media: venus: Rework recovery mechanism
authorStanimir Varbanov <stanimir.varbanov@linaro.org>
Thu, 30 Jul 2020 11:46:31 +0000 (13:46 +0200)
committerMauro Carvalho Chehab <mchehab+huawei@kernel.org>
Tue, 1 Sep 2020 12:13:29 +0000 (14:13 +0200)
After power domains and clock restructuring the recovery for
sdm845 and v4 did not work properly. Fix that by reworking the
recovery function and the sequence.

Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Reviewed-by: Fritz Koenig <frkoenig@chromium.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
drivers/media/platform/qcom/venus/core.c
drivers/media/platform/qcom/venus/hfi_venus.c

index 5fd70f8..6a5c18c 100644 (file)
@@ -6,6 +6,7 @@
 #include <linux/init.h>
 #include <linux/interconnect.h>
 #include <linux/ioctl.h>
+#include <linux/delay.h>
 #include <linux/list.h>
 #include <linux/module.h>
 #include <linux/of_device.h>
@@ -40,13 +41,7 @@ static void venus_event_notify(struct venus_core *core, u32 event)
        mutex_unlock(&core->lock);
 
        disable_irq_nosync(core->irq);
-
-       /*
-        * Delay recovery to ensure venus has completed any pending cache
-        * operations. Without this sleep, we see device reset when firmware is
-        * unloaded after a system error.
-        */
-       schedule_delayed_work(&core->work, msecs_to_jiffies(100));
+       schedule_delayed_work(&core->work, msecs_to_jiffies(10));
 }
 
 static const struct hfi_core_ops venus_core_ops = {
@@ -59,23 +54,30 @@ static void venus_sys_error_handler(struct work_struct *work)
                        container_of(work, struct venus_core, work.work);
        int ret = 0;
 
-       dev_warn(core->dev, "system error has occurred, starting recovery!\n");
-
        pm_runtime_get_sync(core->dev);
 
        hfi_core_deinit(core, true);
-       hfi_destroy(core);
+
+       dev_warn(core->dev, "system error has occurred, starting recovery!\n");
+
        mutex_lock(&core->lock);
+
+       while (pm_runtime_active(core->dev_dec) || pm_runtime_active(core->dev_enc))
+               msleep(10);
+
        venus_shutdown(core);
 
        pm_runtime_put_sync(core->dev);
 
+       while (core->pmdomains[0] && pm_runtime_active(core->pmdomains[0]))
+               usleep_range(1000, 1500);
+
+       hfi_destroy(core);
        ret |= hfi_create(core, &venus_core_ops);
 
        pm_runtime_get_sync(core->dev);
 
        ret |= venus_boot(core);
-
        ret |= hfi_core_resume(core, true);
 
        enable_irq(core->irq);
index 0d88550..3392fd1 100644 (file)
@@ -986,13 +986,6 @@ static void venus_process_msg_sys_error(struct venus_hfi_device *hdev,
 
        venus_set_state(hdev, VENUS_STATE_DEINIT);
 
-       /*
-        * Once SYS_ERROR received from HW, it is safe to halt the AXI.
-        * With SYS_ERROR, Venus FW may have crashed and HW might be
-        * active and causing unnecessary transactions. Hence it is
-        * safe to stop all AXI transactions from venus subsystem.
-        */
-       venus_halt_axi(hdev);
        venus_sfr_print(hdev);
 }
 
@@ -1009,10 +1002,6 @@ static irqreturn_t venus_isr_thread(struct venus_core *core)
        res = hdev->core->res;
        pkt = hdev->pkt_buf;
 
-       if (hdev->irq_status & WRAPPER_INTR_STATUS_A2HWD_MASK) {
-               venus_sfr_print(hdev);
-               hfi_process_watchdog_timeout(core);
-       }
 
        while (!venus_iface_msgq_read(hdev, pkt)) {
                msg_ret = hfi_process_msg_packet(core, pkt);