| Message ID | 20220302012931.4107196-2-bjorn.andersson@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2,1/2] dt-bindings: display: msm: Add optional resets | expand |
On Wed, 2 Mar 2022 at 04:27, Bjorn Andersson <bjorn.andersson@linaro.org> wrote: > > It's typical for the bootloader to bring up the display for showing a > boot splash or efi framebuffer. But in some cases the kernel driver ends > up only partially configuring (in particular) the DPU, which might > result in e.g. that two different data paths attempts to push data to > the interface - with resulting graphical artifacts. > > Naturally the end goal would be to inherit the bootloader's > configuration and provide the user with a glitch free handover from the > boot configuration to a running DPU. > > But as implementing seamless transition from the bootloader > configuration to the running OS will be a considerable effort, start by > simply resetting the entire MDSS to its power-on state, to avoid the > partial configuration. > > Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org> > --- > > Changes since v1: > - Rather than trying to deconfigure individual pieces of the DPU, reset the > entire block. > > drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c | 18 ++++++++++++++++++ > drivers/gpu/drm/msm/msm_drv.c | 4 ++++ > drivers/gpu/drm/msm/msm_kms.h | 1 + > 3 files changed, 23 insertions(+) > > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c > index b10ca505f9ac..419eaaefe606 100644 > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c > @@ -7,6 +7,7 @@ > #include <linux/irqchip.h> > #include <linux/irqdesc.h> > #include <linux/irqchip/chained_irq.h> > +#include <linux/reset.h> > #include "dpu_kms.h" > > #define to_dpu_mdss(x) container_of(x, struct dpu_mdss, base) > @@ -31,6 +32,7 @@ struct dpu_mdss { > void __iomem *mmio; > struct clk_bulk_data *clocks; > size_t num_clocks; > + struct reset_control *reset; > struct dpu_irq_controller irq_controller; > }; > > @@ -197,10 +199,18 @@ static void dpu_mdss_destroy(struct msm_mdss *mdss) > dpu_mdss->mmio = NULL; > } > > +static int dpu_mdss_reset(struct msm_mdss *mdss) > +{ > + struct dpu_mdss *dpu_mdss = to_dpu_mdss(mdss); > + > + return reset_control_reset(dpu_mdss->reset); > +} > + > static const struct msm_mdss_funcs mdss_funcs = { > .enable = dpu_mdss_enable, > .disable = dpu_mdss_disable, > .destroy = dpu_mdss_destroy, > + .reset = dpu_mdss_reset, > }; > > int dpu_mdss_init(struct platform_device *pdev) > @@ -227,6 +237,13 @@ int dpu_mdss_init(struct platform_device *pdev) > } > dpu_mdss->num_clocks = ret; > > + dpu_mdss->reset = devm_reset_control_get_optional_exclusive(&pdev->dev, NULL); > + if (IS_ERR(dpu_mdss->reset)) { > + ret = PTR_ERR(dpu_mdss->reset); > + DPU_ERROR("failed to acquire mdss reset, ret=%d", ret); > + goto reset_parse_err; > + } > + > dpu_mdss->base.dev = &pdev->dev; > dpu_mdss->base.funcs = &mdss_funcs; > > @@ -252,6 +269,7 @@ int dpu_mdss_init(struct platform_device *pdev) > irq_error: > _dpu_mdss_irq_domain_fini(dpu_mdss); > irq_domain_error: > +reset_parse_err: > clk_parse_err: > if (dpu_mdss->mmio) > devm_iounmap(&pdev->dev, dpu_mdss->mmio); > diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c > index 129fa841ac22..7595f83da3f1 100644 > --- a/drivers/gpu/drm/msm/msm_drv.c > +++ b/drivers/gpu/drm/msm/msm_drv.c > @@ -388,6 +388,10 @@ static int msm_drm_init(struct device *dev, const struct drm_driver *drv) > if (ret) > return ret; > > + /* Issue a reset of the entire MDSS */ > + if (priv->mdss && priv->mdss->funcs->reset) > + priv->mdss->funcs->reset(priv->mdss); > + I think this is incorrect. In this way reset happens after all subdevice are probed. They might have programmed some state of the corresponding block. The clocks are already registered, so the clock framework will be out of sync. I think the reset should happen before calling of_platform_populate(), so the device state is consistent with the driver. Also see the https://git.linaro.org/people/dmitry.baryshkov/kernel.git/log/?h=dpu-mdss-rework, which reworks the mdss driver and mdss probing. > /* Bind all our sub-components: */ > ret = component_bind_all(dev, ddev); > if (ret) > diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h > index 2a4f0526cb98..716a34fca1cd 100644 > --- a/drivers/gpu/drm/msm/msm_kms.h > +++ b/drivers/gpu/drm/msm/msm_kms.h > @@ -205,6 +205,7 @@ struct msm_mdss_funcs { > int (*enable)(struct msm_mdss *mdss); > int (*disable)(struct msm_mdss *mdss); > void (*destroy)(struct msm_mdss *mdss); > + int (*reset)(struct msm_mdss *mdss); > }; > > struct msm_mdss { > -- > 2.33.1 >
