accel/ivpu: Add PM support
authorJacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Tue, 17 Jan 2023 09:27:23 +0000 (10:27 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Thu, 19 Jan 2023 10:12:08 +0000 (11:12 +0100)
  - Implement cold and warm firmware boot flows
  - Add hang recovery support
  - Add runtime power management support

Co-developed-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Signed-off-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20230117092723.60441-8-jacek.lawrynowicz@linux.intel.com
drivers/accel/ivpu/Makefile
drivers/accel/ivpu/ivpu_drv.c
drivers/accel/ivpu/ivpu_drv.h
drivers/accel/ivpu/ivpu_fw.c
drivers/accel/ivpu/ivpu_hw_mtl.c
drivers/accel/ivpu/ivpu_ipc.c
drivers/accel/ivpu/ivpu_job.c
drivers/accel/ivpu/ivpu_mmu.c
drivers/accel/ivpu/ivpu_pm.c [new file with mode: 0644]
drivers/accel/ivpu/ivpu_pm.h [new file with mode: 0644]

index e61ece53a9ae654019d7ba1b8896fd9ce52c7e6d..80f1fb3548ae826d45fa7eb902ac5b62076520d6 100644 (file)
@@ -10,6 +10,7 @@ intel_vpu-y := \
        ivpu_job.o \
        ivpu_jsm_msg.o \
        ivpu_mmu.o \
-       ivpu_mmu_context.o
+       ivpu_mmu_context.o \
+       ivpu_pm.o
 
 obj-$(CONFIG_DRM_ACCEL_IVPU) += intel_vpu.o
\ No newline at end of file
index c893c8ac6bedc63780215261e77044e4d16e837a..2bc2f1b90671a9dbf81419872fa008539ecbb6ba 100644 (file)
@@ -24,6 +24,7 @@
 #include "ivpu_jsm_msg.h"
 #include "ivpu_mmu.h"
 #include "ivpu_mmu_context.h"
+#include "ivpu_pm.h"
 
 #ifndef DRIVER_VERSION_STR
 #define DRIVER_VERSION_STR __stringify(DRM_IVPU_DRIVER_MAJOR) "." \
@@ -462,6 +463,10 @@ static int ivpu_dev_init(struct ivpu_device *vdev)
        if (!vdev->ipc)
                return -ENOMEM;
 
+       vdev->pm = drmm_kzalloc(&vdev->drm, sizeof(*vdev->pm), GFP_KERNEL);
+       if (!vdev->pm)
+               return -ENOMEM;
+
        vdev->hw->ops = &ivpu_hw_mtl_ops;
        vdev->platform = IVPU_PLATFORM_INVALID;
        vdev->context_xa_limit.min = IVPU_GLOBAL_CONTEXT_MMU_SSID + 1;
@@ -521,6 +526,12 @@ static int ivpu_dev_init(struct ivpu_device *vdev)
                goto err_fw_fini;
        }
 
+       ret = ivpu_pm_init(vdev);
+       if (ret) {
+               ivpu_err(vdev, "Failed to initialize PM: %d\n", ret);
+               goto err_ipc_fini;
+       }
+
        ret = ivpu_job_done_thread_init(vdev);
        if (ret) {
                ivpu_err(vdev, "Failed to initialize job done thread: %d\n", ret);
@@ -539,6 +550,8 @@ static int ivpu_dev_init(struct ivpu_device *vdev)
                goto err_job_done_thread_fini;
        }
 
+       ivpu_pm_enable(vdev);
+
        return 0;
 
 err_job_done_thread_fini:
@@ -559,6 +572,7 @@ err_xa_destroy:
 
 static void ivpu_dev_fini(struct ivpu_device *vdev)
 {
+       ivpu_pm_disable(vdev);
        ivpu_shutdown(vdev);
        ivpu_job_done_thread_fini(vdev);
        ivpu_ipc_fini(vdev);
@@ -611,11 +625,25 @@ static void ivpu_remove(struct pci_dev *pdev)
        ivpu_dev_fini(vdev);
 }
 
