summaryrefslogtreecommitdiff
path: root/drivers/accel/ivpu
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/accel/ivpu')
-rw-r--r--drivers/accel/ivpu/ivpu_drv.c1
-rw-r--r--drivers/accel/ivpu/ivpu_drv.h15
-rw-r--r--drivers/accel/ivpu/ivpu_hw.c4
-rw-r--r--drivers/accel/ivpu/ivpu_hw_ip.c1
-rw-r--r--drivers/accel/ivpu/ivpu_job.c81
-rw-r--r--drivers/accel/ivpu/ivpu_pm.c15
6 files changed, 81 insertions, 36 deletions
diff --git a/drivers/accel/ivpu/ivpu_drv.c b/drivers/accel/ivpu/ivpu_drv.c
index 0e7748c5e117..3d6d52492536 100644
--- a/drivers/accel/ivpu/ivpu_drv.c
+++ b/drivers/accel/ivpu/ivpu_drv.c
@@ -704,6 +704,7 @@ static struct pci_device_id ivpu_pci_ids[] = {
{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_ARL) },
{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_LNL) },
{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_PTL_P) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_WCL) },
{ }
};
MODULE_DEVICE_TABLE(pci, ivpu_pci_ids);
diff --git a/drivers/accel/ivpu/ivpu_drv.h b/drivers/accel/ivpu/ivpu_drv.h
index 5497e7030e91..62ab1c654e63 100644
--- a/drivers/accel/ivpu/ivpu_drv.h
+++ b/drivers/accel/ivpu/ivpu_drv.h
@@ -1,6 +1,6 @@
/* SPDX-License-Identifier: GPL-2.0-only */
/*
- * Copyright (C) 2020-2024 Intel Corporation
+ * Copyright (C) 2020-2025 Intel Corporation
*/
#ifndef __IVPU_DRV_H__
@@ -26,6 +26,7 @@
#define PCI_DEVICE_ID_ARL 0xad1d
#define PCI_DEVICE_ID_LNL 0x643e
#define PCI_DEVICE_ID_PTL_P 0xb03e
+#define PCI_DEVICE_ID_WCL 0xfd3e
#define IVPU_HW_IP_37XX 37
#define IVPU_HW_IP_40XX 40
@@ -165,6 +166,7 @@ struct ivpu_device {
int boot;
int jsm;
int tdr;
+ int inference;
int autosuspend;
int d0i3_entry_msg;
int state_dump_msg;
@@ -207,10 +209,11 @@ extern bool ivpu_force_snoop;
#define IVPU_TEST_MODE_D0I3_MSG_ENABLE BIT(5)
#define IVPU_TEST_MODE_MIP_DISABLE BIT(6)
#define IVPU_TEST_MODE_DISABLE_TIMEOUTS BIT(8)
-#define IVPU_TEST_MODE_TURBO BIT(9)
-#define IVPU_TEST_MODE_CLK_RELINQ_DISABLE BIT(10)
-#define IVPU_TEST_MODE_CLK_RELINQ_ENABLE BIT(11)
-#define IVPU_TEST_MODE_D0I2_DISABLE BIT(12)
+#define IVPU_TEST_MODE_TURBO_ENABLE BIT(9)
+#define IVPU_TEST_MODE_TURBO_DISABLE BIT(10)
+#define IVPU_TEST_MODE_CLK_RELINQ_DISABLE BIT(11)
+#define IVPU_TEST_MODE_CLK_RELINQ_ENABLE BIT(12)
+#define IVPU_TEST_MODE_D0I2_DISABLE BIT(13)
extern int ivpu_test_mode;
struct ivpu_file_priv *ivpu_file_priv_get(struct ivpu_file_priv *file_priv);
@@ -240,6 +243,7 @@ static inline int ivpu_hw_ip_gen(struct ivpu_device *vdev)
case PCI_DEVICE_ID_LNL:
return IVPU_HW_IP_40XX;
case PCI_DEVICE_ID_PTL_P:
+ case PCI_DEVICE_ID_WCL:
return IVPU_HW_IP_50XX;
default:
dump_stack();
@@ -256,6 +260,7 @@ static inline int ivpu_hw_btrs_gen(struct ivpu_device *vdev)
return IVPU_HW_BTRS_MTL;
case PCI_DEVICE_ID_LNL:
case PCI_DEVICE_ID_PTL_P:
+ case PCI_DEVICE_ID_WCL:
return IVPU_HW_BTRS_LNL;
