[PATCH RESEND V1 0/4] accel/amdxdna: Add AIE4 debug buffer support

[PATCH RESEND V1 0/4] accel/amdxdna: Add AIE4 debug buffer support

[David Zhang]

This series adds AIE4 debug buffer support, which the xrt-smi validate
gemm test uses to read the per-core results, and the AIE4 CERT firmware
version query.

- Add NPU10 and NPU11 devices.
- Bump the driver version to 1.0.
- Add AIE4 debug buffer support.
- Add AIE4 CERT firmware version query.

This series applies on top of drm-misc-next.

David Zhang (4):
  accel/amdxdna: Add NPU10 and NPU11 devices
  accel/amdxdna: Bump driver version to 1.0
  accel/amdxdna: Add AIE4 debug buffer support
  accel/amdxdna: Add AIE4 CERT firmware version query

 drivers/accel/amdxdna/aie4_ctx.c        | 284 +++++++++++++++++++++++-
 drivers/accel/amdxdna/aie4_message.c    |  20 ++
 drivers/accel/amdxdna/aie4_msg_priv.h   |  35 +++
 drivers/accel/amdxdna/aie4_pci.c        |   7 +
 drivers/accel/amdxdna/aie4_pci.h        |  10 +
 drivers/accel/amdxdna/amdxdna_ctx.c     |   4 +-
 drivers/accel/amdxdna/amdxdna_pci_drv.c |  19 +-
 drivers/accel/amdxdna/amdxdna_pci_drv.h |   6 +
 drivers/accel/amdxdna/npu3_regs.c       |  78 +++++++
 include/uapi/drm/amdxdna_accel.h        |  51 +++++
 10 files changed, 510 insertions(+), 4 deletions(-)


base-commit: e32e83a6c5ec031a7f59d4794113d8484b220254
-- 
2.34.1

[PATCH RESEND V1 1/4] accel/amdxdna: Add NPU10 and NPU11 devices

[David Zhang]

NPU3 revisions 0x14 and 0x15 are NPU10 and NPU11. Add their classic,
PF and VF devices with their firmware files and report each by its own
name. They otherwise use the NPU3 settings.

Signed-off-by: David Zhang <yidong.zhang@amd.com>
---
 drivers/accel/amdxdna/amdxdna_pci_drv.c | 14 +++++
 drivers/accel/amdxdna/amdxdna_pci_drv.h |  6 ++
 drivers/accel/amdxdna/npu3_regs.c       | 78 +++++++++++++++++++++++++
 3 files changed, 98 insertions(+)

diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c
index f233eb70a295..685fe1a7dd56 100644
--- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
+++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
@@ -37,6 +37,14 @@ MODULE_FIRMWARE("amdnpu/17f1_13/npu.sbin");
 MODULE_FIRMWARE("amdnpu/17f1_13/cert.sbin");
 MODULE_FIRMWARE("amdnpu/17f2_13/npu.sbin");
 MODULE_FIRMWARE("amdnpu/17f2_13/cert.sbin");
+MODULE_FIRMWARE("amdnpu/17f1_14/npu.sbin");
+MODULE_FIRMWARE("amdnpu/17f1_14/cert.sbin");
+MODULE_FIRMWARE("amdnpu/17f2_14/npu.sbin");
+MODULE_FIRMWARE("amdnpu/17f2_14/cert.sbin");
+MODULE_FIRMWARE("amdnpu/17f1_15/npu.sbin");
+MODULE_FIRMWARE("amdnpu/17f1_15/cert.sbin");
+MODULE_FIRMWARE("amdnpu/17f2_15/npu.sbin");
+MODULE_FIRMWARE("amdnpu/17f2_15/cert.sbin");
 
 /*
  * 0.0: Initial version
@@ -81,6 +89,12 @@ static const struct amdxdna_device_id amdxdna_ids[] = {
 	{ 0x17f1, 0x13, &dev_npu3_classic_info },
 	{ 0x17f2, 0x13, &dev_npu3_pf_info },
 	{ 0x17f3, 0x13, &dev_npu3_vf_info },
+	{ 0x17f1, 0x14, &dev_npu10_classic_info },
+	{ 0x17f2, 0x14, &dev_npu10_pf_info },
+	{ 0x17f3, 0x14, &dev_npu10_vf_info },
+	{ 0x17f1, 0x15, &dev_npu11_classic_info },
+	{ 0x17f2, 0x15, &dev_npu11_pf_info },
+	{ 0x17f3, 0x15, &dev_npu11_vf_info },
 	{0}
 };
 
diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.h b/drivers/accel/amdxdna/amdxdna_pci_drv.h
index 27593190dcf6..5ca32c5b1de0 100644
--- a/drivers/accel/amdxdna/amdxdna_pci_drv.h
+++ b/drivers/accel/amdxdna/amdxdna_pci_drv.h
@@ -187,6 +187,12 @@ extern const struct amdxdna_dev_info dev_npu1_info;
 extern const struct amdxdna_dev_info dev_npu3_classic_info;
 extern const struct amdxdna_dev_info dev_npu3_pf_info;
 extern const struct amdxdna_dev_info dev_npu3_vf_info;
+extern const struct amdxdna_dev_info dev_npu10_classic_info;
+extern const struct amdxdna_dev_info dev_npu10_pf_info;
+extern const struct amdxdna_dev_info dev_npu10_vf_info;
+extern const struct amdxdna_dev_info dev_npu11_classic_info;
+extern const struct amdxdna_dev_info dev_npu11_pf_info;
+extern const struct amdxdna_dev_info dev_npu11_vf_info;
 extern const struct amdxdna_dev_info dev_npu4_info;
 extern const struct amdxdna_dev_info dev_npu5_info;
 extern const struct amdxdna_dev_info dev_npu6_info;
diff --git a/drivers/accel/amdxdna/npu3_regs.c b/drivers/accel/amdxdna/npu3_regs.c
index cff99ddd3a55..4d38aeaea6d0 100644
--- a/drivers/accel/amdxdna/npu3_regs.c
+++ b/drivers/accel/amdxdna/npu3_regs.c
@@ -197,3 +197,81 @@ const struct amdxdna_dev_info dev_npu3_classic_info = {
 	.cert_feature_tbl	= npu3_cert_feature_table,
 	.ops			= &aie4_classic_ops,
 };
+
+const struct amdxdna_dev_info dev_npu10_pf_info = {
+	.mbox_bar		= NPU3_MBOX_BAR,
+	.sram_bar		= NPU3_MBOX_BUFFER_BAR,
+	.psp_bar		= NPU3_PSP_BAR_INDEX,
+	.smu_bar		= NPU3_SMU_BAR_INDEX,
+	.default_vbnv		= "RyzenAI-npu10-pf",
+	.device_type		= AMDXDNA_DEV_TYPE_PF,
+	.dev_priv		= &npu3_dev_priv,
+	.fw_feature_tbl		= npu3_fw_feature_table,
+	.cert_feature_tbl	= npu3_cert_feature_table,
+	.ops			= &aie4_pf_ops,
+};
+
+const struct amdxdna_dev_info dev_npu10_vf_info = {
+	.mbox_bar		= NPU3_MBOX_BAR,
+	.sram_bar		= NPU3_MBOX_BUFFER_BAR,
+	.doorbell_bar		= NPU3_DOORBELL_BAR,
+	.default_vbnv		= "RyzenAI-npu10-vf",
+	.device_type		= AMDXDNA_DEV_TYPE_UMQ,
+	.dev_priv		= &npu3_dev_vf_priv,
+	.fw_feature_tbl		= npu3_fw_feature_table,
+	.cert_feature_tbl	= npu3_cert_feature_table,
+	.ops			= &aie4_vf_ops,
+};
+
+const struct amdxdna_dev_info dev_npu10_classic_info = {
+	.mbox_bar		= NPU3_MBOX_BAR,
+	.sram_bar		= NPU3_MBOX_BUFFER_BAR,
+	.psp_bar		= NPU3_PSP_BAR_INDEX,
+	.smu_bar		= NPU3_SMU_BAR_INDEX,
+	.doorbell_bar		= NPU3_DOORBELL_BAR,
+	.default_vbnv		= "RyzenAI-npu10",
+	.device_type		= AMDXDNA_DEV_TYPE_UMQ,
+	.dev_priv		= &npu3_dev_priv,
+	.fw_feature_tbl		= npu3_fw_feature_table,
+	.cert_feature_tbl	= npu3_cert_feature_table,
+	.ops			= &aie4_classic_ops,
+};
+
+const struct amdxdna_dev_info dev_npu11_pf_info = {
+	.mbox_bar		= NPU3_MBOX_BAR,
+	.sram_bar		= NPU3_MBOX_BUFFER_BAR,
+	.psp_bar		= NPU3_PSP_BAR_INDEX,
+	.smu_bar		= NPU3_SMU_BAR_INDEX,
+	.default_vbnv		= "RyzenAI-npu11-pf",
+	.device_type		= AMDXDNA_DEV_TYPE_PF,
+	.dev_priv		= &npu3_dev_priv,
+	.fw_feature_tbl		= npu3_fw_feature_table,
+	.cert_feature_tbl	= npu3_cert_feature_table,
+	.ops			= &aie4_pf_ops,
+};
+
+const struct amdxdna_dev_info dev_npu11_vf_info = {
+	.mbox_bar		= NPU3_MBOX_BAR,
+	.sram_bar		= NPU3_MBOX_BUFFER_BAR,
+	.doorbell_bar		= NPU3_DOORBELL_BAR,
+	.default_vbnv		= "RyzenAI-npu11-vf",
+	.device_type		= AMDXDNA_DEV_TYPE_UMQ,
+	.dev_priv		= &npu3_dev_vf_priv,
+	.fw_feature_tbl		= npu3_fw_feature_table,
+	.cert_feature_tbl	= npu3_cert_feature_table,
+	.ops			= &aie4_vf_ops,
+};
+
+const struct amdxdna_dev_info dev_npu11_classic_info = {
+	.mbox_bar		= NPU3_MBOX_BAR,
+	.sram_bar		= NPU3_MBOX_BUFFER_BAR,
+	.psp_bar		= NPU3_PSP_BAR_INDEX,
+	.smu_bar		= NPU3_SMU_BAR_INDEX,
+	.doorbell_bar		= NPU3_DOORBELL_BAR,
+	.default_vbnv		= "RyzenAI-npu11",
+	.device_type		= AMDXDNA_DEV_TYPE_UMQ,
+	.dev_priv		= &npu3_dev_priv,
+	.fw_feature_tbl		= npu3_fw_feature_table,
+	.cert_feature_tbl	= npu3_cert_feature_table,
+	.ops			= &aie4_classic_ops,
+};
-- 
2.34.1

[PATCH RESEND V1 2/4] accel/amdxdna: Bump driver version to 1.0

[David Zhang]

The driver supports AIE4 devices for the official release. Bump the
driver version to 1.0.

