From: Yair Shachar <yair.shac...@amd.com>
This patch adds the skeleton H/W debugger module support. This code
enables registration and unregistration of a single HSA process at a
time.
The module saves the process's pasid and use it to verify that only the
registered process is allowed to execute debugger operations through the
kernel driver.
Signed-off-by: Yair Shachar <yair.shachar at amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay at gmail.com>
---
drivers/gpu/drm/amd/amdkfd/Makefile | 3 +-
drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c | 142 ++++++++++
drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h | 193 ++++++++++++++
drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c | 135 ++++++++++
drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h | 287 ++++++++++++++++++++
drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h | 290 +++++++++++++++++++++
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 5 +
.../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 2 +-
9 files changed, 1058 insertions(+), 2 deletions(-)
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h
diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile
b/drivers/gpu/drm/amd/amdkfd/Makefile
index 652d254..2855115 100644
--- a/drivers/gpu/drm/amd/amdkfd/Makefile
+++ b/drivers/gpu/drm/amd/amdkfd/Makefile
@@ -12,6 +12,7 @@ amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o
kfd_topology.o \
kfd_kernel_queue_vi.o kfd_packet_manager.o \
kfd_process_queue_manager.o kfd_device_queue_manager.o \
kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \
- kfd_interrupt.o kfd_events.o cik_event_interrupt.o
+ kfd_interrupt.o kfd_events.o cik_event_interrupt.o \
+ kfd_dbgdev.o kfd_dbgmgr.o
obj-$(CONFIG_HSA_AMD) += amdkfd.o
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
new file mode 100644
index 0000000..eed4a83
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c
@@ -0,0 +1,142 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/log2.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/mutex.h>
+#include <linux/device.h>
+
+#include "kfd_pm4_headers.h"
+#include "kfd_pm4_headers_diq.h"
+#include "kfd_kernel_queue.h"
+#include "kfd_priv.h"
+#include "kfd_pm4_opcodes.h"
+#include "cik_regs.h"
+#include "kfd_dbgmgr.h"
+#include "kfd_dbgdev.h"
+#include "kfd_device_queue_manager.h"
+#include "../../radeon/cik_reg.h"
+
+static void dbgdev_address_watch_disable_nodiq(struct kfd_dev *dev)
+{
+ BUG_ON(!dev || !dev->kfd2kgd);
+
+ dev->kfd2kgd->address_watch_disable(dev->kgd);
+}
+
+static int dbgdev_register_nodiq(struct kfd_dbgdev *dbgdev)
+{
+ BUG_ON(!dbgdev);
+
+ /*
+ * no action is needed in this case,
+ * just make sure diq will not be used
+ */
+
+ dbgdev->kq = NULL;
+
+ return 0;
+}
+
+static int dbgdev_register_diq(struct kfd_dbgdev *dbgdev)
+{
+ struct queue_properties properties;
+ unsigned int qid;
+ struct kernel_queue *kq = NULL;
+ int status;
+
+ BUG_ON(!dbgdev || !dbgdev->pqm || !dbgdev->dev);
+
+ status = pqm_create_queue(dbgdev->pqm, dbgdev->dev, NULL,
+ &properties, 0, KFD_QUEUE_TYPE_DIQ,
+ &qid);
+
+ if (status) {
+ pr_err("amdkfd: Failed to create DIQ\n");
+ return status;
+ }
+
+ pr_debug("DIQ Created with queue id: %d\n", qid);
+
+ kq = pqm_get_kernel_queue(dbgdev->pqm, qid);
+
+ if (kq == NULL) {
+ pr_err("amdkfd: Error getting DIQ\n");
