it works?
[ auto_gate = read(RK_MMU_AUTO_GATING); auto_gate |= BIT(31); write(...) ]

Thanks -- that's clearly the right shape (read-modify-write, before paging is enabled, keeping the reset value instead of my clobbering 0x2).

I rebuilt v7.1-rc6 (with the rocket RK3568 series + your per-device-ops work) using your bit-31 version and tested it on a ROCK 3B: the NPU IOMMU comes up and services the NPU's DMA cleanly -- the NPU probes, attaches its domain, and runs repeated conv submissions with no DMA_READ_ERROR and no page-walk stall. No regression from the write.

To be precise about what I can and can't show: I tested both ways on v7.1-rc6 -- with your bit-31 write, and on the reset value (0x3) -- and the NPU IOMMU services
the NPU's reads with zero faults in both cases (no DMA_READ_ERROR, no page-walk stall). So I don't have a failing baseline here that bit-31 visibly fixes. Is the
AUTO_GATING write needed on current mainline, or only under conditions I'm not reproducing (a particular traffic pattern / silicon rev)? I'll keep the patch in your form unless you'd prefer to drop it.

One question so I document it correctly: what does bit 31 of RK_MMU_AUTO_GATING control on the v2 block -- is it a master "disable internal auto clock-gating" for the page-table walker (i.e. so a TLB-miss walk's AXI master keeps its clock to completion)? The RK3568 TRM I have doesn't cover the IOMMU registers, so a one-line description would let me write an accurate comment.

Kind regards,
Midgy

Quoted 71 lines · Chaoyi Chen

Hello Midgy,

Quoted 26 lines · Midgy BALON

On the RK356x v1 IOMMU, RK_MMU_AUTO_GATING resets to 0x3. Bit 1 enables auto clock-gating of the page-table walker, so the walker's AXI master loses its clock between transactions; a TLB-miss page walk then never completes and the IOMMU is left stuck (PAGING_ENABLED, never IDLE).

Clear bit 1 (keeping bit 0, the slave-port gate) once paging is enabled so the walker keeps its clock. This is required for the RK3568 NPU MMU.

---
 drivers/iommu/rockchip-iommu.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
drivers/iommu/rockchip-iommu.c +5 -0
index 4da80136933c4..e3d8b6e9ca12b 100644
@@ -953,6 +953,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
      ret = rk_iommu_enable_paging(iommu);
 
+     if (!ret) {
+             /*
+              * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+              * auto clock-gating; the walker's AXI master then loses its clock
+              * between transactions and a TLB-miss page walk never completes,

drivers/iommu/rockchip-iommu.c static int rk_iommu_enable(struct rk_iommu *iommu) · line 970

Quoted 10 lines
drivers/iommu/rockchip-iommu.c +7 -0
+              * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+              * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+              */
+             for (i = 0; i < iommu->num_mmu; i++)
+                     rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);
+     }
+
 out_disable_stall:
      rk_iommu_disable_stall(iommu);
 out_disable_clocks:

As I said, it is v2. Could you please try using the code below instead and see if it works? Thank you.

drivers/iommu/rockchip-iommu.c +5 -1
index 0013cf196c57..89e3a83a0251 100644
@@ -930,6 +930,7 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
        struct iommu_domain *domain = iommu->domain;
        struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
        int ret, i;
+       u32 auto_gate;
 
        ret = clk_bulk_enable(iommu->num_clocks, iommu->clocks);
        if (ret)
@@ -948,6 +949,10 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
                               rk_ops->mk_dtentries(rk_domain->dt_dma));
                rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_ZAP_CACHE);
                rk_iommu_write(iommu->bases[i], RK_MMU_INT_MASK, RK_MMU_IRQ_MASK);
+
+               auto_gate = rk_iommu_read(iommu->bases[i], RK_MMU_AUTO_GATING);
+               auto_gate |= BIT(31);
+               rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, auto_gate);
        }
 
        ret = rk_iommu_enable_paging(iommu);
 
--

Best,
Chaoyi

#16 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

Hi Midgy,

Quoted 25 lines · Midgy Balon

Hi Chaoyi,

Quoted 3 lines

As I said, it is v2. Could you please try using the code below instead and see if it works?
[ auto_gate = read(RK_MMU_AUTO_GATING); auto_gate |= BIT(31); write(...) ]

Thanks -- that's clearly the right shape (read-modify-write, before paging is enabled, keeping the reset value instead of my clobbering 0x2).

I rebuilt v7.1-rc6 (with the rocket RK3568 series + your per-device-ops work) using your bit-31 version and tested it on a ROCK 3B: the NPU IOMMU comes up and services the NPU's DMA cleanly -- the NPU probes, attaches its domain, and runs repeated conv submissions with no DMA_READ_ERROR and no page-walk stall. No regression from the write.

To be precise about what I can and can't show: I tested both ways on v7.1-rc6 -- with your bit-31 write, and on the reset value (0x3) -- and the NPU IOMMU services
the NPU's reads with zero faults in both cases (no DMA_READ_ERROR, no page-walk stall). So I don't have a failing baseline here that bit-31 visibly fixes. Is the
AUTO_GATING write needed on current mainline, or only under conditions I'm not reproducing (a particular traffic pattern / silicon rev)? I'll keep the patch in your form unless you'd prefer to drop it.

Quoted 6 lines

One question so I document it correctly: what does bit 31 of RK_MMU_AUTO_GATING control on the v2 block -- is it a master "disable internal auto clock-gating" for the page-table walker (i.e. so a TLB-miss walk's AXI master keeps its clock to completion)? The RK3568 TRM I have doesn't cover the IOMMU registers, so a one-line description would let me write an accurate comment.

Glad to hear this works. Please refer to the commit below.

[0]: https://github.com/rockchip-linux/kernel/commit/7f8158fb41b5cc8e738aaeebc3637c50ebd74cae
[1]: https://github.com/rockchip-linux/kernel/commit/6a355e5f9a2069a2309e240791bc3aad63b7324e

Signature · 2 lines
Best, 
Chaoyi
#17 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

Quoted 35 lines · Chaoyi Chen

Hi Midgy,

Quoted 25 lines · Midgy Balon

Hi Chaoyi,

Quoted 3 lines

As I said, it is v2. Could you please try using the code below instead and see if it works?
[ auto_gate = read(RK_MMU_AUTO_GATING); auto_gate |= BIT(31); write(...) ]

Thanks -- that's clearly the right shape (read-modify-write, before paging is enabled, keeping the reset value instead of my clobbering 0x2).

I rebuilt v7.1-rc6 (with the rocket RK3568 series + your per-device-ops work) using your bit-31 version and tested it on a ROCK 3B: the NPU IOMMU comes up and services the NPU's DMA cleanly -- the NPU probes, attaches its domain, and runs repeated conv submissions with no DMA_READ_ERROR and no page-walk stall. No regression from the write.

To be precise about what I can and can't show: I tested both ways on v7.1-rc6 -- with your bit-31 write, and on the reset value (0x3) -- and the NPU IOMMU services
the NPU's reads with zero faults in both cases (no DMA_READ_ERROR, no page-walk stall). So I don't have a failing baseline here that bit-31 visibly fixes. Is the
AUTO_GATING write needed on current mainline, or only under conditions I'm not reproducing (a particular traffic pattern / silicon rev)? I'll keep the patch in your form unless you'd prefer to drop it.

Quoted 6 lines

One question so I document it correctly: what does bit 31 of RK_MMU_AUTO_GATING control on the v2 block -- is it a master "disable internal auto clock-gating" for the page-table walker (i.e. so a TLB-miss walk's AXI master keeps its clock to completion)? The RK3568 TRM I have doesn't cover the IOMMU registers, so a one-line description would let me write an accurate comment.

Quoted 5 lines

Glad to hear this works. Please refer to the commit below.

[0]: https://github.com/rockchip-linux/kernel/commit/7f8158fb41b5cc8e738aaeebc3637c50ebd74cae
[1]: https://github.com/rockchip-linux/kernel/commit/6a355e5f9a2069a2309e240791bc3aad63b7324e

It looks like RGA needs this patch too, and it has already been merged :).

https://lore.kernel.org/all/[email protected]/

Signature · 2 lines
Best, 
Chaoyi
#18 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

Hi Chaoyi,

Thanks again for the AUTO_GATING fix and the MMU-v2 clarification — both are in v4.

The mainline rocket driver now probes, powers, and **submits** cleanly on the RK3568 NPU. The
last blocker is purely functional: **a submitted conv never computes** — and I've now localized
it precisely enough that I think it's a question only Rockchip can answer.

## The localization: the CNA weight-fetch DMA never fires

I captured the **working vendor rknpu** running our exact conv (conv2d, 16→128, 5×5 s2) and
diffed its pipeline against rocket running the same conv, both driven to the **identical
engaged state**:

- both units reach S_STATUS=0x0c, S_POINTER=0x1000e (executer bit16 set), TASKST=0xf000,
PC AMOUNTS=0x88;
- the CNA config registers are byte-identical (CONV_CON*, DATA_SIZE*, WEIGHT_SIZE*, CBUF).

Yet the data-amount counters diverge completely:

| counter | vendor `rknpu` (correct output) | mainline `rocket` (no output)
|
|---|---|---|
| `wt_rd` (0x803c) | **0x1900** | **0** |
| `dt_wr` (0x8034) | **0x6400** (= 204800 B, full) | **0** |
| `dt_rd` (0x8038) | 0x32c9 | 0x8b (regcmd fetch only) |

So on identical engaged HW with identical CNA config, the **vendor's CNA issues its weight-fetch
DMA and the DPU writes the full output; rocket's CNA never issues the weight read** — the MAC
array starves and the DPU writes nothing.

## What I''ve already ruled out

- **Command stream / weights / input:** byte-exact replay of the vendor's captured conv2d
regcmd **+ weight BO + input BO** through the rocket UABI → still no
compute. Content is not
the gate.
- **state_init:** RK3568's rk356x_rknpu_config.state_init = NULL (only RK3576 has one); we
also tested the RK3576 0x1024=0x80000000 / S_POINTER 0→1→0x1e ritual
on RK3568 anyway →
no change.
- **Reset / power / clocks / IOMMU:** rocket mirrors rknpu_soft_reset (srst assert/deassert,
SCMI/PVTPLL two-step, NoC de-idle at PMU 0xfdd90000, IOMMU re-establish);
clocks match the
vendor (SCMI 1 GHz, CLK_NPU/aclk 200 MHz); AUTO_GATING = your bit-31 fix.
- **Non-NPU-block writes (captured live, vendor + rocket same boot):** the only raw non-NPU
writes the vendor rknpu makes are the NoC/PMU de-idle
(NOC_AUTO=0xfffb,
BUS_IDLE_ST=0x1ea, PWR_STATUS=0x7f) — rocket makes them
**identically** (0x1ea/0xfffb).
No npu_grf read-margin and no memory-repair on RK3568 (those exist only
on RK3576/RK3588).
So GRF/CRU/PMU/NoC are matched too.

## The question

**With the executer engaged (S_STATUS=0x0c, S_POINTER bit16 set) and the CNA regcmd config
landed, what triggers the CNA's weight-fetch DMA?** The vendor rknpu reaches wt_rd=0x1900
from a clean submit; rocket, same SoC / engage / regcmd, stays wt_rd=0.

I've now matched every software-visible surface I can find — NPU-block job writes (byte-exact),
non-NPU writes (NoC/PMU/GRF, above), regcmd+weights+input (Method B replay), power/clocks/reset/
IOMMU — and the vendor still fetches weights while rocket doesn't. So my remaining hypotheses
are all HW-internal, and I'd value your read on which:
- an **ordering/timing** requirement between engage and the weight-fetch trigger that a
byte-identical-but-differently-sequenced submit doesn't satisfy;
- an internal CNA/CBUF **sequencer or credit/handshake** state that the vendor's power-on or
first-job path leaves armed and that isn't a CPU-visible register;
- or a genuinely undocumented CNA condition (a CBUF/DCOMP/weight-DMA enable) needed before the
fetch will issue.

Is any of these something the RK3568 NPU is known to require? Even a pointer to "the weight-fetch
won't start unless X" would unblock the mainline driver.

I would be happy to run any targeted register capture on the board.

Thanks a lot,
Midgy

Le lun. 8 juin 2026 à 05:40, Chaoyi Chen <[email protected]> a écrit :

Quoted 68 lines · Chaoyi Chen
Quoted 6 lines

Hi Midgy,

Quoted 3 lines · Midgy Balon

Hi Chaoyi,

Quoted 1 line

As I said, it is v2. Could you please try using the code below instead

and

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

see if it works?
[ auto_gate = read(RK_MMU_AUTO_GATING); auto_gate |= BIT(31);

write(...) ]

Quoted 2 lines
Quoted 2 lines

Thanks -- that's clearly the right shape (read-modify-write, before

paging is

Quoted 3 lines
Quoted 3 lines

enabled, keeping the reset value instead of my clobbering 0x2).

I rebuilt v7.1-rc6 (with the rocket RK3568 series + your per-device-ops

work)

Quoted 1 line
Quoted 1 line

using your bit-31 version and tested it on a ROCK 3B: the NPU IOMMU

comes up and

Quoted 1 line
Quoted 1 line

services the NPU's DMA cleanly -- the NPU probes, attaches its domain,

and runs

Quoted 1 line
Quoted 1 line

repeated conv submissions with no DMA_READ_ERROR and no page-walk

stall. No

Quoted 3 lines
Quoted 3 lines

regression from the write.

To be precise about what I can and can't show: I tested both ways on

v7.1-rc6 --

Quoted 3 lines
Quoted 3 lines

with your bit-31 write, and on the reset value (0x3) -- and the NPU IOMMU services
the NPU's reads with zero faults in both cases (no DMA_READ_ERROR, no

page-walk

Quoted 3 lines
Quoted 3 lines

stall). So I don't have a failing baseline here that bit-31 visibly fixes. Is the
AUTO_GATING write needed on current mainline, or only under conditions

I'm not

Quoted 1 line
Quoted 1 line

reproducing (a particular traffic pattern / silicon rev)? I'll keep the

patch in

Quoted 3 lines
Quoted 3 lines

your form unless you'd prefer to drop it.

One question so I document it correctly: what does bit 31 of

RK_MMU_AUTO_GATING

Quoted 1 line
Quoted 1 line

control on the v2 block -- is it a master "disable internal auto

clock-gating"

Quoted 1 line
Quoted 1 line

for the page-table walker (i.e. so a TLB-miss walk's AXI master keeps

its clock

Quoted 1 line
Quoted 1 line

to completion)? The RK3568 TRM I have doesn't cover the IOMMU

registers, so a

Quoted 6 lines
Quoted 2 lines

one-line description would let me write an accurate comment.

Glad to hear this works. Please refer to the commit below.

[0]:

https://github.com/rockchip-linux/kernel/commit/7f8158fb41b5cc8e738aaeebc3637c50ebd74cae

Quoted 1 line

[1]:

https://github.com/rockchip-linux/kernel/commit/6a355e5f9a2069a2309e240791bc3aad63b7324e

Quoted 1 line

It looks like RGA needs this patch too, and it has already been merged :).

https://lore.kernel.org/all/[email protected]/

--
Best,
Chaoyi

#19 lore raw

Re: [RFC PATCH v3 6/9] iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU

In reply to Chaoyi Chen

Hi Chaoyi,

Sorry for the slow follow-up — and thanks again for the pointers, that closes this out cleanly.

I read the two rockchip-linux commits (7f8158fb, 6a355e5f) and the merged patch you mentioned: "iommu/rockchip: disable fetch dte time limit" (Sven Püschel, From: Simon Xue, [email protected]), which Joerg applied on 2026-06-02. It adds DISABLE_FETCH_DTE_TIME_LIMIT (BIT(31)) and does exactly the auto_gate |= BIT(31) read-modify-write you suggested, in rk_iommu_enable() before paging is enabled, for every Rockchip IOMMU. So bit 31 disables the DTE-fetch time limit (without it the newer revision blocks after four consecutive master DTE fetches racing the CPU's IOTLB zaps); the RK3568/RK3588 erratum is the same one as RV1126/RV1109. That's the accurate description I was missing — thank you.

Since that fix is already upstream, I'll drop the AUTO_GATING change from my series entirely rather than carry a
duplicate: v4 will rebase on top of it and the NPU MMU picks up BIT(31) for free. For the record, on the ROCK 3B the NPU IOMMU services the NPU's reads with zero faults both with BIT(31) and on the reset value (0x3), so this is a no-op on my hardware/workload — but rebasing past the merged patch is the right thing regardless.

So my v4 IOMMU dependency list becomes: your per-device-ops series, plus the merged "disable fetch dte time limit" patch. My series no longer touches the IOMMU driver itself.

(The convolution output is still uniform zero-point / the job times out — that's the separate NPU compute-completion issue, unrelated to this.)

Thanks again, and kinds regards
Midgy

Quoted 50 lines · Chaoyi Chen
Quoted 35 lines · Chaoyi Chen

Hi Midgy,

Quoted 25 lines · Midgy Balon

Hi Chaoyi,

Quoted 3 lines

As I said, it is v2. Could you please try using the code below instead and see if it works?
[ auto_gate = read(RK_MMU_AUTO_GATING); auto_gate |= BIT(31); write(...) ]

Thanks -- that's clearly the right shape (read-modify-write, before paging is enabled, keeping the reset value instead of my clobbering 0x2).

I rebuilt v7.1-rc6 (with the rocket RK3568 series + your per-device-ops work) using your bit-31 version and tested it on a ROCK 3B: the NPU IOMMU comes up and services the NPU's DMA cleanly -- the NPU probes, attaches its domain, and runs repeated conv submissions with no DMA_READ_ERROR and no page-walk stall. No regression from the write.

To be precise about what I can and can't show: I tested both ways on v7.1-rc6 -- with your bit-31 write, and on the reset value (0x3) -- and the NPU IOMMU services
the NPU's reads with zero faults in both cases (no DMA_READ_ERROR, no page-walk stall). So I don't have a failing baseline here that bit-31 visibly fixes. Is the
AUTO_GATING write needed on current mainline, or only under conditions I'm not reproducing (a particular traffic pattern / silicon rev)? I'll keep the patch in your form unless you'd prefer to drop it.

Quoted 6 lines

One question so I document it correctly: what does bit 31 of RK_MMU_AUTO_GATING control on the v2 block -- is it a master "disable internal auto clock-gating" for the page-table walker (i.e. so a TLB-miss walk's AXI master keeps its clock to completion)? The RK3568 TRM I have doesn't cover the IOMMU registers, so a one-line description would let me write an accurate comment.

Quoted 5 lines

Glad to hear this works. Please refer to the commit below.

[0]: https://github.com/rockchip-linux/kernel/commit/7f8158fb41b5cc8e738aaeebc3637c50ebd74cae
[1]: https://github.com/rockchip-linux/kernel/commit/6a355e5f9a2069a2309e240791bc3aad63b7324e

It looks like RGA needs this patch too, and it has already been merged :).

https://lore.kernel.org/all/[email protected]/

--
Best,
Chaoyi

#20 lore raw

RFC PATCH v3 7/9 dt-bindings: npu: rockchip,rk3588-rknn-core: Add RK3568

In reply to Midgy BALON

The RK3568 carries a single core of the same NVDLA-derived NPU IP as the RK3588. Add its compatible.

On RK3568 the NPU NOC bus-idle and power gating are controlled through the system PMU rather than a dedicated register block, so add a rockchip,pmu phandle to that syscon. The RK3568 NPU has no dedicated SRAM rail, so sram-supply is required only on RK3588.

---
 .../npu/rockchip,rk3588-rknn-core.yaml         | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +17 -1
index caca2a4903cd1..af9936b32e9fe 100644
@@ -21,6 +21,7 @@ properties:
 
   compatible:
     enum:
+      - rockchip,rk3568-rknn-core
       - rockchip,rk3588-rknn-core
 
   reg:
@@ -50,6 +51,13 @@ properties:
 
   npu-supply: true
 
+  rockchip,pmu:
+    $ref: /schemas/types.yaml#/definitions/phandle
+    description:
+      Phandle to the PMU syscon.  On RK3568 the NPU's NOC bus-idle and
+      power gating are controlled through the PMU; this points to that
+      syscon so those registers can be reached.
+
   power-domains:
     maxItems: 1
 
@@ -75,7 +83,15 @@ required:
   - resets
   - reset-names
   - npu-supply
-  - sram-supply
+
+if:
+  properties:
+    compatible:
+      contains:
+        const: rockchip,rk3588-rknn-core
+then:
+  required:
+    - sram-supply
 
 additionalProperties: false
 
Signature · 1 line
2.39.5
#21 lore raw

Re: [RFC PATCH v3 7/9] dt-bindings: npu: rockchip,rk3588-rknn-core: Add RK3568

Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:

New issues:
- [High] The value written to RK_MMU_AUTO_GATING sets bit 1 instead of clearing it, contradicting the inline comment and hardware workaround logic.
- [High] The error handling path for noc_init() failure leaks the IOMMU group reference and leaves the device in an unbalanced PM runtime state.
- [High] A malformed device tree with excessive NPU core nodes triggers an out-of-bounds write in rocket_probe() array access.
- [High] The NOC bus de-idle sequence via the PMU is missing from the runtime PM resume path.
- [Medium] The rockchip,pmu property is functionally mandatory for RK3568 but is not added to the schema's conditional required list.

Pre-existing issues:
- [High] The clock names in the core->clks array are uninitialized before being passed to devm_clk_bulk_get(), causing the framework to fetch the wrong clocks. --

commit 6c61da961bed3360a99f625d4a3b1599e7bfe685
Author: Midgy BALON <[email protected]>

dt-bindings: npu: rockchip, rk3588-rknn-core: Add RK3568

The RK3568 carries a single core of the same NVDLA-derived NPU IP as the RK3588. Add its compatible. On RK3568 the NPU NOC bus-idle and power gating are controlled through the system PMU rather than a dedicated register block.

Quoted 4 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +0 -0
index caca2a4903cd1..af9936b32e9fe 100644

[ ... ]

Quoted 4 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +0 -0
@@ -75,7 +83,15 @@ required:
   - resets
   - reset-names
   - npu-supply

Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml required: · line 94

Quoted 10 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +9 -1
-  - sram-supply
+
+if:
+  properties:
+    compatible:
+      contains:
+        const: rockchip,rk3588-rknn-core
+then:
+  required:
+    - sram-supply

[Severity: Medium]
Does the missing rockchip,pmu property from the conditional required block cause incomplete device trees to pass validation?

The rockchip,pmu property is functionally mandatory for the rockchip,rk3568-rknn-core variant as the driver unconditionally fails to probe without it. Missing this mandatory property here could allow invalid device trees to go undetected.

Quoted 6 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +0 -0
 additionalProperties: false
 
drivers/accel/rocket/rocket_core.c +0 -0
index b3b2fa9ba645a..634f78dfe2887 100644

[ ... ]

drivers/accel/rocket/rocket_core.c int rocket_core_init(struct rocket_core *core) · line 99

Quoted 12 lines
drivers/accel/rocket/rocket_core.c +8 -0
@@ -83,6 +89,15 @@ int rocket_core_init(struct rocket_core *core)
 		return err;
 	}
 
+	if (core->soc_data->noc_init) {
+		err = core->soc_data->noc_init(core);
+		if (err) {
+			pm_runtime_put_sync(dev);
+			rocket_job_fini(core);
+			return err;
+		}
+	}

[Severity: High]
Does this error path leak the IOMMU group reference and leave the device in an unbalanced PM runtime state?

By returning directly after pm_runtime_put_sync() and rocket_job_fini(), the code misses calling rocket_core_fini() which handles iommu_group_put(), pm_runtime_disable(), and pm_runtime_dont_use_autosuspend().

