| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <44b04aeafc7ae414f2a4b51f46f6524ec6babd5e.1597833138.git.mchehab+huawei@kernel.org>
Date: Wed, 19 Aug 2020 13:45:32 +0200
From: Mauro Carvalho Chehab <mchehab+huawei@...nel.org>
To: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc: linuxarm@...wei.com, mauro.chehab@...wei.com,
Liwei Cai <cailiwei@...ilicon.com>,
Mauro Carvalho Chehab <mchehab+huawei@...nel.org>,
John Stultz <john.stultz@...aro.org>,
Manivannan Sadhasivam <mani@...nel.org>,
Daniel Vetter <daniel@...ll.ch>,
dri-devel <dri-devel@...ts.freedesktop.org>,
Xiubin Zhang <zhangxiubin1@...wei.com>,
Wanchun Zheng <zhengwanchun@...ilicon.com>,
Chen Feng <puck.chen@...ilicon.com>,
devel@...verdev.osuosl.org, linux-kernel@...r.kernel.org
Subject: [PATCH 04/49] staging: hikey9xx/gpu: resolve the performance issue by interrupt mechanism
From: Liwei Cai <cailiwei@...ilicon.com>
There is an error at wait for vactive end flags, waiting
vactive flag in 1ms maybe too rough, but it's not good to
control the waiting grain size, there is no way to get the
waiting unit, so the interrupt mechanism is the best way to
solve this problem.
Each frame would report hardware interrupt, implement the interrupt
service to get vactive end interrupt, and fb_post return to tell
gpu render next framebuffer.
Signed-off-by: Wanchun Zheng <zhengwanchun@...ilicon.com>
Signed-off-by: Liwei Cai <cailiwei@...ilicon.com>
Signed-off-by: John Stultz <john.stultz@...aro.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@...nel.org>
---
drivers/staging/hikey9xx/gpu/kirin_dpe_reg.h | 4 +-
.../hikey9xx/gpu/kirin_drm_dpe_utils.c | 3 +-
drivers/staging/hikey9xx/gpu/kirin_drm_dss.c | 14 ++---
.../hikey9xx/gpu/kirin_drm_overlay_utils.c | 56 ++++---------------
4 files changed, 20 insertions(+), 57 deletions(-)
diff --git a/drivers/staging/hikey9xx/gpu/kirin_dpe_reg.h b/drivers/staging/hikey9xx/gpu/kirin_dpe_reg.h
index 9fad9ef942bd..adaa71f6dcd5 100644
--- a/drivers/staging/hikey9xx/gpu/kirin_dpe_reg.h
+++ b/drivers/staging/hikey9xx/gpu/kirin_dpe_reg.h
@@ -2948,8 +2948,8 @@ struct dss_hw_ctx {
bool power_on;
int irq;
- wait_queue_head_t vactive0_start_wq;
- u32 vactive0_start_flag;
+ wait_queue_head_t vactive0_end_wq;
+ u32 vactive0_end_flag;
ktime_t vsync_timestamp;
ktime_t vsync_timestamp_prev;
diff --git a/drivers/staging/hikey9xx/gpu/kirin_drm_dpe_utils.c b/drivers/staging/hikey9xx/gpu/kirin_drm_dpe_utils.c
index 2d6809b72b42..2a13bbd772b7 100644
--- a/drivers/staging/hikey9xx/gpu/kirin_drm_dpe_utils.c
+++ b/drivers/staging/hikey9xx/gpu/kirin_drm_dpe_utils.c
@@ -554,8 +554,7 @@ void dpe_interrupt_unmask(struct dss_crtc *acrtc)
outp32(dss_base + GLB_CPU_PDP_INT_MSK, unmask);