+ pqm_destroy_queue(dbgdev->pqm, qid);
+ return -EFAULT;
+ }
+
+ dbgdev->kq = kq;
+
+ return status;
+}
+
+static int dbgdev_unregister_nodiq(struct kfd_dbgdev *dbgdev)
+{
+ BUG_ON(!dbgdev || !dbgdev->dev);
+
+ /* disable watch address */
+ dbgdev_address_watch_disable_nodiq(dbgdev->dev);
+ return 0;
+}
+
+static int dbgdev_unregister_diq(struct kfd_dbgdev *dbgdev)
+{
+ /* todo - disable address watch */
+ int status;
+
+ BUG_ON(!dbgdev || !dbgdev->pqm || !dbgdev->kq);
+
+ status = pqm_destroy_queue(dbgdev->pqm,
+ dbgdev->kq->queue->properties.queue_id);
+ dbgdev->kq = NULL;
+
+ return status;
+}
+
+void kfd_dbgdev_init(struct kfd_dbgdev *pdbgdev, struct kfd_dev *pdev,
+ enum DBGDEV_TYPE type)
+{
+ BUG_ON(!pdbgdev || !pdev);
+
+ pdbgdev->dev = pdev;
+ pdbgdev->kq = NULL;
+ pdbgdev->type = type;
+ pdbgdev->pqm = NULL;
+
+ switch (type) {
+ case DBGDEV_TYPE_NODIQ:
+ pdbgdev->dbgdev_register = dbgdev_register_nodiq;
+ pdbgdev->dbgdev_unregister = dbgdev_unregister_nodiq;
+ break;
+ case DBGDEV_TYPE_DIQ:
+ default:
+ pdbgdev->dbgdev_register = dbgdev_register_diq;
+ pdbgdev->dbgdev_unregister = dbgdev_unregister_diq;
+ break;
+ }
+
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h
b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h
new file mode 100644
index 0000000..4b0dd5a
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h
@@ -0,0 +1,193 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef KFD_DBGDEV_H_
+#define KFD_DBGDEV_H_
+
+enum {
+ SQ_CMD_VMID_OFFSET = 28,
+ ADDRESS_WATCH_CNTL_OFFSET = 24
+};
+
+enum {
+ PRIV_QUEUE_SYNC_TIME_MS = 200
+};
+
+/* CONTEXT reg space definition */
+enum {
+ CONTEXT_REG_BASE = 0xA000,
+ CONTEXT_REG_END = 0xA400,
+ CONTEXT_REG_SIZE = CONTEXT_REG_END - CONTEXT_REG_BASE
+};
+
+/* USER CONFIG reg space definition */
+enum {
+ USERCONFIG_REG_BASE = 0xC000,
+ USERCONFIG_REG_END = 0x10000,
+ USERCONFIG_REG_SIZE = USERCONFIG_REG_END - USERCONFIG_REG_BASE
+};
+
+/* CONFIG reg space definition */
+enum {
+ CONFIG_REG_BASE = 0x2000, /* in dwords */
+ CONFIG_REG_END = 0x2B00,
+ CONFIG_REG_SIZE = CONFIG_REG_END - CONFIG_REG_BASE
+};
+
+/* SH reg space definition */
+enum {
+ SH_REG_BASE = 0x2C00,
+ SH_REG_END = 0x3000,
+ SH_REG_SIZE = SH_REG_END - SH_REG_BASE
+};
+
+enum SQ_IND_CMD_CMD {
+ SQ_IND_CMD_CMD_NULL = 0x00000000,
+ SQ_IND_CMD_CMD_HALT = 0x00000001,
+ SQ_IND_CMD_CMD_RESUME = 0x00000002,
+ SQ_IND_CMD_CMD_KILL = 0x00000003,
+ SQ_IND_CMD_CMD_DEBUG = 0x00000004,
+ SQ_IND_CMD_CMD_TRAP = 0x00000005,
+};
+
+enum SQ_IND_CMD_MODE {
+ SQ_IND_CMD_MODE_SINGLE = 0x00000000,
+ SQ_IND_CMD_MODE_BROADCAST = 0x00000001,
+ SQ_IND_CMD_MODE_BROADCAST_QUEUE = 0x00000002,
+ SQ_IND_CMD_MODE_BROADCAST_PIPE = 0x00000003,
+ SQ_IND_CMD_MODE_BROADCAST_ME = 0x00000004,
+};
+
+union SQ_IND_INDEX_BITS {
+ struct {
+ uint32_t wave_id:4;
+ uint32_t simd_id:2;
+ uint32_t thread_id:6;
+ uint32_t:1;
+ uint32_t force_read:1;
+ uint32_t read_timeout:1;
+ uint32_t unindexed:1;
+ uint32_t index:16;
+
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+union SQ_IND_CMD_BITS {
+ struct {
+ uint32_t data:32;
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+union SQ_CMD_BITS {
+ struct {
+ uint32_t cmd:3;
+ uint32_t:1;
+ uint32_t mode:3;
+ uint32_t check_vmid:1;
+ uint32_t trap_id:3;
+ uint32_t:5;
+ uint32_t wave_id:4;