+static const struct dev_pm_ops ivpu_drv_pci_pm = {
+       SET_SYSTEM_SLEEP_PM_OPS(ivpu_pm_suspend_cb, ivpu_pm_resume_cb)
+       SET_RUNTIME_PM_OPS(ivpu_pm_runtime_suspend_cb, ivpu_pm_runtime_resume_cb, NULL)
+};
+
+static const struct pci_error_handlers ivpu_drv_pci_err = {
+       .reset_prepare = ivpu_pm_reset_prepare_cb,
+       .reset_done = ivpu_pm_reset_done_cb,
+};
+
 static struct pci_driver ivpu_pci_driver = {
        .name = KBUILD_MODNAME,
        .id_table = ivpu_pci_ids,
        .probe = ivpu_probe,
        .remove = ivpu_remove,
+       .driver = {
+               .pm = &ivpu_drv_pci_pm,
+       },
+       .err_handler = &ivpu_drv_pci_err,
 };
 
 module_pci_driver(ivpu_pci_driver);
index 5e63c70c47b3811a549363a3688bc2e71c2427e8..f47b4965db2e33ced85211672116e7c7a0cc3aad 100644 (file)
@@ -76,6 +76,7 @@ struct ivpu_hw_info;
 struct ivpu_mmu_info;
 struct ivpu_fw_info;
 struct ivpu_ipc_info;
+struct ivpu_pm_info;
 
 struct ivpu_device {
        struct drm_device drm;
@@ -89,6 +90,7 @@ struct ivpu_device {
        struct ivpu_mmu_info *mmu;
        struct ivpu_fw_info *fw;
        struct ivpu_ipc_info *ipc;
+       struct ivpu_pm_info *pm;
 
        struct ivpu_mmu_context gctx;
        struct xarray context_xa;
index 4baa0767a10dcbebe15b4900f6582d4d834b55a2..b463c24adb70d0bdfde1e17ee59deae57e4137eb 100644 (file)
@@ -14,6 +14,7 @@
 #include "ivpu_gem.h"
 #include "ivpu_hw.h"
 #include "ivpu_ipc.h"
+#include "ivpu_pm.h"
 
 #define FW_GLOBAL_MEM_START    (2ull * SZ_1G)
 #define FW_GLOBAL_MEM_END      (3ull * SZ_1G)
@@ -361,9 +362,12 @@ void ivpu_fw_boot_params_setup(struct ivpu_device *vdev, struct vpu_boot_params
        /* In case of warm boot we only have to reset the entrypoint addr */
        if (!ivpu_fw_is_cold_boot(vdev)) {
                boot_params->save_restore_ret_address = 0;
+               vdev->pm->is_warmboot = true;
                return;
        }
 
+       vdev->pm->is_warmboot = false;
+
        boot_params->magic = VPU_BOOT_PARAMS_MAGIC;
        boot_params->vpu_id = to_pci_dev(vdev->drm.dev)->bus->number;
        boot_params->frequency = ivpu_hw_reg_pll_freq_get(vdev);
index 55caf5479f5fcca789d8f5e744cd15577bea4251..b59b1f472b4069ba581628d7a99ff3137686fb5c 100644 (file)
@@ -10,6 +10,7 @@
 #include "ivpu_hw.h"
 #include "ivpu_ipc.h"
 #include "ivpu_mmu.h"
+#include "ivpu_pm.h"
 