On Tue 01 Mar 17:47 PST 2022, Dmitry Baryshkov wrote: > On Wed, 2 Mar 2022 at 04:27, Bjorn Andersson <bjorn.andersson@linaro.org> wrote: > > > > It's typical for the bootloader to bring up the display for showing a > > boot splash or efi framebuffer. But in some cases the kernel driver ends > > up only partially configuring (in particular) the DPU, which might > > result in e.g. that two different data paths attempts to push data to > > the interface - with resulting graphical artifacts. > > > > Naturally the end goal would be to inherit the bootloader's > > configuration and provide the user with a glitch free handover from the > > boot configuration to a running DPU. > > > > But as implementing seamless transition from the bootloader > > configuration to the running OS will be a considerable effort, start by > > simply resetting the entire MDSS to its power-on state, to avoid the > > partial configuration. > > > > Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org> > > --- > > > > Changes since v1: > > - Rather than trying to deconfigure individual pieces of the DPU, reset the > > entire block. > > > > drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c | 18 ++++++++++++++++++ > > drivers/gpu/drm/msm/msm_drv.c | 4 ++++ > > drivers/gpu/drm/msm/msm_kms.h | 1 + > > 3 files changed, 23 insertions(+) > > > > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c > > index b10ca505f9ac..419eaaefe606 100644 > > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c > > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c > > @@ -7,6 +7,7 @@ > > #include <linux/irqchip.h> > > #include <linux/irqdesc.h> > > #include <linux/irqchip/chained_irq.h> > > +#include <linux/reset.h> > > #include "dpu_kms.h" > > > > #define to_dpu_mdss(x) container_of(x, struct dpu_mdss, base) > > @@ -31,6 +32,7 @@ struct dpu_mdss { > > void __iomem *mmio; > > struct clk_bulk_data *clocks; > > size_t num_clocks; > > + struct reset_control *reset; > > struct dpu_irq_controller irq_controller; > > }; > > > > @@ -197,10 +199,18 @@ static void dpu_mdss_destroy(struct msm_mdss *mdss) > > dpu_mdss->mmio = NULL; > > } > > > > +static int dpu_mdss_reset(struct msm_mdss *mdss) > > +{ > > + struct dpu_mdss *dpu_mdss = to_dpu_mdss(mdss); > > + > > + return reset_control_reset(dpu_mdss->reset); > > +} > > + > > static const struct msm_mdss_funcs mdss_funcs = { > > .enable = dpu_mdss_enable, > > .disable = dpu_mdss_disable, > > .destroy = dpu_mdss_destroy, > > + .reset = dpu_mdss_reset, > > }; > > > > int dpu_mdss_init(struct platform_device *pdev) > > @@ -227,6 +237,13 @@ int dpu_mdss_init(struct platform_device *pdev) > > } > > dpu_mdss->num_clocks = ret; > > > > + dpu_mdss->reset = devm_reset_control_get_optional_exclusive(&pdev->dev, NULL); > > + if (IS_ERR(dpu_mdss->reset)) { > > + ret = PTR_ERR(dpu_mdss->reset); > > + DPU_ERROR("failed to acquire mdss reset, ret=%d", ret); > > + goto reset_parse_err; > > + } > > + > > dpu_mdss->base.dev = &pdev->dev; > > dpu_mdss->base.funcs = &mdss_funcs; > > > > @@ -252,6 +269,7 @@ int dpu_mdss_init(struct platform_device *pdev) > > irq_error: > > _dpu_mdss_irq_domain_fini(dpu_mdss); > > irq_domain_error: > > +reset_parse_err: > > clk_parse_err: > > if (dpu_mdss->mmio) > > devm_iounmap(&pdev->dev, dpu_mdss->mmio); > > diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c > > index 129fa841ac22..7595f83da3f1 100644 > > --- a/drivers/gpu/drm/msm/msm_drv.c > > +++ b/drivers/gpu/drm/msm/msm_drv.c > > @@ -388,6 +388,10 @@ static int msm_drm_init(struct device *dev, const struct drm_driver *drv) > > if (ret) > > return ret; > > > > + /* Issue a reset of the entire MDSS */ > > + if (priv->mdss && priv->mdss->funcs->reset) > > + priv->mdss->funcs->reset(priv->mdss); > > + > > I think this is incorrect. In this way reset happens after all > subdevice are probed. They might have programmed some state of the > corresponding block. The clocks are already registered, so the clock > framework will be out of sync. I went back and forth through the drivers and I believe at least the idea is that we probe all the drivers, which will acquire some resources. Then in bind() we actually start to access the hardware (and acquire more resources, for some reason). > I think the reset should happen before calling of_platform_populate(), > so the device state is consistent with the driver. > Perhaps I'm misunderstanding the component framework, but I was under the impression that if any of the subcomponents fails to probe because of lacking resources, this could be printed on the efifb before we reset the hardware. Making errors slightly more user friendly. I.e. in the timeframe between of_platform_populate() and component_bind_all() below... But if you believe I'm incorrect on the assumptions about the hardware not being accessed before this point, I can move the reset before of_platform_populate() - this is the last piece needed to have functional eDP on sc8180x. > Also see the https://git.linaro.org/people/dmitry.baryshkov/kernel.git/log/?h=dpu-mdss-rework, > which reworks the mdss driver and mdss probing. > There seems to be some room for reducing duplication between the two drivers, so this seems reasonable. Regards, Bjorn > > /* Bind all our sub-components: */ > > ret = component_bind_all(dev, ddev); > > if (ret) > > diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h > > index 2a4f0526cb98..716a34fca1cd 100644 > > --- a/drivers/gpu/drm/msm/msm_kms.h > > +++ b/drivers/gpu/drm/msm/msm_kms.h > > @@ -205,6 +205,7 @@ struct msm_mdss_funcs { > > int (*enable)(struct msm_mdss *mdss); > > int (*disable)(struct msm_mdss *mdss); > > void (*destroy)(struct msm_mdss *mdss); > > + int (*reset)(struct msm_mdss *mdss); > > }; > > > > struct msm_mdss { > > -- > > 2.33.1 > > > > > -- > With best wishes > Dmitry
On 02/03/2022 05:44, Bjorn Andersson wrote: > On Tue 01 Mar 17:47 PST 2022, Dmitry Baryshkov wrote: > >> On Wed, 2 Mar 2022 at 04:27, Bjorn Andersson <bjorn.andersson@linaro.org> wrote: >>> >>> It's typical for the bootloader to bring up the display for showing a >>> boot splash or efi framebuffer. But in some cases the kernel driver ends >>> up only partially configuring (in particular) the DPU, which might >>> result in e.g. that two different data paths attempts to push data to >>> the interface - with resulting graphical artifacts. >>> >>> Naturally the end goal would be to inherit the bootloader's >>> configuration and provide the user with a glitch free handover from the >>> boot configuration to a running DPU. >>> >>> But as implementing seamless transition from the bootloader >>> configuration to the running OS will be a considerable effort, start by >>> simply resetting the entire MDSS to its power-on state, to avoid the >>> partial configuration. >>> >>> Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org> >>> --- >>> >>> Changes since v1: >>> - Rather than trying to deconfigure individual pieces of the DPU, reset the >>> entire block. >>> >>> drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c | 18 ++++++++++++++++++ >>> drivers/gpu/drm/msm/msm_drv.c | 4 ++++ >>> drivers/gpu/drm/msm/msm_kms.h | 1 + >>> 3 files changed, 23 insertions(+) >>> >>> diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c >>> index b10ca505f9ac..419eaaefe606 100644 >>> --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c >>> +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c >>> @@ -7,6 +7,7 @@ >>> #include <linux/irqchip.h> >>> #include <linux/irqdesc.h> >>> #include <linux/irqchip/chained_irq.h> >>> +#include <linux/reset.h> >>> #include "dpu_kms.h" >>> >>> #define to_dpu_mdss(x) container_of(x, struct dpu_mdss, base) >>> @@ -31,6 +32,7 @@ struct dpu_mdss { >>> void __iomem *mmio; >>> struct clk_bulk_data *clocks; >>> size_t num_clocks; >>> + struct