default:
dump_stack();
diff --git a/drivers/accel/ivpu/ivpu_hw.c b/drivers/accel/ivpu/ivpu_hw.c
index 633160470c93..08dcc31b56f4 100644
--- a/drivers/accel/ivpu/ivpu_hw.c
+++ b/drivers/accel/ivpu/ivpu_hw.c
@@ -94,12 +94,14 @@ static void timeouts_init(struct ivpu_device *vdev)
vdev->timeout.boot = -1;
vdev->timeout.jsm = -1;
vdev->timeout.tdr = -1;
+ vdev->timeout.inference = -1;
vdev->timeout.autosuspend = -1;
vdev->timeout.d0i3_entry_msg = -1;
} else if (ivpu_is_fpga(vdev)) {
vdev->timeout.boot = 50;
vdev->timeout.jsm = 15000;
vdev->timeout.tdr = 30000;
+ vdev->timeout.inference = 900000;
vdev->timeout.autosuspend = -1;
vdev->timeout.d0i3_entry_msg = 500;
vdev->timeout.state_dump_msg = 10000;
@@ -107,6 +109,7 @@ static void timeouts_init(struct ivpu_device *vdev)
vdev->timeout.boot = 50;
vdev->timeout.jsm = 500;
vdev->timeout.tdr = 10000;
+ vdev->timeout.inference = 300000;
vdev->timeout.autosuspend = 100;
vdev->timeout.d0i3_entry_msg = 100;
vdev->timeout.state_dump_msg = 10;
@@ -114,6 +117,7 @@ static void timeouts_init(struct ivpu_device *vdev)
vdev->timeout.boot = 1000;
vdev->timeout.jsm = 500;
vdev->timeout.tdr = 2000;
+ vdev->timeout.inference = 60000;
if (ivpu_hw_ip_gen(vdev) == IVPU_HW_IP_37XX)
vdev->timeout.autosuspend = 10;
else
diff --git a/drivers/accel/ivpu/ivpu_hw_ip.c b/drivers/accel/ivpu/ivpu_hw_ip.c
index 823f6a57dc54..2bf9882ab52e 100644
--- a/drivers/accel/ivpu/ivpu_hw_ip.c
+++ b/drivers/accel/ivpu/ivpu_hw_ip.c
@@ -683,6 +683,7 @@ static void pwr_island_delay_set(struct ivpu_device *vdev)
return;
switch (ivpu_device_id(vdev)) {
+ case PCI_DEVICE_ID_WCL:
case PCI_DEVICE_ID_PTL_P:
post = high ? 18 : 0;
post1 = 0;
diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c
index fae8351aa330..060f1fc031d3 100644
--- a/drivers/accel/ivpu/ivpu_job.c
+++ b/drivers/accel/ivpu/ivpu_job.c
@@ -1,6 +1,6 @@
// SPDX-License-Identifier: GPL-2.0-only
/*
- * Copyright (C) 2020-2024 Intel Corporation
+ * Copyright (C) 2020-2025 Intel Corporation
*/
#include <drm/drm_file.h>
@@ -100,6 +100,43 @@ err_free_cmdq:
return NULL;
}
+/**
+ * ivpu_cmdq_get_entry_count - Calculate the number of entries in the command queue.
+ * @cmdq: Pointer to the command queue structure.
+ *
+ * Returns the number of entries that can fit in the command queue memory.
+ */
+static inline u32 ivpu_cmdq_get_entry_count(struct ivpu_cmdq *cmdq)
+{
+ size_t size = ivpu_bo_size(cmdq->mem) - sizeof(struct vpu_job_queue_header);
+
+ return size / sizeof(struct vpu_job_queue_entry);
+}
+
+/**
+ * ivpu_cmdq_get_flags - Get command queue flags based on input flags and test mode.
+ * @vdev: Pointer to the ivpu device structure.
+ * @flags: Input flags to determine the command queue flags.
+ *
+ * Returns the calculated command queue flags, considering both the input flags
+ * and the current test mode settings.