Quoted 8 lines
drivers/accel/rocket/rocket_core.c +1 -0
+
 	version = rocket_pc_readl(core, VERSION);
 	version += rocket_pc_readl(core, VERSION_NUM) & 0xffff;
 
drivers/accel/rocket/rocket_device.c +0 -0
index 46e6ee1e72c5f..6186f4faa3a2a 100644

[ ... ]

drivers/accel/rocket/rocket_device.c struct rocket_device *rocket_device_init(struct platform_device *pdev, · line 29

Quoted 11 lines
drivers/accel/rocket/rocket_device.c +2 -5
@@ -27,17 +25,14 @@ struct rocket_device *rocket_device_init(struct platform_device *pdev,
 	ddev = &rdev->ddev;
 	dev_set_drvdata(dev, rdev);
 
-	for_each_compatible_node(core_node, NULL, "rockchip,rk3588-rknn-core")
-		if (of_device_is_available(core_node))
-			num_cores++;
-
-	rdev->cores = devm_kcalloc(dev, num_cores, sizeof(*rdev->cores), GFP_KERNEL);
+	rdev->cores = devm_kcalloc(dev, soc_data->num_cores, sizeof(*rdev->cores),
+				   GFP_KERNEL);

[Severity: High]
Can this code overflow rdev->cores[] if the device tree contains more NPU nodes than soc_data->num_cores?

With this change, the array size is fixed to soc_data->num_cores, but rocket_probe() in drivers/accel/rocket/rocket_drv.c is still called for every matching node. If there are too many nodes in the device tree, it appears rocket_probe() will write out of bounds here:

drivers/accel/rocket/rocket_drv.c:rocket_probe() {
    ...
	unsigned int core = rdev->num_cores;
    ...
	rdev->cores[core].rdev = rdev;
	rdev->cores[core].dev = &pdev->dev;
	rdev->cores[core].index = core;
    ...
}
Quoted 7 lines
drivers/accel/rocket/rocket_device.c +0 -0
 	if (!rdev->cores)
 		return ERR_PTR(-ENOMEM);
 
drivers/accel/rocket/rocket_drv.c +0 -0
index 8bbbce594883e..5a72d0b5f4dff 100644

[ ... ]

Quoted 15 lines
drivers/accel/rocket/rocket_drv.c +11 -0
@@ -213,8 +219,64 @@ static void rocket_remove(struct platform_device *pdev)
 	}
 }
 
+/*
+ * On RK3568 the NPU NOC bus is gated and idle out of reset and must be
+ * powered on and de-idled through the PMU before the NPU is reachable.  PMU
+ * registers use a write-mask protocol: the upper 16 bits enable writes to the
+ * matching lower 16 bits.
+ *
+ * The NPU's high-speed clock is a PVTPLL managed by TF-A via SCMI and must be
+ * running before the NOC acknowledges the de-idle request.  Force a real SCMI
+ * rate change (an intermediate rate defeats the clock framework's
+ * unchanged-rate shortcut) now that the power domain is on and clocks enabled.
+ */

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 242

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +10 -0
+#define ROCKET_RK3568_SCMI_CLK	2
+
+static int rk3568_noc_init(struct rocket_core *core)
+{
+	struct regmap *pmu;
+	unsigned int val;
+	int ret;
+
+	clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 600000000UL);
+	clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 1000000000UL);

[Severity: High]
This is a pre-existing issue, but does this unintentionally alter the system aclk rate?

Looking at rocket_core_init(), the core->clks array is passed to devm_clk_bulk_get() without initializing the id strings:

drivers/accel/rocket/rocket_core.c:rocket_core_init() {
    ...
	err = devm_clk_bulk_get(dev, ARRAY_SIZE(core->clks), core->clks);
    ...
}

Because the id fields are NULL, the framework fetches the first DT clock (aclk) for all elements. Calling clk_set_rate() here seems to alter the system aclk frequency instead of the intended SCMI clock.

Quoted 12 lines
drivers/accel/rocket/rocket_drv.c +12 -0
+
+	pmu = syscon_regmap_lookup_by_phandle(core->dev->of_node, "rockchip,pmu");
+	if (IS_ERR(pmu))
+		return dev_err_probe(core->dev, PTR_ERR(pmu),
+				     "failed to get PMU regmap\n");
+
+	/* Power on the NPU power domain (PWR_GATE_SFTCON bit 1 = 0). */
+	regmap_write(pmu, 0xa0, BIT(1 + 16));
+
+	/* Disable NPU NOC auto-idle (NOC_AUTO_CON0 bit 2). */
+	regmap_write(pmu, 0x70, BIT(2 + 16));
+

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 264

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +10 -0
+	/* Request NPU bus de-idle (BUS_IDLE_SFTCON0 bit 2 = 0). */
+	regmap_write(pmu, 0x50, BIT(2 + 16));
+
+	/* Wait for the bus to report active (BUS_IDLE_ST bit 2 = 0). */
+	ret = regmap_read_poll_timeout(pmu, 0x68, val, !(val & BIT(2)), 10, 1000);
+	if (ret)
+		dev_err(core->dev, "timed out waiting for NPU bus de-idle\n");
+
+	return ret;
+}

[Severity: High]
Is a NOC bus de-idle sequence required in the runtime PM resume path as well?

If the generic power domain (genpd) powers off the NPU during runtime suspend, the NOC bus idle state would reset. Currently, rocket_device_runtime_resume() enables clocks but doesn't repeat this PMU de-idle sequence, which could leave the AXI bus gated and cause subsequent hardware accesses to hang or crash.

Quoted 15 lines
drivers/accel/rocket/rocket_drv.c +1 -0
+
 static const struct rocket_soc_data rk3568_soc_data = {
 	.num_cores = 1,
 	.dma_bits = 32,
 	.noc_init = rk3568_noc_init,
 };
drivers/iommu/rockchip-iommu.c +0 -0
index 0013cf196c573..7d40ecb26123f 100644
@@ -952,6 +952,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
 	ret = rk_iommu_enable_paging(iommu);
 

drivers/iommu/rockchip-iommu.c static int rk_iommu_enable(struct rk_iommu *iommu) · line 964

Quoted 10 lines
drivers/iommu/rockchip-iommu.c +10 -0
+	if (!ret) {
+		/*
+		 * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+		 * auto clock-gating; the walker's AXI master then loses its clock
+		 * between transactions and a TLB-miss page walk never completes,
+		 * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+		 * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+		 */
+		for (i = 0; i < iommu->num_mmu; i++)
+			rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);

[Severity: High]
Will this write operation inadvertently set bit 1 instead of clearing it?

Writing 0x2 (binary 10) sets bit 1 and clears bit 0, which contradicts the inline comment indicating that bit 1 should be cleared and bit 0 kept. This appears to enable the buggy page-walker auto clock-gating and disable the slave-port gate.

Quoted 5 lines
drivers/iommu/rockchip-iommu.c +2 -0
+	}
+
 out_disable_stall:
 	rk_iommu_disable_stall(iommu);
 out_disable_clocks:
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=7
#22 lore raw

Re: [RFC PATCH v3 7/9] dt-bindings: npu: rockchip,rk3588-rknn-core: Add RK3568

Quoted 47 lines · On Thu, Jun 04, 2026 at 02:08:50PM +0000, [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 6 potential issue(s) to consider:

New issues:
- [High] The value written to RK_MMU_AUTO_GATING sets bit 1 instead of clearing it, contradicting the inline comment and hardware workaround logic.
- [High] The error handling path for noc_init() failure leaks the IOMMU group reference and leaves the device in an unbalanced PM runtime state.
- [High] A malformed device tree with excessive NPU core nodes triggers an out-of-bounds write in rocket_probe() array access.
- [High] The NOC bus de-idle sequence via the PMU is missing from the runtime PM resume path.
- [Medium] The rockchip,pmu property is functionally mandatory for RK3568 but is not added to the schema's conditional required list.

Pre-existing issues:
- [High] The clock names in the core->clks array are uninitialized before being passed to devm_clk_bulk_get(), causing the framework to fetch the wrong clocks. --

commit 6c61da961bed3360a99f625d4a3b1599e7bfe685
Author: Midgy BALON <[email protected]>

dt-bindings: npu: rockchip, rk3588-rknn-core: Add RK3568

The RK3568 carries a single core of the same NVDLA-derived NPU IP as the RK3588. Add its compatible. On RK3568 the NPU NOC bus-idle and power gating are controlled through the system PMU rather than a dedicated register block.

Quoted 4 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +0 -0
index caca2a4903cd1..af9936b32e9fe 100644

[ ... ]

Quoted 4 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +0 -0
@@ -75,7 +83,15 @@ required:
   - resets
   - reset-names
   - npu-supply

Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml required: · line 94

Quoted 10 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +9 -1
-  - sram-supply
+
+if:
+  properties:
+    compatible:
+      contains:
+        const: rockchip,rk3588-rknn-core
+then:
+  required:
+    - sram-supply

[Severity: Medium]
Does the missing rockchip,pmu property from the conditional required block cause incomplete device trees to pass validation?

Quoted 4 lines

The rockchip,pmu property is functionally mandatory for the rockchip,rk3568-rknn-core variant as the driver unconditionally fails to probe without it. Missing this mandatory property here could allow invalid device trees to go undetected.

This looks valid. Additionally, the condition should forbid this property on rk3588.
pw-bot: changes-requested

Quoted 4 lines

drivers/accel/rocket/rocket_core.c

Quoted 4 lines
Documentation/devicetree/bindings/npu/rockchip,rk3588-rknn-core.yaml +0 -0
 additionalProperties: false
 
drivers/accel/rocket/rocket_core.c +0 -0

Why is the bot commenting on driver stuff here?

Cheers,
Conor.

Quoted 189 lines
Quoted 3 lines

index b3b2fa9ba645a..634f78dfe2887 100644

drivers/accel/rocket/rocket_core.c +0 -0

[ ... ]

drivers/accel/rocket/rocket_core.c int rocket_core_init(struct rocket_core *core) · line 99

Quoted 12 lines
drivers/accel/rocket/rocket_core.c +8 -0
@@ -83,6 +89,15 @@ int rocket_core_init(struct rocket_core *core)
 		return err;
 	}
 
+	if (core->soc_data->noc_init) {
+		err = core->soc_data->noc_init(core);
+		if (err) {
+			pm_runtime_put_sync(dev);
+			rocket_job_fini(core);
+			return err;
+		}
+	}

[Severity: High]
Does this error path leak the IOMMU group reference and leave the device in an unbalanced PM runtime state?

By returning directly after pm_runtime_put_sync() and rocket_job_fini(), the code misses calling rocket_core_fini() which handles iommu_group_put(), pm_runtime_disable(), and pm_runtime_dont_use_autosuspend().

Quoted 8 lines
drivers/accel/rocket/rocket_core.c +1 -0
+
 	version = rocket_pc_readl(core, VERSION);
 	version += rocket_pc_readl(core, VERSION_NUM) & 0xffff;
 
drivers/accel/rocket/rocket_device.c +0 -0
index 46e6ee1e72c5f..6186f4faa3a2a 100644

[ ... ]

drivers/accel/rocket/rocket_device.c struct rocket_device *rocket_device_init(struct platform_device *pdev, · line 29

Quoted 11 lines
drivers/accel/rocket/rocket_device.c +2 -5
@@ -27,17 +25,14 @@ struct rocket_device *rocket_device_init(struct platform_device *pdev,
 	ddev = &rdev->ddev;
 	dev_set_drvdata(dev, rdev);
 
-	for_each_compatible_node(core_node, NULL, "rockchip,rk3588-rknn-core")
-		if (of_device_is_available(core_node))
-			num_cores++;
-
-	rdev->cores = devm_kcalloc(dev, num_cores, sizeof(*rdev->cores), GFP_KERNEL);
+	rdev->cores = devm_kcalloc(dev, soc_data->num_cores, sizeof(*rdev->cores),
+				   GFP_KERNEL);

[Severity: High]
Can this code overflow rdev->cores[] if the device tree contains more NPU nodes than soc_data->num_cores?

With this change, the array size is fixed to soc_data->num_cores, but rocket_probe() in drivers/accel/rocket/rocket_drv.c is still called for every matching node. If there are too many nodes in the device tree, it appears rocket_probe() will write out of bounds here:

drivers/accel/rocket/rocket_drv.c:rocket_probe() {
    ...
	unsigned int core = rdev->num_cores;
    ...
	rdev->cores[core].rdev = rdev;
	rdev->cores[core].dev = &pdev->dev;
	rdev->cores[core].index = core;
    ...
}
Quoted 7 lines
drivers/accel/rocket/rocket_device.c +0 -0
 	if (!rdev->cores)
 		return ERR_PTR(-ENOMEM);
 
drivers/accel/rocket/rocket_drv.c +0 -0
index 8bbbce594883e..5a72d0b5f4dff 100644

[ ... ]

Quoted 15 lines
drivers/accel/rocket/rocket_drv.c +11 -0
@@ -213,8 +219,64 @@ static void rocket_remove(struct platform_device *pdev)
 	}
 }
 
+/*
+ * On RK3568 the NPU NOC bus is gated and idle out of reset and must be
+ * powered on and de-idled through the PMU before the NPU is reachable.  PMU
+ * registers use a write-mask protocol: the upper 16 bits enable writes to the
+ * matching lower 16 bits.
+ *
+ * The NPU's high-speed clock is a PVTPLL managed by TF-A via SCMI and must be
+ * running before the NOC acknowledges the de-idle request.  Force a real SCMI
+ * rate change (an intermediate rate defeats the clock framework's
+ * unchanged-rate shortcut) now that the power domain is on and clocks enabled.
+ */

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 242

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +10 -0
+#define ROCKET_RK3568_SCMI_CLK	2
+
+static int rk3568_noc_init(struct rocket_core *core)
+{
+	struct regmap *pmu;
+	unsigned int val;
+	int ret;
+
+	clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 600000000UL);
+	clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 1000000000UL);

[Severity: High]
This is a pre-existing issue, but does this unintentionally alter the system aclk rate?

Looking at rocket_core_init(), the core->clks array is passed to devm_clk_bulk_get() without initializing the id strings:

drivers/accel/rocket/rocket_core.c:rocket_core_init() {
    ...
	err = devm_clk_bulk_get(dev, ARRAY_SIZE(core->clks), core->clks);
    ...
}

Because the id fields are NULL, the framework fetches the first DT clock (aclk) for all elements. Calling clk_set_rate() here seems to alter the system aclk frequency instead of the intended SCMI clock.

Quoted 12 lines
drivers/accel/rocket/rocket_drv.c +12 -0
+
+	pmu = syscon_regmap_lookup_by_phandle(core->dev->of_node, "rockchip,pmu");
+	if (IS_ERR(pmu))
+		return dev_err_probe(core->dev, PTR_ERR(pmu),
+				     "failed to get PMU regmap\n");
+
+	/* Power on the NPU power domain (PWR_GATE_SFTCON bit 1 = 0). */
+	regmap_write(pmu, 0xa0, BIT(1 + 16));
+
+	/* Disable NPU NOC auto-idle (NOC_AUTO_CON0 bit 2). */
+	regmap_write(pmu, 0x70, BIT(2 + 16));
+

drivers/accel/rocket/rocket_drv.c static void rocket_remove(struct platform_device *pdev) · line 264

Quoted 10 lines
drivers/accel/rocket/rocket_drv.c +10 -0
+	/* Request NPU bus de-idle (BUS_IDLE_SFTCON0 bit 2 = 0). */
+	regmap_write(pmu, 0x50, BIT(2 + 16));
+
+	/* Wait for the bus to report active (BUS_IDLE_ST bit 2 = 0). */
+	ret = regmap_read_poll_timeout(pmu, 0x68, val, !(val & BIT(2)), 10, 1000);
+	if (ret)
+		dev_err(core->dev, "timed out waiting for NPU bus de-idle\n");
+
+	return ret;
+}

[Severity: High]
Is a NOC bus de-idle sequence required in the runtime PM resume path as well?

If the generic power domain (genpd) powers off the NPU during runtime suspend, the NOC bus idle state would reset. Currently, rocket_device_runtime_resume() enables clocks but doesn't repeat this PMU de-idle sequence, which could leave the AXI bus gated and cause subsequent hardware accesses to hang or crash.

Quoted 15 lines
drivers/accel/rocket/rocket_drv.c +1 -0
+
 static const struct rocket_soc_data rk3568_soc_data = {
 	.num_cores = 1,
 	.dma_bits = 32,
 	.noc_init = rk3568_noc_init,
 };
drivers/iommu/rockchip-iommu.c +0 -0
index 0013cf196c573..7d40ecb26123f 100644
@@ -952,6 +952,18 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
 
 	ret = rk_iommu_enable_paging(iommu);
 

drivers/iommu/rockchip-iommu.c static int rk_iommu_enable(struct rk_iommu *iommu) · line 964

Quoted 10 lines
drivers/iommu/rockchip-iommu.c +10 -0
+	if (!ret) {
+		/*
+		 * RK356x v1 IOMMU: RK_MMU_AUTO_GATING bit 1 enables page-walker
+		 * auto clock-gating; the walker's AXI master then loses its clock
+		 * between transactions and a TLB-miss page walk never completes,
+		 * leaving the IOMMU stuck (PAGING_ENABLED, never IDLE).  Clear
+		 * bit 1 (keep bit 0, the slave-port gate) once paging is enabled.
+		 */
+		for (i = 0; i < iommu->num_mmu; i++)
+			rk_iommu_write(iommu->bases[i], RK_MMU_AUTO_GATING, 0x2);

[Severity: High]
Will this write operation inadvertently set bit 1 instead of clearing it?

Writing 0x2 (binary 10) sets bit 1 and clears bit 0, which contradicts the inline comment indicating that bit 1 should be cleared and bit 0 kept. This appears to enable the buggy page-walker auto clock-gating and disable the slave-port gate.

Quoted 5 lines
drivers/iommu/rockchip-iommu.c +2 -0
+	}
+
 out_disable_stall:
 	rk_iommu_disable_stall(iommu);
 out_disable_clocks:
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=7
#23 lore raw

RFC PATCH v3 8/9 arm64: dts: rockchip: rk356x: Add the NPU and its IOMMU

In reply to Midgy BALON

The RK3568 has an NVDLA-derived NPU at fde40000 with its own IOMMU at fde4b000. Add both nodes (disabled by default) and the NPU power-domain child under the PMU power-controller, and point rockchip,pmu at the PMU syscon that controls the NPU NoC bus-idle.

The power-domain deliberately carries no pm_qos: qos_npu sits behind the NPU NoC, which is gated until the NPU is brought up, so a genpd power-off QoS save would fault reading it.

---
 arch/arm64/boot/dts/rockchip/rk356x-base.dtsi | 38 +++++++++++++++++++
 1 file changed, 38 insertions(+)
arch/arm64/boot/dts/rockchip/rk356x-base.dtsi +38 -0
index 64bdd8b7754b5..50ce5a5e4fc24 100644
@@ -512,6 +512,13 @@ power-domain@RK3568_PD_GPU {
 				#power-domain-cells = <0>;
 			};
 
+			pd_npu: power-domain@RK3568_PD_NPU {
+				reg = <RK3568_PD_NPU>;
+				clocks = <&cru ACLK_NPU_PRE>,
+					 <&cru HCLK_NPU_PRE>;
+				#power-domain-cells = <0>;
+			};
+
 			/* These power domains are grouped by VD_LOGIC */
 			power-domain@RK3568_PD_VI {
 				reg = <RK3568_PD_VI>;
@@ -948,6 +955,37 @@ qos_rga_wr: qos@fe158300 {
 		reg = <0x0 0xfe158300 0x0 0x20>;
 	};
 
+	rknn_core_0: npu@fde40000 {
+		compatible = "rockchip,rk3568-rknn-core";
+		reg = <0x0 0xfde40000 0x0 0x1000>,
+		      <0x0 0xfde41000 0x0 0x1000>,
+		      <0x0 0xfde43000 0x0 0x1000>;
+		reg-names = "pc", "cna", "core";
+		interrupts = <GIC_SPI 151 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&cru ACLK_NPU>, <&cru HCLK_NPU>,
+			 <&scmi_clk SCMI_CLK_NPU>, <&cru PCLK_NPU_PRE>;
+		clock-names = "aclk", "hclk", "npu", "pclk";
+		assigned-clocks = <&scmi_clk SCMI_CLK_NPU>;
+		assigned-clock-rates = <200000000>;
+		resets = <&cru SRST_A_NPU>, <&cru SRST_H_NPU>;
+		reset-names = "srst_a", "srst_h";
+		power-domains = <&power RK3568_PD_NPU>;
+		rockchip,pmu = <&pmu>;
+		iommus = <&rknn_mmu_0>;
+		status = "disabled";
+	};
+
+	rknn_mmu_0: iommu@fde4b000 {
+		compatible = "rockchip,iommu";
+		reg = <0x0 0xfde4b000 0x0 0x40>;
+		interrupts = <GIC_SPI 151 IRQ_TYPE_LEVEL_HIGH>;
+		clock-names = "aclk", "iface";
+		clocks = <&cru ACLK_NPU>, <&cru HCLK_NPU>;
+		power-domains = <&power RK3568_PD_NPU>;
+		#iommu-cells = <0>;
+		status = "disabled";
+	};
+
 	qos_npu: qos@fe180000 {
 		compatible = "rockchip,rk3568-qos", "syscon";
 		reg = <0x0 0xfe180000 0x0 0x20>;
Signature · 1 line
2.39.5
#24 lore raw

Re: [RFC PATCH v3 8/9] arm64: dts: rockchip: rk356x: Add the NPU and its IOMMU

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

New issues:
- [High] The NPU node simultaneously defines power-domains and rockchip,pmu, causing the generic power domain framework and the NPU driver to concurrently manipulate the same PMU power and idle registers.
- [Low] Device tree nodes npu@fde40000 and iommu@fde4b000 are not sorted by unit address.

Pre-existing issues:
- [Critical] The driver calls devm_clk_bulk_get without initializing the .id fields of the clock bulk data array, causing all clock requests to incorrectly default to the first clock in the device tree node. --

commit dff3ba648a490ef1a1c6644d672ca0b9e1d68360
Author: Midgy BALON <[email protected]>

arm64: dts: rockchip: rk356x: Add the NPU and its IOMMU

This commit adds the NPU and its associated IOMMU device tree nodes to the RK3568 base devicetree. It also configures the necessary clocks, resets, and power domain for the NPU.

Quoted 4 lines
arch/arm64/boot/dts/rockchip/rk356x-base.dtsi +0 -0
index 64bdd8b7754b5..50ce5a5e4fc24 100644

[ ... ]

arch/arm64/boot/dts/rockchip/rk356x-base.dtsi qos_rga_wr: qos@fe158300 { · line 958

Quoted 5 lines
arch/arm64/boot/dts/rockchip/rk356x-base.dtsi +1 -0
@@ -948,6 +955,37 @@ qos_rga_wr: qos@fe158300 {
 		reg = <0x0 0xfe158300 0x0 0x20>;
 	};
 
+	rknn_core_0: npu@fde40000 {

[Severity: Low]
This isn't a bug, but should the npu@fde40000 and iommu@fde4b000 nodes be ordered alphanumerically by unit address? They appear to be inserted after qos@fe158300 rather than being placed right before gpu@fde60000.

arch/arm64/boot/dts/rockchip/rk356x-base.dtsi qos_rga_wr: qos@fe158300 { · line 966

Quoted 8 lines
arch/arm64/boot/dts/rockchip/rk356x-base.dtsi +8 -0
+		compatible = "rockchip,rk3568-rknn-core";
+		reg = <0x0 0xfde40000 0x0 0x1000>,
+		      <0x0 0xfde41000 0x0 0x1000>,
+		      <0x0 0xfde43000 0x0 0x1000>;
+		reg-names = "pc", "cna", "core";
+		interrupts = <GIC_SPI 151 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&cru ACLK_NPU>, <&cru HCLK_NPU>,
+			 <&scmi_clk SCMI_CLK_NPU>, <&cru PCLK_NPU_PRE>;

[Severity: Critical]
This is a pre-existing issue, but enabling this node exposes a bug where devm_clk_bulk_get() is called without initializing the .id fields in rocket_core_init().