reset_control *reset; >>> struct dpu_irq_controller irq_controller; >>> }; >>> >>> @@ -197,10 +199,18 @@ static void dpu_mdss_destroy(struct msm_mdss *mdss) >>> dpu_mdss->mmio = NULL; >>> } >>> >>> +static int dpu_mdss_reset(struct msm_mdss *mdss) >>> +{ >>> + struct dpu_mdss *dpu_mdss = to_dpu_mdss(mdss); >>> + >>> + return reset_control_reset(dpu_mdss->reset); >>> +} >>> + >>> static const struct msm_mdss_funcs mdss_funcs = { >>> .enable = dpu_mdss_enable, >>> .disable = dpu_mdss_disable, >>> .destroy = dpu_mdss_destroy, >>> + .reset = dpu_mdss_reset, >>> }; >>> >>> int dpu_mdss_init(struct platform_device *pdev) >>> @@ -227,6 +237,13 @@ int dpu_mdss_init(struct platform_device *pdev) >>> } >>> dpu_mdss->num_clocks = ret; >>> >>> + dpu_mdss->reset = devm_reset_control_get_optional_exclusive(&pdev->dev, NULL); >>> + if (IS_ERR(dpu_mdss->reset)) { >>> + ret = PTR_ERR(dpu_mdss->reset); >>> + DPU_ERROR("failed to acquire mdss reset, ret=%d", ret); >>> + goto reset_parse_err; >>> + } >>> + >>> dpu_mdss->base.dev = &pdev->dev; >>> dpu_mdss->base.funcs = &mdss_funcs; >>> >>> @@ -252,6 +269,7 @@ int dpu_mdss_init(struct platform_device *pdev) >>> irq_error: >>> _dpu_mdss_irq_domain_fini(dpu_mdss); >>> irq_domain_error: >>> +reset_parse_err: >>> clk_parse_err: >>> if (dpu_mdss->mmio) >>> devm_iounmap(&pdev->dev, dpu_mdss->mmio); >>> diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c >>> index 129fa841ac22..7595f83da3f1 100644 >>> --- a/drivers/gpu/drm/msm/msm_drv.c >>> +++ b/drivers/gpu/drm/msm/msm_drv.c >>> @@ -388,6 +388,10 @@ static int msm_drm_init(struct device *dev, const struct drm_driver *drv) >>> if (ret) >>> return ret; >>> >>> + /* Issue a reset of the entire MDSS */ >>> + if (priv->mdss && priv->mdss->funcs->reset) >>> + priv->mdss->funcs->reset(priv->mdss); >>> + >> >> I think this is incorrect. In this way reset happens after all >> subdevice are probed. They might have programmed some state of the >> corresponding block. The clocks are already registered, so the clock >> framework will be out of sync. > > I went back and forth through the drivers and I believe at least the > idea is that we probe all the drivers, which will acquire some > resources. > > Then in bind() we actually start to access the hardware (and acquire > more resources, for some reason). DSI clock init happens in the dsi_clk_init(), called from msm_dsi_host_inti(), dsi_init(), dsi_dev_probe(). But it's not the major problem. All the PHYs do not use component framework, be it the DSI PHY or HDMI PHY. So at the time you call MDSS reset, the PHYs are already initialized, the clocks are registered in the clock subsystem, etc. Performing a reset will put the hardware out of sync with the Linux kernel. So, if you'd like to perform a reset in the msm_drm_init(), we'd have to modify DSI to perform some parts of init later. And to move PHYs to also use the component framework. And I'm not sure if that won't break the dispcc, the way it gets the DSI clocks. > >> I think the reset should happen before calling of_platform_populate(), >> so the device state is consistent with the driver. >> > > Perhaps I'm misunderstanding the component framework, but I was under > the impression that if any of the subcomponents fails to probe because > of lacking resources, this could be printed on the efifb before we reset > the hardware. Making errors slightly more user friendly. Yes, I understand why did you place the call in msm_drm_init(). > > I.e. in the timeframe between of_platform_populate() and > component_bind_all() below... > > > But if you believe I'm incorrect on the assumptions about the hardware > not being accessed before this point, I can move the reset before > of_platform_populate() - this is the last piece needed to have > functional eDP on sc8180x. > >> Also see the https://git.linaro.org/people/dmitry.baryshkov/kernel.git/log/?h=dpu-mdss-rework, >> which reworks the mdss driver and mdss probing. >> > > There seems to be some room for reducing duplication between the two > drivers, so this seems reasonable. > > Regards, > Bjorn > >>> /* Bind all our sub-components: */ >>> ret = component_bind_all(dev, ddev); >>> if (ret) >>> diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h >>> index 2a4f0526cb98..716a34fca1cd 100644 >>> --- a/drivers/gpu/drm/msm/msm_kms.h >>> +++ b/drivers/gpu/drm/msm/msm_kms.h >>> @@ -205,6 +205,7 @@ struct msm_mdss_funcs { >>> int (*enable)(struct msm_mdss *mdss); >>> int (*disable)(struct msm_mdss *mdss); >>> void (*destroy)(struct msm_mdss *mdss); >>> + int (*reset)(struct msm_mdss *mdss); >>> }; >>> >>> struct msm_mdss { >>> -- >>> 2.33.1 >>> >> >> >> -- >> With best wishes >> Dmitry
diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c index b10ca505f9ac..419eaaefe606 100644 --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c @@ -7,6 +7,7 @@ #include <linux/irqchip.h> #include <linux/irqdesc.h> #include <linux/irqchip/chained_irq.h> +#include <linux/reset.h> #include "dpu_kms.h" #define to_dpu_mdss(x) container_of(x, struct dpu_mdss, base) @@ -31,6 +32,7 @@ struct dpu_mdss { void __iomem *mmio; struct clk_bulk_data *clocks; size_t num_clocks; + struct reset_control *reset; struct dpu_irq_controller irq_controller; }; @@ -197,10 +199,18 @@ static void dpu_mdss_destroy(struct msm_mdss *mdss) dpu_mdss->mmio = NULL; } +static int dpu_mdss_reset(struct msm_mdss *mdss) +{ + struct dpu_mdss *dpu_mdss = to_dpu_mdss(mdss); + + return reset_control_reset(dpu_mdss->reset); +} + static const struct msm_mdss_funcs mdss_funcs = { .enable = dpu_mdss_enable, .disable = dpu_mdss_disable, .destroy = dpu_mdss_destroy, + .reset = dpu_mdss_reset, }; int dpu_mdss_init(struct platform_device *pdev) @@ -227,6 +237,13 @@ int dpu_mdss_init(struct platform_device *pdev) } dpu_mdss->num_clocks = ret; + dpu_mdss->reset = devm_reset_control_get_optional_exclusive(&pdev->dev, NULL); + if (IS_ERR(dpu_mdss->reset)) { + ret = PTR_ERR(dpu_mdss->reset); + DPU_ERROR("failed to acquire mdss reset, ret=%d", ret); + goto reset_parse_err; + } + dpu_mdss->base.dev = &pdev->dev; dpu_mdss->base.funcs = &mdss_funcs; @@ -252,6 +269,7 @@ int dpu_mdss_init(struct platform_device *pdev) irq_error: _dpu_mdss_irq_domain_fini(dpu_mdss); irq_domain_error: +reset_parse_err: clk_parse_err: if (dpu_mdss->mmio) devm_iounmap(&pdev->dev, dpu_mdss->mmio); diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index 129fa841ac22..7595f83da3f1 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -388,6 +388,10 @@ static int msm_drm_init(struct device *dev, const struct drm_driver *drv) if (ret) return ret; + /* Issue a reset of the entire MDSS */ + if (priv->mdss && priv->mdss->funcs->reset) + priv->mdss->funcs->reset(priv->mdss); + /* Bind all our sub-components: */ ret = component_bind_all(dev, ddev); if (ret) diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h index 2a4f0526cb98..716a34fca1cd 100644 --- a/drivers/gpu/drm/msm/msm_kms.h +++ b/drivers/gpu/drm/msm/msm_kms.h @@ -205,6 +205,7 @@ struct msm_mdss_funcs { int (*enable)(struct msm_mdss *mdss); int (*disable)(struct msm_mdss *mdss); void (*destroy)(struct msm_mdss *mdss); + int (*reset)(struct msm_mdss *mdss); }; struct msm_mdss {
It's typical for the bootloader to bring up the display for showing a boot splash or efi framebuffer. But in some cases the kernel driver ends up only partially configuring (in particular) the DPU, which might result in e.g. that two different data paths attempts to push data to the interface - with resulting graphical artifacts. Naturally the end goal would be to inherit the bootloader's configuration and provide the user with a glitch free handover from the boot configuration to a running DPU. But as implementing seamless transition from the bootloader configuration to the running OS will be a considerable effort, start by simply resetting the entire MDSS to its power-on state, to avoid the partial configuration. Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org> --- Changes since v1: - Rather than trying to deconfigure individual pieces of the DPU, reset the entire block. drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c | 18 ++++++++++++++++++ drivers/gpu/drm/msm/msm_drv.c | 4 ++++ drivers/gpu/drm/msm/msm_kms.h | 1 + 3 files changed, 23 insertions(+)