+ uint32_t simd_id:2;
+ uint32_t:2;
+ uint32_t queue_id:3;
+ uint32_t:1;
+ uint32_t vm_id:4;
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+union SQ_IND_DATA_BITS {
+ struct {
+ uint32_t data:32;
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+union GRBM_GFX_INDEX_BITS {
+ struct {
+ uint32_t instance_index:8;
+ uint32_t sh_index:8;
+ uint32_t se_index:8;
+ uint32_t:5;
+ uint32_t sh_broadcast_writes:1;
+ uint32_t instance_broadcast_writes:1;
+ uint32_t se_broadcast_writes:1;
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+union TCP_WATCH_ADDR_H_BITS {
+ struct {
+ uint32_t addr:16;
+ uint32_t:16;
+
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+union TCP_WATCH_ADDR_L_BITS {
+ struct {
+ uint32_t:6;
+ uint32_t addr:26;
+ } bitfields, bits;
+ uint32_t u32All;
+ signed int i32All;
+ float f32All;
+};
+
+enum {
+ QUEUESTATE__INVALID = 0, /* so by default we'll get invalid state */
+ QUEUESTATE__ACTIVE_COMPLETION_PENDING,
+ QUEUESTATE__ACTIVE
+};
+
+union ULARGE_INTEGER {
+ struct {
+ uint32_t low_part;
+ uint32_t high_part;
+ } u;
+ unsigned long long quad_part;
+};
+
+
+#define KFD_CIK_VMID_START_OFFSET (8)
+#define KFD_CIK_VMID_END_OFFSET (KFD_CIK_VMID_START_OFFSET + (8))
+
+
+void kfd_dbgdev_init(struct kfd_dbgdev *pdbgdev, struct kfd_dev *pdev,
+ enum DBGDEV_TYPE type);
+
+#endif /* KFD_DBGDEV_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c
b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c
new file mode 100644
index 0000000..d8c45a3
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/log2.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/device.h>
+
+#include "kfd_priv.h"
+#include "cik_regs.h"
+#include "kfd_pm4_headers.h"
+#include "kfd_pm4_headers_diq.h"
+#include "kfd_dbgmgr.h"
+#include "kfd_dbgdev.h"
+
+static DEFINE_MUTEX(kfd_dbgmgr_mutex);
+
+struct mutex *get_dbgmgr_mutex(void)
+{
+ return &kfd_dbgmgr_mutex;
+}
+
+
+static void kfd_dbgmgr_uninitialize(struct kfd_dbgmgr *pmgr)
+{
+ BUG_ON(!pmgr);
+
+ kfree(pmgr->dbgdev);
+
+ pmgr->dbgdev = NULL;
+ pmgr->pasid = 0;
+ pmgr->dev = NULL;
+}
+
+void kfd_dbgmgr_destroy(struct kfd_dbgmgr *pmgr)
+{
+ if (pmgr != NULL) {
+ kfd_dbgmgr_uninitialize(pmgr);
+ kfree(pmgr);
+ }
+}
+
+bool kfd_dbgmgr_create(struct kfd_dbgmgr **ppmgr, struct kfd_dev *pdev)
+{
+ enum DBGDEV_TYPE type = DBGDEV_TYPE_DIQ;
+ struct kfd_dbgmgr *new_buff;
+
+ BUG_ON(pdev == NULL);
+ BUG_ON(!pdev->init_complete);
+
+ new_buff = kfd_alloc_struct(new_buff);
+ if (!new_buff) {
+ pr_err("amdkfd: Failed to allocate dbgmgr instance\n");
+ return false;
+ }
+
+ new_buff->pasid = 0;
+ new_buff->dev = pdev;
+ new_buff->dbgdev = kfd_alloc_struct(new_buff->dbgdev);
+ if (!new_buff->dbgdev) {
+ pr_err("amdkfd: Failed to allocate dbgdev instance\n");
+ kfree(new_buff);
+ return false;
+ }
+
+ /* get actual type of DBGDevice cpsch or not */
+ if (sched_policy == KFD_SCHED_POLICY_NO_HWS)
+ type = DBGDEV_TYPE_NODIQ;
+
+ kfd_dbgdev_init(new_buff->dbgdev, pdev, type);
+ *ppmgr = new_buff;
+
+ return true;
+}
+
+long kfd_dbgmgr_register(struct kfd_dbgmgr *pmgr, struct kfd_process *p)
+{
+ BUG_ON(!p || !pmgr || !pmgr->dbgdev);
+
+ if (pmgr->pasid != 0) {
+ pr_debug("H/W debugger is already active using pasid %d\n",
+ pmgr->pasid);
+ return -EBUSY;
+ }
+
+ /* remember pasid */
+ pmgr->pasid = p->pasid;
+