unmask = ~0;
- unmask &= ~(BIT_VSYNC | BIT_VACTIVE0_START
- | BIT_VACTIVE0_END | BIT_FRM_END | BIT_LDI_UNFLOW);
+ unmask &= ~(BIT_VSYNC | BIT_VACTIVE0_END | BIT_LDI_UNFLOW);
outp32(dss_base + DSS_LDI0_OFFSET + LDI_CPU_ITF_INT_MSK, unmask);
}
diff --git a/drivers/staging/hikey9xx/gpu/kirin_drm_dss.c b/drivers/staging/hikey9xx/gpu/kirin_drm_dss.c
index 62ac1a0648cc..64d0b1979bf5 100644
--- a/drivers/staging/hikey9xx/gpu/kirin_drm_dss.c
+++ b/drivers/staging/hikey9xx/gpu/kirin_drm_dss.c
@@ -167,8 +167,8 @@ static int dss_power_up(struct dss_crtc *acrtc)
dss_inner_clk_common_enable(acrtc);
dpe_interrupt_mask(acrtc);
dpe_interrupt_clear(acrtc);
- //dpe_irq_enable(acrtc);
- //dpe_interrupt_unmask(acrtc);
+ dpe_irq_enable(acrtc);
+ dpe_interrupt_unmask(acrtc);
ctx->power_on = true;
return 0;
@@ -237,9 +237,9 @@ static irqreturn_t dss_irq_handler(int irq, void *data)
isr_s2 &= ~(inp32(dss_base + DSS_LDI0_OFFSET + LDI_CPU_ITF_INT_MSK));
isr_s2_dpp &= ~(inp32(dss_base + DSS_DPP_OFFSET + DPP_INT_MSK));
- if (isr_s2 & BIT_VACTIVE0_START) {
- ctx->vactive0_start_flag++;
- wake_up_interruptible_all(&ctx->vactive0_start_wq);
+ if (isr_s2 & BIT_VACTIVE0_END) {
+ ctx->vactive0_end_flag++;
+ wake_up_interruptible_all(&ctx->vactive0_end_wq);
}
if (isr_s2 & BIT_VSYNC) {
@@ -637,8 +637,8 @@ static int dss_drm_init(struct drm_device *dev)
ctx->screen_size = 0;
ctx->smem_start = 0;
- ctx->vactive0_start_flag = 0;
- init_waitqueue_head(&ctx->vactive0_start_wq);
+ ctx->vactive0_end_flag = 0;
+ init_waitqueue_head(&ctx->vactive0_end_wq);
/*
* plane init
diff --git a/drivers/staging/hikey9xx/gpu/kirin_drm_overlay_utils.c b/drivers/staging/hikey9xx/gpu/kirin_drm_overlay_utils.c
index 917e1a7d7bdf..28778b15512a 100644
--- a/drivers/staging/hikey9xx/gpu/kirin_drm_overlay_utils.c
+++ b/drivers/staging/hikey9xx/gpu/kirin_drm_overlay_utils.c
@@ -30,8 +30,6 @@
#define DSS_CHN_MAX_DEFINE (DSS_COPYBIT_MAX)
-#define TIME_OUT (16)
-
static int mid_array[DSS_CHN_MAX_DEFINE] = {0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x2, 0x1, 0x3, 0x0};
/*
@@ -1065,49 +1063,17 @@ void hisi_dss_unflow_handler(struct dss_hw_ctx *ctx, bool unmask)
outp32(dss_base + DSS_LDI0_OFFSET + LDI_CPU_ITF_INT_MSK, tmp);
}
-void hisi_dss_wait_for_complete(struct dss_hw_ctx *ctx, bool need_clear)
-{
- void __iomem *dss_base;
- u32 tmp = 0;
- u32 isr_s2 = 0;
-
- if (!ctx) {
- DRM_ERROR("ctx is NULL!\n");
- return;
- }
-
- dss_base = ctx->base;
-
- do {
- isr_s2 = inp32(dss_base + DSS_LDI0_OFFSET + LDI_CPU_ITF_INTS);
- if (isr_s2 & BIT_VACTIVE0_END) {
- DRM_DEBUG("hisi_dss_wait_for_complete exit! temp = %d\n", tmp);
- if (need_clear)
- outp32(dss_base + DSS_LDI0_OFFSET + LDI_CPU_ITF_INTS, BIT_VACTIVE0_END);
- break;
