The code in imx_ldb_encoder_mode_set crashes trying to access the crtc->state->state drm_atomic_state pointer if that was previously cleared by drm_atomic_helper_swap_state. Instead of trying to walk all connectors during encoder mode_set to determine the LVDS bus format from panel or external bridge connector display info, store it in imx_crtc_state during encoder atomic_check, where it is already known, and just retrieve it from imx_crtc_state during encoder mode_set.
Fixes: 49f98bc4d44a4 ("drm/imx: store internal bus configuration in crtc state") Signed-off-by: Philipp Zabel p.zabel@pengutronix.de --- drivers/gpu/drm/imx/imx-drm.h | 1 + drivers/gpu/drm/imx/imx-ldb.c | 22 +++++----------------- 2 files changed, 6 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/imx/imx-drm.h b/drivers/gpu/drm/imx/imx-drm.h index bdaa381..5e2f68a 100644 --- a/drivers/gpu/drm/imx/imx-drm.h +++ b/drivers/gpu/drm/imx/imx-drm.h @@ -19,6 +19,7 @@ struct imx_crtc_state { u32 bus_flags; int di_hsync_pin; int di_vsync_pin; + u32 ldb_bus_format; };
static inline struct imx_crtc_state *to_imx_crtc_state(struct drm_crtc_state *s) diff --git a/drivers/gpu/drm/imx/imx-ldb.c b/drivers/gpu/drm/imx/imx-ldb.c index dc2b420..871dff9 100644 --- a/drivers/gpu/drm/imx/imx-ldb.c +++ b/drivers/gpu/drm/imx/imx-ldb.c @@ -262,7 +262,10 @@ static void imx_ldb_encoder_mode_set(struct drm_encoder *encoder, unsigned long serial_clk; unsigned long di_clk = mode->clock * 1000; int mux = drm_of_encoder_active_port_id(imx_ldb_ch->child, encoder); - u32 bus_format = imx_ldb_ch->bus_format; + struct drm_crtc_state *crtc_state = encoder->crtc->state; + struct imx_crtc_state *imx_crtc_state = to_imx_crtc_state(crtc_state); + u32 bus_format = imx_ldb_ch->bus_format ?: + imx_crtc_state->ldb_bus_format;
if (mode->clock > 170000) { dev_warn(ldb->dev, @@ -297,22 +300,6 @@ static void imx_ldb_encoder_mode_set(struct drm_encoder *encoder, ldb->ldb_ctrl &= ~LDB_DI1_VS_POL_ACT_LOW; }
- if (!bus_format) { - struct drm_connector_state *conn_state; - struct drm_connector *connector; - int i; - - for_each_connector_in_state(encoder->crtc->state->state, - connector, conn_state, i) { - struct drm_display_info *di = &connector->display_info; - - if (conn_state->crtc == encoder->crtc && - di->num_bus_formats) { - bus_format = di->bus_formats[0]; - break; - } - } - } imx_ldb_ch_set_bus_format(imx_ldb_ch, bus_format); }
@@ -404,6 +391,7 @@ static int imx_ldb_encoder_atomic_check(struct drm_encoder *encoder,
imx_crtc_state->di_hsync_pin = 2; imx_crtc_state->di_vsync_pin = 3; + imx_crtc_state->ldb_bus_format = bus_format;
return 0; }
Hi Philipp,
This patch's headline doesn't exactly reflect what the patch actually does - retrieve lvds bus format from imx_crtc_state during encoder mode_set.
On Thu, Jul 21, 2016 at 9:25 PM, Philipp Zabel p.zabel@pengutronix.de wrote:
The code in imx_ldb_encoder_mode_set crashes trying to access the crtc->state->state drm_atomic_state pointer if that was previously cleared by drm_atomic_helper_swap_state.
Nit: Providing a crash log might be good.
Instead of trying to walk all connectors during encoder mode_set to determine the LVDS bus format from panel or external bridge connector display info, store it in imx_crtc_state during encoder atomic_check, where it is already known, and just retrieve it from imx_crtc_state during encoder mode_set.