Since the connection ID is NULL, does clk_get() implicitly fall back to of_clk_get(np, 0) for each loop iteration, mapping all four clock array entries to ACLK_NPU?

Later, when rk3568_noc_init() dynamically scales the SCMI clock rate to 1 GHz:

    clk_set_rate(core->clks[ROCKET_RK3568_SCMI_CLK].clk, 1000000000UL);

Does this mistakenly configure the rate of ACLK_NPU instead, which could overclock the AXI bus and lead to immediate hardware instability?

arch/arm64/boot/dts/rockchip/rk356x-base.dtsi qos_rga_wr: qos@fe158300 { · line 973

Quoted 7 lines
arch/arm64/boot/dts/rockchip/rk356x-base.dtsi +7 -0
+		clock-names = "aclk", "hclk", "npu", "pclk";
+		assigned-clocks = <&scmi_clk SCMI_CLK_NPU>;
+		assigned-clock-rates = <200000000>;
+		resets = <&cru SRST_A_NPU>, <&cru SRST_H_NPU>;
+		reset-names = "srst_a", "srst_h";
+		power-domains = <&power RK3568_PD_NPU>;
+		rockchip,pmu = <&pmu>;

[Severity: High]
Could setting both power-domains and rockchip,pmu cause a race condition with TF-A and corrupt hardware state?

The generic power domain framework inherently manages power gating and NoC bus idle requests for this domain, and explicitly protects these PMU transitions with pmu->mutex.

Concurrently, the rocket driver's rk3568_noc_init() issues direct, un-polled regmap_write() operations to the exact same PMU offsets:

    regmap_write(pmu, 0xa0, BIT(1 + 16));
    ...
    regmap_write(pmu, 0x50, BIT(2 + 16));

Does bypassing GenPD's tracked state machine and the pmu->mutex lock create a layering violation here?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=8
#25 lore raw

RFC PATCH v3 9/9 arm64: dts: rockchip: rk3568-rock-3b: Enable the NPU

In reply to Midgy BALON

Enable the NPU and its IOMMU on ROCK 3B.

vdd_npu is marked always-on so the rail is up before genpd de-idles the NPU NoC at power-on: the PMU de-idle handshake needs the rail powered. The PVTPLL compute clock is brought up later by the driver.

---
 arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts +13 -1
index 69001e453732e..7ac780ed313d5 100644
@@ -330,8 +330,10 @@ regulator-state-mem {
 
 			vdd_npu: DCDC_REG4 {
 				regulator-name = "vdd_npu";
+				regulator-always-on;
+				regulator-boot-on;
 				regulator-initial-mode = <0x2>;
-				regulator-min-microvolt = <500000>;
+				regulator-min-microvolt = <825000>;
 				regulator-max-microvolt = <1350000>;
 				regulator-ramp-delay = <6001>;
 
@@ -787,3 +789,13 @@ vp0_out_hdmi: endpoint@ROCKCHIP_VOP2_EP_HDMI0 {
 		remote-endpoint = <&hdmi_in_vp0>;
 	};
 };
+
+&rknn_core_0 {
+	npu-supply = <&vdd_npu>;
+	status = "okay";
+};
+
+&rknn_mmu_0 {
+	status = "okay";
+};
+
Signature · 1 line
2.39.5
#26 lore raw
In reply to Midgy BALON

Hello Midgy,

Quoted 25 lines · Midgy BALON

RFC, not for merge. End-to-end inference does not produce correct output yet (see Status), so per the v2 discussion this is a request for design feedback. It now probes, attaches, and submits cleanly on a stock v7.1-rc6 tree; what remains is one hardware-internal issue.

The RK3568 has a single NVDLA-derived NPU core, the same IP family as the RK3588 NPU the driver already supports; the register layout matches. The RK3568 differences are a 32-bit NPU AXI/IOMMU (vs 40-bit) and explicit PVTPLL/PMU bring-up to power and de-idle the NPU before it is reachable.

Patches:
1-2 rocket: per-SoC data struct, then derive DMA width and core count
from match data (refactors, no functional change).
3 rocket: RK3568 SoC data + PVTPLL/PMU/NOC bring-up.
4 rocket: reset the NPU before detaching the IOMMU on a job timeout
(the detach otherwise stalls a wedged AXI master and WARNs).
5 rocket: keep the IOMMU domain attached across jobs instead of
re-attaching per job (the per-job rk_iommu handshake on the idle NPU MMU is slow and noisy).
6 iommu/rockchip: clear AUTO_GATING bit 1 on the RK356x v1 IOMMU so
the page-walker keeps its clock (else a TLB-miss walk never completes).
7 dt-bindings: add the RK3568 NPU compatible.
8-9 arm64 dts: add the NPU and its IOMMU, and enable them on ROCK 3B.

Quoted 6 lines

Dependency. The NPU MMU is rockchip-iommu v1 (32-bit) while the rest of the RK3568 uses v2 (40-bit). They cannot coexist until the driver carries per-device ops; this series is developed on top of Simon Xue's "iommu/rockchip: Drop global rk_ops in favor of per-device ops" [1]. Without it the NPU IOMMU fails to probe on a full RK3568 boot.

Hmmm. If I understand correctly, the NPU IOMMU should be v2 rather than v1, implying it should support 40-bit PAs. Nevertheless, please note that the upper limit for DTE is 32 bits.

Quoted 7 lines

Power bring-up. The NPU is brought up through the power-domain layer (no driver hack): the NPU power-domain keeps its clocks but drops the pm_qos phandle (qos_npu sits behind the gated NPU NoC, so genpd's power-off QoS save faults reading it), and vdd_npu is marked always-on so the rail is up before genpd de-idles the NoC at power-on. The PMU de-idle then ACKs without PVTPLL running; PVTPLL is only needed for compute.

Can these operations not be completed via the pmdomain driver? If some operations are controlled by TF-A, are you using open source TF-A? Thank you.

Quoted 58 lines

Status. On v7.1-rc6 the driver probes, creates /dev/accel/accel0, attaches an IOMMU domain, and submits jobs; the program controller fetches and broadcasts the command list. Inference output is still wrong, and the cause is split across three layers:
- kernel (this series): the RK3568 differences appear handled;
- mesa/Teflon userspace: still emits RK3588-tuned config, wrong for

    RK3568 (to be filed separately on mesa-dev);

- hardware: with corrected config the NPU's DMA reads the full input and weight tensors (confirmed via its DMA bandwidth counters), but the MAC/output stage never completes, the job times out, and the output stays at the buffer's zero-point. I have not found the missing step; it is not in the command list (replaying the vendor's byte-exact command list behaves the same). Pointers welcome, especially from anyone with RK3568 NPU experience.

Known residual. On the first IOMMU attach the NPU MMU is idle with paging already enabled; the rk_iommu stall/reset handshake does not complete in that state and logs one burst of timeouts before the (kept) domain settles. It is harmless here because the job times out regardless, but it points at an idle-MMU reconfiguration corner the rk_iommu code does not handle on this block.

[1] https://lore.kernel.org/linux-rockchip/[email protected]/

Changes since v2:
- Tagged RFC; now tested on a stock v7.1-rc6 tree.
- Bring-up moved into the power-domain/DT layer (no initcall hack).
- Added the IOMMU detach-on-timeout and attach-once driver fixes.
- Split the driver patch (Heiko): soc_data / match-data / RK3568.
- Derive DMA width and core count from match data; drop the DT rescans.
- Binding describes the hardware; added the missing $ref on rockchip,pmu.
- Disclosed the per-device-ops IOMMU dependency.

Midgy BALON (9):
accel: rocket: Introduce per-SoC rocket_soc_data
accel: rocket: Derive DMA width and core count from match data
accel: rocket: Add RK3568 SoC support
accel: rocket: Reset the NPU before detaching the IOMMU on timeout
accel: rocket: Keep the IOMMU domain attached across jobs iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU
dt-bindings: npu: rockchip,rk3588-rknn-core: Add RK3568
arm64: dts: rockchip: rk356x: Add the NPU and its IOMMU
arm64: dts: rockchip: rk3568-rock-3b: Enable the NPU

 .../npu/rockchip,rk3588-rknn-core.yaml        | 18 ++++-
 .../boot/dts/rockchip/rk3568-rock-3b.dts      | 14 +++-
 arch/arm64/boot/dts/rockchip/rk356x-base.dtsi | 38 +++++++++++
 drivers/accel/rocket/rocket_core.c            | 22 ++++++-
 drivers/accel/rocket/rocket_core.h            | 19 ++++++
 drivers/accel/rocket/rocket_device.c          | 15 ++---
 drivers/accel/rocket/rocket_device.h          |  3 +-
 drivers/accel/rocket/rocket_drv.c             | 66 ++++++++++++++++++-
 drivers/accel/rocket/rocket_job.c             | 35 ++++++++--
 drivers/iommu/rockchip-iommu.c                | 12 ++++
 10 files changed, 219 insertions(+), 23 deletions(-)

base-commit: 52c800fdcf11888ebeb50c3d707f782cc15b66eb

Signature · 2 lines
Best, 
Chaoyi
#27 lore raw

Hi Chaoyi,

Thanks a lot for looking at this -- input from Rockchip is exactly what this series needs.

Quoted 3 lines

Hmmm. If I understand correctly, the NPU IOMMU should be v2 rather than v1, implying it should support 40-bit PAs. Nevertheless, please note that the upper limit for DTE is 32 bits.

Understood, and that 32-bit-DTE note is the crux of the trouble I had, so let me lay out what I see and ask how you'd prefer to solve it.

The mainline node is already v2 (rockchip,rk3568-iommu in rk356x-base.dtsi). The problem on this 8 GiB board: with the v2 ops the page-table allocations (gfp_flags == 0) can land above 4 GiB, so the DTE ends up > 32 bits and the NPU's first translation faults with DMA_READ_ERROR. To work around that I had switched the NPU MMU to the v1 compatible (rockchip,iommu), whose ops set GFP_DMA32 and keep the DTE sub-4 GiB. That works in isolation, but because the driver keeps a single global rk_ops, a v1 NPU MMU then trips WARN_ON(rk_ops != ops) against the SoC's v2 instances (VOP/VDEC), which is why I based the series on Simon's per-device-ops work.

So my question: with per-device ops in place, what's the intended way to keep the NPU MMU on v2 *and* cap its DTE at 32 bits on boards with >4 GiB of RAM? A v2 ops variant carrying GFP_DMA32 for this device, or is there a register/ config bit that constrains the DTE address? I'd rather follow the Rockchip intent here than carry the v1 workaround. (Simon, cc'd -- this is right next to your per-device-ops series.)

Quoted 2 lines

Can these operations not be completed via the pmdomain driver? If some operations are controlled by TF-A, are you using open source TF-A?

Most of it is in pmdomain already. Power-on and NoC de-idle are done by the RK3568 NPU power domain (genpd) at power-on -- the driver no longer pokes the PMU directly. Two things remain outside it:

- vdd_npu: I mark it regulator-always-on in DT rather than wiring it as the domain's domain-supply, because as a domain-supply it created a device-link to the I2C PMIC (rk809) and genpd's power-off QoS-save path then hung reading the NPU QoS registers behind the (gated) NoC. If there's a clean way to let genpd own vdd_npu without that I2C ordering deadlock I'd much prefer that -- pointers welcome.

- the NPU compute clock (PVTPLL): set from the driver via SCMI, and only needed for actual compute, not for bring-up.

One more pmdomain observation from testing, possibly relevant to how the NPU domain should be modelled: the domain's power-off/on cycle doesn't reliably re-de-idle the NoC. If the NPU is probed after genpd has already powered the (unused) domain off, the power-on de-idle fails ("failed to set idle on domain 'npu'") and the NPU IOMMU then takes an external abort on its first MMIO access. Probing the NPU before the unused-domain power-off, or marking the domain always-on, both avoid it. Is the NoC de-idle expected to work on a genpd re-power here, or should this domain effectively stay on?

On TF-A: yes -- bl31 is built from upstream arm-trusted-firmware (github.com/ARM-software/arm-trusted-firmware, RK3568 platform), providing PSCI and the SCMI clock service. The only closed blob in the boot chain is Rockchip's DDR init (rkbin), which is the standard situation for mainline RK356x.

Kind regards,
Midgy

Quoted 114 lines · Chaoyi Chen

Hello Midgy,

Quoted 25 lines · Midgy BALON

RFC, not for merge. End-to-end inference does not produce correct output yet (see Status), so per the v2 discussion this is a request for design feedback. It now probes, attaches, and submits cleanly on a stock v7.1-rc6 tree; what remains is one hardware-internal issue.

The RK3568 has a single NVDLA-derived NPU core, the same IP family as the RK3588 NPU the driver already supports; the register layout matches. The RK3568 differences are a 32-bit NPU AXI/IOMMU (vs 40-bit) and explicit PVTPLL/PMU bring-up to power and de-idle the NPU before it is reachable.

Patches:
1-2 rocket: per-SoC data struct, then derive DMA width and core count
from match data (refactors, no functional change).
3 rocket: RK3568 SoC data + PVTPLL/PMU/NOC bring-up.
4 rocket: reset the NPU before detaching the IOMMU on a job timeout
(the detach otherwise stalls a wedged AXI master and WARNs).
5 rocket: keep the IOMMU domain attached across jobs instead of
re-attaching per job (the per-job rk_iommu handshake on the idle NPU MMU is slow and noisy).
6 iommu/rockchip: clear AUTO_GATING bit 1 on the RK356x v1 IOMMU so
the page-walker keeps its clock (else a TLB-miss walk never completes).
7 dt-bindings: add the RK3568 NPU compatible.
8-9 arm64 dts: add the NPU and its IOMMU, and enable them on ROCK 3B.

Quoted 6 lines

Dependency. The NPU MMU is rockchip-iommu v1 (32-bit) while the rest of the RK3568 uses v2 (40-bit). They cannot coexist until the driver carries per-device ops; this series is developed on top of Simon Xue's "iommu/rockchip: Drop global rk_ops in favor of per-device ops" [1]. Without it the NPU IOMMU fails to probe on a full RK3568 boot.

Hmmm. If I understand correctly, the NPU IOMMU should be v2 rather than v1, implying it should support 40-bit PAs. Nevertheless, please note that the upper limit for DTE is 32 bits.

Quoted 7 lines

Power bring-up. The NPU is brought up through the power-domain layer (no driver hack): the NPU power-domain keeps its clocks but drops the pm_qos phandle (qos_npu sits behind the gated NPU NoC, so genpd's power-off QoS save faults reading it), and vdd_npu is marked always-on so the rail is up before genpd de-idles the NoC at power-on. The PMU de-idle then ACKs without PVTPLL running; PVTPLL is only needed for compute.

Can these operations not be completed via the pmdomain driver? If some operations are controlled by TF-A, are you using open source TF-A? Thank you.

Quoted 57 lines

Status. On v7.1-rc6 the driver probes, creates /dev/accel/accel0, attaches an IOMMU domain, and submits jobs; the program controller fetches and broadcasts the command list. Inference output is still wrong, and the cause is split across three layers:
- kernel (this series): the RK3568 differences appear handled;
- mesa/Teflon userspace: still emits RK3588-tuned config, wrong for

    RK3568 (to be filed separately on mesa-dev);

- hardware: with corrected config the NPU's DMA reads the full input and weight tensors (confirmed via its DMA bandwidth counters), but the MAC/output stage never completes, the job times out, and the output stays at the buffer's zero-point. I have not found the missing step; it is not in the command list (replaying the vendor's byte-exact command list behaves the same). Pointers welcome, especially from anyone with RK3568 NPU experience.

Known residual. On the first IOMMU attach the NPU MMU is idle with paging already enabled; the rk_iommu stall/reset handshake does not complete in that state and logs one burst of timeouts before the (kept) domain settles. It is harmless here because the job times out regardless, but it points at an idle-MMU reconfiguration corner the rk_iommu code does not handle on this block.

[1] https://lore.kernel.org/linux-rockchip/[email protected]/

Changes since v2:
- Tagged RFC; now tested on a stock v7.1-rc6 tree.
- Bring-up moved into the power-domain/DT layer (no initcall hack).
- Added the IOMMU detach-on-timeout and attach-once driver fixes.
- Split the driver patch (Heiko): soc_data / match-data / RK3568.
- Derive DMA width and core count from match data; drop the DT rescans.
- Binding describes the hardware; added the missing $ref on rockchip,pmu.
- Disclosed the per-device-ops IOMMU dependency.

Midgy BALON (9):
accel: rocket: Introduce per-SoC rocket_soc_data
accel: rocket: Derive DMA width and core count from match data
accel: rocket: Add RK3568 SoC support
accel: rocket: Reset the NPU before detaching the IOMMU on timeout
accel: rocket: Keep the IOMMU domain attached across jobs iommu/rockchip: Clear AUTO_GATING bit 1 on the RK356x v1 IOMMU
dt-bindings: npu: rockchip,rk3588-rknn-core: Add RK3568
arm64: dts: rockchip: rk356x: Add the NPU and its IOMMU
arm64: dts: rockchip: rk3568-rock-3b: Enable the NPU

 .../npu/rockchip,rk3588-rknn-core.yaml        | 18 ++++-
 .../boot/dts/rockchip/rk3568-rock-3b.dts      | 14 +++-
 arch/arm64/boot/dts/rockchip/rk356x-base.dtsi | 38 +++++++++++
 drivers/accel/rocket/rocket_core.c            | 22 ++++++-
 drivers/accel/rocket/rocket_core.h            | 19 ++++++
 drivers/accel/rocket/rocket_device.c          | 15 ++---
 drivers/accel/rocket/rocket_device.h          |  3 +-
 drivers/accel/rocket/rocket_drv.c             | 66 ++++++++++++++++++-
 drivers/accel/rocket/rocket_job.c             | 35 ++++++++--
 drivers/iommu/rockchip-iommu.c                | 12 ++++
 10 files changed, 219 insertions(+), 23 deletions(-)
Quoted 1 line

base-commit: 52c800fdcf11888ebeb50c3d707f782cc15b66eb

--
Best,
Chaoyi

#28 lore raw

Hi Midgy,

Quoted 22 lines · Midgy Balon

Hi Chaoyi,

Thanks a lot for looking at this -- input from Rockchip is exactly what this series needs.

Quoted 3 lines

Hmmm. If I understand correctly, the NPU IOMMU should be v2 rather than v1, implying it should support 40-bit PAs. Nevertheless, please note that the upper limit for DTE is 32 bits.

Understood, and that 32-bit-DTE note is the crux of the trouble I had, so let me lay out what I see and ask how you'd prefer to solve it.

The mainline node is already v2 (rockchip,rk3568-iommu in rk356x-base.dtsi). The problem on this 8 GiB board: with the v2 ops the page-table allocations (gfp_flags == 0) can land above 4 GiB, so the DTE ends up > 32 bits and the NPU's first translation faults with DMA_READ_ERROR. To work around that I had switched the NPU MMU to the v1 compatible (rockchip,iommu), whose ops set GFP_DMA32 and keep the DTE sub-4 GiB. That works in isolation, but because the driver keeps a single global rk_ops, a v1 NPU MMU then trips WARN_ON(rk_ops != ops) against the SoC's v2 instances (VOP/VDEC), which is why I based the series on Simon's per-device-ops work.

Quoted 7 lines

So my question: with per-device ops in place, what's the intended way to keep the NPU MMU on v2 *and* cap its DTE at 32 bits on boards with >4 GiB of RAM? A v2 ops variant carrying GFP_DMA32 for this device, or is there a register/ config bit that constrains the DTE address? I'd rather follow the Rockchip intent here than carry the v1 workaround. (Simon, cc'd -- this is right next to your per-device-ops series.)

If Simon's method works, please use it :)

Quoted 7 lines
Quoted 2 lines

Can these operations not be completed via the pmdomain driver? If some operations are controlled by TF-A, are you using open source TF-A?

Most of it is in pmdomain already. Power-on and NoC de-idle are done by the RK3568 NPU power domain (genpd) at power-on -- the driver no longer pokes the PMU directly. Two things remain outside it:

Quoted 7 lines

- vdd_npu: I mark it regulator-always-on in DT rather than wiring it as the domain's domain-supply, because as a domain-supply it created a device-link to the I2C PMIC (rk809) and genpd's power-off QoS-save path then hung reading the NPU QoS registers behind the (gated) NoC. If there's a clean way to let genpd own vdd_npu without that I2C ordering deadlock I'd much prefer that -- pointers welcome.

Please refer to the patch below regarding the RK3588 NPU pmdomain. In short, you need to set a "need_regulator" for the RK3568 NPU pmdomain.

https://lore.kernel.org/all/[email protected]/

Quoted 12 lines

- the NPU compute clock (PVTPLL): set from the driver via SCMI, and only needed for actual compute, not for bring-up.

One more pmdomain observation from testing, possibly relevant to how the NPU domain should be modelled: the domain's power-off/on cycle doesn't reliably re-de-idle the NoC. If the NPU is probed after genpd has already powered the (unused) domain off, the power-on de-idle fails ("failed to set idle on domain 'npu'") and the NPU IOMMU then takes an external abort on its first MMIO access. Probing the NPU before the unused-domain power-off, or marking the domain always-on, both avoid it. Is the NoC de-idle expected to work on a genpd re-power here, or should this domain effectively stay on?

Not quite sure what's going on with PVTPLL and NOC. Maybe @Finley knows about this?

Quoted 4 lines

On TF-A: yes -- bl31 is built from upstream arm-trusted-firmware (github.com/ARM-software/arm-trusted-firmware, RK3568 platform), providing PSCI and the SCMI clock service. The only closed blob in the boot chain is Rockchip's DDR init (rkbin), which is the standard situation for mainline RK356x.

Signature · 2 lines
Best, 
Chaoyi
#29 lore raw

Hello Chaoyi,

Thanks -- this is exactly what I needed.

- v2/DTE: will do. I'll keep building on Simon's per-device-ops series -- with that in place the NPU MMU can use the 32-bit-DTE ops (the per-ops GFP_DMA32 that's already in mainline) without the global rk_ops conflict. I'll
keep it as
a stated dependency of the v4 cover letter.

- vdd_npu: I'll switch the RK3568 NPU
power domain to need_regulator + domain-supply = <&vdd_npu> and drop the regulator-always-on workaround. I suspect that's also the right fix for the power-off/on de-idle issue I described -- the always-on was really
just papering
over the domain not being modelled with a regulator. I'll confirm on
the board.

- AUTO_GATING: thanks for the commit references -- I'll keep the bit-31 read-modify-write form with your Suggested-by and write the comment
from those.
For the record: on v7.1-rc6 the NPU MMU also completes translations
on the reset
value (I couldn't reproduce a page-walk stall without the write), so I'll note in the commit that it matches the vendor clock-gating handling rather than fixing a failure I can reproduce here -- happy to drop it if the iommu maintainers would prefer.

- PVTPLL/NoC: I'll follow up with Finley. First I'll check whether the need_regulator change resolves the NoC re-power de-idle on its own;
if it still
I'll bring him the details (the genpd power-on de-idle ack and the BUS_IDLE_ST state).

I'll send a v4 with these. Thanks again for the quick, detailed answers.

Kind regards,
Midgy

Quoted 80 lines · Chaoyi Chen

Hi Midgy,

Quoted 22 lines · Midgy Balon

Hi Chaoyi,

Thanks a lot for looking at this -- input from Rockchip is exactly what this series needs.

Quoted 3 lines

Hmmm. If I understand correctly, the NPU IOMMU should be v2 rather than v1, implying it should support 40-bit PAs. Nevertheless, please note that the upper limit for DTE is 32 bits.