+ */
+static u32 ivpu_cmdq_get_flags(struct ivpu_device *vdev, u32 flags)
+{
+ u32 cmdq_flags = 0;
+
+ if ((flags & DRM_IVPU_CMDQ_FLAG_TURBO) && (ivpu_hw_ip_gen(vdev) >= IVPU_HW_IP_40XX))
+ cmdq_flags |= VPU_JOB_QUEUE_FLAGS_TURBO_MODE;
+
+ /* Test mode can override the TURBO flag coming from the application */
+ if (ivpu_test_mode & IVPU_TEST_MODE_TURBO_ENABLE)
+ cmdq_flags |= VPU_JOB_QUEUE_FLAGS_TURBO_MODE;
+ if (ivpu_test_mode & IVPU_TEST_MODE_TURBO_DISABLE)
+ cmdq_flags &= ~VPU_JOB_QUEUE_FLAGS_TURBO_MODE;
+
+ return cmdq_flags;
+}
+
static void ivpu_cmdq_free(struct ivpu_file_priv *file_priv, struct ivpu_cmdq *cmdq)
{
ivpu_preemption_buffers_free(file_priv->vdev, file_priv, cmdq);
@@ -107,8 +144,7 @@ static void ivpu_cmdq_free(struct ivpu_file_priv *file_priv, struct ivpu_cmdq *c
kfree(cmdq);
}
-static struct ivpu_cmdq *ivpu_cmdq_create(struct ivpu_file_priv *file_priv, u8 priority,
- bool is_legacy)
+static struct ivpu_cmdq *ivpu_cmdq_create(struct ivpu_file_priv *file_priv, u8 priority, u32 flags)
{
struct ivpu_device *vdev = file_priv->vdev;
struct ivpu_cmdq *cmdq = NULL;
@@ -121,10 +157,6 @@ static struct ivpu_cmdq *ivpu_cmdq_create(struct ivpu_file_priv *file_priv, u8 p
ivpu_err(vdev, "Failed to allocate command queue\n");
return NULL;
}
-
- cmdq->priority = priority;
- cmdq->is_legacy = is_legacy;
-
ret = xa_alloc_cyclic(&file_priv->cmdq_xa, &cmdq->id, cmdq, file_priv->cmdq_limit,
&file_priv->cmdq_id_next, GFP_KERNEL);
if (ret < 0) {
@@ -132,7 +164,15 @@ static struct ivpu_cmdq *ivpu_cmdq_create(struct ivpu_file_priv *file_priv, u8 p
goto err_free_cmdq;
}
- ivpu_dbg(vdev, JOB, "Command queue %d created, ctx %d\n", cmdq->id, file_priv->ctx.id);
+ cmdq->entry_count = ivpu_cmdq_get_entry_count(cmdq);
+ cmdq->priority = priority;
+
+ cmdq->jobq = (struct vpu_job_queue *)ivpu_bo_vaddr(cmdq->mem);
+ cmdq->jobq->header.engine_idx = VPU_ENGINE_COMPUTE;
+ cmdq->jobq->header.flags = ivpu_cmdq_get_flags(vdev, flags);
+
+ ivpu_dbg(vdev, JOB, "Command queue %d created, ctx %d, flags 0x%08x\n",
+ cmdq->id, file_priv->ctx.id, cmdq->jobq->header.flags);
return cmdq;
err_free_cmdq:
@@ -188,27 +228,14 @@ static int ivpu_register_db(struct ivpu_file_priv *file_priv, struct ivpu_cmdq *
return ret;
}
-static void ivpu_cmdq_jobq_init(struct ivpu_device *vdev, struct vpu_job_queue *jobq)
+static void ivpu_cmdq_jobq_reset(struct ivpu_device *vdev, struct vpu_job_queue *jobq)
{
- jobq->header.engine_idx = VPU_ENGINE_COMPUTE;
jobq->header.head = 0;
jobq->header.tail = 0;
- if (ivpu_test_mode & IVPU_TEST_MODE_TURBO) {
- ivpu_dbg(vdev, JOB, "Turbo mode enabled");
- jobq->header.flags = VPU_JOB_QUEUE_FLAGS_TURBO_MODE;
- }
-
wmb(); /* Flush WC buffer for jobq->header */
}
-static inline u32 ivpu_cmdq_get_entry_count(struct ivpu_cmdq *cmdq)
-{
- size_t size = ivpu_bo_size(cmdq->mem) - sizeof(struct vpu_job_queue_header);
-
- return size / sizeof(struct vpu_job_queue_entry);
-}
-
static int ivpu_cmdq_register(struct ivpu_file_priv *file_priv, struct ivpu_cmdq *cmdq)
{
struct ivpu_device *vdev = file_priv->vdev;