+ /* provide the pqm for diq generation */
+ pmgr->dbgdev->pqm = &p->pqm;
+
+ /* activate the actual registering */
+ pmgr->dbgdev->dbgdev_register(pmgr->dbgdev);
+
+ return 0;
+}
+
+long kfd_dbgmgr_unregister(struct kfd_dbgmgr *pmgr, struct kfd_process *p)
+{
+ BUG_ON(!p || !pmgr || !pmgr->dbgdev);
+
+ /* Is the requests coming from the already registered process? */
+ if (pmgr->pasid != p->pasid) {
+ pr_debug("H/W debugger is not registered by calling pasid %d\n",
+ p->pasid);
+ return -EINVAL;
+ }
+
+ pmgr->dbgdev->dbgdev_unregister(pmgr->dbgdev);
+
+ pmgr->pasid = 0;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h
b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h
new file mode 100644
index 0000000..4ea361c
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h
@@ -0,0 +1,287 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef KFD_DBGMGR_H_
+#define KFD_DBGMGR_H_
+
+#include "kfd_priv.h"
+
+/* must align with hsakmttypes definition */
+#pragma pack(push, 4)
+
+enum HSA_DBG_WAVEOP {
+ HSA_DBG_WAVEOP_HALT = 1, /* Halts a wavefront */
+ HSA_DBG_WAVEOP_RESUME = 2, /* Resumes a wavefront */
+ HSA_DBG_WAVEOP_KILL = 3, /* Kills a wavefront */
+ HSA_DBG_WAVEOP_DEBUG = 4, /* Causes wavefront to enter
+ debug mode */
+ HSA_DBG_WAVEOP_TRAP = 5, /* Causes wavefront to take
+ a trap */
+ HSA_DBG_NUM_WAVEOP = 5,
+ HSA_DBG_MAX_WAVEOP = 0xFFFFFFFF
+};
+
+enum HSA_DBG_WAVEMODE {
+ /* send command to a single wave */
+ HSA_DBG_WAVEMODE_SINGLE = 0,
+ /*
+ * Broadcast to all wavefronts of all processes is not
+ * supported for HSA user mode
+ */
+
+ /* send to waves within current process */
+ HSA_DBG_WAVEMODE_BROADCAST_PROCESS = 2,
+ /* send to waves within current process on CU */
+ HSA_DBG_WAVEMODE_BROADCAST_PROCESS_CU = 3,
+ HSA_DBG_NUM_WAVEMODE = 3,
+ HSA_DBG_MAX_WAVEMODE = 0xFFFFFFFF
+};
+
+enum HSA_DBG_WAVEMSG_TYPE {
+ HSA_DBG_WAVEMSG_AUTO = 0,
+ HSA_DBG_WAVEMSG_USER = 1,
+ HSA_DBG_WAVEMSG_ERROR = 2,
+ HSA_DBG_NUM_WAVEMSG,
+ HSA_DBG_MAX_WAVEMSG = 0xFFFFFFFF
+};
+
+enum HSA_DBG_WATCH_MODE {
+ HSA_DBG_WATCH_READ = 0, /* Read operations only */
+ HSA_DBG_WATCH_NONREAD = 1, /* Write or Atomic operations only */
+ HSA_DBG_WATCH_ATOMIC = 2, /* Atomic Operations only */
+ HSA_DBG_WATCH_ALL = 3, /* Read, Write or Atomic operations */
+ HSA_DBG_WATCH_NUM,
+ HSA_DBG_WATCH_SIZE = 0xFFFFFFFF
+};
+
+/* This structure is hardware specific and may change in the future */
+struct HsaDbgWaveMsgAMDGen2 {
+ union {
+ struct ui32 {
+ uint32_t UserData:8; /* user data */
+ uint32_t ShaderArray:1; /* Shader array */
+ uint32_t Priv:1; /* Privileged */
+ uint32_t Reserved0:4; /* This field is reserved,
+ should be 0 */
+ uint32_t WaveId:4; /* wave id */
+ uint32_t SIMD:2; /* SIMD id */
+ uint32_t HSACU:4; /* Compute unit */
+ uint32_t ShaderEngine:2;/* Shader engine */
+ uint32_t MessageType:2; /* see HSA_DBG_WAVEMSG_TYPE */
+ uint32_t Reserved1:4; /* This field is reserved,
+ should be 0 */
+ } ui32;
+ uint32_t Value;
+ };
+ uint32_t Reserved2;
+};
+
+union HsaDbgWaveMessageAMD {
+ struct HsaDbgWaveMsgAMDGen2 WaveMsgInfoGen2;
+ /* for future HsaDbgWaveMsgAMDGen3; */
+};
+
+struct HsaDbgWaveMessage {
+ void *MemoryVA; /* ptr to associated host-accessible data */
+ union HsaDbgWaveMessageAMD DbgWaveMsg;
+};
+
+/*
+ * TODO: This definitions to be MOVED to kfd_event, once it is implemented.