Signed-off-by: David Zhang <yidong.zhang@amd.com>
---
 drivers/accel/amdxdna/amdxdna_pci_drv.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c
index 685fe1a7dd56..0dc156a1c0c0 100644
--- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
+++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
@@ -58,9 +58,10 @@ MODULE_FIRMWARE("amdnpu/17f2_15/cert.sbin");
  * 0.8: Support BO usage query
  * 0.9: Add new device type AMDXDNA_DEV_TYPE_PF
  * 0.10: Support AIE4 UMQ
+ * 1.0: Official release with AIE4 device support
  */
-#define AMDXDNA_DRIVER_MAJOR		0
-#define AMDXDNA_DRIVER_MINOR		10
+#define AMDXDNA_DRIVER_MAJOR		1
+#define AMDXDNA_DRIVER_MINOR		0
 
 /*
  * Bind the driver base on (vendor_id, device_id) pair and later use the
-- 
2.34.1

[PATCH RESEND V1 3/4] accel/amdxdna: Add AIE4 debug buffer support

[David Zhang]

Let user space attach a debug buffer to an AIE4 hardware context through
the hwctx config ioctl. The buffer stays attached across suspend and
resume: the driver removes it from the firmware context before suspend
and configures it again on resume.

Signed-off-by: David Zhang <yidong.zhang@amd.com>
---
 drivers/accel/amdxdna/aie4_ctx.c      | 284 +++++++++++++++++++++++++-
 drivers/accel/amdxdna/aie4_message.c  |  20 ++
 drivers/accel/amdxdna/aie4_msg_priv.h |  35 ++++
 drivers/accel/amdxdna/aie4_pci.c      |   4 +
 drivers/accel/amdxdna/aie4_pci.h      |  10 +
 drivers/accel/amdxdna/amdxdna_ctx.c   |   4 +-
 include/uapi/drm/amdxdna_accel.h      |  50 +++++
 7 files changed, 405 insertions(+), 2 deletions(-)

diff --git a/drivers/accel/amdxdna/aie4_ctx.c b/drivers/accel/amdxdna/aie4_ctx.c
index a3f0a69b93b7..972e924dc1b3 100644
--- a/drivers/accel/amdxdna/aie4_ctx.c
+++ b/drivers/accel/amdxdna/aie4_ctx.c
@@ -9,6 +9,7 @@
 #include <drm/drm_gem_shmem_helper.h>
 #include <drm/drm_print.h>
 #include <drm/gpu_scheduler.h>
+#include <linux/bitfield.h>
 #include <linux/hmm.h>
 #include <linux/iommu.h>
 #include <linux/types.h>
@@ -130,6 +131,70 @@ static void aie4_msg_destroy_context(struct amdxdna_dev_hdl *ndev, u32 hw_contex
 		XDNA_WARN(xdna, "destroy ctx id %d failed %d", hw_context_id, ret);
 }
 
+/* Release the debug buffer of @type, taken on attach. */
+static void aie4_hwctx_put_fw_buf(struct amdxdna_hwctx *hwctx, u32 type)
+{
+	struct amdxdna_hwctx_priv *priv = hwctx->priv;
+
+	priv->fw_buf_bo[type]->assigned_hwctx = AMDXDNA_INVALID_CTX_HANDLE;
+	/* Drop the reference taken when the buffer was attached. */
+	amdxdna_gem_put_obj(priv->fw_buf_bo[type]);
+	priv->fw_buf_bo[type] = NULL;
+}
+
+static int aie4_fw_buf_property(u32 type)
+{
+	switch (type) {
+	case AMDXDNA_FW_BUF_LOG:
+		return AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_LOG_BUFFER;
+	case AMDXDNA_FW_BUF_DEBUG:
+		return AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_DEBUG_BUFFER;
+	case AMDXDNA_FW_BUF_TRACE:
+		return AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_TRACE_BUFFER;
+	case AMDXDNA_FW_BUF_DBG_Q:
+		return AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_DEBUG_QUEUE;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+/* Release all attached debug buffers. */
+static void aie4_hwctx_put_fw_bufs(struct amdxdna_hwctx *hwctx)
+{
+	struct amdxdna_hwctx_priv *priv = hwctx->priv;
+	u32 type;
+
+	for (type = 0; type < ARRAY_SIZE(priv->fw_buf_bo); type++) {
+		if (priv->fw_buf_bo[type])
+			aie4_hwctx_put_fw_buf(hwctx, type);
+	}
+}
+
+/*
+ * Configure the attached debug buffers on the firmware context, or remove
+ * them from it. Suspend and resume recreate the firmware context, so the
+ * buffers are removed first and configured again afterwards.
+ */
+static void aie4_hwctx_cfg_fw_bufs(struct amdxdna_hwctx *hwctx, bool attach)
+{
+	struct aie4_msg_context_config_cert_logging none = { };
+	struct amdxdna_hwctx_priv *priv = hwctx->priv;
+	struct amdxdna_dev *xdna = hwctx->client->xdna;
+	u32 type;
+	int ret;
+
+	for (type = 0; type < ARRAY_SIZE(priv->fw_buf_bo); type++) {
+		if (!priv->fw_buf_bo[type])
+			continue;
+		ret = aie4_configure_hw_context_cert_log(xdna->dev_handle, priv->hw_ctx_id,
+							 aie4_fw_buf_property(type),
+							 attach ? &priv->fw_buf_cl[type] : &none);
+		if (ret)
+			XDNA_WARN(xdna, "%s debug buffer type %u on %s failed, ret %d",
+				  attach ? "Configure" : "Remove", type, hwctx->name, ret);
+	}
+}
+
 static u8 aie4_parse_priority_to_dev(u32 priority)
 {
 	switch (priority) {
@@ -198,6 +263,9 @@ int aie4_hwctx_create(struct amdxdna_hwctx *hwctx)
 	hwctx->start_col = 0;
 	hwctx->num_col = ndev->total_col;
 