Understood, and that 32-bit-DTE note is the crux of the trouble I had, so let me lay out what I see and ask how you'd prefer to solve it.

The mainline node is already v2 (rockchip,rk3568-iommu in rk356x-base.dtsi). The problem on this 8 GiB board: with the v2 ops the page-table allocations (gfp_flags == 0) can land above 4 GiB, so the DTE ends up > 32 bits and the NPU's first translation faults with DMA_READ_ERROR. To work around that I had switched the NPU MMU to the v1 compatible (rockchip,iommu), whose ops set GFP_DMA32 and keep the DTE sub-4 GiB. That works in isolation, but because the driver keeps a single global rk_ops, a v1 NPU MMU then trips WARN_ON(rk_ops != ops) against the SoC's v2 instances (VOP/VDEC), which is why I based the series on Simon's per-device-ops work.

Quoted 7 lines

So my question: with per-device ops in place, what's the intended way to keep the NPU MMU on v2 *and* cap its DTE at 32 bits on boards with >4 GiB of RAM? A v2 ops variant carrying GFP_DMA32 for this device, or is there a register/ config bit that constrains the DTE address? I'd rather follow the Rockchip intent here than carry the v1 workaround. (Simon, cc'd -- this is right next to your per-device-ops series.)

If Simon's method works, please use it :)

Quoted 7 lines
Quoted 2 lines

Can these operations not be completed via the pmdomain driver? If some operations are controlled by TF-A, are you using open source TF-A?

Most of it is in pmdomain already. Power-on and NoC de-idle are done by the RK3568 NPU power domain (genpd) at power-on -- the driver no longer pokes the PMU directly. Two things remain outside it:

Quoted 7 lines

- vdd_npu: I mark it regulator-always-on in DT rather than wiring it as the domain's domain-supply, because as a domain-supply it created a device-link to the I2C PMIC (rk809) and genpd's power-off QoS-save path then hung reading the NPU QoS registers behind the (gated) NoC. If there's a clean way to let genpd own vdd_npu without that I2C ordering deadlock I'd much prefer that -- pointers welcome.

Please refer to the patch below regarding the RK3588 NPU pmdomain. In short, you need to set a "need_regulator" for the RK3568 NPU pmdomain.

https://lore.kernel.org/all/[email protected]/

Quoted 12 lines

- the NPU compute clock (PVTPLL): set from the driver via SCMI, and only needed for actual compute, not for bring-up.

One more pmdomain observation from testing, possibly relevant to how the NPU domain should be modelled: the domain's power-off/on cycle doesn't reliably re-de-idle the NoC. If the NPU is probed after genpd has already powered the (unused) domain off, the power-on de-idle fails ("failed to set idle on domain 'npu'") and the NPU IOMMU then takes an external abort on its first MMIO access. Probing the NPU before the unused-domain power-off, or marking the domain always-on, both avoid it. Is the NoC de-idle expected to work on a genpd re-power here, or should this domain effectively stay on?

Not quite sure what's going on with PVTPLL and NOC. Maybe @Finley knows about this?

Quoted 4 lines

On TF-A: yes -- bl31 is built from upstream arm-trusted-firmware (github.com/ARM-software/arm-trusted-firmware, RK3568 platform), providing PSCI and the SCMI clock service. The only closed blob in the boot chain is Rockchip's DDR init (rkbin), which is the standard situation for mainline RK356x.

--
Best,
Chaoyi

#30 lore raw

Hello Chaoyi,

Following up on the need_regulator suggestion -- I implemented and tested it on the
board, and unfortunately it doesn't avoid the deadlock on RK3568; it moves it from
boot to the NPU job submit.

What I did: gave the RK3568 NPU power domain a regulator (a DOMAIN_M_R variant with
need_regulator = true), wired domain-supply = <&vdd_npu>, and dropped the regulator-always-on workaround.

Boot is now clean and the NPU probes, but there is a warning during boot:

rockchip-pm-domain ...: Failed to create device link (0x180) with supplier 0-0020 for .../power-domain@6

(0-0020 is the rk809 PMIC that supplies vdd_npu.) Then on the first NPU job submit the board hard-hangs with an RCU stall:

  rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
  rcu:     3-...!: (1 GPs behind) ...
  rcu: rcu_preempt kthread starved for 5115 jiffies! ... RCU_GP_WAIT_FQS(5)
  rcu: Unless rcu_preempt kthread gets sufficient CPU time, OOM is now expected

My reading: vdd_npu is on the rk809 *I2C* PMIC, so when genpd enables/disables the
regulator during the NPU's runtime-PM power transition, the I2C transfer runs in a
context that starves RCU and the box freezes. (I suspect need_regulator is fine on
the RK3588 NPU because its supply isn't behind an I2C PMIC.) The always-on workaround avoids this precisely because genpd never touches the I2C regulator in
that path.

So: for an NPU domain whose supply is an I2C PMIC, is there a supported way to let
genpd own the regulator without performing the I2C op in the power-transition path
(a deferred/async regulator enable, or a flag), or should RK3568 keep vdd_npu as regulator-always-on? For v4 I'll keep always-on unless there's a cleaner path.

Thanks,
Midgy

Quoted 121 lines · Midgy Balon

Hello Chaoyi,

Thanks -- this is exactly what I needed.

- v2/DTE: will do. I'll keep building on Simon's per-device-ops series -- with that in place the NPU MMU can use the 32-bit-DTE ops (the per-ops GFP_DMA32 that's already in mainline) without the global rk_ops conflict. I'll
keep it as
a stated dependency of the v4 cover letter.

- vdd_npu: I'll switch the RK3568 NPU
power domain to need_regulator + domain-supply = <&vdd_npu> and drop the regulator-always-on workaround. I suspect that's also the right fix for the power-off/on de-idle issue I described -- the always-on was really
just papering
over the domain not being modelled with a regulator. I'll confirm on
the board.

- AUTO_GATING: thanks for the commit references -- I'll keep the bit-31 read-modify-write form with your Suggested-by and write the comment
from those.
For the record: on v7.1-rc6 the NPU MMU also completes translations
on the reset
value (I couldn't reproduce a page-walk stall without the write), so I'll note in the commit that it matches the vendor clock-gating handling rather than fixing a failure I can reproduce here -- happy to drop it if the iommu maintainers would prefer.

- PVTPLL/NoC: I'll follow up with Finley. First I'll check whether the need_regulator change resolves the NoC re-power de-idle on its own;
if it still
I'll bring him the details (the genpd power-on de-idle ack and the BUS_IDLE_ST state).

I'll send a v4 with these. Thanks again for the quick, detailed answers.

Kind regards,
Midgy

Quoted 80 lines · Chaoyi Chen

Hi Midgy,

Quoted 22 lines · Midgy Balon

Hi Chaoyi,

Thanks a lot for looking at this -- input from Rockchip is exactly what this series needs.

Quoted 3 lines

Hmmm. If I understand correctly, the NPU IOMMU should be v2 rather than v1, implying it should support 40-bit PAs. Nevertheless, please note that the upper limit for DTE is 32 bits.

Understood, and that 32-bit-DTE note is the crux of the trouble I had, so let me lay out what I see and ask how you'd prefer to solve it.

The mainline node is already v2 (rockchip,rk3568-iommu in rk356x-base.dtsi). The problem on this 8 GiB board: with the v2 ops the page-table allocations (gfp_flags == 0) can land above 4 GiB, so the DTE ends up > 32 bits and the NPU's first translation faults with DMA_READ_ERROR. To work around that I had switched the NPU MMU to the v1 compatible (rockchip,iommu), whose ops set GFP_DMA32 and keep the DTE sub-4 GiB. That works in isolation, but because the driver keeps a single global rk_ops, a v1 NPU MMU then trips WARN_ON(rk_ops != ops) against the SoC's v2 instances (VOP/VDEC), which is why I based the series on Simon's per-device-ops work.

Quoted 7 lines

So my question: with per-device ops in place, what's the intended way to keep the NPU MMU on v2 *and* cap its DTE at 32 bits on boards with >4 GiB of RAM? A v2 ops variant carrying GFP_DMA32 for this device, or is there a register/ config bit that constrains the DTE address? I'd rather follow the Rockchip intent here than carry the v1 workaround. (Simon, cc'd -- this is right next to your per-device-ops series.)

If Simon's method works, please use it :)

Quoted 7 lines
Quoted 2 lines

Can these operations not be completed via the pmdomain driver? If some operations are controlled by TF-A, are you using open source TF-A?

Most of it is in pmdomain already. Power-on and NoC de-idle are done by the RK3568 NPU power domain (genpd) at power-on -- the driver no longer pokes the PMU directly. Two things remain outside it:

Quoted 7 lines

- vdd_npu: I mark it regulator-always-on in DT rather than wiring it as the domain's domain-supply, because as a domain-supply it created a device-link to the I2C PMIC (rk809) and genpd's power-off QoS-save path then hung reading the NPU QoS registers behind the (gated) NoC. If there's a clean way to let genpd own vdd_npu without that I2C ordering deadlock I'd much prefer that -- pointers welcome.

Please refer to the patch below regarding the RK3588 NPU pmdomain. In short, you need to set a "need_regulator" for the RK3568 NPU pmdomain.

https://lore.kernel.org/all/[email protected]/

Quoted 12 lines

- the NPU compute clock (PVTPLL): set from the driver via SCMI, and only needed for actual compute, not for bring-up.

One more pmdomain observation from testing, possibly relevant to how the NPU domain should be modelled: the domain's power-off/on cycle doesn't reliably re-de-idle the NoC. If the NPU is probed after genpd has already powered the (unused) domain off, the power-on de-idle fails ("failed to set idle on domain 'npu'") and the NPU IOMMU then takes an external abort on its first MMIO access. Probing the NPU before the unused-domain power-off, or marking the domain always-on, both avoid it. Is the NoC de-idle expected to work on a genpd re-power here, or should this domain effectively stay on?

Not quite sure what's going on with PVTPLL and NOC. Maybe @Finley knows about this?

Quoted 4 lines

On TF-A: yes -- bl31 is built from upstream arm-trusted-firmware (github.com/ARM-software/arm-trusted-firmware, RK3568 platform), providing PSCI and the SCMI clock service. The only closed blob in the boot chain is Rockchip's DDR init (rkbin), which is the standard situation for mainline RK356x.

--
Best,
Chaoyi

#31 lore raw

Hi Midgy,

Quoted 26 lines · Midgy Balon

Hello Chaoyi,

Following up on the need_regulator suggestion -- I implemented and tested it on the
board, and unfortunately it doesn't avoid the deadlock on RK3568; it moves it from
boot to the NPU job submit.

What I did: gave the RK3568 NPU power domain a regulator (a DOMAIN_M_R variant with
need_regulator = true), wired domain-supply = <&vdd_npu>, and dropped the regulator-always-on workaround.

Boot is now clean and the NPU probes, but there is a warning during boot:

rockchip-pm-domain ...: Failed to create device link (0x180) with supplier 0-0020 for .../power-domain@6

(0-0020 is the rk809 PMIC that supplies vdd_npu.) Then on the first NPU job submit the board hard-hangs with an RCU stall:

  rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
  rcu:     3-...!: (1 GPs behind) ...
  rcu: rcu_preempt kthread starved for 5115 jiffies! ... RCU_GP_WAIT_FQS(5)
  rcu: Unless rcu_preempt kthread gets sufficient CPU time, OOM is now expected
Quoted 11 lines

My reading: vdd_npu is on the rk809 *I2C* PMIC, so when genpd enables/disables the
regulator during the NPU's runtime-PM power transition, the I2C transfer runs in a
context that starves RCU and the box freezes. (I suspect need_regulator is fine on
the RK3588 NPU because its supply isn't behind an I2C PMIC.) The always-on workaround avoids this precisely because genpd never touches the I2C regulator in
that path.

No, they are all controlled by RK809.

And This looks werid. Is your rocket driver compiled as a module? Please try compiling it as a module. When is the above error printed? Please provide the complete boot log.

Quoted 7 lines

So: for an NPU domain whose supply is an I2C PMIC, is there a supported way to let
genpd own the regulator without performing the I2C op in the power-transition path
(a deferred/async regulator enable, or a flag), or should RK3568 keep vdd_npu as regulator-always-on? For v4 I'll keep always-on unless there's a cleaner path.

Signature · 7 lines
Best, 
Chaoyi

_______________________________________________
Linux-rockchip mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-rockchip
#32 lore raw

Hello Chaoyi,

You were right - building rocket as a module fixes it. Thanks for the pointer.

I rebuilt with CONFIG_DRM_ACCEL_ROCKET=m (everything else the same:
need_regulator on
the RK3568 NPU power domain via a DOMAIN_M_R variant, domain-supply = <&vdd_npu>, and the
regulator-always-on workaround dropped). The board now boots cleanly and, more importantly,
an NPU job submit no longer hangs: I ran the test workload five times with no RCU stall and
no freeze.

So with rocket=m the need_regulator approach works on RK3568, and I'll keep it for v4
(domain-supply + need_regulator, instead of marking vdd_npu always-on). rocket=m is the
normal configuration anyway; my earlier hang came from building it =y in a self-contained
image, so it probed in the initcalls (around 2 s) and the genpd -> I2C-PMIC regulator
transition ran before the system was ready. As a module it loads from udev much later
(~6.8 s here), after the I2C controller and regulator core are fully up.

On your question of when the device-link error is printed - it is at power-domain
controller probe, not at the rocket probe:

  [    2.700618] vdd_npu: Bringing 500000uV into 825000-825000uV
  [    2.749637] rockchip-pm-domain fdd90000.power-management:power-controller:
                 Failed to create device link (0x180) with supplier 0-0020 for
                 /power-management@fdd90000/power-controller/power-domain@6
  [    2.945955] platform fde40000.npu: Adding to iommu group 3
  ...
  [    6.840374] rocket: loading out-of-tree module taints kernel.
  [    6.877647] [drm] Initialized rocket 0.0.0 for rknn on minor 0
  [    6.879950] rocket fde40000.npu: Rockchip NPU core 0 version: 0

So the device-link to the rk809 PMIC (0-0020) fails to form at ~2.75 s, well before rocket
loads at ~6.8 s. It is non-fatal here - the vdd_npu rail is brought up by the regulator core
and all jobs run - and there is no "failed to get ack on domain npu" NoC warning this boot
(the always-on kernel had one). The complete boot log is attached.

Two notes / one question:
- This boot used fw_devlink=permissive on the command line. Is the "Failed to create device
link ... supplier 0-0020" at pmdomain probe expected/benign, or is
there a clean way to make
it order correctly (so it also works without permissive, and a =y
build wouldn't deadlock in
the initcalls)?
- (The convolution output is still uniform zero-point / the job times out - that is the
separate NPU compute-completion issue, unrelated to the power-domain
work. Finley, that is
the one I flagged earlier re PVTPLL/NoC.)

Kind regards,
Midgy

Quoted 56 lines · Chaoyi Chen

Hi Midgy,

Quoted 26 lines · Midgy Balon

Hello Chaoyi,

Following up on the need_regulator suggestion -- I implemented and tested it on the
board, and unfortunately it doesn't avoid the deadlock on RK3568; it moves it from
boot to the NPU job submit.

What I did: gave the RK3568 NPU power domain a regulator (a DOMAIN_M_R variant with
need_regulator = true), wired domain-supply = <&vdd_npu>, and dropped the regulator-always-on workaround.

Boot is now clean and the NPU probes, but there is a warning during boot:

rockchip-pm-domain ...: Failed to create device link (0x180) with supplier 0-0020 for .../power-domain@6

(0-0020 is the rk809 PMIC that supplies vdd_npu.) Then on the first NPU job submit the board hard-hangs with an RCU stall:

  rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
  rcu:     3-...!: (1 GPs behind) ...
  rcu: rcu_preempt kthread starved for 5115 jiffies! ... RCU_GP_WAIT_FQS(5)
  rcu: Unless rcu_preempt kthread gets sufficient CPU time, OOM is now expected
Quoted 11 lines

My reading: vdd_npu is on the rk809 *I2C* PMIC, so when genpd enables/disables the
regulator during the NPU's runtime-PM power transition, the I2C transfer runs in a
context that starves RCU and the box freezes. (I suspect need_regulator is fine on
the RK3588 NPU because its supply isn't behind an I2C PMIC.) The always-on workaround avoids this precisely because genpd never touches the I2C regulator in
that path.

No, they are all controlled by RK809.

And This looks werid. Is your rocket driver compiled as a module? Please try compiling it as a module. When is the above error printed? Please provide the complete boot log.

Quoted 7 lines

So: for an NPU domain whose supply is an I2C PMIC, is there a supported way to let
genpd own the regulator without performing the I2C op in the power-transition path
(a deferred/async regulator enable, or a flag), or should RK3568 keep vdd_npu as regulator-always-on? For v4 I'll keep always-on unless there's a cleaner path.

Quoted 3 lines

--
Best,
Chaoyi

sudo reboot
Password:

Login incorrect
rock-3b login: radxa
Password:
Linux rock-3b 6.19.0-rc5-00003-gb16c04e5e619-dirty #65 SMP PREEMPT Thu May 21 11:16:54 CEST 2026 aarch64

The programs included with the Debian GNU/Linux system are free software; the exact distribution terms for each program are described in the individual files in /usr/share/doc/*/copyright.

Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent permitted by applicable law.
Last login: Mon Jun 8 14:14:44 CEST 2026 from 10.160.1.15 on pts/1

[?2004hradxa@rock-3b:~$ [ 1800.807070] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
[ 1800.811885] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
[ 1800.813239] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
[ 1800.814408] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
[ 1800.815613] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
[ 1800.817175] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
[ 1800.821487] hdmi-audio-codec hdmi-audio-codec.11.auto: HDMI: Unknown ELD version 0
dsudo reboot
[?2004l
         Stopping Session 42 of user radxa.
[  OK  ] Removed slice system-modprobe.slice.
[  OK  ] Removed slice system-ssh.slice.
[  OK  ] Stopped target Graphical Interface.
[  OK  ] Stopped target Multi-User System.
[  OK  ] Stopped target Login Prompts.
[  OK  ] Stopped target Remote Encrypted Volumes.
[  OK  ] Stopped target Sound Card.
[  OK  ] Stopped target Timers.
[  OK  ] Stopped Daily apt upgrade and clean activities.
[  OK  ] Stopped Daily apt download activities.
[  OK  ] Stopped Periodic ext4 Onli…ata Check for All Filesystems.
[  OK  ] Stopped Discard unused blocks once a week.
[  OK  ] Stopped Daily man-db regeneration.
[  OK  ] Stopped Daily Cleanup of Temporary Directories.
[  OK  ] Stopped target System Time Synchronized.
[  OK  ] Stopped target System Time Set.
[  OK  ] Stopped target Hardware activated USB gadget.
[  OK  ] Closed Load/Save RF Kill Switch Status /dev/rfkill Watch.
         Unmounting /config...
         Stopping Save/Restore Sound Card State...
         Stopping Avahi mDNS/DNS-SD Stack...
         Stopping Getty on tty1...
         Stopping Netdata, X-Ray Vi…on for your infrastructure!...
         Stopping Enable adbd on supported Radxa products...
         Stopping Enable USB Ethern…on supported Radxa products...
         Stopping Serial Getty on ttyS2...
         Stopping LSB: Set sysfs variables from /etc/sysfs.conf...
         Stopping Journal Service for Namespace netdata...
         Stopping Load/Save Random Seed...
         Stopping Tailscale node agent...
[  OK  ] Stopped Avahi mDNS/DNS-SD Stack.
[  OK  ] Stopped Getty on tty1.
[  OK  ] Stopped Serial Getty on ttyS2.
[  OK  ] Stopped Journal Service for Namespace netdata.
[  OK  ] Unmounted /config.
[  OK  ] Stopped Save/Restore Sound Card State.
[  OK  ] Stopped Load/Save Random Seed.
[  OK  ] Stopped Session 42 of user radxa.
[  OK  ] Stopped LSB: Set sysfs variables from /etc/sysfs.conf.
[  OK  ] Removed slice system-getty.slice.
[  OK  ] Removed slice system-serial\x2dgetty.slice.
         Stopping User Login Management...
         Stopping User Manager for UID 1000...
[  OK  ] Stopped User Manager for UID 1000.
[  OK  ] Stopped User Login Management.
         Stopping User Runtime Directory /run/user/1000...
[  OK  ] Unmounted /run/user/1000.
[  OK  ] Stopped User Runtime Directory /run/user/1000.
[  OK  ] Removed slice User Slice of UID 1000.
         Stopping Permit User Sessions...
[  OK  ] Stopped Permit User Sessions.
[  OK  ] Stopped target Remote File Systems.
[  OK  ] Stopped Tailscale node agent.
[  OK  ] Stopped Netdata, X-Ray Vision for your infrastructure!.
[  OK  ] Stopped target Network is Online.
[  OK  ] Stopped target Host and Network Name Lookups.
[  OK  ] Stopped Network Manager Wait Online.
[  OK  ] Stopped Wait for Network to be Configured.
[ 1809.095948] dwc3 fcc00000.usb: request 00000000d5d2d033 was not queued to ep0out
[ 1809.143743] unloading
[  OK  ] Stopped Enable USB Ethernet on supported Radxa products.
[  OK  ] Stopped Enable adbd on supported Radxa products.
[  OK  ] Stopped target Network.
         Stopping Network Manager...
         Stopping Raise network interfaces...
         Stopping Network Name Resolution...
         Stopping WPA supplicant...
[  OK  ] Stopped Raise network interfaces.
[  OK  ] Stopped Network Name Resolution.
         Stopping Network Service...
[  OK  ] Stopped Network Service.
[ 1809.276260] wlp1s0: deauthenticating from 38:ff:36:bb:04:ac by local choice (Reason: 3=DEAUTH_LEAVING)
[  OK  ] Stopped WPA supplicant.
[  OK  ] Stopped Network Manager.
[  OK  ] Stopped target Network (Pre).
         Stopping D-Bus System Message Bus...
[  OK  ] Stopped D-Bus System Message Bus.
[  OK  ] Stopped target Basic System.
[  OK  ] Stopped target Paths.
[  OK  ] Stopped target Slices.
[  OK  ] Removed slice User and Session Slice.
[  OK  ] Stopped target Sockets.
[  OK  ] Closed Avahi mDNS/DNS-SD Stack Activation Socket.
[  OK  ] Closed D-Bus System Message Bus Socket.
[  OK  ] Closed OpenBSD Secure Shell server socket.
[  OK  ] Closed Journal Varlink Socket for Namespace netdata.
[  OK  ] Removed slice system-syste…\x2djournald\x2dvarlink.slice.
[  OK  ] Closed Journal Socket for Namespace netdata.
[  OK  ] Removed slice system-systemd\x2djournald.slice.
[  OK  ] Stopped target System Initialization.
[  OK  ] Stopped target Local Encrypted Volumes.
[  OK  ] Stopped Dispatch Password …ts to Console Directory Watch.
[  OK  ] Stopped Forward Password R…uests to Wall Directory Watch.
[  OK  ] Stopped target Swap.
         Deactivating swap /mnt/nvme/swapfile...
[  OK  ] Stopped Apply Kernel Variables.
[  OK  ] Stopped Load Kernel Modules.
         Stopping Network Time Synchronization...
         Stopping Update UTMP about System Boot/Shutdown...
[  OK  ] Stopped Network Time Synchronization.
[  OK  ] Deactivated swap /mnt/nvme/swapfile.
         Unmounting /mnt/nvme...
[  OK  ] Stopped Update UTMP about System Boot/Shutdown.
[  OK  ] Stopped Create Volatile Files and Directories.
[  OK  ] Stopped target Local File Systems.
[  OK  ] Unset automount boot-efi.automount.
[  OK  ] Unset automount config.automount.
[ 1811.430689] EXT4-fs (nvme0n1): unmounting filesystem 0d9000fd-1edf-455d-9058-56e0855a1edb.
[  OK  ] Unmounted /mnt/nvme.
[  OK  ] Reached target Unmount All Filesystems.
[  OK  ] Stopped File System Check …d-1edf-455d-9058-56e0855a1edb.
[  OK  ] Removed slice system-systemd\x2dfsck.slice.
[  OK  ] Stopped target Local File Systems (Pre).
[  OK  ] Stopped Create Static Device Nodes in /dev.
[  OK  ] Stopped Create System Users.
[  OK  ] Stopped Remount Root and Kernel File Systems.
[  OK  ] Reached target Shutdown.
[  OK  ] Reached target Final Step.
[  OK  ] Finished Reboot.
[  OK  ] Reached target Reboot.
[ 1811.697989] watchdog: watchdog0: watchdog did not stop!
[ 1811.815097] systemd-shutdown[1]: Syncing filesystems and block devices.
[ 1811.829482] systemd-shutdown[1]: Sending SIGTERM to remaining processes...
[ 1811.853568] systemd-journald[204]: Received SIGTERM from PID 1 (systemd-shutdow).
[ 1811.899622] systemd-shutdown[1]: Sending SIGKILL to remaining processes...
[ 1811.913025] systemd-shutdown[1]: Using hardware watchdog 'Synopsys DesignWare Watchdog', version 0, device /dev/watchdog
[ 1811.917532] systemd-shutdown[1]: Unmounting file systems.
[ 1811.920424] [11189]: Remounting '/' read-only in with options '(null)'.
[ 1811.983886] EXT4-fs (mmcblk0p3): re-mounted d210e617-6a4c-4771-b955-ddd835a32d2b ro.
[ 1812.006902] systemd-shutdown[1]: All filesystems unmounted.
[ 1812.007480] systemd-shutdown[1]: Deactivating swaps.
[ 1812.008275] systemd-shutdown[1]: All swaps deactivated.
[ 1812.008802] systemd-shutdown[1]: Detaching loop devices.
[ 1812.015813] systemd-shutdown[1]: All loop devices detached.
[ 1812.016332] systemd-shutdown[1]: Stopping MD devices.
[ 1812.017106] systemd-shutdown[1]: All MD devices stopped.
[ 1812.017591] systemd-shutdown[1]: Detaching DM devices.
[ 1812.018320] systemd-shutdown[1]: All DM devices detached.
[ 1812.018817] systemd-shutdown[1]: All filesystems, swaps, loop devices, MD devices and DM devices detached.
[ 1812.027209] systemd-shutdown[1]: Syncing filesystems and block devices.
[ 1812.031042] systemd-shutdown[1]: Rebooting.
[ 1812.186088] kvm: exiting hardware virtualization
[ 1812.186550] reboot: Restarting system
DDR 03ea844c5d typ 24/09/03-10:42:57,fwver: v1.23
In
wdqs_if: 0x1010100
LP4/4x derate en, other dram:1x trefi
ddrconfig:7
MID:0xff
LPDDR4X, 324MHz
BW=32 Col=10 Bk=8 CS0 Row=17 CS1 Row=17 CS=2 Die BW=16 Size=8192MB
tdqss_lf: cs0 dqs0: -24ps, dqs1: -96ps, dqs2: -48ps, dqs3: -168ps, 
tdqss_lf: cs1 dqs0: 24ps, dqs1: -72ps, dqs2: -48ps, dqs3: -144ps, 
tdqss_hf: cs0 dqs0: -24ps, dqs1: -96ps, dqs2: -48ps, dqs3: -168ps, 
tdqss_hf: cs1 dqs0: 24ps, dqs1: -72ps, dqs2: -48ps, dqs3: -144ps, 