 #define TILE_FUSE_ENABLE_BOTH       0x0
 #define TILE_FUSE_ENABLE_UPPER      0x1
@@ -921,6 +922,8 @@ static void ivpu_hw_mtl_irq_disable(struct ivpu_device *vdev)
 static void ivpu_hw_mtl_irq_wdt_nce_handler(struct ivpu_device *vdev)
 {
        ivpu_err_ratelimited(vdev, "WDT NCE irq\n");
+
+       ivpu_pm_schedule_recovery(vdev);
 }
 
 static void ivpu_hw_mtl_irq_wdt_mss_handler(struct ivpu_device *vdev)
@@ -928,11 +931,14 @@ static void ivpu_hw_mtl_irq_wdt_mss_handler(struct ivpu_device *vdev)
        ivpu_err_ratelimited(vdev, "WDT MSS irq\n");
 
        ivpu_hw_wdt_disable(vdev);
+       ivpu_pm_schedule_recovery(vdev);
 }
 
 static void ivpu_hw_mtl_irq_noc_firewall_handler(struct ivpu_device *vdev)
 {
        ivpu_err_ratelimited(vdev, "NOC Firewall irq\n");
+
+       ivpu_pm_schedule_recovery(vdev);
 }
 
 /* Handler for IRQs from VPU core (irqV) */
@@ -970,6 +976,7 @@ static u32 ivpu_hw_mtl_irqv_handler(struct ivpu_device *vdev, int irq)
 static u32 ivpu_hw_mtl_irqb_handler(struct ivpu_device *vdev, int irq)
 {
        u32 status = REGB_RD32(MTL_BUTTRESS_INTERRUPT_STAT) & BUTTRESS_IRQ_MASK;
+       bool schedule_recovery = false;
 
        if (status == 0)
                return 0;
@@ -983,6 +990,7 @@ static u32 ivpu_hw_mtl_irqb_handler(struct ivpu_device *vdev, int irq)
        if (REG_TEST_FLD(MTL_BUTTRESS_INTERRUPT_STAT, ATS_ERR, status)) {
                ivpu_err(vdev, "ATS_ERR irq 0x%016llx", REGB_RD64(MTL_BUTTRESS_ATS_ERR_LOG_0));
                REGB_WR32(MTL_BUTTRESS_ATS_ERR_CLEAR, 0x1);
+               schedule_recovery = true;
        }
 
        if (REG_TEST_FLD(MTL_BUTTRESS_INTERRUPT_STAT, UFI_ERR, status)) {
@@ -993,6 +1001,7 @@ static u32 ivpu_hw_mtl_irqb_handler(struct ivpu_device *vdev, int irq)
                         REG_GET_FLD(MTL_BUTTRESS_UFI_ERR_LOG, AXI_ID, ufi_log),
                         REG_GET_FLD(MTL_BUTTRESS_UFI_ERR_LOG, CQ_ID, ufi_log));
                REGB_WR32(MTL_BUTTRESS_UFI_ERR_CLEAR, 0x1);
+               schedule_recovery = true;
        }
 
        /*
@@ -1005,6 +1014,9 @@ static u32 ivpu_hw_mtl_irqb_handler(struct ivpu_device *vdev, int irq)
        /* Re-enable global interrupt */
        REGB_WR32(MTL_BUTTRESS_GLOBAL_INT_MASK, 0x0);
 
+       if (schedule_recovery)
+               ivpu_pm_schedule_recovery(vdev);
+
        return status;
 }
 
index c756476bfc5b17b4c8b5bd736992bd720ec83188..3adcfa80fc0e556a9e48c08e1a8869474d5a18a0 100644 (file)
@@ -14,6 +14,7 @@
 #include "ivpu_hw_reg_io.h"
 #include "ivpu_ipc.h"
 #include "ivpu_jsm_msg.h"
+#include "ivpu_pm.h"
 
 #define IPC_MAX_RX_MSG 128
 #define IS_KTHREAD()   (get_current()->flags & PF_KTHREAD)
@@ -294,18 +295,27 @@ int ivpu_ipc_send_receive(struct ivpu_device *vdev, struct vpu_jsm_msg *req,
 {
        struct vpu_jsm_msg hb_req = { .type = VPU_JSM_MSG_QUERY_ENGINE_HB };
        struct vpu_jsm_msg hb_resp;
-       int ret;
+       int ret, hb_ret;
+
+       ret = ivpu_rpm_get(vdev);
+       if (ret < 0)
+               return ret;
 
        ret = ivpu_ipc_send_receive_internal(vdev, req, expected_resp_type, resp,
                                             channel, timeout_ms);
        if (ret != -ETIMEDOUT)
-               return ret;
+               goto rpm_put;
 