+	/* Pairs with aie4_hwctx_destroy(): re-attach debug buffers after resume or reset. */
+	aie4_hwctx_cfg_fw_bufs(hwctx, true);
+
 	/* Set up driver doorbell and return invalid offset to userspace. */
 	mutex_lock(&priv->io_lock);
 	ret = aie4_doorbell_setup(hwctx, &resp);
@@ -256,8 +324,15 @@ void aie4_hwctx_destroy(struct amdxdna_hwctx *hwctx, enum aie4_hwctx_flags flags
 	if (has_error)
 		wake_up_all(&priv->job_list_wq);
 
-	if (flags != AIE4_HWCTX_DISCONNECT && priv->hw_ctx_id != CTX_INVALID_ID)
+	if (flags != AIE4_HWCTX_DISCONNECT && priv->hw_ctx_id != CTX_INVALID_ID) {
+		/*
+		 * Pairs with aie4_hwctx_create(): detach debug buffers.
+		 * On error, skip it; firmware drops them with the context.
+		 */
+		if (flags != AIE4_HWCTX_ERROR)
+			aie4_hwctx_cfg_fw_bufs(hwctx, false);
 		aie4_msg_destroy_context(ndev, priv->hw_ctx_id);
+	}
 
 	if (cert_comp)
 		aie4_put_cert_comp(cert_comp);
@@ -430,6 +505,8 @@ void aie4_hwctx_fini(struct amdxdna_hwctx *hwctx)
 		aie4_hwctx_wait_for_running(hwctx);
 		destroy_workqueue(priv->job_work_q);
 	}
+	/* The context is gone, so release its debug buffers. */
+	aie4_hwctx_put_fw_bufs(hwctx);
 	aie4_hwctx_umq_fini(hwctx);
 	mutex_destroy(&priv->io_lock);
 	kfree(hwctx->priv);
@@ -1284,3 +1361,208 @@ int aie4_cmd_submit(struct amdxdna_hwctx *hwctx, struct amdxdna_sched_job *job,
 	mmput(job->mm);
 	return ret;
 }
+
+/*
+ * Attach or detach the debug buffer @meta describes. @meta is a kernel copy
+ * of the user descriptor.
+ *
+ * The buffer stays attached until user space removes it or frees the
+ * hardware context. After suspend, resume or context recovery, the driver
+ * configures it again on the new firmware context.
+ */
+static int aie4_hwctx_cfg_fw_buf(struct amdxdna_hwctx *hwctx,
+				 const struct amdxdna_fw_buffer_metadata *meta,
+				 u32 size, bool attach)
+{
+	struct aie4_msg_context_config_cert_logging cl = { };
+	struct amdxdna_hwctx_priv *priv = hwctx->priv;
+	struct amdxdna_client *client = hwctx->client;
+	struct amdxdna_dev *xdna = client->xdna;
+	struct amdxdna_dev_hdl *ndev = xdna->dev_handle;
+	struct amdxdna_gem_obj *payload_bo;
+	u32 prev_size = 0;
+	u64 base_addr;
+	u32 property;
+	u32 seen = 0;	/* uC indices in the descriptor */
+	u32 used = 0;	/* uC indices with a slice size above 0 */
+	int ret;
+	int i;
+
+	if (!meta || size < sizeof(*meta)) {
+		XDNA_DBG(xdna, "Invalid debug buffer descriptor, size %u", size);
+		return -EINVAL;
+	}
+
+	if (XDNA_MBZ_DBG(xdna, meta->pad, sizeof(meta->pad)))
+		return -EINVAL;
+
+	ret = aie4_fw_buf_property(meta->buf_type);
+	if (ret < 0) {
+		XDNA_DBG(xdna, "Unsupported buf_type %u", meta->buf_type);
+		return ret;
+	}
+	property = ret;
+
+	/*
+	 * On attach, this reference is held until detach or context teardown.
+	 * The device reaches the buffer through the process address space
+	 * (PASID), so the BO is not pinned or fenced: if the process changes
+	 * that mapping, only its own jobs are affected.
+	 */
+	payload_bo = amdxdna_gem_get_obj(client, meta->bo_handle, AMDXDNA_BO_SHARE);
+	if (!payload_bo) {
+		XDNA_DBG(xdna, "Get bo %u failed", meta->bo_handle);
+		return -EINVAL;
+	}
+
+	if (!attach) {
+		if (payload_bo != priv->fw_buf_bo[meta->buf_type]) {
+			XDNA_DBG(xdna, "BO %u is not attached as debug buffer type %u on %s",
+				 meta->bo_handle, meta->buf_type, hwctx->name);
+			ret = -EINVAL;
+			goto put_obj;
+		}
+
+		/*
+		 * While the context has a firmware context, remove the buffer
+		 * from it. After a reset or resume that could not recreate it,
+		 * there is no firmware context, so only drop the driver's
+		 * reference to the BO.
+		 */
+		if (priv->hw_ctx_id != CTX_INVALID_ID) {
+			ret = aie4_configure_hw_context_cert_log(ndev, priv->hw_ctx_id,
+								 property, &cl);
+			/* Firmware may still use it, so keep it attached. */
+			if (ret)
+				goto put_obj;
+		}
+