change to: 324MHz
PHY drv:clk:36,ca:36,DQ:29,odt:240
vrefinner:16%, vrefout:41%
dram drv:40,odt:0
clk skew:0x62

change to: 528MHz
PHY drv:clk:36,ca:36,DQ:29,odt:240
vrefinner:16%, vrefout:41%
dram drv:40,odt:0
clk skew:0x58

change to: 780MHz
PHY drv:clk:36,ca:36,DQ:29,odt:60
vrefinner:16%, vrefout:41%
dram drv:40,odt:0
clk skew:0x58
rx vref: 14.6%
tx vref: 34.0%

change to: 1560MHz(final freq)
PHY drv:clk:36,ca:36,DQ:29,odt:60
vrefinner:16%, vrefout:22%
dram drv:40,odt:80
vref_ca:00000071
clk skew:0x26
rx vref: 15.6%
tx vref: 22.8%
cs 0:
rdtrn RS:
DQS0:0x30, DQS1:0x31, DQS2:0x33, DQS3:0x2c,

min  : 0xd  0xe 0x10  0xe  0x1  0x2  0x8  0x5 , 0xa  0x7  0x2  0x3  0xc  0xb  0xd  0xa ,
      0x11  0xf  0xa  0xa  0x2  0x2  0x2  0x6 , 0xc  0x7  0x6  0x3 0x10 0x11  0xd 0x11 ,
mid  :0x26 0x26 0x29 0x27 0x1b 0x1c 0x22 0x1f ,0x24 0x21 0x1b 0x1c 0x26 0x25 0x26 0x25 ,
      0x2c 0x2a 0x24 0x24 0x1c 0x1b 0x1b 0x20 ,0x24 0x20 0x1e 0x1a 0x27 0x29 0x25 0x29 ,
max  :0x3f 0x3f 0x43 0x40 0x35 0x36 0x3c 0x39 ,0x3e 0x3c 0x35 0x36 0x41 0x3f 0x3f 0x40 ,
      0x47 0x45 0x3e 0x3e 0x36 0x35 0x34 0x3b ,0x3d 0x39 0x37 0x32 0x3f 0x41 0x3e 0x42 ,
range:0x32 0x31 0x33 0x32 0x34 0x34 0x34 0x34 ,0x34 0x35 0x33 0x33 0x35 0x34 0x32 0x36 ,
      0x36 0x36 0x34 0x34 0x34 0x33 0x32 0x35 ,0x31 0x32 0x31 0x2f 0x2f 0x30 0x31 0x31 ,
wrtrn RS:
DQS0:0x22, DQS1:0x13, DQS2:0x1d, DQS3:0x5, 
min  :0x76 0x79 0x7c 0x78 0x6c 0x6f 0x73 0x72 0x72 ,0x63 0x5f 0x59 0x59 0x65 0x63 0x64 0x62 0x5e ,
      0x6a 0x6a 0x66 0x64 0x5d 0x5c 0x5c 0x60 0x63 ,0x55 0x51 0x51 0x4c 0x58 0x5a 0x55 0x5b 0x53 ,
mid  :0x91 0x93 0x95 0x92 0x84 0x88 0x8c 0x8b 0x8a ,0x7c 0x79 0x73 0x73 0x7d 0x7b 0x7c 0x7b 0x77 ,
      0x85 0x84 0x7f 0x7e 0x76 0x75 0x75 0x7a 0x7b ,0x70 0x6b 0x6b 0x66 0x73 0x73 0x6f 0x75 0x6c ,
max  :0xac 0xae 0xaf 0xad 0x9c 0xa1 0xa5 0xa5 0xa2 ,0x96 0x94 0x8e 0x8d 0x96 0x94 0x95 0x94 0x90 ,
      0xa1 0x9f 0x98 0x98 0x90 0x8e 0x8f 0x94 0x94 ,0x8b 0x85 0x86 0x80 0x8f 0x8d 0x89 0x8f 0x86 ,
range:0x36 0x35 0x33 0x35 0x30 0x32 0x32 0x33 0x30 ,0x33 0x35 0x35 0x34 0x31 0x31 0x31 0x32 0x32 ,
      0x37 0x35 0x32 0x34 0x33 0x32 0x33 0x34 0x31 ,0x36 0x34 0x35 0x34 0x37 0x33 0x34 0x34 0x33 ,
cs 1:
rdtrn RS:
DQS0:0x30, DQS1:0x31, DQS2:0x33, DQS3:0x2c, 
min  : 0xd  0xe 0x10  0xe  0x1  0x2  0x8  0x5 , 0xa  0x7  0x2  0x3  0xc  0xb  0xd  0xa ,
      0x11  0xf  0xa  0xa  0x2  0x2  0x2  0x6 , 0xc  0x7  0x6  0x3 0x10 0x11  0xd 0x11 ,
mid  :0x26 0x26 0x29 0x27 0x1b 0x1c 0x22 0x1f ,0x24 0x21 0x1b 0x1c 0x26 0x25 0x26 0x25 ,
      0x2c 0x2a 0x24 0x24 0x1c 0x1b 0x1b 0x20 ,0x24 0x20 0x1e 0x1a 0x27 0x29 0x25 0x29 ,
max  :0x3f 0x3f 0x43 0x40 0x35 0x36 0x3c 0x39 ,0x3e 0x3c 0x35 0x36 0x41 0x3f 0x3f 0x40 ,
      0x47 0x45 0x3e 0x3e 0x36 0x35 0x34 0x3b ,0x3d 0x39 0x37 0x32 0x3f 0x41 0x3e 0x42 ,
range:0x32 0x31 0x33 0x32 0x34 0x34 0x34 0x34 ,0x34 0x35 0x33 0x33 0x35 0x34 0x32 0x36 ,
      0x36 0x36 0x34 0x34 0x34 0x33 0x32 0x35 ,0x31 0x32 0x31 0x2f 0x2f 0x30 0x31 0x31 ,
wrtrn RS:
DQS0:0x22, DQS1:0x13, DQS2:0x1d, DQS3:0x5, 
min  :0x76 0x79 0x7c 0x78 0x6c 0x6f 0x73 0x72 0x72 ,0x63 0x5f 0x59 0x59 0x65 0x63 0x64 0x62 0x5e ,
      0x6a 0x6a 0x66 0x64 0x5d 0x5c 0x5c 0x60 0x63 ,0x55 0x51 0x51 0x4c 0x58 0x5a 0x55 0x5b 0x53 ,
mid  :0x91 0x93 0x95 0x92 0x84 0x88 0x8c 0x8b 0x8a ,0x7c 0x79 0x73 0x73 0x7d 0x7b 0x7c 0x7b 0x77 ,
      0x85 0x84 0x7f 0x7e 0x76 0x75 0x75 0x7a 0x7b ,0x70 0x6b 0x6b 0x66 0x73 0x73 0x6f 0x75 0x6c ,
max  :0xac 0xae 0xaf 0xad 0x9c 0xa1 0xa5 0xa5 0xa2 ,0x96 0x94 0x8e 0x8d 0x96 0x94 0x95 0x94 0x90 ,
      0xa1 0x9f 0x98 0x98 0x90 0x8e 0x8f 0x94 0x94 ,0x8b 0x85 0x86 0x80 0x8f 0x8d 0x89 0x8f 0x86 ,
range:0x36 0x35 0x33 0x35 0x30 0x32 0x32 0x33 0x30 ,0x33 0x35 0x35 0x34 0x31 0x31 0x31 0x32 0x32 ,
      0x37 0x35 0x32 0x34 0x33 0x32 0x33 0x34 0x31 ,0x36 0x34 0x35 0x34 0x37 0x33 0x34 0x34 0x33 ,
CBT RS:
cs:0 min  :0x43 0x39 0x38 0x2d 0x36 0x27 0x3c ,0x44 0x36 0x37 0x2a 0x35 0x2a 0x3d ,
cs:0 mid  :0x7d 0x7d 0x71 0x71 0x70 0x6c 0x69 ,0x7d 0x7a 0x70 0x6d 0x6e 0x6e 0x69 ,
cs:0 max  :0xb7 0xc1 0xab 0xb5 0xab 0xb1 0x97 ,0xb7 0xbf 0xa9 0xb0 0xa7 0xb2 0x96 ,
cs:0 range:0x74 0x88 0x73 0x88 0x75 0x8a 0x5b ,0x73 0x89 0x72 0x86 0x72 0x88 0x59 ,
cs:1 min  :0x42 0x3e 0x39 0x33 0x38 0x2f 0x40 ,0x43 0x3c 0x35 0x2f 0x37 0x30 0x3f ,
cs:1 mid  :0x7f 0x7f 0x75 0x72 0x74 0x6e 0x6e ,0x7f 0x7c 0x72 0x6e 0x73 0x6f 0x6e ,
cs:1 max  :0xbd 0xc0 0xb2 0xb2 0xb0 0xad 0x9c ,0xbc 0xbc 0xb0 0xae 0xaf 0xae 0x9d ,

cs:1 range:0x7b 0x82 0x79 0x7f 0x78 0x7e 0x5c ,0x79 0x80 0x7b 0x7f 0x78 0x7e 0x5e , out

<debug_uart>
dmc
pinctrl
serial@fe660000

U-Boot SPL 2025.07-rc4-dirty (Mar 25 2026 - 16:29:49 +0100) mmc@fe310000
clock-controller@fdd20000 clock-controller@fdd00000
mmc@fe2b0000
Trying to boot from MMC1
## Checking hash(es) for config config-1 ... OK
## Checking hash(es) for Image atf-1 ... sha256+ OK ## Checking hash(es) for Image u-boot ... sha256+ OK ## Checking hash(es) for Image fdt-1 ... sha256+ OK ## Checking hash(es) for Image atf-2 ... sha256+ OK ## Checking hash(es) for Image atf-3 ... sha256+ OK

NOTICE:  BL31: v2.14.0(release):8dae086
NOTICE:  BL31: Built : 14:26:06, Mar 25 2026
NOTICE:  BL31: Rockchip release version: v1.0

<debug_uart>
A
B
C
D
E
F
G
H
pinctrl
serial@fe660000

U-Boot 2025.07-rc4-dirty (Mar 25 2026 - 16:29:49 +0100)

clock-controller@fdd20000
clock-controller@fdd00000
Model: Radxa ROCK 3B
nvmem@fe38c000
SoC: RK3568
I
DRAM: dmc
J
8 GiB (total 7.7 GiB)
io-domains
clock-controller@fdd00000
pinctrl
i2c@fdd40000
clock-controller@fdd20000
pmic@20
PMIC: RK809 (on=0x02, off=0x00)
LDO_REG6
LDO_REG4
LDO_REG5
DCDC_REG5
SWITCH_REG1
DCDC_REG1
DCDC_REG2
DCDC_REG3
LDO_REG2
LDO_REG3
LDO_REG7
LDO_REG8
SWITCH_REG2
led-0
gpio@fdd60000
serial@fe660000

Core:  340 devices, 33 uclasses, devicetree: separate
MMC:   mmc@fe310000
mmc@fe2b0000
mmc@fe2b0000: 1, mmc@fe310000: 0
Loading Environment from nowhere... OK
In:    serial@fe660000
Out:   serial@fe660000
Err:   serial@fe660000
Model: Radxa ROCK 3B

nvmem@fe38c000
SoC: RK3568
saradc@fe720000
reset
Net: ethernet@fe010000
gpio@fe760000
ethernet@fe2a0000
eth1: ethernet@fe010000, eth0: ethernet@fe2a0000
Hit any key to stop autoboot: 2  1  0
bootstd
Scanning for bootflows in all bootdevs

Seq  Method       State   Uclass    Part  Name                      Filename
---  -----------  ------  --------  ----  ------------------------  ----------------

vbe_simple
vbe_simple
Scanning global bootmeth 'efi_mgr':
7mail protected]
Card did not respond to voltage select! : -110
[email protected]
rng@fe388000
psci
0 efi_mgr ready (none) 0 <NULL>
** Booting bootflow '<NULL>' with efi_mgr
Loading Boot0000 'mmc 0' failed
EFI boot manager: Cannot load any image
Boot failed (err=-14)
Scanning bootdev '[email protected]':
[email protected]
Card did not respond to voltage select! : -110
Scanning bootdev '[email protected]':
1 extlinux ready mmc 3 [email protected] /boot/extlinux/extlinux.conf
** Booting bootflow '[email protected]_3' with extlinux
U-Boot menu
1: Mainline 6.19 NPU IOMMU (default)
2: Mainline 6.19 NPU non-IOMMU (fallback)
3: Mainline 6.19 Rocket accel driver (test)
4: Mainline 7.1-rc6 Rocket NPU (test)
5: Mainline 7.1-rc6 Rocket NPU + Chaoyi AUTO_GATING BIT(31) (test)
6: Mainline 7.1-rc6 NPU need_regulator + rocket=m (test)
Enter choice: 6
6: Mainline 7.1-rc6 NPU need_regulator + rocket=m (test)
Retrieving file: /boot/Image-7.1-needreg-m
Retrieving file: /boot/initrd.img-7.1.0-rc6-00007-g043be7a551c4
append: root=UUID=d210e617-6a4c-4771-b955-ddd835a32d2b rw rootwait earlycon console=ttyFIQ0,1500000n8 console=ttyS2,1500000n8 clk_ignore_unused cma=128M kernel.panic=5 fw_devlink=permissive Retrieving file: /boot/rk3568-rock-3b-7.1-needreg-m.dtb
## Flattened Device Tree blob at 12000000
Booting using the fdt blob at 0x12000000
Working FDT set to 12000000
Loading Ramdisk to eb2c0000, end ecead483 ... OK
Loading Device Tree to 00000000eb2ac000, end 00000000eb2bf1e3 ... OK
Working FDT set to eb2ac000

Starting kernel ...