-       ret = ivpu_ipc_send_receive_internal(vdev, &hb_req, VPU_JSM_MSG_QUERY_ENGINE_HB_DONE,
-                                            &hb_resp, VPU_IPC_CHAN_ASYNC_CMD, vdev->timeout.jsm);
-       if (ret == -ETIMEDOUT)
+       hb_ret = ivpu_ipc_send_receive_internal(vdev, &hb_req, VPU_JSM_MSG_QUERY_ENGINE_HB_DONE,
+                                               &hb_resp, VPU_IPC_CHAN_ASYNC_CMD,
+                                               vdev->timeout.jsm);
+       if (hb_ret == -ETIMEDOUT) {
                ivpu_hw_diagnose_failure(vdev);
+               ivpu_pm_schedule_recovery(vdev);
+       }
 
+rpm_put:
+       ivpu_rpm_put(vdev);
        return ret;
 }
 
index 23cdb3b3f1f380f082838cc291bd30d1109f4810..3276bd9107b4ac8cc8c3c08637988b8b1b116f4d 100644 (file)
@@ -17,6 +17,7 @@
 #include "ivpu_ipc.h"
 #include "ivpu_job.h"
 #include "ivpu_jsm_msg.h"
+#include "ivpu_pm.h"
 
 #define CMD_BUF_IDX         0
 #define JOB_ID_JOB_MASK             GENMASK(7, 0)
@@ -270,6 +271,9 @@ static void job_release(struct kref *ref)
 
        ivpu_dbg(vdev, KREF, "Job released: id %u\n", job->job_id);
        kfree(job);
+
+       /* Allow the VPU to get suspended, must be called after ivpu_file_priv_put() */
+       ivpu_rpm_put(vdev);
 }
 
 static void job_put(struct ivpu_job *job)
@@ -286,11 +290,16 @@ ivpu_create_job(struct ivpu_file_priv *file_priv, u32 engine_idx, u32 bo_count)
        struct ivpu_device *vdev = file_priv->vdev;
        struct ivpu_job *job;
        size_t buf_size;
+       int ret;
+
+       ret = ivpu_rpm_get(vdev);
+       if (ret < 0)
+               return NULL;
 
        buf_size = sizeof(*job) + bo_count * sizeof(struct ivpu_bo *);
        job = kzalloc(buf_size, GFP_KERNEL);
        if (!job)
-               return NULL;
+               goto err_rpm_put;
 
        kref_init(&job->ref);
 
@@ -311,6 +320,8 @@ ivpu_create_job(struct ivpu_file_priv *file_priv, u32 engine_idx, u32 bo_count)
 
 err_free_job:
        kfree(job);
+err_rpm_put:
+       ivpu_rpm_put(vdev);
        return NULL;
 }
 
@@ -565,6 +576,7 @@ static int ivpu_job_done_thread(void *arg)
                        if (jobs_submitted && !xa_empty(&vdev->submitted_jobs_xa)) {
                                ivpu_err(vdev, "TDR detected, timeout %d ms", timeout);
                                ivpu_hw_diagnose_failure(vdev);
+                               ivpu_pm_schedule_recovery(vdev);
                        }
                }
        }
index ff5ef6da1fd3b8b142704160d4eb75a39d249bd7..694e978aba6637bbc979e06343a32a1f36dda30b 100644 (file)
@@ -11,6 +11,7 @@
 #include "ivpu_hw_reg_io.h"
 #include "ivpu_mmu.h"
 #include "ivpu_mmu_context.h"
+#include "ivpu_pm.h"
 