+ *
+ * HSA sync primitive, Event and HW Exception notification API definitions.
+ * The API functions allow the runtime to define a so-called sync-primitive,
+ * a SW object combining a user-mode provided "syncvar" and a scheduler event
+ * that can be signaled through a defined GPU interrupt. A syncvar is
+ * a process virtual memory location of a certain size that can be accessed
+ * by CPU and GPU shader code within the process to set and query the content
+ * within that memory. The definition of the content is determined by the HSA
+ * runtime and potentially GPU shader code interfacing with the HSA runtime.
+ * The syncvar values may be commonly written through an PM4 WRITE_DATA packet
+ * in the user mode instruction stream. The OS scheduler event is typically
+ * associated and signaled by an interrupt issued by the GPU, but other HSA
+ * system interrupt conditions from other HW (e.g. IOMMUv2) may be surfaced
+ * by the KFD by this mechanism, too. */
+
+/* these are the new definitions for events */
+enum HSA_EVENTTYPE {
+ HSA_EVENTTYPE_SIGNAL = 0, /* user-mode generated GPU signal */
+ HSA_EVENTTYPE_NODECHANGE = 1, /* HSA node change (attach/detach) */
+ HSA_EVENTTYPE_DEVICESTATECHANGE = 2, /* HSA device state change
+ (start/stop) */
+ HSA_EVENTTYPE_HW_EXCEPTION = 3, /* GPU shader exception event */
+ HSA_EVENTTYPE_SYSTEM_EVENT = 4, /* GPU SYSCALL with parameter info */
+ HSA_EVENTTYPE_DEBUG_EVENT = 5, /* GPU signal for debugging */
+ HSA_EVENTTYPE_PROFILE_EVENT = 6,/* GPU signal for profiling */
+ HSA_EVENTTYPE_QUEUE_EVENT = 7, /* GPU signal queue idle state
+ (EOP pm4) */
+ /* ... */
+ HSA_EVENTTYPE_MAXID,
+ HSA_EVENTTYPE_TYPE_SIZE = 0xFFFFFFFF
+};
+
+/* Sub-definitions for various event types: Syncvar */
+struct HsaSyncVar {
+ union SyncVar {
+ void *UserData; /* pointer to user mode data */
+ uint64_t UserDataPtrValue; /* 64bit compatibility of value */
+ } SyncVar;
+ uint64_t SyncVarSize;
+};
+
+/* Sub-definitions for various event types: NodeChange */
+
+enum HSA_EVENTTYPE_NODECHANGE_FLAGS {
+ HSA_EVENTTYPE_NODECHANGE_ADD = 0,
+ HSA_EVENTTYPE_NODECHANGE_REMOVE = 1,
+ HSA_EVENTTYPE_NODECHANGE_SIZE = 0xFFFFFFFF
+};
+
+struct HsaNodeChange {
+ /* HSA node added/removed on the platform */
+ enum HSA_EVENTTYPE_NODECHANGE_FLAGS Flags;
+};
+
+/* Sub-definitions for various event types: DeviceStateChange */
+enum HSA_EVENTTYPE_DEVICESTATECHANGE_FLAGS {
+ /* device started (and available) */
+ HSA_EVENTTYPE_DEVICESTATUSCHANGE_START = 0,
+ /* device stopped (i.e. unavailable) */
+ HSA_EVENTTYPE_DEVICESTATUSCHANGE_STOP = 1,
+ HSA_EVENTTYPE_DEVICESTATUSCHANGE_SIZE = 0xFFFFFFFF
+};
+
+enum HSA_DEVICE {
+ HSA_DEVICE_CPU = 0,
+ HSA_DEVICE_GPU = 1,
+ MAX_HSA_DEVICE = 2
+};
+
+struct HsaDeviceStateChange {
+ uint32_t NodeId; /* F-NUMA node that contains the device */
+ enum HSA_DEVICE Device; /* device type: GPU or CPU */
+ enum HSA_EVENTTYPE_DEVICESTATECHANGE_FLAGS Flags; /* event flags */
+};
+
+struct HsaEventData {
+ enum HSA_EVENTTYPE EventType; /* event type */
+ union EventData {
+ /*
+ * return data associated with HSA_EVENTTYPE_SIGNAL
+ * and other events
+ */
+ struct HsaSyncVar SyncVar;
+
+ /* data associated with HSA_EVENTTYPE_NODE_CHANGE */
+ struct HsaNodeChange NodeChangeState;
+
+ /* data associated with HSA_EVENTTYPE_DEVICE_STATE_CHANGE */
+ struct HsaDeviceStateChange DeviceState;
+ } EventData;
+