[    0.000000] Booting Linux on physical CPU 0x0000000000 [0x412fd050]
[    0.000000] Linux version 7.1.0-rc6-chaoyi-00011-ga31e2e6fae27 (radxa@rock-3b) (gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for Debian) 2.35.2) #5 SMP PREEMPT Tue Jun  9 12:40:22 CEST 2026
[    0.000000] KASLR enabled
[    0.000000] Machine model: Radxa ROCK 3B
[    0.000000] efi: UEFI not found.
[    0.000000] earlycon: uart0 at MMIO32 0x00000000fe660000 (options '1500000n8')
[    0.000000] printk: legacy bootconsole [uart0] enabled
[    0.000000] OF: reserved mem: 0x000000000010f000..0x000000000010f0ff (0 KiB) nomap non-reusable shmem@10f000
[    0.000000] NUMA: Faking a node at [mem 0x0000000000200000-0x00000001ffffffff]
[    0.000000] NODE_DATA(0) allocated [mem 0x1ff01df80-0x1ff02067f]
[    0.000000] cma: Reserved 128 MiB at 0x00000000e3200000
[    0.000000] psci: probing for conduit method from DT.
[    0.000000] psci: PSCIv1.1 detected in firmware.
[    0.000000] psci: Using standard PSCI v0.2 function IDs
[    0.000000] psci: MIGRATE_INFO_TYPE not supported.
[    0.000000] psci: SMC Calling Convention v1.5
[    0.000000] Zone ranges:
[    0.000000]   DMA      [mem 0x0000000000200000-0x00000000ffffffff]
[    0.000000]   DMA32    empty
[    0.000000]   Normal   [mem 0x0000000100000000-0x00000001ffffffff]
[    0.000000] Movable zone start for each node
[    0.000000] Early memory node ranges
[    0.000000]   node   0: [mem 0x0000000000200000-0x00000000efffffff]
[    0.000000]   node   0: [mem 0x0000000100000000-0x00000001ffffffff]
[    0.000000] Initmem setup node 0 [mem 0x0000000000200000-0x00000001ffffffff]
[    0.000000] On node 0, zone DMA: 512 pages in unavailable ranges
[    0.000000] percpu: Embedded 26 pages/cpu s67288 r8192 d31016 u106496
[    0.000000] Detected VIPT I-cache on CPU0
[    0.000000] CPU features: detected: GICv3 CPU interface
[    0.000000] CPU features: detected: Virtualization Host Extensions
[    0.000000] CPU features: kernel page table isolation forced ON by KASLR
[    0.000000] CPU features: detected: Kernel page table isolation (KPTI)
[    0.000000] CPU features: detected: ARM errata 1165522, 1319367, or 1530923
[    0.000000] alternatives: applying boot alternatives
[    0.000000] Kernel command line: root=UUID=d210e617-6a4c-4771-b955-ddd835a32d2b rw rootwait earlycon console=ttyFIQ0,1500000n8 console=ttyS2,1500000n8 clk_ignore_unused cma=128M kernel.panic=5 fw_devlink=permissive
[    0.000000] printk: log buffer data + meta data: 131072 + 458752 = 589824 bytes
[    0.000000] Dentry cache hash table entries: 1048576 (order: 11, 8388608 bytes, linear)
[    0.000000] Inode-cache hash table entries: 524288 (order: 10, 4194304 bytes, linear)
[    0.000000] software IO TLB: area num 4.
[    0.000000] software IO TLB: mapped [mem 0x00000000df200000-0x00000000e3200000] (64MB)
[    0.000000] Fallback order for Node 0: 0 
[    0.000000] Built 1 zonelists, mobility grouping on.  Total pages: 2031104
[    0.000000] Policy zone: Normal
[    0.000000] mem auto-init: stack:off, heap alloc:off, heap free:off
[    0.000000] SLUB: HWalign=64, Order=0-3, MinObjects=0, CPUs=4, Nodes=1
[    0.000000] rcu: Preemptible hierarchical RCU implementation.
[    0.000000] rcu: 	RCU event tracing is enabled.
[    0.000000] rcu: 	RCU restricting CPUs from NR_CPUS=512 to nr_cpu_ids=4.
[    0.000000] 	Trampoline variant of Tasks RCU enabled.
[    0.000000] 	Tracing variant of Tasks RCU enabled.
[    0.000000] rcu: RCU calculated value of scheduler-enlistment delay is 25 jiffies.
[    0.000000] rcu: Adjusting geometry for rcu_fanout_leaf=16, nr_cpu_ids=4
[    0.000000] RCU Tasks: Setting shift to 2 and lim to 1 rcu_task_cb_adjust=1 rcu_task_cpu_ids=4.
[    0.000000] NR_IRQS: 64, nr_irqs: 64, preallocated irqs: 0
[    0.000000] GIC: enabling workaround for GICv3: non-coherent attribute
[    0.000000] GICv3: GIC: Using split EOI/Deactivate mode
[    0.000000] GICv3: 320 SPIs implemented
[    0.000000] GICv3: 0 Extended SPIs implemented
[    0.000000] GICv3: MBI range [296:319]
[    0.000000] GICv3: Using MBI frame 0x00000000fd410000
[    0.000000] Root IRQ handler: gic_handle_irq
[    0.000000] GICv3: GICv3 features: 16 PPIs
[    0.000000] GICv3: GICD_CTLR.DS=0, SCR_EL3.FIQ=0
[    0.000000] GICv3: CPU0: found redistributor 0 region 0:0x00000000fd460000
[    0.000000] ITS [mem 0xfd440000-0xfd45ffff]
[    0.000000] GIC: enabling workaround for ITS: Rockchip erratum RK3568002
[    0.000000] GIC: enabling workaround for ITS: non-coherent attribute
[    0.000000] ITS@0x00000000fd440000: allocated 8192 Devices @410000 (indirect, esz 8, psz 64K, shr 0)
[    0.000000] ITS@0x00000000fd440000: allocated 32768 Interrupt Collections @420000 (flat, esz 2, psz 64K, shr 0)
[    0.000000] ITS: using cache flushing for cmd queue
[    0.000000] GICv3: using LPI property table @0x0000000000430000
[    0.000000] GIC: using cache flushing for LPI property table
[    0.000000] GICv3: CPU0: using allocated LPI pending table @0x0000000000440000
[    0.000000] rcu: srcu_init: Setting srcu_struct sizes based on contention.
[    0.000000] arch_timer: cp15 timer running at 24.00MHz (phys).
[    0.000000] clocksource: arch_sys_counter: mask: 0xffffffffffffff max_cycles: 0x588fe9dc0, max_idle_ns: 440795202592 ns
[    0.000001] sched_clock: 56 bits at 24MHz, resolution 41ns, wraps every 4398046511097ns
[    0.004345] Console: colour dummy device 80x25
[    0.004930] Calibrating delay loop (skipped), value calculated using timer frequency.. 48.00 BogoMIPS (lpj=96000)
[    0.005946] pid_max: default: 32768 minimum: 301
[    0.006823] Mount-cache hash table entries: 16384 (order: 5, 131072 bytes, linear)
[    0.007612] Mountpoint-cache hash table entries: 16384 (order: 5, 131072 bytes, linear)
[    0.008917] VFS: Finished mounting rootfs on nullfs
[    0.012680] rcu: Hierarchical SRCU implementation.
[    0.013171] rcu: 	Max phase no-delay instances is 1000.
[    0.014123] Timer migration: 1 hierarchy levels; 8 children per group; 1 crossnode level
[    0.015214] fsl-mc MSI: msi-controller@fd440000 domain created
[    0.020987] EFI services will not be available.
[    0.021934] smp: Bringing up secondary CPUs ...
[    0.023391] Detected VIPT I-cache on CPU1
[    0.023553] GICv3: CPU1: found redistributor 100 region 0:0x00000000fd480000
[    0.023577] GICv3: CPU1: using allocated LPI pending table @0x0000000000450000
[    0.023640] CPU1: Booted secondary processor 0x0000000100 [0x412fd050]
[    0.024763] Detected VIPT I-cache on CPU2
[    0.024907] GICv3: CPU2: found redistributor 200 region 0:0x00000000fd4a0000
[    0.024929] GICv3: CPU2: using allocated LPI pending table @0x0000000000460000
[    0.024978] CPU2: Booted secondary processor 0x0000000200 [0x412fd050]
[    0.026138] Detected VIPT I-cache on CPU3
[    0.026279] GICv3: CPU3: found redistributor 300 region 0:0x00000000fd4c0000
[    0.026303] GICv3: CPU3: using allocated LPI pending table @0x0000000000470000
[    0.026352] CPU3: Booted secondary processor 0x0000000300 [0x412fd050]
[    0.026541] smp: Brought up 1 node, 4 CPUs
[    0.034241] SMP: Total of 4 processors activated.
[    0.034710] CPU: All CPU(s) started at EL2
[    0.035116] CPU features: detected: 32-bit EL0 Support
[    0.035620] CPU features: detected: 32-bit EL1 Support
[    0.036127] CPU features: detected: Data cache clean to the PoU not required for I/D coherence
[    0.036969] CPU features: detected: Common not Private translations
[    0.037582] CPU features: detected: CRC32 instructions
[    0.038125] CPU features: detected: RCpc load-acquire (LDAPR)
[    0.038694] CPU features: detected: LSE atomic instructions
[    0.039244] CPU features: detected: Privileged Access Never
[    0.039791] CPU features: detected: PMUv3
[    0.040186] CPU features: detected: RAS Extension Support
[    0.040719] CPU features: detected: XNX
[    0.041102] CPU features: detected: Speculative Store Bypassing Safe (SSBS)
[    0.041855] alternatives: applying system-wide alternatives
[    0.047418] CPU features: detected: ICV_DIR_EL1 trapping
[    0.048557] Memory: 7683552K/8124416K available (20736K kernel code, 5018K rwdata, 14920K rodata, 12480K init, 724K bss, 305152K reserved, 131072K cma-reserved)
[    0.053188] devtmpfs: initialized
[    0.076194] clocksource: jiffies: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 7645041785100000 ns
[    0.077193] posixtimers hash table entries: 2048 (order: 3, 32768 bytes, linear)
[    0.077991] futex hash table entries: 1024 (65536 bytes on 1 NUMA nodes, total 64 KiB, linear).
[    0.080642] 2G module region forced by RANDOMIZE_MODULE_REGION_FULL
[    0.081302] 0 pages in range for non-PLT usage
[    0.081310] 510752 pages in range for PLT usage
[    0.087203] DMI: not present or invalid.
[    0.092012] NET: Registered PF_NETLINK/PF_ROUTE protocol family
[    0.094182] DMA: preallocated 1024 KiB GFP_KERNEL pool for atomic allocations
[    0.095295] DMA: preallocated 1024 KiB GFP_KERNEL|GFP_DMA pool for atomic allocations
[    0.096163] audit: initializing netlink subsys (disabled)
[    0.096954] audit: type=2000 audit(0.092:1): state=initialized audit_enabled=0 res=1
[    0.102019] thermal_sys: Registered thermal governor 'step_wise'
[    0.102040] thermal_sys: Registered thermal governor 'power_allocator'
[    0.102800] cpuidle: using governor menu
[    0.104380] hw-breakpoint: found 6 breakpoint and 4 watchpoint registers.
[    0.105274] ASID allocator initialised with 32768 entries
[    0.111040] Serial: AMBA PL011 UART driver
[    0.140432] /vop@fe040000: Fixed dependency cycle(s) with /hdmi@fe0a0000
[    0.141206] /hdmi@fe0a0000: Fixed dependency cycle(s) with /vop@fe040000
[    0.157642] /pcie@fe260000: Fixed dependency cycle(s) with /pcie@fe260000/legacy-interrupt-controller
[    0.178928] rockchip-gpio fdd60000.gpio: probed /pinctrl/gpio@fdd60000
[    0.180729] rockchip-gpio fe740000.gpio: probed /pinctrl/gpio@fe740000
[    0.182380] rockchip-gpio fe750000.gpio: probed /pinctrl/gpio@fe750000
[    0.184237] rockchip-gpio fe760000.gpio: probed /pinctrl/gpio@fe760000
[    0.185880] rockchip-gpio fe770000.gpio: probed /pinctrl/gpio@fe770000
[    0.192768] /pcie@fe280000: Fixed dependency cycle(s) with /pcie@fe280000/legacy-interrupt-controller
[    0.196426] /hdmi@fe0a0000: Fixed dependency cycle(s) with /hdmi-con
[    0.197174] /hdmi-con: Fixed dependency cycle(s) with /hdmi@fe0a0000
[    0.207741] HugeTLB: registered 1.00 GiB page size, pre-allocated 0 pages
[    0.208425] HugeTLB: 0 KiB vmemmap can be freed for a 1.00 GiB page
[    0.209043] HugeTLB: registered 32.0 MiB page size, pre-allocated 0 pages
[    0.209708] HugeTLB: 0 KiB vmemmap can be freed for a 32.0 MiB page
[    0.210324] HugeTLB: registered 2.00 MiB page size, pre-allocated 0 pages
[    0.210986] HugeTLB: 0 KiB vmemmap can be freed for a 2.00 MiB page
[    0.211638] HugeTLB: registered 64.0 KiB page size, pre-allocated 0 pages
[    0.212304] HugeTLB: 0 KiB vmemmap can be freed for a 64.0 KiB page
[    0.216832] ACPI: Interpreter disabled.
[    0.226115] iommu: Default domain type: Translated
[    0.226612] iommu: DMA domain TLB invalidation policy: strict mode
[    0.229120] SCSI subsystem initialized
[    0.230219] usbcore: registered new interface driver usbfs
[    0.230824] usbcore: registered new interface driver hub
[    0.231419] usbcore: registered new device driver usb
[    0.234769] pps_core: LinuxPPS API ver. 1 registered
[    0.235266] pps_core: Software ver. 5.3.6 - Copyright 2005-2007 Rodolfo Giometti <[email protected]>
[    0.236205] PTP clock support registered
[    0.236952] EDAC MC: Ver: 3.0.0
[    0.238218] scmi_core: SCMI protocol bus registered
[    0.241592] FPGA manager framework
[    0.244246] vgaarb: loaded
[    0.245545] clocksource: Switched to clocksource arch_sys_counter
[    0.246553] VFS: Disk quotas dquot_6.6.0
[    0.246977] VFS: Dquot-cache hash table entries: 512 (order 0, 4096 bytes)
[    0.248448] pnp: PnP ACPI: disabled
[    0.262673] NET: Registered PF_INET protocol family
[    0.263673] IP idents hash table entries: 131072 (order: 8, 1048576 bytes, linear)
[    0.271846] tcp_listen_portaddr_hash hash table entries: 4096 (order: 4, 65536 bytes, linear)
[    0.272763] Table-perturb hash table entries: 65536 (order: 6, 262144 bytes, linear)
[    0.273581] TCP established hash table entries: 65536 (order: 7, 524288 bytes, linear)
[    0.274889] TCP bind hash table entries: 65536 (order: 9, 2097152 bytes, linear)
[    0.277519] TCP: Hash tables configured (established 65536 bind 65536)
[    0.278369] UDP hash table entries: 4096 (order: 6, 262144 bytes, linear)
[    0.279544] NET: Registered PF_UNIX/PF_LOCAL protocol family
[    0.280868] RPC: Registered named UNIX socket transport module.
[    0.281459] RPC: Registered udp transport module.
[    0.281967] RPC: Registered tcp transport module.
[    0.282433] RPC: Registered tcp-with-tls transport module.
[    0.282973] RPC: Registered tcp NFSv4.1 backchannel transport module.
[    0.285066] PCI: CLS 0 bytes, default 64
[    0.285751] Unpacking initramfs...
[    0.294394] kvm [1]: nv: 570 coarse grained trap handlers
[    0.295328] kvm [1]: nv: 710 fine grained trap handlers
[    0.296530] kvm [1]: IPA Size Limit: 40 bits
[    0.296999] kvm [1]: GICv3: no GICV resource entry
[    0.297475] kvm [1]: disabling GICv2 emulation
[    0.297989] kvm [1]: GIC system register CPU interface enabled
[    0.298599] kvm [1]: vgic interrupt IRQ9
[    0.299039] kvm [1]: VHE mode initialized successfully
[    0.302206] Initialise system trusted keyrings
[    0.303030] workingset: timestamp_bits=42 (anon: 38) max_order=21 bucket_order=0 (anon: 0)
[    0.304530] squashfs: version 4.0 (2009/01/31) Phillip Lougher
[    0.305699] NFS: Registering the id_resolver key type
[    0.306250] Key type id_resolver registered
[    0.306668] Key type id_legacy registered
[    0.307105] nfs4filelayout_init: NFSv4 File Layout Driver Registering...
[    0.307767] nfs4flexfilelayout_init: NFSv4 Flexfile Layout Driver Registering...
[    0.308845] 9p: Installing v9fs 9p2000 file system support
[    0.378356] Key type asymmetric registered
[    0.378791] Asymmetric key parser 'x509' registered
[    0.379447] Block layer SCSI generic (bsg) driver version 0.4 loaded (major 244)
[    0.380179] io scheduler mq-deadline registered
[    0.380628] io scheduler kyber registered
[    0.381094] io scheduler bfq registered
[    2.017986] Freeing initrd memory: 28596K
[    2.025771] ledtrig-cpu: registered to indicate activity on CPUs
[    2.036309] phy phy-fe8c0000.phy.3: lane number 0, val 1
[    2.036934] rockchip-dw-pcie 3c0800000.pcie: host bridge /pcie@fe280000 ranges:
[    2.037737] rockchip-dw-pcie 3c0800000.pcie:       IO 0x00f0100000..0x00f01fffff -> 0x00f0100000
[    2.038623] rockchip-dw-pcie 3c0800000.pcie:      MEM 0x00f0200000..0x00f1ffffff -> 0x00f0200000
[    2.039516] rockchip-dw-pcie 3c0800000.pcie:      MEM 0x0380000000..0x03bfffffff -> 0x0380000000
[    2.048377] rockchip-dw-pcie 3c0800000.pcie: iATU: unroll T, 8 ob, 8 ib, align 64K, limit 8G
[    2.349576] rockchip-dw-pcie 3c0800000.pcie: PCIe Gen.3 x2 link up
[    2.350587] rockchip-dw-pcie 3c0800000.pcie: PCI host bridge to bus 0002:20
[    2.351289] pci_bus 0002:20: root bus resource [bus 20-2f]
[    2.351840] pci_bus 0002:20: root bus resource [io  0x0000-0xfffff] (bus address [0xf0100000-0xf01fffff])
[    2.352778] pci_bus 0002:20: root bus resource [mem 0xf0200000-0xf1ffffff]
[    2.353456] pci_bus 0002:20: root bus resource [mem 0x380000000-0x3bfffffff]
[    2.354237] pci 0002:20:00.0: [1d87:3566] type 01 class 0x060400 PCIe Root Port
[    2.354983] pci 0002:20:00.0: ROM [mem 0x00000000-0x0000ffff pref]
[    2.355597] pci 0002:20:00.0: PCI bridge to [bus 01-ff]
[    2.356120] pci 0002:20:00.0:   bridge window [io  0x0000-0x0fff]
[    2.356722] pci 0002:20:00.0:   bridge window [mem 0x00000000-0x000fffff]
[    2.357397] pci 0002:20:00.0:   bridge window [mem 0x00000000-0x000fffff 64bit pref]
[    2.358247] pci 0002:20:00.0: supports D1 D2
[    2.358677] pci 0002:20:00.0: PME# supported from D0 D1 D3hot
[    2.364773] pci 0002:20:00.0: Primary bus is hard wired to 0
[    2.365345] pci 0002:20:00.0: bridge configuration invalid ([bus 01-ff]), reconfiguring
[    2.366452] pci 0002:21:00.0: [10ec:5765] type 00 class 0x010802 PCIe Endpoint
[    2.367380] pci 0002:21:00.0: BAR 0 [mem 0x00000000-0x00003fff 64bit]
[    2.368045] pci 0002:21:00.0: BAR 5 [mem 0x00000000-0x00001fff]
[    2.369416] pci 0002:21:00.0: 15.752 Gb/s available PCIe bandwidth, limited by 8.0 GT/s PCIe x2 link at 0002:20:00.0 (capable of 31.504 Gb/s with 8.0 GT/s PCIe x4 link)
[    2.377603] pci 0002:21:00.0: ASPM: default states L1
[    2.378208] pci_bus 0002:21: busn_res: [bus 21-2f] end is updated to 21
[    2.378898] pci 0002:20:00.0: bridge window [mem 0xf0200000-0xf02fffff]: assigned
[    2.379643] pci 0002:20:00.0: ROM [mem 0xf0300000-0xf030ffff pref]: assigned
[    2.380348] pci 0002:21:00.0: BAR 0 [mem 0xf0200000-0xf0203fff 64bit]: assigned
[    2.381111] pci 0002:21:00.0: BAR 5 [mem 0xf0204000-0xf0205fff]: assigned
[    2.381824] pci 0002:20:00.0: PCI bridge to [bus 21]
[    2.382325] pci 0002:20:00.0:   bridge window [mem 0xf0200000-0xf02fffff]
[    2.383002] pci_bus 0002:20: resource 4 [io  0x0000-0xfffff]
[    2.383562] pci_bus 0002:20: resource 5 [mem 0xf0200000-0xf1ffffff]
[    2.384181] pci_bus 0002:20: resource 6 [mem 0x380000000-0x3bfffffff]
[    2.384817] pci_bus 0002:21: resource 1 [mem 0xf0200000-0xf02fffff]
[    2.390163] pcieport 0002:20:00.0: PME: Signaling with IRQ 31
[    2.391304] pcieport 0002:20:00.0: AER: enabled with IRQ 32
[    2.461045] dma-pl330 fe530000.dma-controller: Loaded driver for PL330 DMAC-241330
[    2.461857] dma-pl330 fe530000.dma-controller: 	DBUFF-128x8bytes Num_Chans-8 Num_Peri-32 Num_Events-16
[    2.465804] dma-pl330 fe550000.dma-controller: Loaded driver for PL330 DMAC-241330
[    2.466560] dma-pl330 fe550000.dma-controller: 	DBUFF-128x8bytes Num_Chans-8 Num_Peri-32 Num_Events-16
[    2.506402] Serial: 8250/16550 driver, 4 ports, IRQ sharing enabled
[    2.512775] printk: legacy console [ttyS2] disabled
[    2.513873] fe660000.serial: ttyS2 at MMIO 0xfe660000 (irq = 37, base_baud = 1500000) is a 16550A
[    2.514882] printk: legacy console [ttyS2] enabled
[    2.514882] printk: legacy console [ttyS2] enabled
[    2.515784] printk: legacy bootconsole [uart0] disabled
[    2.515784] printk: legacy bootconsole [uart0] disabled
[    2.523685] msm_serial: driver initialized
[    2.525677] SuperH (H)SCI(F) driver initialized
[    2.526792] STM32 USART driver initialized
[    2.538263] random: crng init done
[    2.538643] platform fdea0000.video-codec: Adding to iommu group 0
[    2.541147] platform fdee0000.video-codec: Adding to iommu group 1
[    2.543525] platform fe040000.vop: Adding to iommu group 2
[    2.545348] Error: Driver 'efi-framebuffer' is already registered, aborting...
[    2.556173] loop: module loaded
[    2.560241] megasas: 07.734.00.00-rc1
[    2.563304] nvme nvme0: pci function 0002:21:00.0
[    2.563774] nvme 0002:21:00.0: enabling device (0000 -> 0002)
[    2.581994] tun: Universal TUN/TAP device driver, 1.6
[    2.585959] thunder_xcv, ver 1.0
[    2.586353] thunder_bgx, ver 1.0
[    2.586717] nicpf, ver 1.0
[    2.591410] e1000: Intel(R) PRO/1000 Network Driver
[    2.591864] e1000: Copyright (c) 1999-2006 Intel Corporation.
[    2.592455] e1000e: Intel(R) PRO/1000 Network Driver
[    2.592904] e1000e: Copyright(c) 1999 - 2015 Intel Corporation.
[    2.593517] igb: Intel(R) Gigabit Ethernet Network Driver
[    2.594043] igb: Copyright (c) 2007-2014 Intel Corporation.
[    2.594615] igbvf: Intel(R) Gigabit Virtual Function Network Driver
[    2.595175] igbvf: Copyright (c) 2009 - 2012 Intel Corporation.
[    2.596815] sky2: driver version 1.30
[    2.601784] rk_gmac-dwmac fe010000.ethernet: IRQ sfty not found
[    2.604039] rk_gmac-dwmac fe2a0000.ethernet: IRQ sfty not found
[    2.606627] usbcore: registered new device driver r8152-cfgselector
[    2.607267] usbcore: registered new interface driver r8152
[    2.607825] usbcore: registered new interface driver asix
[    2.608408] usbcore: registered new interface driver ax88179_178a
[    2.609582] VFIO - User Level meta-driver version: 0.3
[    2.624316] ehci-platform fd800000.usb: EHCI Host Controller
[    2.624342] ohci-platform fd840000.usb: Generic Platform OHCI controller
[    2.624869] ehci-platform fd800000.usb: new USB bus registered, assigned bus number 1
[    2.625452] ohci-platform fd840000.usb: new USB bus registered, assigned bus number 2
[    2.625481] usbcore: registered new interface driver usb-storage
[    2.626353] ehci-platform fd800000.usb: irq 46, io mem 0xfd800000
[    2.627002] ohci-platform fd840000.usb: irq 47, io mem 0xfd840000
[    2.636410] i2c_dev: i2c /dev entries driver
[    2.637620] ehci-platform fd800000.usb: USB 2.0 started, EHCI 1.00
[    2.639265] hub 1-0:1.0: USB hub found
[    2.639666] hub 1-0:1.0: 1 port detected
[    2.643281] fan53555-regulator 0-001c: FAN53555 Option[12] Rev[15] Detected!
[    2.686641] hub 2-0:1.0: USB hub found
[    2.687081] hub 2-0:1.0: 1 port detected
[    2.700618] vdd_npu: Bringing 500000uV into 825000-825000uV
[    2.701202] nvme nvme0: passthrough uses implicit buffer lengths
[    2.709720] vdda0v9_image: Bringing 600000uV into 900000-900000uV
[    2.714524] nvme nvme0: allocated 64 MiB host memory buffer (16 segments).
[    2.737292] vcca1v8_image: Bringing 600000uV into 1800000-1800000uV
[    2.749637] rockchip-pm-domain fdd90000.power-management:power-controller: Failed to create device link (0x180) with supplier 0-0020 for /power-management@fdd90000/power-controller/power-domain@6
[    2.750502] nvme nvme0: 4/0/0 default/read/poll queues
[    2.773746] nvme nvme0: Ignoring bogus Namespace Identifiers
[    2.788100] dwmmc_rockchip fe2b0000.mmc: IDMAC supports 32-bit address mode.
[    2.788831] dwmmc_rockchip fe2b0000.mmc: Using internal DMA controller.
[    2.789432] dwmmc_rockchip fe2b0000.mmc: Version ID is 270a
[    2.790037] dwmmc_rockchip fe2b0000.mmc: DW MMC controller at irq 80,32 bit host data width,256 deep fifo
[    2.802482] arm-scmi arm-scmi.6.auto: Using scmi_smc_transport
[    2.803032] arm-scmi arm-scmi.6.auto: SCMI max-rx-timeout: 30ms / max-msg-size: 104bytes / max-msg: 20
[    2.804109] scmi_protocol scmi_dev.1: Enabled polling mode TX channel - prot_id:16
[    2.805142] arm-scmi arm-scmi.6.auto: SCMI Notifications - Core Enabled.
[    2.805740] mmc_host mmc1: Bus speed = 375000Hz (req 400000Hz, actual 375000HZ div = 0)
[    2.806552] arm-scmi arm-scmi.6.auto: SCMI Protocol v2.0 'rockchip:' Firmware version 0x0
[    2.807391] arm-scmi arm-scmi.6.auto: Enabling SCMI Quirk [quirk_clock_rates_triplet_out_of_spec]
[    2.810214] SMCCC: SOC_ID: ARCH_SOC_ID not implemented, skipping ....
[    2.815909] usbcore: registered new interface driver usbhid
[    2.816421] usbhid: USB HID core driver
[    2.825563] mmc0: SDHCI controller on fe310000.mmc [fe310000.mmc] using ADMA
[    2.829265] hw perfevents: enabled with armv8_cortex_a55 PMU driver, 7 (0,8000003f) counters available
[    2.849585] NET: Registered PF_INET6 protocol family
[    2.851218] Segment Routing with IPv6
[    2.851605] In-situ OAM (IOAM) with IPv6
[    2.852047] NET: Registered PF_PACKET protocol family
[    2.852619] 9pnet: Installing 9P2000 support
[    2.853103] Key type dns_resolver registered
[    2.878282] registered taskstats version 1
[    2.878959] Loading compiled-in X.509 certificates
[    2.885591] usb 1-1: new high-speed USB device number 2 using ehci-platform
[    2.889469] sdhci-dwcmshc fe310000.mmc: Can't reduce the clock below 52MHz in HS200/HS400 mode
[    2.890328] sdhci-dwcmshc fe310000.mmc: Can't reduce the clock below 52MHz in HS200/HS400 mode
[    2.891109] sdhci-dwcmshc fe310000.mmc: Can't reduce the clock below 52MHz in HS200/HS400 mode
[    2.892871] mmc0: new HS200 MMC card at address 0001
[    2.894399] mmcblk0: mmc0:0001 BJTD4R 29.1 GiB
[    2.894880] Demotion targets for Node 0: null
[    2.899582]  mmcblk0: p1 p2 p3
[    2.901085] mmcblk0boot0: mmc0:0001 BJTD4R 4.00 MiB
[    2.904012] mmcblk0boot1: mmc0:0001 BJTD4R 4.00 MiB
[    2.907113] mmcblk0rpmb: mmc0:0001 BJTD4R 4.00 MiB, chardev (509:0)
[    2.945955] platform fde40000.npu: Adding to iommu group 3
[    2.950953] rk_gmac-dwmac fe010000.ethernet: IRQ sfty not found
[    2.952440] rk_gmac-dwmac fe010000.ethernet: clock input or output? (input).
[    2.953079] rk_gmac-dwmac fe010000.ethernet: Can not read property: tx_delay.
[    2.953754] rk_gmac-dwmac fe010000.ethernet: set tx_delay to 0x30
[    2.954304] rk_gmac-dwmac fe010000.ethernet: Can not read property: rx_delay.
[    2.954936] rk_gmac-dwmac fe010000.ethernet: set rx_delay to 0x10
[    2.955485] rk_gmac-dwmac fe010000.ethernet: integrated PHY? (no).
[    2.956079] rk_gmac-dwmac fe010000.ethernet: clock input from PHY
[    2.961643] rk_gmac-dwmac fe010000.ethernet: init for RGMII_ID
[    2.962534] rk_gmac-dwmac fe010000.ethernet: User ID: 0x30, Synopsys ID: 0x51
[    2.963181] rk_gmac-dwmac fe010000.ethernet: 	DWMAC4/5
[    2.963645] rk_gmac-dwmac fe010000.ethernet: DMA HW capability register supported
[    2.964309] rk_gmac-dwmac fe010000.ethernet: Active PHY interface: RGMII (1)
[    2.964933] rk_gmac-dwmac fe010000.ethernet: RX Checksum Offload Engine supported
[    2.965638] rk_gmac-dwmac fe010000.ethernet: TX Checksum insertion supported
[    2.966267] rk_gmac-dwmac fe010000.ethernet: Wake-Up On Lan supported
[    2.966926] rk_gmac-dwmac fe010000.ethernet: Enable RX Mitigation via HW Watchdog Timer
[    2.967641] rk_gmac-dwmac fe010000.ethernet: Enabled RFS Flow TC (entries=10)
[    2.968277] rk_gmac-dwmac fe010000.ethernet: TSO supported
[    2.968766] rk_gmac-dwmac fe010000.ethernet: TSO feature enabled
[    2.969301] rk_gmac-dwmac fe010000.ethernet: Using 32/32 bits DMA host/device width
[    3.035000] hub 1-1:1.0: USB hub found
[    3.035550] hub 1-1:1.0: 4 ports detected
[    3.073444] rk_gmac-dwmac fe2a0000.ethernet: IRQ sfty not found
[    3.075209] rk_gmac-dwmac fe2a0000.ethernet: clock input or output? (input).
[    3.075852] rk_gmac-dwmac fe2a0000.ethernet: Can not read property: tx_delay.
[    3.076485] rk_gmac-dwmac fe2a0000.ethernet: set tx_delay to 0x30
[    3.077027] rk_gmac-dwmac fe2a0000.ethernet: Can not read property: rx_delay.
[    3.077695] rk_gmac-dwmac fe2a0000.ethernet: set rx_delay to 0x10
[    3.078252] rk_gmac-dwmac fe2a0000.ethernet: integrated PHY? (no).
[    3.078851] rk_gmac-dwmac fe2a0000.ethernet: clock input from PHY
[    3.084415] rk_gmac-dwmac fe2a0000.ethernet: init for RGMII_ID
[    3.085289] rk_gmac-dwmac fe2a0000.ethernet: User ID: 0x30, Synopsys ID: 0x51
[    3.085968] rk_gmac-dwmac fe2a0000.ethernet: 	DWMAC4/5
[    3.086436] rk_gmac-dwmac fe2a0000.ethernet: DMA HW capability register supported
[    3.087099] rk_gmac-dwmac fe2a0000.ethernet: Active PHY interface: RGMII (1)
[    3.087724] rk_gmac-dwmac fe2a0000.ethernet: RX Checksum Offload Engine supported
[    3.088385] rk_gmac-dwmac fe2a0000.ethernet: TX Checksum insertion supported
[    3.089008] rk_gmac-dwmac fe2a0000.ethernet: Wake-Up On Lan supported
[    3.089683] rk_gmac-dwmac fe2a0000.ethernet: Enable RX Mitigation via HW Watchdog Timer
[    3.090398] rk_gmac-dwmac fe2a0000.ethernet: Enabled RFS Flow TC (entries=10)
[    3.091034] rk_gmac-dwmac fe2a0000.ethernet: TSO supported
[    3.091521] rk_gmac-dwmac fe2a0000.ethernet: TSO feature enabled
[    3.092056] rk_gmac-dwmac fe2a0000.ethernet: Using 32/32 bits DMA host/device width
[    3.228427] cfg80211: Loading compiled-in X.509 certificates for regulatory database
[    3.244306] Loaded X.509 cert 'sforshee: 00b28ddf47aef9cea7'
[    3.245444] Loaded X.509 cert 'wens: 61c038651aabdcf94bd0ac7ff06c7248db18c600'
[    3.246270] faux_driver regulatory: Direct firmware load for regulatory.db failed with error -2
[    3.246879] clk: Not disabling unused clocks
[    3.247042] cfg80211: failed to load regulatory.db
[    3.247425] PM: genpd: Disabling unused power domains
[    3.248577] dw-apb-uart fe660000.serial: forbid DMA for kernel console
[    3.254849] Freeing unused kernel memory: 12480K
[    3.255399] Run /init as init process