+		/* Release the attach reference; put_obj drops the lookup one. */
+		aie4_hwctx_put_fw_buf(hwctx, meta->buf_type);
+		XDNA_DBG(xdna, "Detached debug buffer type %u BO %u from %s",
+			 meta->buf_type, meta->bo_handle, hwctx->name);
+		ret = 0;
+		goto put_obj;
+	}
+
+	if (!meta->num_ucs || meta->num_ucs > AIE4_MAX_NUM_CERTS ||
+	    size < struct_size(meta, uc_info, meta->num_ucs)) {
+		XDNA_DBG(xdna, "Invalid num_ucs %u for size %u", meta->num_ucs, size);
+		ret = -EINVAL;
+		goto put_obj;
+	}
+
+	/* One buffer of each type per context. */
+	if (priv->fw_buf_bo[meta->buf_type]) {
+		XDNA_DBG(xdna, "Debug buffer type %u already attached on %s",
+			 meta->buf_type, hwctx->name);
+		ret = -EBUSY;
+		goto put_obj;
+	}
+
+	if (payload_bo->assigned_hwctx != AMDXDNA_INVALID_CTX_HANDLE) {
+		XDNA_DBG(xdna, "BO %u already attached to a context",
+			 meta->bo_handle);
+		ret = -EBUSY;
+		goto put_obj;
+	}
+
+	if (priv->hw_ctx_id == CTX_INVALID_ID) {
+		XDNA_ERR(xdna, "No firmware context to attach debug buffer type %u on %s",
+			 meta->buf_type, hwctx->name);
+		ret = -ENODEV;
+		goto put_obj;
+	}
+
+	base_addr = amdxdna_gem_dev_addr(payload_bo);
+	if (base_addr == AMDXDNA_INVALID_ADDR) {
+		XDNA_DBG(xdna, "Debug buffer type %u payload BO %u is not mapped",
+			 meta->buf_type, meta->bo_handle);
+		ret = -EINVAL;
+		goto put_obj;
+	}
+
+	for (i = 0; i < meta->num_ucs; i++) {
+		u32 slice_size = meta->uc_info[i].size;
+		u32 index = meta->uc_info[i].index;
+		u32 next_size;
+
+		if (index >= AIE4_MAX_NUM_CERTS) {
+			XDNA_DBG(xdna, "Invalid uc index %u", index);
+			ret = -EINVAL;
+			goto put_obj;
+		}
+
+		if (seen & BIT(index)) {
+			XDNA_DBG(xdna, "Duplicate uc index %u", index);
+			ret = -EINVAL;
+			goto put_obj;
+		}
+		seen |= BIT(index);
+
+		if (!slice_size)
+			continue;
+
+		if (check_add_overflow(prev_size, slice_size, &next_size) ||
+		    next_size > payload_bo->mem.size) {
+			XDNA_DBG(xdna, "uc[%u] slice 0x%x at 0x%x overflows bo size %lu",
+				 index, slice_size, prev_size, payload_bo->mem.size);
+			ret = -EINVAL;
+			goto put_obj;
+		}
+
+		cl.info[index].paddr = base_addr + prev_size;
+		cl.info[index].size = slice_size;
+		used |= BIT(index);
+		prev_size = next_size;
+	}
+
+	if (!used) {
+		XDNA_DBG(xdna, "Debug buffer type %u: all uc slices are empty",
+			 meta->buf_type);
+		ret = -EINVAL;
+		goto put_obj;
+	}
+
+	/* cl.info[] is indexed by uC, so cover up to the highest one used. */
+	cl.num = FIELD_PREP(AIE4_MSG_CERT_LOG_NUM, fls(used));
+	ret = aie4_configure_hw_context_cert_log(ndev, priv->hw_ctx_id, property, &cl);
+	if (ret)
+		goto put_obj;
+
+	/* Keep the lookup reference. */
+	payload_bo->assigned_hwctx = hwctx->id;
+	priv->fw_buf_bo[meta->buf_type] = payload_bo;
+	priv->fw_buf_cl[meta->buf_type] = cl;
+	XDNA_DBG(xdna, "Attached debug buffer type %u BO %u to %s",
+		 meta->buf_type, meta->bo_handle, hwctx->name);
+	return 0;
+
+put_obj:
+	amdxdna_gem_put_obj(payload_bo);
+	return ret;
+}
+
+int aie4_hwctx_config(struct amdxdna_hwctx *hwctx, u32 type, u64 value,
+		      void *buf, u32 size)
+{
+	struct amdxdna_dev *xdna = hwctx->client->xdna;
+
+	drm_WARN_ON(&xdna->ddev, !mutex_is_locked(&xdna->dev_lock));
+
+	switch (type) {
+	case DRM_AMDXDNA_HWCTX_ASSIGN_FW_BUF:
+		return aie4_hwctx_cfg_fw_buf(hwctx, buf, size, true);
+	case DRM_AMDXDNA_HWCTX_REMOVE_FW_BUF:
+		return aie4_hwctx_cfg_fw_buf(hwctx, buf, size, false);
+	default:
+		XDNA_DBG(xdna, "Not supported type %d", type);
+		return -EOPNOTSUPP;
+	}
+}
+
+int aie4_hwctx_sync_debug_bo(struct amdxdna_hwctx *hwctx, u32 debug_bo_hdl)
+{
+	/* NO-OP */
+	return 0;
+}
diff --git a/drivers/accel/amdxdna/aie4_message.c b/drivers/accel/amdxdna/aie4_message.c
index 060e199f4006..bc862cbe68f3 100644
--- a/drivers/accel/amdxdna/aie4_message.c
+++ b/drivers/accel/amdxdna/aie4_message.c
@@ -326,3 +326,23 @@ int aie4_set_ctx_hysteresis(struct amdxdna_dev_hdl *ndev, u32 timeout_us)
 