 #define IVPU_MMU_IDR0_REF              0x080f3e0f
 #define IVPU_MMU_IDR0_REF_SIMICS       0x080f3e1f
@@ -814,6 +815,7 @@ static u32 *ivpu_mmu_get_event(struct ivpu_device *vdev)
 
 void ivpu_mmu_irq_evtq_handler(struct ivpu_device *vdev)
 {
+       bool schedule_recovery = false;
        u32 *event;
        u32 ssid;
 
@@ -823,9 +825,14 @@ void ivpu_mmu_irq_evtq_handler(struct ivpu_device *vdev)
                ivpu_mmu_dump_event(vdev, event);
 
                ssid = FIELD_GET(IVPU_MMU_EVT_SSID_MASK, event[0]);
-               if (ssid != IVPU_GLOBAL_CONTEXT_MMU_SSID)
+               if (ssid == IVPU_GLOBAL_CONTEXT_MMU_SSID)
+                       schedule_recovery = true;
+               else
                        ivpu_mmu_user_context_mark_invalid(vdev, ssid);
        }
+
+       if (schedule_recovery)
+               ivpu_pm_schedule_recovery(vdev);
 }
 
 void ivpu_mmu_irq_gerr_handler(struct ivpu_device *vdev)
diff --git a/drivers/accel/ivpu/ivpu_pm.c b/drivers/accel/ivpu/ivpu_pm.c
new file mode 100644 (file)
index 0000000..553bcbd
--- /dev/null
@@ -0,0 +1,329 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020-2023 Intel Corporation
+ */
+
+#include <linux/highmem.h>
+#include <linux/moduleparam.h>
+#include <linux/pci.h>
+#include <linux/pm_runtime.h>
+#include <linux/reboot.h>
+
+#include "vpu_boot_api.h"
+#include "ivpu_drv.h"
+#include "ivpu_hw.h"
+#include "ivpu_fw.h"
+#include "ivpu_ipc.h"
+#include "ivpu_job.h"
+#include "ivpu_mmu.h"
+#include "ivpu_pm.h"
+
+static bool ivpu_disable_recovery;
+module_param_named_unsafe(disable_recovery, ivpu_disable_recovery, bool, 0644);
+MODULE_PARM_DESC(disable_recovery, "Disables recovery when VPU hang is detected");
+
+#define PM_RESCHEDULE_LIMIT     5
+
+static void ivpu_pm_prepare_cold_boot(struct ivpu_device *vdev)
+{
+       struct ivpu_fw_info *fw = vdev->fw;
+
+       ivpu_cmdq_reset_all_contexts(vdev);
+       ivpu_ipc_reset(vdev);
+       ivpu_fw_load(vdev);
+       fw->entry_point = fw->cold_boot_entry_point;
+}
+
+static void ivpu_pm_prepare_warm_boot(struct ivpu_device *vdev)
+{
+       struct ivpu_fw_info *fw = vdev->fw;
+       struct vpu_boot_params *bp = fw->mem->kvaddr;
+
+       if (!bp->save_restore_ret_address) {
+               ivpu_pm_prepare_cold_boot(vdev);
+               return;
+       }
+
+       ivpu_dbg(vdev, FW_BOOT, "Save/restore entry point %llx", bp->save_restore_ret_address);
+       fw->entry_point = bp->save_restore_ret_address;
+}
+
+static int ivpu_suspend(struct ivpu_device *vdev)
+{
+       int ret;
+
+       ret = ivpu_shutdown(vdev);
+       if (ret) {
+               ivpu_err(vdev, "Failed to shutdown VPU: %d\n", ret);
+               return ret;
+       }
+
+       return ret;
+}
+
+static int ivpu_resume(struct ivpu_device *vdev)
+{
+       int ret;
+
+retry:
+       ret = ivpu_hw_power_up(vdev);
+       if (ret) {
+               ivpu_err(vdev, "Failed to power up HW: %d\n", ret);
+               return ret;
+       }
+
+       ret = ivpu_mmu_enable(vdev);