Loading, please wait...
Starting version 247.3-7+deb11u4
Begin: Loading essential drivers ... done.
Begin: Running /scripts/init-premount ... done.
Begin: Mounting root file system ... Begin: Running /scripts/local-top ... done.
Begin: Running /scripts/local-premount ... done.
Begin: Will now check root file system ... fsck from util-linux 2.36.1 [/sbin/fsck.ext4 (1) -- /dev/mmcblk0p3] fsck.ext4 -a -C0 /dev/mmcblk0p3
rootfs: clean, 297753/1855392 files, 4946612/7548923 blocks
done.

[    4.081211] EXT4-fs (mmcblk0p3): mounted filesystem d210e617-6a4c-4771-b955-ddd835a32d2b r/w with ordered data mode. Quota mode: none.
done.
Begin: Running /scripts/local-bottom ... done.
Begin: Running /scripts/init-bottom ... done.
[    4.488149] systemd[1]: System time before build time, advancing clock.
[    4.545777] systemd[1]: systemd 247.3-7+deb11u4 running in system mode. (+PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ +LZ4 +ZSTD +SECCOMP +BLKID +ELFUTILS +KMOD +IDN2 -IDN +PCRE2 default-hierarchy=unified)
[    4.549900] systemd[1]: Detected architecture arm64.

Welcome to Debian GNU/Linux 11 (bullseye)!

[    4.563093] systemd[1]: Set hostname to <rock-3b>.
[    4.712335] block mmcblk0: the capability attribute has been deprecated.
[    5.090660] systemd[1]: Queued start job for default target Graphical Interface.
[    5.131927] systemd[1]: Created slice system-getty.slice.
[  OK  ] Created slice system-getty.slice.
[    5.134792] systemd[1]: Created slice system-modprobe.slice.
[  OK  ] Created slice system-modprobe.slice.
[    5.143177] systemd[1]: Created slice system-serial\x2dgetty.slice.
[  OK  ] Created slice system-serial\x2dgetty.slice.
[    5.151765] systemd[1]: Created slice system-systemd\x2dfsck.slice.
[  OK  ] Created slice system-systemd\x2dfsck.slice.
[    5.163903] systemd[1]: Created slice system-systemd\x2djournald.slice.
[  OK  ] Created slice system-systemd\x2djournald.slice.
[    5.175785] systemd[1]: Created slice system-systemd\x2djournald\x2dvarlink.slice.
[  OK  ] Created slice system-syste…\x2djournald\x2dvarlink.slice.
[    5.183362] systemd[1]: Created slice User and Session Slice.
[  OK  ] Created slice User and Session Slice.
[    5.190135] systemd[1]: Started Dispatch Password Requests to Console Directory Watch.
[  OK  ] Started Dispatch Password …ts to Console Directory Watch.
[    5.202212] systemd[1]: Started Forward Password Requests to Wall Directory Watch.
[  OK  ] Started Forward Password R…uests to Wall Directory Watch.
[    5.210105] systemd[1]: Condition check resulted in Arbitrary Executable File Formats File System Automount Point being skipped.
[    5.211439] systemd[1]: Reached target Local Encrypted Volumes.
[  OK  ] Reached target Local Encrypted Volumes.
[    5.222078] systemd[1]: Reached target Network (Pre).
[  OK  ] Reached target Network (Pre).
[    5.229939] systemd[1]: Reached target Paths.
[  OK  ] Reached target Paths.
[    5.238057] systemd[1]: Reached target Remote Encrypted Volumes.
[  OK  ] Reached target Remote Encrypted Volumes.
[    5.245910] systemd[1]: Reached target Remote File Systems.
[  OK  ] Reached target Remote File Systems.
[    5.247592] systemd[1]: Reached target Slices.
[  OK  ] Reached target Slices.
[    5.250927] systemd[1]: Listening on fsck to fsckd communication Socket.
[  OK  ] Listening on fsck to fsckd communication Socket.
[    5.258649] systemd[1]: Listening on initctl Compatibility Named Pipe.
[  OK  ] Listening on initctl Compatibility Named Pipe.
[    5.267782] systemd[1]: Listening on Journal Audit Socket.
[  OK  ] Listening on Journal Audit Socket.
[    5.270864] systemd[1]: systemd-journald-dev-log.socket: SO_PASSSEC failed: Operation not supported
[    5.271966] systemd[1]: Listening on Journal Socket (/dev/log).
[  OK  ] Listening on Journal Socket (/dev/log).
[    5.274918] systemd[1]: systemd-journald.socket: SO_PASSSEC failed: Operation not supported
[    5.276098] systemd[1]: systemd-journald.socket: SO_PASSSEC failed: Operation not supported
[    5.277020] systemd[1]: Listening on Journal Socket.
[  OK  ] Listening on Journal Socket.
[    5.280707] systemd[1]: Listening on Network Service Netlink Socket.
[  OK  ] Listening on Network Service Netlink Socket.
[    5.291657] systemd[1]: Listening on udev Control Socket.
[  OK  ] Listening on udev Control Socket.
[    5.294591] systemd[1]: Listening on udev Kernel Socket.
[  OK  ] Listening on udev Kernel Socket.
[    5.321932] systemd[1]: Mounting Huge Pages File System...
         Mounting Huge Pages File System...
[    5.334172] systemd[1]: Mounting POSIX Message Queue File System...
         Mounting POSIX Message Queue File System...
[    5.339943] systemd[1]: Mounting Kernel Debug File System...
         Mounting Kernel Debug File System...
[    5.346080] systemd[1]: Condition check resulted in Kernel Trace File System being skipped.
[    5.351151] systemd[1]: Starting Wait for network to be configured by ifupdown...
         Starting Wait for network to be configured by ifupdown...
[    5.357911] systemd[1]: Condition check resulted in Create list of static device nodes for the current kernel being skipped.
[    5.363081] systemd[1]: Starting Load Kernel Module configfs...
         Starting Load Kernel Module configfs...
[    5.368299] systemd[1]: Starting Load Kernel Module drm...
         Starting Load Kernel Module drm...
[    5.376882] systemd[1]: Starting Load Kernel Module fuse...
         Starting Load Kernel Module fuse...
[    5.386967] systemd[1]: Condition check resulted in Set Up Additional Binary Formats being skipped.
[    5.387937] systemd[1]: Condition check resulted in File System Check on Root Device being skipped.
[    5.398865] systemd[1]: Starting Journal Service...
         Starting Journal Service...
[    5.407719] systemd[1]: Starting Load Kernel Modules...
         Starting Load Kernel Modules...
[    5.412414] systemd[1]: Starting Remount Root and Kernel File Systems...
         Starting Remount Root and Kernel File Systems...
[    5.429850] systemd[1]: Starting Coldplug All udev Devices...
         Starting Coldplug All udev Devices...
[    5.447220] systemd[1]: Mounted Huge Pages File System.
[  OK  ] Mounted Huge Pages File System.
[    5.449061] systemd[1]: Mounted POSIX Message Queue File System.
[  OK  ] Mounted POSIX Message Queue File System.
[    5.453688] EXT4-fs (mmcblk0p3): re-mounted d210e617-6a4c-4771-b955-ddd835a32d2b.
[    5.459654] systemd[1]: Mounted Kernel Debug File System.
[  OK  ] Mounted Kernel Debug File System.
[    5.467229] systemd[1]: Finished Wait for network to be configured by ifupdown.
[  OK  ] Finished Wait for network to be configured by ifupdown.
[    5.474892] systemd[1]: [email protected]: Succeeded.
[    5.476453] systemd[1]: Finished Load Kernel Module configfs.
[  OK  ] Finished Load Kernel Module configfs.
[    5.486629] systemd[1]: [email protected]: Succeeded.
[    5.488053] systemd[1]: Finished Load Kernel Module drm.
[  OK  ] Finished Load Kernel Module drm.
[    5.498681] systemd[1]: [email protected]: Succeeded.
[    5.500190] systemd[1]: Finished Load Kernel Module fuse.
[  OK  ] Finished Load Kernel Module fuse.
[    5.511197] systemd[1]: Finished Load Kernel Modules.
[  OK  ] Finished Load Kernel Modules.
[    5.523376] systemd[1]: Finished Remount Root and Kernel File Systems.
[  OK  ] Finished Remount Root and Kernel File Systems.
[    5.534599] systemd[1]: Condition check resulted in FUSE Control File System being skipped.
[    5.550113] systemd[1]: Mounting Kernel Configuration File System...
         Mounting Kernel Configuration File System...
[    5.556704] systemd[1]: Condition check resulted in Rebuild Hardware Database being skipped.
[    5.557696] systemd[1]: Condition check resulted in Platform Persistent Storage Archival being skipped.
[    5.561425] systemd[1]: Starting Load/Save Random Seed...
         Starting Load/Save Random Seed...
[    5.572806] systemd[1]: Starting Apply Kernel Variables...
         Starting Apply Kernel Variables...
[    5.584810] systemd[1]: Starting Create System Users...
         Starting Create System Users...
[    5.596109] systemd[1]: Started Journal Service.
[  OK  ] Started Journal Service.
[  OK  ] Mounted Kernel Configuration File System.
[  OK  ] Finished Load/Save Random Seed.
[  OK  ] Finished Apply Kernel Variables.
         Starting Flush Journal to Persistent Storage...
[  OK  ] Finished Create System Users.
[    5.637155] systemd-journald[230]: Received client request to flush runtime journal.
[    5.640079] systemd-journald[230]: File /var/log/journal/f26be486655e4e559a1282889eb20124/system.journal corrupted or uncleanly shut down, renaming and replacing.
         Starting Create Static Device Nodes in /dev...
[  OK  ] Finished Create Static Device Nodes in /dev.
[  OK  ] Reached target Local File Systems (Pre).
[  OK  ] Set up automount boot-efi.automount.
[  OK  ] Set up automount config.automount.
[  OK  ] Reached target Local File Systems.

Starting Rule-based Manage…for Device Events and Files...
[ OK ] Finished Flush Journal to Persistent Storage.
Starting Create Volatile Files and Directories...
[ OK ] Started Rule-based Manager for Device Events and Files.

         Starting Network Service...
[  OK  ] Finished Create Volatile Files and Directories.
[  OK  ] Started Entropy Daemon based on the HAVEGE algorithm.

Starting Network Time Synchronization... Starting Update UTMP about System Boot/Shutdown...
[ OK ] Finished Coldplug All udev Devices.
Starting Helper to synchronize boot up for ifupdown...

[  OK  ] Finished Helper to synchronize boot up for ifupdown.
[  OK  ] Finished Update UTMP about System Boot/Shutdown.
         Starting Raise network interfaces...
[  OK  ] Started Network Service.
         Starting Wait for Network to be Configured...
         Starting Network Name Resolution...
[  OK  ] Started Network Time Synchronization.
[  OK  ] Reached target System Time Set.
[  OK  ] Reached target System Time Synchronized.
[  OK  ] Finished Raise network interfaces.
[  OK  ] Started Network Name Resolution.
[    6.840374] rocket: loading out-of-tree module taints kernel.
[    6.877647] [drm] Initialized rocket 0.0.0 for rknn on minor 0
[    6.879950] rocket fde40000.npu: Rockchip NPU core 0 version: 0
[  OK  ] Found device EDILOCA EN605 512GB.
         Starting File System Check…1edf-455d-9058-56e0855a1edb...
[  OK  ] Started File System Check Daemon to report status.
[  OK  ] Listening on Load/Save RF …itch Status /dev/rfkill Watch.
[  OK  ] Finished File System Check…d-1edf-455d-9058-56e0855a1edb.
[  OK  ] Found device /dev/ttyS2.
         Mounting /mnt/nvme...
[    7.209346] EXT4-fs (nvme0n1): mounted filesystem 0d9000fd-1edf-455d-9058-56e0855a1edb r/w with ordered data mode. Quota mode: none.
[  OK  ] Mounted /mnt/nvme.
         Activating swap /mnt/nvme/swapfile...
[    7.236341] Adding 10485756k swap on /mnt/nvme/swapfile.  Priority:-1 extents:10 across:19374076k SS
[  OK  ] Activated swap /mnt/nvme/swapfile.
[  OK  ] Reached target Swap.
[  OK  ] Reached target System Initialization.
[  OK  ] Started Daily apt download activities.
[  OK  ] Started Daily apt upgrade and clean activities.
[  OK  ] Started Periodic ext4 Onli…ata Check for All Filesystems.
[  OK  ] Started Discard unused blocks once a week.
[  OK  ] Started Daily man-db regeneration.
[  OK  ] Started Daily Cleanup of Temporary Directories.
[  OK  ] Reached target Timers.
[  OK  ] Listening on Avahi mDNS/DNS-SD Stack Activation Socket.
[  OK  ] Listening on D-Bus System Message Bus Socket.
[  OK  ] Listening on OpenBSD Secure Shell server socket.
[  OK  ] Listening on Journal Varli… Socket for Namespace netdata.
[  OK  ] Listening on Journal Socket for Namespace netdata.
[  OK  ] Reached target Sockets.
[  OK  ] Reached target Basic System.
         Starting Avahi mDNS/DNS-SD Stack...

[ OK ] Started D-Bus System Message Bus.
Starting Network Manager...
Starting Remove Stale Onli…t4 Metadata Check Snapshots...
[ OK ] Started rsetup configuration service.

         Starting LSB: Set sysfs variables from /etc/sysfs.conf...
         Starting User Login Management...
         Starting WPA supplicant...
         Starting Linux zramswap setup...
[  OK  ] Started Avahi mDNS/DNS-SD Stack.
         Mounting /config...
[    7.640448] FAT-fs (mmcblk0p1): Volume was not properly unmounted. Some data may be corrupt. Please run fsck.
[FAILED] Failed to start Linux zramswap setup.
See 'systemctl status zramswap.service' for details.
[  OK  ] Mounted /config.
[  OK  ] Started WPA supplicant.
[  OK  ] Started LSB: Set sysfs variables from /etc/sysfs.conf.
[  OK  ] Started User Login Management.
[  OK  ] Finished Remove Stale Onli…ext4 Metadata Check Snapshots.
[  OK  ] Started Network Manager.
[  OK  ] Reached target Network.
         Starting Network Manager Wait Online...
         Starting dnsmasq - A light…DHCP and caching DNS server...
[  OK  ] Started Enable adbd on supported Radxa products.
[  OK  ] Started Enable USB Ethernet on supported Radxa products.
         Starting Permit User Sessions...
         Starting Tailscale node agent...
[  OK  ] Finished Permit User Sessions.
[  OK  ] Started Getty on tty1.
[  OK  ] Started Serial Getty on ttyS2.
[  OK  ] Reached target Login Prompts.
         Starting Hostname Service...
[FAILED] Failed to start dnsmasq - …t DHCP and caching DNS server.
See 'systemctl status dnsmasq.service' for details.
[  OK  ] Reached target Host and Network Name Lookups.
[  OK  ] Started Hostname Service.
         Starting Network Manager Script Dispatcher Service...
[    8.192053] rk_gmac-dwmac fe010000.ethernet eth0: Register MEM_TYPE_PAGE_POOL RxQ-0
[  OK  ] Started Network Manager Script Dispatcher Service.
[    8.221724] rk_gmac-dwmac fe010000.ethernet eth0: PHY [stmmac-1:01] driver [RTL8211F Gigabit Ethernet] (irq=POLL)
[    8.224669] dwmac4: Master AXI performs any burst length
[    8.225174] rk_gmac-dwmac fe010000.ethernet eth0: No Safety Features support found
[    8.227524] rk_gmac-dwmac fe010000.ethernet eth0: IEEE 1588-2008 Advanced Timestamp supported
[    8.229192] rk_gmac-dwmac fe010000.ethernet eth0: registered PTP clock
[    8.229923] rk_gmac-dwmac fe010000.ethernet eth0: configuring for phy/rgmii-id link mode
[    8.284289] rk_gmac-dwmac fe2a0000.ethernet eth1: Register MEM_TYPE_PAGE_POOL RxQ-0
[  OK  ] Started Tailscale node agent.
[    8.313571] rk_gmac-dwmac fe2a0000.ethernet eth1: PHY [stmmac-0:01] driver [RTL8211F Gigabit Ethernet] (irq=POLL)
[    8.325571] dwmac4: Master AXI performs any burst length
[    8.326067] rk_gmac-dwmac fe2a0000.ethernet eth1: No Safety Features support found
[    8.327175] rk_gmac-dwmac fe2a0000.ethernet eth1: IEEE 1588-2008 Advanced Timestamp supported
[    8.328451] rk_gmac-dwmac fe2a0000.ethernet eth1: registered PTP clock
[    8.329047] rk_gmac-dwmac fe2a0000.ethernet eth1: configuring for phy/rgmii-id link mode
[  OK  ] Finished Wait for Network to be Configured.
[   12.359432] rk_gmac-dwmac fe010000.ethernet eth0: Link is Up - 1Gbps/Full - flow control rx/tx

Debian GNU/Linux 11 rock-3b ttyS2

rock-3b login: [ 17.158640] platform sdio-pwrseq: deferred probe pending: pwrseq_simple: reset control not ready

[   17.159481] platform 3c0000000.pcie: deferred probe pending: rockchip-dw-pcie: failed to initialize the phy
[   17.160361] platform fcc00000.usb: deferred probe pending: dwc3: failed to initialize core
[   17.161108] platform fd000000.usb: deferred probe pending: dwc3: failed to initialize core
[   17.161993] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fd000000.usb
[   17.162945] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fcc00000.usb
[   17.163884] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fde60000.gpu
[   17.164821] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fdea0000.video-codec
[   17.165877] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fdeb0000.rga
[   17.166824] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fdee0000.video-codec
[   17.167821] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fe040000.vop
[   17.168757] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to fe0a0000.hdmi
[   17.169795] rockchip-pm-domain fdd90000.power-management:power-controller: sync_state() pending due to 3c0000000.pcie

rock-3b login: radxa
Password:
Linux rock-3b 7.1.0-rc6-chaoyi-00011-ga31e2e6fae27 #5 SMP PREEMPT Tue Jun 9 12:40:22 CEST 2026 aarch64

The programs included with the Debian GNU/Linux system are free software; the exact distribution terms for each program are described in the individual files in /usr/share/doc/*/copyright.

Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent permitted by applicable law.
Last login: Tue Jun 9 12:55:51 CEST 2026 on ttyS2

[?2004hradxa@rock-3b:~$ lsmod | grep rocket
lsmod | grep rocket
[?2004l
rocket                 24576  0
[?2004hradxa@rock-3b:~$ sudo dmesg -C; python3 ~/npu-debug/teflon_test.py
sudo dmesg -C; python3 ~/npu-debug/teflon_test.py
[?2004l
Input: [ 1 80 80 16] <class 'numpy.uint8'>
Output: [  1  40  40 128] <class 'numpy.uint8'>
[  136.899509] rocket fde40000.npu: NPU job timed out
Run 0: elapsed=529.6ms out_sum=26214400
[  137.443518] rocket fde40000.npu: NPU job timed out
Run 1: elapsed=538.0ms out_sum=26214400
[  137.987451] rocket fde40000.npu: NPU job timed out
Run 2: elapsed=543.3ms out_sum=26214400
[  138.531447] rocket fde40000.npu: NPU job timed out
Run 3: elapsed=540.3ms out_sum=26214400
[  139.075418] rocket fde40000.npu: NPU job timed out
Run 4: elapsed=541.1ms out_sum=26214400
Done
[?2004hradxa@rock-3b:~$ 
#33 lore raw

Hi Midgy,

Quoted 48 lines · Midgy Balon

Hello Chaoyi,

You were right - building rocket as a module fixes it. Thanks for the pointer.