@@ -219,10 +246,7 @@ static int ivpu_cmdq_register(struct ivpu_file_priv *file_priv, struct ivpu_cmdq
if (cmdq->db_id)
return 0;
- cmdq->entry_count = ivpu_cmdq_get_entry_count(cmdq);
- cmdq->jobq = (struct vpu_job_queue *)ivpu_bo_vaddr(cmdq->mem);
-
- ivpu_cmdq_jobq_init(vdev, cmdq->jobq);
+ ivpu_cmdq_jobq_reset(vdev, cmdq->jobq);
if (vdev->fw->sched_mode == VPU_SCHEDULING_MODE_HW) {
ret = ivpu_hws_cmdq_init(file_priv, cmdq, VPU_ENGINE_COMPUTE, cmdq->priority);
@@ -291,9 +315,10 @@ static struct ivpu_cmdq *ivpu_cmdq_acquire_legacy(struct ivpu_file_priv *file_pr
break;
if (!cmdq) {
- cmdq = ivpu_cmdq_create(file_priv, priority, true);
+ cmdq = ivpu_cmdq_create(file_priv, priority, 0);
if (!cmdq)
return NULL;
+ cmdq->is_legacy = true;
}
return cmdq;
@@ -891,7 +916,7 @@ int ivpu_cmdq_create_ioctl(struct drm_device *dev, void *data, struct drm_file *
mutex_lock(&file_priv->lock);
- cmdq = ivpu_cmdq_create(file_priv, ivpu_job_to_jsm_priority(args->priority), false);
+ cmdq = ivpu_cmdq_create(file_priv, ivpu_job_to_jsm_priority(args->priority), args->flags);
if (cmdq)
args->cmdq_id = cmdq->id;
diff --git a/drivers/accel/ivpu/ivpu_pm.c b/drivers/accel/ivpu/ivpu_pm.c
index ea30db181cd7..eacda1dbe840 100644
--- a/drivers/accel/ivpu/ivpu_pm.c
+++ b/drivers/accel/ivpu/ivpu_pm.c
@@ -33,8 +33,11 @@ static unsigned long ivpu_tdr_timeout_ms;
module_param_named(tdr_timeout_ms, ivpu_tdr_timeout_ms, ulong, 0644);
MODULE_PARM_DESC(tdr_timeout_ms, "Timeout for device hang detection, in milliseconds, 0 - default");
+static unsigned long ivpu_inference_timeout_ms;
+module_param_named(inference_timeout_ms, ivpu_inference_timeout_ms, ulong, 0644);
+MODULE_PARM_DESC(inference_timeout_ms, "Inference maximum duration, in milliseconds, 0 - default");
+
#define PM_RESCHEDULE_LIMIT 5
-#define PM_TDR_HEARTBEAT_LIMIT 30
static void ivpu_pm_prepare_cold_boot(struct ivpu_device *vdev)
{
@@ -191,6 +194,10 @@ static void ivpu_job_timeout_work(struct work_struct *work)
{
struct ivpu_pm_info *pm = container_of(work, struct ivpu_pm_info, job_timeout_work.work);
struct ivpu_device *vdev = pm->vdev;
+ unsigned long timeout_ms = ivpu_tdr_timeout_ms ? ivpu_tdr_timeout_ms : vdev->timeout.tdr;
+ unsigned long inference_timeout_ms = ivpu_inference_timeout_ms ? ivpu_inference_timeout_ms :
+ vdev->timeout.inference;
+ u64 inference_max_retries;
u64 heartbeat;
if (ivpu_jsm_get_heartbeat(vdev, 0, &heartbeat) || heartbeat <= vdev->fw->last_heartbeat) {
@@ -198,8 +205,10 @@ static void ivpu_job_timeout_work(struct work_struct *work)
goto recovery;
}
- if (atomic_fetch_inc(&vdev->job_timeout_counter) > PM_TDR_HEARTBEAT_LIMIT) {
- ivpu_err(vdev, "Job timeout detected, heartbeat limit exceeded\n");
+ inference_max_retries = DIV_ROUND_UP(inference_timeout_ms, timeout_ms);
+ if (atomic_fetch_inc(&vdev->job_timeout_counter) >= inference_max_retries) {
+ ivpu_err(vdev, "Job timeout detected, heartbeat limit (%lld) exceeded\n",
+ inference_max_retries);
goto recovery;
}
generated by cgit (git 2.34.1) at 2025-07-31 12:03:26 +0000