+       if (ret) {
+               ivpu_err(vdev, "Failed to resume MMU: %d\n", ret);
+               ivpu_hw_power_down(vdev);
+               return ret;
+       }
+
+       ret = ivpu_boot(vdev);
+       if (ret) {
+               ivpu_mmu_disable(vdev);
+               ivpu_hw_power_down(vdev);
+               if (!ivpu_fw_is_cold_boot(vdev)) {
+                       ivpu_warn(vdev, "Failed to resume the FW: %d. Retrying cold boot..\n", ret);
+                       ivpu_pm_prepare_cold_boot(vdev);
+                       goto retry;
+               } else {
+                       ivpu_err(vdev, "Failed to resume the FW: %d\n", ret);
+               }
+       }
+
+       return ret;
+}
+
+static void ivpu_pm_recovery_work(struct work_struct *work)
+{
+       struct ivpu_pm_info *pm = container_of(work, struct ivpu_pm_info, recovery_work);
+       struct ivpu_device *vdev =  pm->vdev;
+       char *evt[2] = {"IVPU_PM_EVENT=IVPU_RECOVER", NULL};
+       int ret;
+
+       ret = pci_reset_function(to_pci_dev(vdev->drm.dev));
+       if (ret)
+               ivpu_err(vdev, "Failed to reset VPU: %d\n", ret);
+
+       kobject_uevent_env(&vdev->drm.dev->kobj, KOBJ_CHANGE, evt);
+}
+
+void ivpu_pm_schedule_recovery(struct ivpu_device *vdev)
+{
+       struct ivpu_pm_info *pm = vdev->pm;
+
+       if (ivpu_disable_recovery) {
+               ivpu_err(vdev, "Recovery not available when disable_recovery param is set\n");
+               return;
+       }
+
+       if (ivpu_is_fpga(vdev)) {
+               ivpu_err(vdev, "Recovery not available on FPGA\n");
+               return;
+       }
+
+       /* Schedule recovery if it's not in progress */
+       if (atomic_cmpxchg(&pm->in_reset, 0, 1) == 0) {
+               ivpu_hw_irq_disable(vdev);
+               queue_work(system_long_wq, &pm->recovery_work);
+       }
+}
+
+int ivpu_pm_suspend_cb(struct device *dev)
+{
+       struct drm_device *drm = dev_get_drvdata(dev);
+       struct ivpu_device *vdev = to_ivpu_device(drm);
+       int ret;
+
+       ivpu_dbg(vdev, PM, "Suspend..\n");
+
+       ret = ivpu_suspend(vdev);
+       if (ret && vdev->pm->suspend_reschedule_counter) {
+               ivpu_dbg(vdev, PM, "Failed to enter idle, rescheduling suspend, retries left %d\n",
+                        vdev->pm->suspend_reschedule_counter);
+               pm_schedule_suspend(dev, vdev->timeout.reschedule_suspend);
+               vdev->pm->suspend_reschedule_counter--;
+               return -EBUSY;
+       } else if (!vdev->pm->suspend_reschedule_counter) {
+               ivpu_warn(vdev, "Failed to enter idle, force suspend\n");
+               ivpu_pm_prepare_cold_boot(vdev);
+       } else {
+               ivpu_pm_prepare_warm_boot(vdev);
+       }
+
+       vdev->pm->suspend_reschedule_counter = PM_RESCHEDULE_LIMIT;
+
+       pci_save_state(to_pci_dev(dev));
+       pci_set_power_state(to_pci_dev(dev), PCI_D3hot);
+
+       ivpu_dbg(vdev, PM, "Suspend done.\n");
+
+       return ret;
+}
+
+int ivpu_pm_resume_cb(struct device *dev)