- } else {
- msleep(1);
- tmp++;
- }
- } while (tmp < TIME_OUT);
-
- if (tmp == TIME_OUT) {
- isr_s2 = inp32(dss_base + DSS_LDI0_OFFSET + LDI_CPU_ITF_INTS);
- DRM_INFO("wait vactive0_end timeout: isr_s2 = 0x%x\n", isr_s2);
- }
-}
-#if 0
-static int hisi_vactive0_start_config(struct dss_hw_ctx *ctx)
+static int hisi_dss_wait_for_complete(struct dss_hw_ctx *ctx)
{
int ret = 0;
u32 times = 0;
- u32 prev_vactive0_start = 0;
+ u32 prev_vactive0_end = 0;
- prev_vactive0_start = ctx->vactive0_start_flag;
+ prev_vactive0_end = ctx->vactive0_end_flag;
REDO:
- ret = wait_event_interruptible_timeout(ctx->vactive0_start_wq,
- (prev_vactive0_start != ctx->vactive0_start_flag),
+ ret = wait_event_interruptible_timeout(ctx->vactive0_end_wq,
+ (prev_vactive0_end != ctx->vactive0_end_flag),
msecs_to_jiffies(300));
if (ret == -ERESTARTSYS) {
if (times < 50) {
@@ -1118,7 +1084,7 @@ static int hisi_vactive0_start_config(struct dss_hw_ctx *ctx)
}
if (ret <= 0) {
- DRM_ERROR("wait_for vactive0_start_flag timeout! ret=%d.\n", ret);
+ DRM_ERROR("wait_for vactive0_end_flag timeout! ret=%d.\n", ret);
ret = -ETIMEDOUT;
} else {
@@ -1127,7 +1093,6 @@ static int hisi_vactive0_start_config(struct dss_hw_ctx *ctx)
return ret;
}
-#endif
void hisi_fb_pan_display(struct drm_plane *plane)
{
@@ -1195,8 +1160,6 @@ void hisi_fb_pan_display(struct drm_plane *plane)
vbp = mode->vtotal - mode->vsync_end;
vsw = mode->vsync_end - mode->vsync_start;
- hisi_dss_wait_for_complete(ctx, true);
-
hisi_dss_mctl_mutex_lock(ctx);
hisi_dss_aif_ch_config(ctx, chn_idx);
hisi_dss_mif_config(ctx, chn_idx, mmu_enable);
@@ -1209,9 +1172,10 @@ void hisi_fb_pan_display(struct drm_plane *plane)
hisi_dss_mctl_ov_config(ctx, chn_idx);
hisi_dss_mctl_sys_config(ctx, chn_idx);
hisi_dss_mctl_mutex_unlock(ctx);
+ hisi_dss_unflow_handler(ctx, true);
enable_ldi(acrtc);
- hisi_dss_wait_for_complete(ctx, false);
+ hisi_dss_wait_for_complete(ctx);
}
void hisi_dss_online_play(struct drm_plane *plane, drm_dss_layer_t *layer)
@@ -1257,7 +1221,6 @@ void hisi_dss_online_play(struct drm_plane *plane, drm_dss_layer_t *layer)
vfp = mode->vsync_start - mode->vdisplay;
vbp = mode->vtotal - mode->vsync_end;
vsw = mode->vsync_end - mode->vsync_start;
- hisi_dss_wait_for_complete(ctx, true);
hisi_dss_mctl_mutex_lock(ctx);
hisi_dss_aif_ch_config(ctx, chn_idx);
@@ -1271,7 +1234,8 @@ void hisi_dss_online_play(struct drm_plane *plane, drm_dss_layer_t *layer)
hisi_dss_mctl_ov_config(ctx, chn_idx);
hisi_dss_mctl_sys_config(ctx, chn_idx);
hisi_dss_mctl_mutex_unlock(ctx);
+ hisi_dss_unflow_handler(ctx, true);
enable_ldi(acrtc);
- hisi_dss_wait_for_complete(ctx, false);
+ hisi_dss_wait_for_complete(ctx);
}
--
2.26.2
Powered by blists - more mailing lists