 	return ret;
 }
+
+int aie4_configure_hw_context_cert_log(struct amdxdna_dev_hdl *ndev,
+				       u32 hw_context_id, u32 property,
+				       const struct aie4_msg_context_config_cert_logging *cl)
+{
+	DECLARE_AIE_MSG(aie4_msg_configure_hw_context, AIE4_MSG_OP_CONFIGURE_HW_CONTEXT);
+	struct amdxdna_dev *xdna = ndev->aie.xdna;
+	int ret;
+
+	req.hw_context_id = hw_context_id;
+	req.property = property;
+	req.cert_logging = *cl;
+
+	ret = aie_send_mgmt_msg_wait(&ndev->aie, &msg);
+	if (ret)
+		XDNA_ERR(xdna, "Configure debug buffer failed, ctx %u property %u ret %d",
+			 hw_context_id, property, ret);
+
+	return ret;
+}
diff --git a/drivers/accel/amdxdna/aie4_msg_priv.h b/drivers/accel/amdxdna/aie4_msg_priv.h
index 566f6abc9e2d..977f7651341a 100644
--- a/drivers/accel/amdxdna/aie4_msg_priv.h
+++ b/drivers/accel/amdxdna/aie4_msg_priv.h
@@ -22,6 +22,7 @@ enum aie4_msg_opcode {
 	AIE4_MSG_OP_DESTROY_PARTITION                = 0x30002,
 	AIE4_MSG_OP_CREATE_HW_CONTEXT                = 0x30003,
 	AIE4_MSG_OP_DESTROY_HW_CONTEXT               = 0x30004,
+	AIE4_MSG_OP_CONFIGURE_HW_CONTEXT             = 0x30005,
 	AIE4_MSG_OP_AIE_TILE_INFO                    = 0x30006,
 	AIE4_MSG_OP_AIE_VERSION_INFO                 = 0x30007,
 	AIE4_MSG_OP_POWER_OVERRIDE                   = 0x3000B,
@@ -156,6 +157,40 @@ struct aie4_msg_destroy_hw_context_resp {
 	__u16 resvd;
 } __packed;
 
