2020-10-29 08:46:22

by Irui Wang (王瑞)

Subject: [PATCH 1/2] media: mtk-vpu: VPU should be in idle state before system is suspended

VPU should be in idle state before system is suspended
or it will work abnormally like VPU program counter not
in a correct address or VPU reset

Signed-off-by: Irui Wang <[email protected]>
---
drivers/media/platform/mtk-vpu/mtk_vpu.c | 68 ++++++++++++++++++++++++
1 file changed, 68 insertions(+)

diff --git a/drivers/media/platform/mtk-vpu/mtk_vpu.c b/drivers/media/platform/mtk-vpu/mtk_vpu.c
index 36cb9b6131f7..86ab808ba877 100644
--- a/drivers/media/platform/mtk-vpu/mtk_vpu.c
+++ b/drivers/media/platform/mtk-vpu/mtk_vpu.c
@@ -27,6 +27,7 @@

#define INIT_TIMEOUT_MS 2000U
#define IPI_TIMEOUT_MS 2000U
+#define VPU_IDLE_TIMEOUT_MS 1000U
#define VPU_FW_VER_LEN 16

/* maximum program/data TCM (Tightly-Coupled Memory) size */
@@ -57,11 +58,15 @@
#define VPU_DMEM_EXT0_ADDR 0x0014
#define VPU_DMEM_EXT1_ADDR 0x0018
#define HOST_TO_VPU 0x0024
+#define VPU_IDLE_REG 0x002C
+#define VPU_INT_STATUS 0x0034
#define VPU_PC_REG 0x0060
#define VPU_WDT_REG 0x0084

/* vpu inter-processor communication interrupt */
#define VPU_IPC_INT BIT(8)
+/* vpu idle state */
+#define VPU_IDLE_STATE BIT(23)

/**
* enum vpu_fw_type - VPU firmware type
@@ -945,11 +950,74 @@ static int mtk_vpu_remove(struct platform_device *pdev)
return 0;
}

+static int mtk_vpu_suspend(struct device *dev)
+{
+ struct mtk_vpu *vpu = dev_get_drvdata(dev);
+ unsigned long timeout;
+ int ret;
+
+ ret = vpu_clock_enable(vpu);
+ if (ret) {
+ dev_err(dev, "failed to enable vpu clock\n");
+ return ret;
+ }
+
+ mutex_lock(&vpu->vpu_mutex);
+ /* disable vpu timer interrupt */
+ vpu_cfg_writel(vpu, vpu_cfg_readl(vpu, VPU_INT_STATUS) | VPU_IDLE_STATE,
+ VPU_INT_STATUS);
+ /* check if vpu is idle for system suspend */
+ timeout = jiffies + msecs_to_jiffies(VPU_IDLE_TIMEOUT_MS);
+ do {
+ if (time_after(jiffies, timeout)) {
+ dev_err(dev, "vpu idle timeout\n");
+ mutex_unlock(&vpu->vpu_mutex);
+ vpu_clock_disable(vpu);
+ return -EIO;
+ }
+ } while (!vpu_cfg_readl(vpu, VPU_IDLE_REG));
+
+ mutex_unlock(&vpu->vpu_mutex);
+ vpu_clock_disable(vpu);
+ clk_unprepare(vpu->clk);
+
+ return 0;
+}
+
+static int mtk_vpu_resume(struct device *dev)
+{
+ struct mtk_vpu *vpu = dev_get_drvdata(dev);
+ int ret;
+
+ clk_prepare(vpu->clk);
+ ret = vpu_clock_enable(vpu);
+ if (ret) {
+ dev_err(dev, "failed to enable vpu clock\n");
+ return ret;
+ }
+
+ mutex_lock(&vpu->vpu_mutex);
+ /* enable vpu timer interrupt */
+ vpu_cfg_writel(vpu,
+ vpu_cfg_readl(vpu, VPU_INT_STATUS) & ~(VPU_IDLE_STATE),
+ VPU_INT_STATUS);
+ mutex_unlock(&vpu->vpu_mutex);
+ vpu_clock_disable(vpu);
+
+ return 0;
+}
+
+static const struct dev_pm_ops mtk_vpu_pm = {
+ .suspend = mtk_vpu_suspend,
+ .resume = mtk_vpu_resume,
+};
+
static struct platform_driver mtk_vpu_driver = {
.probe = mtk_vpu_probe,
.remove = mtk_vpu_remove,
.driver = {
.name = "mtk_vpu",
+ .pm = &mtk_vpu_pm,
.of_match_table = mtk_vpu_match,
},
};
--
2.25.1

2020-11-25 02:51:16

by Alexandre Courbot

Subject: Re: [PATCH 1/2] media: mtk-vpu: VPU should be in idle state before system is suspended

On Thu, Oct 29, 2020 at 10:17 AM Irui Wang <[email protected]> wrote:
>
> VPU should be in idle state before system is suspended
> or it will work abnormally like VPU program counter not
> in a correct address or VPU reset
>
> Signed-off-by: Irui Wang <[email protected]>

FWIW,

Reviewed-by: Alexandre Courbot <[email protected]>