Fixes: 49f98bc4d44a4 ("drm/imx: store internal bus configuration in crtc state") Signed-off-by: Philipp Zabel p.zabel@pengutronix.de
drivers/gpu/drm/imx/imx-drm.h | 1 + drivers/gpu/drm/imx/imx-ldb.c | 22 +++++----------------- 2 files changed, 6 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/imx/imx-drm.h b/drivers/gpu/drm/imx/imx-drm.h index bdaa381..5e2f68a 100644 --- a/drivers/gpu/drm/imx/imx-drm.h +++ b/drivers/gpu/drm/imx/imx-drm.h @@ -19,6 +19,7 @@ struct imx_crtc_state { u32 bus_flags; int di_hsync_pin; int di_vsync_pin;
u32 ldb_bus_format;
Pretty much a hack here, I think. Comparing to the other members of imx_crtc_state, ldb_bus_format is less likely to be a member, since crtc just doesn't like to know the LVDS bus format...
};
static inline struct imx_crtc_state *to_imx_crtc_state(struct drm_crtc_state *s) diff --git a/drivers/gpu/drm/imx/imx-ldb.c b/drivers/gpu/drm/imx/imx-ldb.c index dc2b420..871dff9 100644 --- a/drivers/gpu/drm/imx/imx-ldb.c +++ b/drivers/gpu/drm/imx/imx-ldb.c @@ -262,7 +262,10 @@ static void imx_ldb_encoder_mode_set(struct drm_encoder *encoder, unsigned long serial_clk; unsigned long di_clk = mode->clock * 1000; int mux = drm_of_encoder_active_port_id(imx_ldb_ch->child, encoder);
u32 bus_format = imx_ldb_ch->bus_format;
struct drm_crtc_state *crtc_state = encoder->crtc->state;
struct imx_crtc_state *imx_crtc_state = to_imx_crtc_state(crtc_state);
u32 bus_format = imx_ldb_ch->bus_format ?:
imx_crtc_state->ldb_bus_format;
Instead of introducing imx_crtc_state->ldb_bus_format to get the lvds bus format, you may get the connector via crtc_state->connector_mask(verify connector->encoder to be the encoder we are talking about). connector_mask is maintained well by the atomic core I assume.
Regards, Liu Ying
if (mode->clock > 170000) { dev_warn(ldb->dev,
@@ -297,22 +300,6 @@ static void imx_ldb_encoder_mode_set(struct drm_encoder *encoder, ldb->ldb_ctrl &= ~LDB_DI1_VS_POL_ACT_LOW; }
if (!bus_format) {
struct drm_connector_state *conn_state;
struct drm_connector *connector;
int i;
for_each_connector_in_state(encoder->crtc->state->state,
connector, conn_state, i) {
struct drm_display_info *di = &connector->display_info;
if (conn_state->crtc == encoder->crtc &&
di->num_bus_formats) {
bus_format = di->bus_formats[0];
break;
}
}
} imx_ldb_ch_set_bus_format(imx_ldb_ch, bus_format);
}
@@ -404,6 +391,7 @@ static int imx_ldb_encoder_atomic_check(struct drm_encoder *encoder,
imx_crtc_state->di_hsync_pin = 2; imx_crtc_state->di_vsync_pin = 3;
imx_crtc_state->ldb_bus_format = bus_format; return 0;
}
2.8.1
Hi Liu,
thank you for your comments.
Am Freitag, den 22.07.2016, 12:01 +0800 schrieb Ying Liu:
Hi Philipp,
This patch's headline doesn't exactly reflect what the patch actually does - retrieve lvds bus format from imx_crtc_state during encoder mode_set.
Yes. I initially stored ldb_bus_format in imx_ldb_ch, then fixed that, and forgot to change the subject.
On Thu, Jul 21, 2016 at 9:25 PM, Philipp Zabel p.zabel@pengutronix.de wrote:
The code in imx_ldb_encoder_mode_set crashes trying to access the crtc->state->state drm_atomic_state pointer if that was previously cleared by drm_atomic_helper_swap_state.
Nit: Providing a crash log might be good.
Will do.
Instead of trying to walk all connectors during encoder mode_set to determine the LVDS bus format from panel or external bridge connector display info, store it in imx_crtc_state during encoder atomic_check, where it is already known, and just retrieve it from imx_crtc_state during encoder mode_set.