+enum aie4_msg_configure_hw_context_property {
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_PRIORITY_BAND,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_SCHEDULING,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_DPM,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_LOG_BUFFER,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_DEBUG_BUFFER,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_TRACE_BUFFER,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_CERT_DEBUG_QUEUE,
+	AIE4_CONFIGURE_HW_CONTEXT_PROPERTY_HANDLE,
+};
+
+#define AIE4_MAX_NUM_CERTS	6
+
+struct aie4_msg_context_config_cert_logging_info {
+	__u64 paddr;
+	__u32 size;
+} __packed;
+
+struct aie4_msg_context_config_cert_logging {
+#define AIE4_MSG_CERT_LOG_NUM	GENMASK(7, 0)
+	__u32 num;
+	struct aie4_msg_context_config_cert_logging_info info[AIE4_MAX_NUM_CERTS];
+} __packed;
+
+struct aie4_msg_configure_hw_context_req {
+	__u32 hw_context_id;
+	__u32 property;
+	struct aie4_msg_context_config_cert_logging cert_logging;
+} __packed;
+
+struct aie4_msg_configure_hw_context_resp {
+	enum aie4_msg_status status;
+} __packed;
+
 struct aie4_tile_info {
 	__u32 size;
 	__u16 major;
diff --git a/drivers/accel/amdxdna/aie4_pci.c b/drivers/accel/amdxdna/aie4_pci.c
index 02dcb2f813d4..101f0f27b466 100644
--- a/drivers/accel/amdxdna/aie4_pci.c
+++ b/drivers/accel/amdxdna/aie4_pci.c
@@ -1393,6 +1393,8 @@ const struct amdxdna_dev_ops aie4_vf_ops = {
 	.hwctx_init		= aie4_hwctx_init,
 	.hwctx_stop		= aie4_hwctx_stop,
 	.hwctx_fini		= aie4_hwctx_fini,
+	.hwctx_config		= aie4_hwctx_config,
+	.hwctx_sync_debug_bo	= aie4_hwctx_sync_debug_bo,
 	.cmd_submit		= aie4_cmd_submit,
 	.cmd_wait		= aie4_cmd_wait,
 	.get_aie_info		= aie4_get_info,
@@ -1408,6 +1410,8 @@ const struct amdxdna_dev_ops aie4_classic_ops = {
 	.hwctx_init		= aie4_hwctx_init,
 	.hwctx_stop		= aie4_hwctx_stop,
 	.hwctx_fini		= aie4_hwctx_fini,
+	.hwctx_config		= aie4_hwctx_config,
+	.hwctx_sync_debug_bo	= aie4_hwctx_sync_debug_bo,
 	.cmd_submit		= aie4_cmd_submit,
 	.cmd_wait		= aie4_cmd_wait,
 	.get_aie_info		= aie4_get_info,
diff --git a/drivers/accel/amdxdna/aie4_pci.h b/drivers/accel/amdxdna/aie4_pci.h
index fdc55e498246..158e68199952 100644
--- a/drivers/accel/amdxdna/aie4_pci.h
+++ b/drivers/accel/amdxdna/aie4_pci.h
@@ -49,6 +49,10 @@ struct amdxdna_hwctx_priv {
 	struct cert_comp                *cert_comp;
 	u32                             hw_ctx_id;
 	bool                            has_error;
+	/* Attached debug buffers, held until removal or fini. */
+	struct amdxdna_gem_obj          *fw_buf_bo[AMDXDNA_FW_BUF_NUM];
+	/* Their configuration, sent again when the context is created. */
+	struct aie4_msg_context_config_cert_logging fw_buf_cl[AMDXDNA_FW_BUF_NUM];
 
 	/* Direct and indirect packet storage aliasing umq_bo. */
 	u64                             write_index;
@@ -142,6 +146,9 @@ int aie4_msg_set_power_mode(struct amdxdna_dev_hdl *ndev, u8 power_mode);
 int aie4_set_runtime_cfg(struct amdxdna_dev_hdl *ndev, u32 type,
 			 const void *data, size_t size);
 int aie4_set_ctx_hysteresis(struct amdxdna_dev_hdl *ndev, u32 timeout_us);
+int aie4_configure_hw_context_cert_log(struct amdxdna_dev_hdl *ndev,
+				       u32 hw_context_id, u32 property,
+				       const struct aie4_msg_context_config_cert_logging *cl);
 u32 aie4_msg_pasid(struct amdxdna_client *client);
 
 /* aie4_ctx.c */
@@ -161,6 +168,9 @@ int aie4_hwctx_create(struct amdxdna_hwctx *hwctx);
 void aie4_hwctx_destroy(struct amdxdna_hwctx *hwctx, enum aie4_hwctx_flags);
 int aie4_hwctx_wait_for_running(struct amdxdna_hwctx *hwctx);
 void aie4_hwctx_resume_jobs(struct amdxdna_hwctx *hwctx);
+int aie4_hwctx_config(struct amdxdna_hwctx *hwctx, u32 type, u64 value,
+		      void *buf, u32 size);
+int aie4_hwctx_sync_debug_bo(struct amdxdna_hwctx *hwctx, u32 debug_bo_hdl);
 
 /* aie4_pci.c */
 void aie4_hwctx_suspend_all(struct amdxdna_dev_hdl *ndev, int clean_jobs);
diff --git a/drivers/accel/amdxdna/amdxdna_ctx.c b/drivers/accel/amdxdna/amdxdna_ctx.c
index f4004630dc1b..0a730a637180 100644
--- a/drivers/accel/amdxdna/amdxdna_ctx.c
+++ b/drivers/accel/amdxdna/amdxdna_ctx.c
@@ -365,9 +365,11 @@ int amdxdna_drm_config_hwctx_ioctl(struct drm_device *dev, void *data, struct dr
 
 	switch (args->param_type) {
 	case DRM_AMDXDNA_HWCTX_CONFIG_CU:
+	case DRM_AMDXDNA_HWCTX_ASSIGN_FW_BUF:
+	case DRM_AMDXDNA_HWCTX_REMOVE_FW_BUF:
 		/* For those types that param_val is pointer */
 		if (buf_size > PAGE_SIZE) {
-			XDNA_ERR(xdna, "Config CU param buffer too large");
+			XDNA_ERR(xdna, "Config param buffer too large");
 			return -E2BIG;
 		}
 