+ /* the following data entries are internal to the KFD & thunk itself */
+
+ /* internal thunk store for Event data (OsEventHandle) */
+ uint64_t HWData1;
+ /* internal thunk store for Event data (HWAddress) */
+ uint64_t HWData2;
+ /* internal thunk store for Event data (HWData) */
+ uint32_t HWData3;
+};
+
+struct HsaEventDescriptor {
+ /* event type to allocate */
+ enum HSA_EVENTTYPE EventType;
+ /* H-NUMA node containing GPU device that is event source */
+ uint32_t NodeId;
+ /* pointer to user mode syncvar data, syncvar->UserDataPtrValue
+ * may be NULL
+ */
+ struct HsaSyncVar SyncVar;
+};
+
+struct HsaEvent {
+ uint32_t EventId;
+ struct HsaEventData EventData;
+};
+
+#pragma pack(pop)
+
+enum DBGDEV_TYPE {
+ DBGDEV_TYPE_ILLEGAL = 0,
+ DBGDEV_TYPE_NODIQ = 1,
+ DBGDEV_TYPE_DIQ = 2,
+ DBGDEV_TYPE_TEST = 3
+};
+
+struct dbg_address_watch_info {
+ struct kfd_process *process;
+ enum HSA_DBG_WATCH_MODE *watch_mode;
+ uint64_t *watch_address;
+ uint64_t *watch_mask;
+ struct HsaEvent *watch_event;
+ uint32_t num_watch_points;
+};
+
+struct dbg_wave_control_info {
+ struct kfd_process *process;
+ uint32_t trapId;
+ enum HSA_DBG_WAVEOP operand;
+ enum HSA_DBG_WAVEMODE mode;
+ struct HsaDbgWaveMessage dbgWave_msg;
+};
+
+struct kfd_dbgdev {
+
+ /* The device that owns this data. */
+ struct kfd_dev *dev;
+
+ /* kernel queue for DIQ */
+ struct kernel_queue *kq;
+
+ /* a pointer to the pqm of the calling process */
+ struct process_queue_manager *pqm;
+
+ /* type of debug device ( DIQ, non DIQ, etc. ) */
+ enum DBGDEV_TYPE type;
+
+ /* virtualized function pointers to device dbg */
+ int (*dbgdev_register)(struct kfd_dbgdev *dbgdev);
+ int (*dbgdev_unregister)(struct kfd_dbgdev *dbgdev);
+
+};
+
+struct kfd_dbgmgr {
+ unsigned int pasid;
+ struct kfd_dev *dev;
+ struct kfd_dbgdev *dbgdev;
+};
+
+/* prototypes for debug manager functions */
+struct mutex *get_dbgmgr_mutex(void);
+void kfd_dbgmgr_destroy(struct kfd_dbgmgr *pmgr);
+bool kfd_dbgmgr_create(struct kfd_dbgmgr **ppmgr, struct kfd_dev *pdev);
+long kfd_dbgmgr_register(struct kfd_dbgmgr *pmgr, struct kfd_process *p);
+long kfd_dbgmgr_unregister(struct kfd_dbgmgr *pmgr, struct kfd_process *p);
+
+#endif /* KFD_DBGMGR_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index f1f86db..1d1e2e9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -37,6 +37,7 @@ static const struct kfd_device_info kaveri_device_info = {
.max_no_of_hqd = 24,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
+ .num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED
};
@@ -296,6 +297,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
goto dqm_start_error;
}
+ kfd->dbgmgr = NULL;
+
kfd->init_complete = true;
dev_info(kfd_device, "added device (%x:%x)\n", kfd->pdev->vendor,
kfd->pdev->device);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h
b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h
new file mode 100644
index 0000000..a0ff348
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h
@@ -0,0 +1,290 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef KFD_PM4_HEADERS_DIQ_H_
+#define KFD_PM4_HEADERS_DIQ_H_
+
+/*--------------------_INDIRECT_BUFFER-------------------- */
+
+#ifndef _PM4__INDIRECT_BUFFER_DEFINED
+#define _PM4__INDIRECT_BUFFER_DEFINED
+enum _INDIRECT_BUFFER_cache_policy_enum {
+ cache_policy___indirect_buffer__lru = 0,
+ cache_policy___indirect_buffer__stream = 1,
+ cache_policy___indirect_buffer__bypass = 2
+};