Fixes: 49f98bc4d44a4 ("drm/imx: store internal bus configuration in crtc state") Signed-off-by: Philipp Zabel p.zabel@pengutronix.de
drivers/gpu/drm/imx/imx-drm.h | 1 + drivers/gpu/drm/imx/imx-ldb.c | 22 +++++----------------- 2 files changed, 6 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/imx/imx-drm.h b/drivers/gpu/drm/imx/imx-drm.h index bdaa381..5e2f68a 100644 --- a/drivers/gpu/drm/imx/imx-drm.h +++ b/drivers/gpu/drm/imx/imx-drm.h @@ -19,6 +19,7 @@ struct imx_crtc_state { u32 bus_flags; int di_hsync_pin; int di_vsync_pin;
u32 ldb_bus_format;
Pretty much a hack here, I think. Comparing to the other members of imx_crtc_state, ldb_bus_format is less likely to be a member, since crtc just doesn't like to know the LVDS bus format...
I suppose I should keep the connector walk then. I'll send a new version.
[...]
Instead of introducing imx_crtc_state->ldb_bus_format to get the lvds bus format, you may get the connector via crtc_state->connector_mask(verify connector->encoder to be the encoder we are talking about). connector_mask is maintained well by the atomic core I assume.
I could also use drm_for_each_connector and just keep the (connector->state->crtc == encoder->crtc) check in the loop.
regards Philipp
Hi Philipp,
On Fri, Jul 22, 2016 at 4:59 PM, Philipp Zabel p.zabel@pengutronix.de wrote:
Hi Liu,
thank you for your comments.
Am Freitag, den 22.07.2016, 12:01 +0800 schrieb Ying Liu:
Hi Philipp,
This patch's headline doesn't exactly reflect what the patch actually does - retrieve lvds bus format from imx_crtc_state during encoder mode_set.
Yes. I initially stored ldb_bus_format in imx_ldb_ch, then fixed that, and forgot to change the subject.
On Thu, Jul 21, 2016 at 9:25 PM, Philipp Zabel p.zabel@pengutronix.de wrote:
The code in imx_ldb_encoder_mode_set crashes trying to access the crtc->state->state drm_atomic_state pointer if that was previously cleared by drm_atomic_helper_swap_state.
Nit: Providing a crash log might be good.
Will do.
Instead of trying to walk all connectors during encoder mode_set to determine the LVDS bus format from panel or external bridge connector display info, store it in imx_crtc_state during encoder atomic_check, where it is already known, and just retrieve it from imx_crtc_state during encoder mode_set.
Fixes: 49f98bc4d44a4 ("drm/imx: store internal bus configuration in crtc state") Signed-off-by: Philipp Zabel p.zabel@pengutronix.de
drivers/gpu/drm/imx/imx-drm.h | 1 + drivers/gpu/drm/imx/imx-ldb.c | 22 +++++----------------- 2 files changed, 6 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/imx/imx-drm.h b/drivers/gpu/drm/imx/imx-drm.h index bdaa381..5e2f68a 100644 --- a/drivers/gpu/drm/imx/imx-drm.h +++ b/drivers/gpu/drm/imx/imx-drm.h @@ -19,6 +19,7 @@ struct imx_crtc_state { u32 bus_flags; int di_hsync_pin; int di_vsync_pin;
u32 ldb_bus_format;
Pretty much a hack here, I think. Comparing to the other members of imx_crtc_state, ldb_bus_format is less likely to be a member, since crtc just doesn't like to know the LVDS bus format...
I suppose I should keep the connector walk then. I'll send a new version.
[...]
Instead of introducing imx_crtc_state->ldb_bus_format to get the lvds bus format, you may get the connector via crtc_state->connector_mask(verify connector->encoder to be the encoder we are talking about). connector_mask is maintained well by the atomic core I assume.
I could also use drm_for_each_connector and just keep the (connector->state->crtc == encoder->crtc) check in the loop.
It looks (connector->encoder == encoder) is simpler.
Regards, Liu Ying
regards Philipp
dri-devel@lists.freedesktop.org