diff --git a/include/uapi/drm/amdxdna_accel.h b/include/uapi/drm/amdxdna_accel.h
index 51a507561df6..0e41c45ba720 100644
--- a/include/uapi/drm/amdxdna_accel.h
+++ b/include/uapi/drm/amdxdna_accel.h
@@ -135,6 +135,56 @@ enum amdxdna_drm_config_hwctx_param {
 	DRM_AMDXDNA_HWCTX_CONFIG_CU,
 	DRM_AMDXDNA_HWCTX_ASSIGN_DBG_BUF,
 	DRM_AMDXDNA_HWCTX_REMOVE_DBG_BUF,
+	DRM_AMDXDNA_HWCTX_ASSIGN_FW_BUF = 4,
+	DRM_AMDXDNA_HWCTX_REMOVE_FW_BUF,
+};
+
+/**
+ * enum amdxdna_fw_buf_type - Debug buffer type.
+ * @AMDXDNA_FW_BUF_DEBUG: Debug buffer.
+ * @AMDXDNA_FW_BUF_TRACE: Trace buffer.
+ * @AMDXDNA_FW_BUF_DBG_Q: Debug queue.
+ * @AMDXDNA_FW_BUF_LOG:   Log buffer.
+ * @AMDXDNA_FW_BUF_NUM:   Number of debug buffer types.
+ */
+enum amdxdna_fw_buf_type {
+	AMDXDNA_FW_BUF_DEBUG = 0,
+	AMDXDNA_FW_BUF_TRACE,
+	AMDXDNA_FW_BUF_DBG_Q,
+	AMDXDNA_FW_BUF_LOG,
+	AMDXDNA_FW_BUF_NUM,
+};
+
+/**
+ * struct amdxdna_uc_info_entry - per-uC slice descriptor in
+ *                                struct amdxdna_fw_buffer_metadata.
+ * @index: uC index. The payload BO is sliced per-uC; @index must be unique
+ *         within an array and stay below the device's CERT uC count.
+ * @size:  Per-uC slice size in bytes. Entries with size 0 are skipped.
+ *         At least one entry must have a size greater than 0.
+ */
+struct amdxdna_uc_info_entry {
+	__u32 index;
+	__u32 size;
+};
+
+/**
+ * struct amdxdna_fw_buffer_metadata - per-hwctx debug buffer description.
+ * @buf_type:   One of enum amdxdna_fw_buf_type.
+ * @num_ucs:    Number of valid entries in @uc_info.
+ * @pad:        MBZ. Reserved for ABI growth.
+ * @bo_handle:  Handle of the payload BO (AMDXDNA_BO_SHARE) holding the
+ *              concatenated per-uC slices.
+ * @command_id: User tag for trace correlation.
+ * @uc_info:    Per-uC slice descriptors, in payload BO layout order.
+ */
+struct amdxdna_fw_buffer_metadata {
+	__u8 buf_type;
+	__u8 num_ucs;
+	__u8 pad[2];
+	__u32 bo_handle;
+	__u64 command_id;
+	struct amdxdna_uc_info_entry uc_info[] __counted_by(num_ucs);
 };
 
 /**
-- 
2.34.1

[PATCH RESEND V1 4/4] accel/amdxdna: Add AIE4 CERT firmware version query

[David Zhang]

Enable the CERT firmware version query ioctl on AIE4.

Signed-off-by: David Zhang <yidong.zhang@amd.com>
---
 drivers/accel/amdxdna/aie4_pci.c | 3 +++
 include/uapi/drm/amdxdna_accel.h | 1 +
 2 files changed, 4 insertions(+)

diff --git a/drivers/accel/amdxdna/aie4_pci.c b/drivers/accel/amdxdna/aie4_pci.c
index 101f0f27b466..5b1d61dd0d8a 100644
--- a/drivers/accel/amdxdna/aie4_pci.c
+++ b/drivers/accel/amdxdna/aie4_pci.c
@@ -835,6 +835,9 @@ static int aie4_get_info(struct amdxdna_client *client, struct amdxdna_drm_get_i
 	case DRM_AMDXDNA_QUERY_FIRMWARE_VERSION:
 		ret = amdxdna_get_firmware_version(client, args, &xdna->fw_ver);
 		break;
+	case DRM_AMDXDNA_QUERY_CERT_FIRMWARE_VERSION:
+		ret = amdxdna_get_firmware_version(client, args, &ndev->cert_version);
+		break;
 	case DRM_AMDXDNA_GET_POWER_MODE:
 		ret = aie4_get_power_mode(client, args);
 		break;
diff --git a/include/uapi/drm/amdxdna_accel.h b/include/uapi/drm/amdxdna_accel.h
index 0e41c45ba720..6369d82e5067 100644
--- a/include/uapi/drm/amdxdna_accel.h
+++ b/include/uapi/drm/amdxdna_accel.h
@@ -525,6 +525,7 @@ enum amdxdna_drm_get_param {
 	DRM_AMDXDNA_GET_FORCE_PREEMPT_STATE,
 	DRM_AMDXDNA_QUERY_RESOURCE_INFO,
 	DRM_AMDXDNA_GET_FRAME_BOUNDARY_PREEMPT_STATE,
+	DRM_AMDXDNA_QUERY_CERT_FIRMWARE_VERSION = 14,
 };
 
 /**
-- 
2.34.1