+
+enum {
+ IT_INDIRECT_BUFFER_PASID = 0x5C
+};
+
+struct pm4__indirect_buffer_pasid {
+ union {
+ union PM4_MES_TYPE_3_HEADER header; /* header */
+ unsigned int ordinal1;
+ };
+
+ union {
+ struct {
+ unsigned int reserved1:2;
+ unsigned int ib_base_lo:30;
+ } bitfields2;
+ unsigned int ordinal2;
+ };
+
+ union {
+ struct {
+ unsigned int ib_base_hi:16;
+ unsigned int reserved2:16;
+ } bitfields3;
+ unsigned int ordinal3;
+ };
+
+ union {
+ unsigned int control;
+ unsigned int ordinal4;
+ };
+
+ union {
+ struct {
+ unsigned int pasid:10;
+ unsigned int reserved4:22;
+ } bitfields5;
+ unsigned int ordinal5;
+ };
+
+};
+
+#endif
+
+/*--------------------_RELEASE_MEM-------------------- */
+
+#ifndef _PM4__RELEASE_MEM_DEFINED
+#define _PM4__RELEASE_MEM_DEFINED
+enum _RELEASE_MEM_event_index_enum {
+ event_index___release_mem__end_of_pipe = 5,
+ event_index___release_mem__shader_done = 6
+};
+
+enum _RELEASE_MEM_cache_policy_enum {
+ cache_policy___release_mem__lru = 0,
+ cache_policy___release_mem__stream = 1,
+ cache_policy___release_mem__bypass = 2
+};
+
+enum _RELEASE_MEM_dst_sel_enum {
+ dst_sel___release_mem__memory_controller = 0,
+ dst_sel___release_mem__tc_l2 = 1,
+ dst_sel___release_mem__queue_write_pointer_register = 2,
+ dst_sel___release_mem__queue_write_pointer_poll_mask_bit = 3
+};
+
+enum _RELEASE_MEM_int_sel_enum {
+ int_sel___release_mem__none = 0,
+ int_sel___release_mem__send_interrupt_only = 1,
+ int_sel___release_mem__send_interrupt_after_write_confirm = 2,
+ int_sel___release_mem__send_data_after_write_confirm = 3
+};
+
+enum _RELEASE_MEM_data_sel_enum {
+ data_sel___release_mem__none = 0,
+ data_sel___release_mem__send_32_bit_low = 1,
+ data_sel___release_mem__send_64_bit_data = 2,
+ data_sel___release_mem__send_gpu_clock_counter = 3,
+ data_sel___release_mem__send_cp_perfcounter_hi_lo = 4,
+ data_sel___release_mem__store_gds_data_to_memory = 5
+};
+
+struct pm4__release_mem {
+ union {
+ union PM4_MES_TYPE_3_HEADER header; /*header */
+ unsigned int ordinal1;
+ };
+
+ union {
+ struct {
+ unsigned int event_type:6;
+ unsigned int reserved1:2;
+ enum _RELEASE_MEM_event_index_enum event_index:4;
+ unsigned int tcl1_vol_action_ena:1;
+ unsigned int tc_vol_action_ena:1;
+ unsigned int reserved2:1;
+ unsigned int tc_wb_action_ena:1;
+ unsigned int tcl1_action_ena:1;
+ unsigned int tc_action_ena:1;
+ unsigned int reserved3:6;
+ unsigned int atc:1;
+ enum _RELEASE_MEM_cache_policy_enum cache_policy:2;
+ unsigned int reserved4:5;
+ } bitfields2;
+ unsigned int ordinal2;
+ };
+
+ union {
+ struct {
+ unsigned int reserved5:16;
+ enum _RELEASE_MEM_dst_sel_enum dst_sel:2;
+ unsigned int reserved6:6;
+ enum _RELEASE_MEM_int_sel_enum int_sel:3;
+ unsigned int reserved7:2;
+ enum _RELEASE_MEM_data_sel_enum data_sel:3;
+ } bitfields3;
+ unsigned int ordinal3;
+ };
+
+ union {
+ struct {
+ unsigned int reserved8:2;
+ unsigned int address_lo_32b:30;
+ } bitfields4;
+ struct {
+ unsigned int reserved9:3;
+ unsigned int address_lo_64b:29;
+ } bitfields5;
+ unsigned int ordinal4;
+ };
+
+ unsigned int address_hi;
+
+ unsigned int data_lo;
+
+ unsigned int data_hi;
+
+};
+#endif
+
+
+/*--------------------_SET_CONFIG_REG-------------------- */
+
+#ifndef _PM4__SET_CONFIG_REG_DEFINED
+#define _PM4__SET_CONFIG_REG_DEFINED
+
+struct pm4__set_config_reg {
+ union {
+ union PM4_MES_TYPE_3_HEADER header; /*header */
+ unsigned int ordinal1;
+ };
+