+{
+       struct drm_device *drm = dev_get_drvdata(dev);
+       struct ivpu_device *vdev = to_ivpu_device(drm);
+       int ret;
+
+       ivpu_dbg(vdev, PM, "Resume..\n");
+
+       pci_set_power_state(to_pci_dev(dev), PCI_D0);
+       pci_restore_state(to_pci_dev(dev));
+
+       ret = ivpu_resume(vdev);
+       if (ret)
+               ivpu_err(vdev, "Failed to resume: %d\n", ret);
+
+       ivpu_dbg(vdev, PM, "Resume done.\n");
+
+       return ret;
+}
+
+int ivpu_pm_runtime_suspend_cb(struct device *dev)
+{
+       struct drm_device *drm = dev_get_drvdata(dev);
+       struct ivpu_device *vdev = to_ivpu_device(drm);
+       int ret;
+
+       ivpu_dbg(vdev, PM, "Runtime suspend..\n");
+
+       if (!ivpu_hw_is_idle(vdev) && vdev->pm->suspend_reschedule_counter) {
+               ivpu_dbg(vdev, PM, "Failed to enter idle, rescheduling suspend, retries left %d\n",
+                        vdev->pm->suspend_reschedule_counter);
+               pm_schedule_suspend(dev, vdev->timeout.reschedule_suspend);
+               vdev->pm->suspend_reschedule_counter--;
+               return -EAGAIN;
+       }
+
+       ret = ivpu_suspend(vdev);
+       if (ret)
+               ivpu_err(vdev, "Failed to set suspend VPU: %d\n", ret);
+
+       if (!vdev->pm->suspend_reschedule_counter) {
+               ivpu_warn(vdev, "VPU failed to enter idle, force suspended.\n");
+               ivpu_pm_prepare_cold_boot(vdev);
+       } else {
+               ivpu_pm_prepare_warm_boot(vdev);
+       }
+
+       vdev->pm->suspend_reschedule_counter = PM_RESCHEDULE_LIMIT;
+
+       ivpu_dbg(vdev, PM, "Runtime suspend done.\n");
+
+       return 0;
+}
+
+int ivpu_pm_runtime_resume_cb(struct device *dev)
+{
+       struct drm_device *drm = dev_get_drvdata(dev);
+       struct ivpu_device *vdev = to_ivpu_device(drm);
+       int ret;
+
+       ivpu_dbg(vdev, PM, "Runtime resume..\n");
+
+       ret = ivpu_resume(vdev);
+       if (ret)
+               ivpu_err(vdev, "Failed to set RESUME state: %d\n", ret);
+
+       ivpu_dbg(vdev, PM, "Runtime resume done.\n");
+
+       return ret;
+}
+
+int ivpu_rpm_get(struct ivpu_device *vdev)
+{
+       int ret;
+
+       ivpu_dbg(vdev, RPM, "rpm_get count %d\n", atomic_read(&vdev->drm.dev->power.usage_count));
+
+       ret = pm_runtime_resume_and_get(vdev->drm.dev);
+       if (!drm_WARN_ON(&vdev->drm, ret < 0))
+               vdev->pm->suspend_reschedule_counter = PM_RESCHEDULE_LIMIT;
+
+       return ret;
+}
+
+void ivpu_rpm_put(struct ivpu_device *vdev)
+{
+       ivpu_dbg(vdev, RPM, "rpm_put count %d\n", atomic_read(&vdev->drm.dev->power.usage_count));
+
+       pm_runtime_mark_last_busy(vdev->drm.dev);
+       pm_runtime_put_autosuspend(vdev->drm.dev);
+}
+
+void ivpu_pm_reset_prepare_cb(struct pci_dev *pdev)
+{
+       struct ivpu_device *vdev = pci_get_drvdata(pdev);
+
+       pm_runtime_get_sync(vdev->drm.dev);
+
+       ivpu_dbg(vdev, PM, "Pre-reset..\n");
+       atomic_set(&vdev->pm->in_reset, 1);
+       ivpu_shutdown(vdev);