I rebuilt with CONFIG_DRM_ACCEL_ROCKET=m (everything else the same:
need_regulator on
the RK3568 NPU power domain via a DOMAIN_M_R variant, domain-supply = <&vdd_npu>, and the
regulator-always-on workaround dropped). The board now boots cleanly and, more importantly,
an NPU job submit no longer hangs: I ran the test workload five times with no RCU stall and
no freeze.

So with rocket=m the need_regulator approach works on RK3568, and I'll keep it for v4
(domain-supply + need_regulator, instead of marking vdd_npu always-on). rocket=m is the
normal configuration anyway; my earlier hang came from building it =y in a self-contained
image, so it probed in the initcalls (around 2 s) and the genpd -> I2C-PMIC regulator
transition ran before the system was ready. As a module it loads from udev much later
(~6.8 s here), after the I2C controller and regulator core are fully up.

On your question of when the device-link error is printed - it is at power-domain
controller probe, not at the rocket probe:

  [    2.700618] vdd_npu: Bringing 500000uV into 825000-825000uV
  [    2.749637] rockchip-pm-domain fdd90000.power-management:power-controller:
                 Failed to create device link (0x180) with supplier 0-0020 for
                 /power-management@fdd90000/power-controller/power-domain@6
  [    2.945955] platform fde40000.npu: Adding to iommu group 3
  ...
  [    6.840374] rocket: loading out-of-tree module taints kernel.
  [    6.877647] [drm] Initialized rocket 0.0.0 for rknn on minor 0
  [    6.879950] rocket fde40000.npu: Rockchip NPU core 0 version: 0

So the device-link to the rk809 PMIC (0-0020) fails to form at ~2.75 s, well before rocket
loads at ~6.8 s. It is non-fatal here - the vdd_npu rail is brought up by the regulator core
and all jobs run - and there is no "failed to get ack on domain npu" NoC warning this boot
(the always-on kernel had one). The complete boot log is attached.

Quoted 8 lines

Two notes / one question:
- This boot used fw_devlink=permissive on the command line. Is the "Failed to create device
link ... supplier 0-0020" at pmdomain probe expected/benign, or is
there a clean way to make
it order correctly (so it also works without permissive, and a =y
build wouldn't deadlock in
the initcalls)?

We encountered the same issue on the RK3588 NPU before. And it was resolved with the following patch at that time.

https://lore.kernel.org/all/[email protected]/

Please compare the differences in NPU pmdomain and DTS configuration between the RK3568 and RK3588.

Quoted 9 lines

- (The convolution output is still uniform zero-point / the job times out - that is the
separate NPU compute-completion issue, unrelated to the power-domain
work. Finley, that is
the one I flagged earlier re PVTPLL/NoC.)

Kind regards,
Midgy

Signature · 2 lines
Best, 
Chaoyi
#34 lore raw

Hi,

Quoted 60 lines · Chaoyi Chen

Hi Midgy,

Quoted 48 lines · Midgy Balon

Hello Chaoyi,

You were right - building rocket as a module fixes it. Thanks for the pointer.

I rebuilt with CONFIG_DRM_ACCEL_ROCKET=m (everything else the same:
need_regulator on
the RK3568 NPU power domain via a DOMAIN_M_R variant, domain-supply = <&vdd_npu>, and the
regulator-always-on workaround dropped). The board now boots cleanly and, more importantly,
an NPU job submit no longer hangs: I ran the test workload five times with no RCU stall and
no freeze.

So with rocket=m the need_regulator approach works on RK3568, and I'll keep it for v4
(domain-supply + need_regulator, instead of marking vdd_npu always-on). rocket=m is the
normal configuration anyway; my earlier hang came from building it =y in a self-contained
image, so it probed in the initcalls (around 2 s) and the genpd -> I2C-PMIC regulator
transition ran before the system was ready. As a module it loads from udev much later
(~6.8 s here), after the I2C controller and regulator core are fully up.

On your question of when the device-link error is printed - it is at power-domain
controller probe, not at the rocket probe:

  [    2.700618] vdd_npu: Bringing 500000uV into 825000-825000uV
  [    2.749637] rockchip-pm-domain fdd90000.power-management:power-controller:
                 Failed to create device link (0x180) with supplier 0-0020 for
                 /power-management@fdd90000/power-controller/power-domain@6
  [    2.945955] platform fde40000.npu: Adding to iommu group 3
  ...
  [    6.840374] rocket: loading out-of-tree module taints kernel.
  [    6.877647] [drm] Initialized rocket 0.0.0 for rknn on minor 0
  [    6.879950] rocket fde40000.npu: Rockchip NPU core 0 version: 0

So the device-link to the rk809 PMIC (0-0020) fails to form at ~2.75 s, well before rocket
loads at ~6.8 s. It is non-fatal here - the vdd_npu rail is brought up by the regulator core
and all jobs run - and there is no "failed to get ack on domain npu" NoC warning this boot
(the always-on kernel had one). The complete boot log is attached.

Quoted 8 lines

Two notes / one question:
- This boot used fw_devlink=permissive on the command line. Is the "Failed to create device
link ... supplier 0-0020" at pmdomain probe expected/benign, or is
there a clean way to make
it order correctly (so it also works without permissive, and a =y
build wouldn't deadlock in
the initcalls)?

Quoted 7 lines

We encountered the same issue on the RK3588 NPU before. And it was resolved with the following patch at that time.

https://lore.kernel.org/all/[email protected]/

Please compare the differences in NPU pmdomain and DTS configuration between the RK3568 and RK3588.

About a month ago on #linux-rockchip we were discussing PM 'stuff':
https://libera.catirclogs.org/linux-rockchip/2026-05-15#39939137;
which references this paste
https://paste.sr.ht/~diederik/89d9f84e22474e837b55286d213b67f03859ce2e
I've since removed the DCDC_REG2 for PineTab2 and the 'fix' should likely be extended to cover all RK3566/RK3568 devices though.

It's what I made at the time hoping to fix a suspend/resume issue when trying upstream TF-A. It didn't fix the issue at the time, but may still be useful/needed and I think it's what Chaoyi hinted at.

Just yesterday, Jonas posted this patch which may be useful/needed too:
https://lore.kernel.org/linux-rockchip/[email protected]/

HTH,
Diederik

Quoted 9 lines
Quoted 9 lines

- (The convolution output is still uniform zero-point / the job times out - that is the
separate NPU compute-completion issue, unrelated to the power-domain
work. Finley, that is
the one I flagged earlier re PVTPLL/NoC.)

Kind regards,
Midgy

#35 lore raw

Hello Chaoyi & Diederik,

I compared the RK3568 and RK3588 NPU power-domain + DTS as you suggested, and it lines up
exactly with what you described.

The difference is the need_regulator capability. RK3588's NPU domain is DOMAIN_RK3588("npu", …, false, true) — the trailing true is regulator/need_regulator.
The mainline RK3568 macro DOMAIN_RK3568(name, pwr, req, wakeup) has no regulator parameter at
all, so RK3568_PD_NPU can't be marked need_regulator. My v4 adds
that: a regulator-capable
RK3568 NPU domain (need_regulator = true) plus domain-supply = <&vdd_npu> on the NPU node —
i.e. the same shape as RK3588.

And the fix you referenced (Frank Zhang's "pmdomain: rockchip: Fix init genpd as GENPD_STATE_ON before regulator ready", plus "quiet regulator error on -EPROBE_DEFER") is
already in my base (v7.1-rc6), so the if (need_regulator) rockchip_pd_power(pd, false)
default-off path is in effect. That's what resolves the actual problem for me: with rocket
built as a module (the normal config), need_regulator on the NPU domain, and those pmdomain
patches in place, the board boots cleanly and NPU jobs run with no RCU stall / no deadlock. My
earlier hang was an artifact of a self-contained rocket=y image probing in the initcalls before
the I2C regulator core was up — as a module it loads ~6.8 s in, well after, so it's gone.

I also went back and checked the fw_devlink=permissive question myself — and good news, it
turns out it is NOT needed. I rebooted the exact same kernel with permissive removed from the
cmdline (strict fw_devlink, the default), and the board boots cleanly, the NPU probes
(rocket fde40000.npu: Rockchip NPU core 0 version: 0), and NPU jobs submit and run five times
in a row with no deadlock and no RCU stall. So strict fw_devlink resolves the NPU/PMIC ordering
fine via deferred probe.

The one remaining thing is cosmetic: at power-domain-controller probe (~2.94 s) I still get,
in BOTH modes (with or without permissive):

rockchip-pm-domain …: Failed to create device link (0x180) with
supplier 0-0020 …power-domain@6

i.e. genpd can't form the link to the rk809 (the I2C PMIC supplying vdd_npu) because the PMIC
isn't registered yet at that point. It's non-fatal — the domain defaults off (Frank's patch),
the rail comes up via the regulator core, the NPU probes a few seconds later, and all jobs run.

One question: on RK3588 with need_regulator, do you also see that "Failed to create device
link … supplier <pmic>" line at pmdomain probe, or does it order cleanly? If RK3588 is clean,
is there a DTS detail (e.g. the regulator's bus/probe order) I should mirror on RK3568 to make
the link form in time — or is this line just expected/harmless and best left as-is?

@Diederik — thanks; the DCDC_REG2 change and Jonas's USB-suspend series look like generally
useful RK356x robustness fixes, though for this specific NPU device-link the need_regulator +
Frank's pmdomain patches seem to be the relevant piece. I'll keep them in mind for suspend.

The convolution-output / compute-completion issue is still separate and open (@Finley — that's
the PVTPLL/NoC one); the power-domain side is in good shape for v4.

Thanks y'all for your help :)

Kind regards,
Midgy

Quoted 99 lines · Diederik de Haas

Hi,

Quoted 60 lines · Chaoyi Chen

Hi Midgy,

Quoted 48 lines · Midgy Balon

Hello Chaoyi,

You were right - building rocket as a module fixes it. Thanks for the pointer.

I rebuilt with CONFIG_DRM_ACCEL_ROCKET=m (everything else the same:
need_regulator on
the RK3568 NPU power domain via a DOMAIN_M_R variant, domain-supply = <&vdd_npu>, and the
regulator-always-on workaround dropped). The board now boots cleanly and, more importantly,
an NPU job submit no longer hangs: I ran the test workload five times with no RCU stall and
no freeze.

So with rocket=m the need_regulator approach works on RK3568, and I'll keep it for v4
(domain-supply + need_regulator, instead of marking vdd_npu always-on). rocket=m is the
normal configuration anyway; my earlier hang came from building it =y in a self-contained
image, so it probed in the initcalls (around 2 s) and the genpd -> I2C-PMIC regulator
transition ran before the system was ready. As a module it loads from udev much later
(~6.8 s here), after the I2C controller and regulator core are fully up.

On your question of when the device-link error is printed - it is at power-domain
controller probe, not at the rocket probe:

  [    2.700618] vdd_npu: Bringing 500000uV into 825000-825000uV
  [    2.749637] rockchip-pm-domain fdd90000.power-management:power-controller:
                 Failed to create device link (0x180) with supplier 0-0020 for
                 /power-management@fdd90000/power-controller/power-domain@6
  [    2.945955] platform fde40000.npu: Adding to iommu group 3
  ...
  [    6.840374] rocket: loading out-of-tree module taints kernel.
  [    6.877647] [drm] Initialized rocket 0.0.0 for rknn on minor 0
  [    6.879950] rocket fde40000.npu: Rockchip NPU core 0 version: 0

So the device-link to the rk809 PMIC (0-0020) fails to form at ~2.75 s, well before rocket
loads at ~6.8 s. It is non-fatal here - the vdd_npu rail is brought up by the regulator core
and all jobs run - and there is no "failed to get ack on domain npu" NoC warning this boot
(the always-on kernel had one). The complete boot log is attached.

Quoted 8 lines

Two notes / one question:
- This boot used fw_devlink=permissive on the command line. Is the "Failed to create device
link ... supplier 0-0020" at pmdomain probe expected/benign, or is
there a clean way to make
it order correctly (so it also works without permissive, and a =y
build wouldn't deadlock in
the initcalls)?

Quoted 7 lines

We encountered the same issue on the RK3588 NPU before. And it was resolved with the following patch at that time.

https://lore.kernel.org/all/[email protected]/

Please compare the differences in NPU pmdomain and DTS configuration between the RK3568 and RK3588.

About a month ago on #linux-rockchip we were discussing PM 'stuff':
https://libera.catirclogs.org/linux-rockchip/2026-05-15#39939137;
which references this paste
https://paste.sr.ht/~diederik/89d9f84e22474e837b55286d213b67f03859ce2e
I've since removed the DCDC_REG2 for PineTab2 and the 'fix' should likely be extended to cover all RK3566/RK3568 devices though.

It's what I made at the time hoping to fix a suspend/resume issue when trying upstream TF-A. It didn't fix the issue at the time, but may still be useful/needed and I think it's what Chaoyi hinted at.

Just yesterday, Jonas posted this patch which may be useful/needed too:
https://lore.kernel.org/linux-rockchip/[email protected]/

HTH,
Diederik

Quoted 9 lines
Quoted 9 lines

- (The convolution output is still uniform zero-point / the job times out - that is the
separate NPU compute-completion issue, unrelated to the power-domain
work. Finley, that is
the one I flagged earlier re PVTPLL/NoC.)

Kind regards,
Midgy

#36 lore raw
Quoted 59 lines · Midgy Balon

Hello Chaoyi & Diederik,

I compared the RK3568 and RK3588 NPU power-domain + DTS as you suggested, and it lines up
exactly with what you described.

The difference is the need_regulator capability. RK3588's NPU domain is DOMAIN_RK3588("npu", …, false, true) — the trailing true is regulator/need_regulator.
The mainline RK3568 macro DOMAIN_RK3568(name, pwr, req, wakeup) has no regulator parameter at
all, so RK3568_PD_NPU can't be marked need_regulator. My v4 adds
that: a regulator-capable
RK3568 NPU domain (need_regulator = true) plus domain-supply = <&vdd_npu> on the NPU node —
i.e. the same shape as RK3588.

And the fix you referenced (Frank Zhang's "pmdomain: rockchip: Fix init genpd as GENPD_STATE_ON before regulator ready", plus "quiet regulator error on -EPROBE_DEFER") is
already in my base (v7.1-rc6), so the if (need_regulator) rockchip_pd_power(pd, false)
default-off path is in effect. That's what resolves the actual problem for me: with rocket
built as a module (the normal config), need_regulator on the NPU domain, and those pmdomain
patches in place, the board boots cleanly and NPU jobs run with no RCU stall / no deadlock. My
earlier hang was an artifact of a self-contained rocket=y image probing in the initcalls before
the I2C regulator core was up — as a module it loads ~6.8 s in, well after, so it's gone.

I also went back and checked the fw_devlink=permissive question myself — and good news, it
turns out it is NOT needed. I rebooted the exact same kernel with permissive removed from the
cmdline (strict fw_devlink, the default), and the board boots cleanly, the NPU probes
(rocket fde40000.npu: Rockchip NPU core 0 version: 0), and NPU jobs submit and run five times
in a row with no deadlock and no RCU stall. So strict fw_devlink resolves the NPU/PMIC ordering
fine via deferred probe.

The one remaining thing is cosmetic: at power-domain-controller probe (~2.94 s) I still get,
in BOTH modes (with or without permissive):

rockchip-pm-domain …: Failed to create device link (0x180) with
supplier 0-0020 …power-domain@6

i.e. genpd can't form the link to the rk809 (the I2C PMIC supplying vdd_npu) because the PMIC
isn't registered yet at that point. It's non-fatal — the domain defaults off (Frank's patch),
the rail comes up via the regulator core, the NPU probes a few seconds later, and all jobs run.

Quoted 8 lines

One question: on RK3588 with need_regulator, do you also see that "Failed to create device
link … supplier <pmic>" line at pmdomain probe, or does it order cleanly? If RK3588 is clean,
is there a DTS detail (e.g. the regulator's bus/probe order) I should mirror on RK3568 to make
the link form in time — or is this line just expected/harmless and best left as-is?

[    2.110935] rockchip-pm-domain fd8d8000.power-management:power-controller: Failed to create device link (0x180) with supplier 2-0042 for /power-management@fd8d8000/power-controller/power-domain@8
[    2.557459] sdhci-dwcmshc fe2e0000.mmc: Can't reduce the clock below 52MHz in HS200/HS400 mode
[    2.647174] rockchip-pm-domain fd8d8000.power-management:power-controller: Failed to create device link (0x180) with supplier 2-0042 for /power-management@fd8d8000/power-controller/power-domain@8
[    2.945089] rockchip-pm-domain fd8d8000.power-management:power-controller: Failed to create device link (0x180) with supplier spi2.0 for /power-management@fd8d8000/power-controller/power-domain@12

8 = NPU; 12 = GPU

on both nanopc-t6-lts and nanopc-t6-plus (both RK3588).
And on a 6.18 dmesg output I have for Rock 5B, I see the ~ same, but then it's 1-0042 instead of 2-0042.

I don't know if it's bad or harmless, but it is consistent.

HTH,
Diederik

Quoted 122 lines

@Diederik — thanks; the DCDC_REG2 change and Jonas's USB-suspend series look like generally
useful RK356x robustness fixes, though for this specific NPU device-link the need_regulator +
Frank's pmdomain patches seem to be the relevant piece. I'll keep them in mind for suspend.

The convolution-output / compute-completion issue is still separate and open (@Finley — that's
the PVTPLL/NoC one); the power-domain side is in good shape for v4.

Thanks y'all for your help :)

Kind regards,
Midgy

Quoted 83 lines · Diederik de Haas

Hi,

Quoted 60 lines · Chaoyi Chen

Hi Midgy,

Quoted 48 lines · Midgy Balon

Hello Chaoyi,

You were right - building rocket as a module fixes it. Thanks for the pointer.

I rebuilt with CONFIG_DRM_ACCEL_ROCKET=m (everything else the same:
need_regulator on
the RK3568 NPU power domain via a DOMAIN_M_R variant, domain-supply = <&vdd_npu>, and the
regulator-always-on workaround dropped). The board now boots cleanly and, more importantly,
an NPU job submit no longer hangs: I ran the test workload five times with no RCU stall and
no freeze.

So with rocket=m the need_regulator approach works on RK3568, and I'll keep it for v4
(domain-supply + need_regulator, instead of marking vdd_npu always-on). rocket=m is the
normal configuration anyway; my earlier hang came from building it =y in a self-contained
image, so it probed in the initcalls (around 2 s) and the genpd -> I2C-PMIC regulator
transition ran before the system was ready. As a module it loads from udev much later
(~6.8 s here), after the I2C controller and regulator core are fully up.

On your question of when the device-link error is printed - it is at power-domain
controller probe, not at the rocket probe:

  [    2.700618] vdd_npu: Bringing 500000uV into 825000-825000uV
  [    2.749637] rockchip-pm-domain fdd90000.power-management:power-controller:
                 Failed to create device link (0x180) with supplier 0-0020 for
                 /power-management@fdd90000/power-controller/power-domain@6
  [    2.945955] platform fde40000.npu: Adding to iommu group 3
  ...
  [    6.840374] rocket: loading out-of-tree module taints kernel.
  [    6.877647] [drm] Initialized rocket 0.0.0 for rknn on minor 0
  [    6.879950] rocket fde40000.npu: Rockchip NPU core 0 version: 0

So the device-link to the rk809 PMIC (0-0020) fails to form at ~2.75 s, well before rocket
loads at ~6.8 s. It is non-fatal here - the vdd_npu rail is brought up by the regulator core
and all jobs run - and there is no "failed to get ack on domain npu" NoC warning this boot
(the always-on kernel had one). The complete boot log is attached.

Quoted 8 lines

Two notes / one question:
- This boot used fw_devlink=permissive on the command line. Is the "Failed to create device
link ... supplier 0-0020" at pmdomain probe expected/benign, or is
there a clean way to make
it order correctly (so it also works without permissive, and a =y
build wouldn't deadlock in
the initcalls)?

Quoted 7 lines

We encountered the same issue on the RK3588 NPU before. And it was resolved with the following patch at that time.

https://lore.kernel.org/all/[email protected]/

Please compare the differences in NPU pmdomain and DTS configuration between the RK3568 and RK3588.

About a month ago on #linux-rockchip we were discussing PM 'stuff':
https://libera.catirclogs.org/linux-rockchip/2026-05-15#39939137;
which references this paste
https://paste.sr.ht/~diederik/89d9f84e22474e837b55286d213b67f03859ce2e
I've since removed the DCDC_REG2 for PineTab2 and the 'fix' should likely be extended to cover all RK3566/RK3568 devices though.

It's what I made at the time hoping to fix a suspend/resume issue when trying upstream TF-A. It didn't fix the issue at the time, but may still be useful/needed and I think it's what Chaoyi hinted at.

Quoted 16 lines

Just yesterday, Jonas posted this patch which may be useful/needed too:
https://lore.kernel.org/linux-rockchip/[email protected]/

HTH,
Diederik

Quoted 9 lines
Quoted 9 lines

- (The convolution output is still uniform zero-point / the job times out - that is the
separate NPU compute-completion issue, unrelated to the power-domain
work. Finley, that is
the one I flagged earlier re PVTPLL/NoC.)

Kind regards,
Midgy

_______________________________________________
Linux-rockchip mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-rockchip

#37 lore raw

Hi,

Quoted 6 lines · Diederik de Haas

On Wed Jun 10, 2026 at 3:36 PM CEST, Midgy Balon wrote:

[    2.110935] rockchip-pm-domain fd8d8000.power-management:power-controller: Failed to create device link (0x180) with supplier 2-0042 for /power-management@fd8d8000/power-controller/power-domain@8
[    2.557459] sdhci-dwcmshc fe2e0000.mmc: Can't reduce the clock below 52MHz in HS200/HS400 mode
[    2.647174] rockchip-pm-domain fd8d8000.power-management:power-controller: Failed to create device link (0x180) with supplier 2-0042 for /power-management@fd8d8000/power-controller/power-domain@8
[    2.945089] rockchip-pm-domain fd8d8000.power-management:power-controller: Failed to create device link (0x180) with supplier spi2.0 for /power-management@fd8d8000/power-controller/power-domain@12
Quoted 7 lines

8 = NPU; 12 = GPU

on both nanopc-t6-lts and nanopc-t6-plus (both RK3588).
And on a 6.18 dmesg output I have for Rock 5B, I see the ~ same, but then it's 1-0042 instead of 2-0042.

I don't know if it's bad or harmless, but it is consistent.

The fw_devlink framework tries to figure out a sensible probe order by analyzing links between devices. The warning is because there is a cyclic dependency. This happens because all power domains are provided by one device (power-controller).

Now if you want to probe the I2C regulator 2-0042, you need the I2C controller and to probe the I2C controller you need the I2C power domain and for that you need the power-controller. But for the power-controller you need 2-0042 (for the NPU power-domain). At this point fw_devlink gives up and prints the warning.

Apart from the warning this results in the kernel missing dependency information, so there might be some extra probe calls ending in -EPROBE_DEFER (which wastes CPU power and delays the boot process).

So it's neither super bad, nor completely harmless. Fixing this properly requires some heavy restructuring of the Rockchip power-controller driver.

Greetings,

-- Sebastian