+ union {
+ struct {
+ unsigned int reg_offset:16;
+ unsigned int reserved1:7;
+ unsigned int vmid_shift:5;
+ unsigned int insert_vmid:1;
+ unsigned int reserved2:3;
+ } bitfields2;
+ unsigned int ordinal2;
+ };
+
+ unsigned int reg_data[1]; /*1..N of these fields */
+
+};
+#endif
+
+/*--------------------_WAIT_REG_MEM-------------------- */
+
+#ifndef _PM4__WAIT_REG_MEM_DEFINED
+#define _PM4__WAIT_REG_MEM_DEFINED
+enum _WAIT_REG_MEM_function_enum {
+ function___wait_reg_mem__always_pass = 0,
+ function___wait_reg_mem__less_than_ref_value = 1,
+ function___wait_reg_mem__less_than_equal_to_the_ref_value = 2,
+ function___wait_reg_mem__equal_to_the_reference_value = 3,
+ function___wait_reg_mem__not_equal_reference_value = 4,
+ function___wait_reg_mem__greater_than_or_equal_reference_value = 5,
+ function___wait_reg_mem__greater_than_reference_value = 6,
+ function___wait_reg_mem__reserved = 7
+};
+
+enum _WAIT_REG_MEM_mem_space_enum {
+ mem_space___wait_reg_mem__register_space = 0,
+ mem_space___wait_reg_mem__memory_space = 1
+};
+
+enum _WAIT_REG_MEM_operation_enum {
+ operation___wait_reg_mem__wait_reg_mem = 0,
+ operation___wait_reg_mem__wr_wait_wr_reg = 1
+};
+
+struct pm4__wait_reg_mem {
+ union {
+ union PM4_MES_TYPE_3_HEADER header; /*header */
+ unsigned int ordinal1;
+ };
+
+ union {
+ struct {
+ enum _WAIT_REG_MEM_function_enum function:3;
+ unsigned int reserved1:1;
+ enum _WAIT_REG_MEM_mem_space_enum mem_space:2;
+ enum _WAIT_REG_MEM_operation_enum operation:2;
+ unsigned int reserved2:24;
+ } bitfields2;
+ unsigned int ordinal2;
+ };
+
+ union {
+ struct {
+ unsigned int reserved3:2;
+ unsigned int memory_poll_addr_lo:30;
+ } bitfields3;
+ struct {
+ unsigned int register_poll_addr:16;
+ unsigned int reserved4:16;
+ } bitfields4;
+ struct {
+ unsigned int register_write_addr:16;
+ unsigned int reserved5:16;
+ } bitfields5;
+ unsigned int ordinal3;
+ };
+
+ union {
+ struct {
+ unsigned int poll_address_hi:16;
+ unsigned int reserved6:16;
+ } bitfields6;
+ struct {
+ unsigned int register_write_addr:16;
+ unsigned int reserved7:16;
+ } bitfields7;
+ unsigned int ordinal4;
+ };
+
+ unsigned int reference;
+
+ unsigned int mask;
+
+ union {
+ struct {
+ unsigned int poll_interval:16;
+ unsigned int reserved8:16;
+ } bitfields8;
+ unsigned int ordinal7;
+ };
+
+};
+#endif
+
+
+#endif /* KFD_PM4_HEADERS_DIQ_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 8e99d2b..12ad585 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -195,6 +195,9 @@ struct kfd_dev {
* from the HW ring into a SW ring.
*/
bool interrupts_active;
+
+ /* Debug manager */
+ struct kfd_dbgmgr *dbgmgr;
};
/* KGD2KFD callbacks */
@@ -649,6 +652,8 @@ int pqm_create_queue(struct process_queue_manager *pqm,
int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid);
int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid,
struct queue_properties *p);
+struct kernel_queue *pqm_get_kernel_queue(struct process_queue_manager *pqm,
+ unsigned int qid);
/* Packet Manager */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 85b7fec..7b69070 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -357,7 +357,7 @@ int pqm_update_queue(struct process_queue_manager *pqm,
unsigned int qid,
return 0;
}
-static __attribute__((unused)) struct kernel_queue *pqm_get_kernel_queue(
+struct kernel_queue *pqm_get_kernel_queue(
struct process_queue_manager *pqm,
unsigned int qid)
{
--
2.1.0