+       ivpu_pm_prepare_cold_boot(vdev);
+       ivpu_jobs_abort_all(vdev);
+       ivpu_dbg(vdev, PM, "Pre-reset done.\n");
+}
+
+void ivpu_pm_reset_done_cb(struct pci_dev *pdev)
+{
+       struct ivpu_device *vdev = pci_get_drvdata(pdev);
+       int ret;
+
+       ivpu_dbg(vdev, PM, "Post-reset..\n");
+       ret = ivpu_resume(vdev);
+       if (ret)
+               ivpu_err(vdev, "Failed to set RESUME state: %d\n", ret);
+       atomic_set(&vdev->pm->in_reset, 0);
+       ivpu_dbg(vdev, PM, "Post-reset done.\n");
+
+       pm_runtime_put_autosuspend(vdev->drm.dev);
+}
+
+int ivpu_pm_init(struct ivpu_device *vdev)
+{
+       struct device *dev = vdev->drm.dev;
+       struct ivpu_pm_info *pm = vdev->pm;
+
+       pm->vdev = vdev;
+       pm->suspend_reschedule_counter = PM_RESCHEDULE_LIMIT;
+
+       atomic_set(&pm->in_reset, 0);
+       INIT_WORK(&pm->recovery_work, ivpu_pm_recovery_work);
+
+       pm_runtime_use_autosuspend(dev);
+
+       if (ivpu_disable_recovery)
+               pm_runtime_set_autosuspend_delay(dev, -1);
+       else if (ivpu_is_silicon(vdev))
+               pm_runtime_set_autosuspend_delay(dev, 100);
+       else
+               pm_runtime_set_autosuspend_delay(dev, 60000);
+
+       return 0;
+}
+
+void ivpu_pm_enable(struct ivpu_device *vdev)
+{
+       struct device *dev = vdev->drm.dev;
+
+       pm_runtime_set_active(dev);
+       pm_runtime_allow(dev);
+       pm_runtime_mark_last_busy(dev);
+       pm_runtime_put_autosuspend(dev);
+
+       ivpu_dbg(vdev, RPM, "Enable RPM count %d\n", atomic_read(&dev->power.usage_count));
+}
+
+void ivpu_pm_disable(struct ivpu_device *vdev)
+{
+       struct device *dev = vdev->drm.dev;
+
+       ivpu_dbg(vdev, RPM, "Disable RPM count %d\n", atomic_read(&dev->power.usage_count));
+
+       pm_runtime_get_noresume(vdev->drm.dev);
+       pm_runtime_forbid(vdev->drm.dev);
+}
diff --git a/drivers/accel/ivpu/ivpu_pm.h b/drivers/accel/ivpu/ivpu_pm.h
new file mode 100644 (file)
index 0000000..dc1b375
--- /dev/null
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2020-2023 Intel Corporation
+ */
+
+#ifndef __IVPU_PM_H__
+#define __IVPU_PM_H__
+
+#include <linux/types.h>
+
+struct ivpu_device;
+
+struct ivpu_pm_info {
+       struct ivpu_device *vdev;
+       struct work_struct recovery_work;
+       atomic_t in_reset;
+       bool is_warmboot;
+       u32 suspend_reschedule_counter;
+};
+
+int ivpu_pm_init(struct ivpu_device *vdev);
+void ivpu_pm_enable(struct ivpu_device *vdev);
+void ivpu_pm_disable(struct ivpu_device *vdev);
+
+int ivpu_pm_suspend_cb(struct device *dev);
+int ivpu_pm_resume_cb(struct device *dev);
+int ivpu_pm_runtime_suspend_cb(struct device *dev);
+int ivpu_pm_runtime_resume_cb(struct device *dev);
+
+void ivpu_pm_reset_prepare_cb(struct pci_dev *pdev);
+void ivpu_pm_reset_done_cb(struct pci_dev *pdev);
+
+int __must_check ivpu_rpm_get(struct ivpu_device *vdev);
+void ivpu_rpm_put(struct ivpu_device *vdev);
+
+void ivpu_pm_schedule_recovery(struct ivpu_device *vdev);
+
+#endif /* __IVPU_PM_H__ */