Skip to contentSkip to conversation

Get EDID late for VGA switcheroo

PATCH Superseded Ramkumar Ramachandra 4 participants 16 messages Jan 7, 2014

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover Get EDID late for VGA switcheroo

Hi,

VGA switcheroo doesn't work on my 2013 MBP, and I'm trying to fix it. From what I've gathered from previous patches, it seems that the EDID is not computed at boottime, because LVDS isn't connected to the i915 card (and is connected to the nouveau card instead). So, here's a series to get switcheroo-reprobe to call get-edid. I think it's a step in the right direction, although I think more stuff is required to fix the issue.

Disclaimer: The series might be utter rubbish, because this is my first attempt at any driver code.

Thanks.

Cc Andreas Heider <[email protected]>
Cc Seth Forshee <[email protected]>

Ramkumar Ramachandra (4):
drm/i915: add support for vga_switcheroo reprobe drm/i915: factor out intel_lvds_get_edid() drm/i915: prepare intel_lvds_get_edid() for multiple calls drm/i915: check LVDS for EDID on GPU switches

 drivers/gpu/drm/i915/i915_dma.c   |   9 ++-
 drivers/gpu/drm/i915/i915_drv.h   |   1 +
 drivers/gpu/drm/i915/intel_drv.h  |   1 +
 drivers/gpu/drm/i915/intel_lvds.c | 112 +++++++++++++++++++++++---------------
 4 files changed, 78 insertions(+), 45 deletions(-)
Signature · 1 line
1.8.5.2.229.g4448466
#1 lore raw

PATCH 1/4 drm/i915: add support for vga_switcheroo reprobe

---
 drivers/gpu/drm/i915/i915_dma.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_dma.c +7 -1
index 5c64842..336a835b 100644
@@ -1265,6 +1265,12 @@ static void i915_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_
 	}
 }
 
+static void i915_switcheroo_reprobe(struct pci_dev *pdev)
+{
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	intel_fbdev_output_poll_changed(dev);
+}
+
 static bool i915_switcheroo_can_switch(struct pci_dev *pdev)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
@@ -1278,7 +1284,7 @@ static bool i915_switcheroo_can_switch(struct pci_dev *pdev)
 
 static const struct vga_switcheroo_client_ops i915_switcheroo_ops = {
 	.set_gpu_state = i915_switcheroo_set_state,
-	.reprobe = NULL,
+	.reprobe = i915_switcheroo_reprobe,
 	.can_switch = i915_switcheroo_can_switch,
 };
 
Signature · 1 line
1.8.5.2.229.g4448466
#2 lore raw

Re: [PATCH 1/4] drm/i915: add support for vga_switcheroo reprobe

Quoted 13 lines · Ramkumar Ramachandra
---
 drivers/gpu/drm/i915/i915_dma.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_dma.c +0 -0
index 5c64842..336a835b 100644

drivers/gpu/drm/i915/i915_dma.c static void i915_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_ · line 1272

Quoted 8 lines
drivers/gpu/drm/i915/i915_dma.c +5 -0
@@ -1265,6 +1265,12 @@ static void i915_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_
 	}
 }
 
+static void i915_switcheroo_reprobe(struct pci_dev *pdev)
+{
+	struct drm_device *dev = pci_get_drvdata(pdev);
+	intel_fbdev_output_poll_changed(dev);
+}

switcheroo should be sending a hotplug notification, which should trigger both this kernel listener and userspace.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#3 lore raw

Re: [PATCH 1/4] drm/i915: add support for vga_switcheroo reprobe

Hi Chris,

Quoted 8 lines · Chris Wilson
Quoted 8 lines
drivers/gpu/drm/i915/i915_dma.c +0 -0
index 5c64842..336a835b 100644
@@ -1265,6 +1265,12 @@ static void i915_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_
      }
 }
 
Quoted 8 lines

drivers/gpu/drm/i915/i915_dma.c static void i915_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_ · line 1272

Quoted 5 lines
drivers/gpu/drm/i915/i915_dma.c +5 -0
+static void i915_switcheroo_reprobe(struct pci_dev *pdev)
+{
+     struct drm_device *dev = pci_get_drvdata(pdev);
+     intel_fbdev_output_poll_changed(dev);
+}

switcheroo should be sending a hotplug notification, which should trigger both this kernel listener and userspace.

If I understand correctly, switcheroo will send a hotplug notification when LVDS is connected to the i915 device, and trigger this reprobe function(), which in a later patch calls get-edid. Right?

#4 lore raw

PATCH 2/4 drm/i915: factor out intel_lvds_get_edid()

In reply to Ramkumar Ramachandra

This code will be reused to support hybrid graphics on some Apple machines that can't get a mode for the LVDS panel at boot, so move it into a new function named intel_lvds_get_edid().

---
 drivers/gpu/drm/i915/i915_drv.h   |   1 +
 drivers/gpu/drm/i915/intel_lvds.c | 104 ++++++++++++++++++++++----------------
 2 files changed, 61 insertions(+), 44 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +1 -0
index 90fcccb..2c055f8 100644
@@ -1496,6 +1496,7 @@ typedef struct drm_i915_private {
 	} wm;
 
 	struct i915_package_c8 pc8;
+	struct drm_connector *int_lvds_connector;
 
 	/* Old dri1 support infrastructure, beware the dragons ya fools entering
 	 * here! */
drivers/gpu/drm/i915/intel_lvds.c +60 -44

146 lines not on this page · raw message

Signature · 1 line
1.8.5.2.229.g4448466
#5 lore raw

PATCH 3/4 drm/i915: prepare intel_lvds_get_edid() for multiple calls

In reply to Ramkumar Ramachandra

intel_lvds_get_edid() needs to be called when switching GPUs, but it currently assumes that it will only be called once and that there's always an LVDS connector present when it's called. Fix this assumptions.

---
 drivers/gpu/drm/i915/intel_lvds.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_lvds.c +9 -1
index 6c09617..8275551 100644
@@ -927,10 +927,18 @@ static bool intel_lvds_get_edid(struct drm_device *dev)
 	struct edid *edid;
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_connector *connector = dev_priv->int_lvds_connector;
-	struct intel_lvds_connector *lvds_connector = to_lvds_connector(connector);
+	struct intel_lvds_connector *lvds_connector;
 	struct drm_display_mode *scan; /* *modes, *bios_mode; */
 	struct drm_display_mode *fixed_mode = NULL;
 
+	if (!connector)
+		return false;
+	lvds_connector = to_lvds_connector(connector);
+
+	/* If we already have an EDID, no need to check again */
+	if (lvds_connector->base.edid)
+		return true;
+
 	/*
 	 * Attempt to get the fixed panel mode from DDC.  Assume that the
 	 * preferred mode is the right one.
Signature · 1 line
1.8.5.2.229.g4448466
#6 lore raw

PATCH 4/4 drm/i915: check LVDS for EDID on GPU switches

In reply to Ramkumar Ramachandra

If the LVDS panel wasn't connected at boot then we won't have an EDID for it. To fix this, call intel_lvds_get_edid() from the vga_switcheroo reprobe callback.

---
 drivers/gpu/drm/i915/i915_dma.c   | 1 +
 drivers/gpu/drm/i915/intel_drv.h  | 1 +
 drivers/gpu/drm/i915/intel_lvds.c | 2 +-
 3 files changed, 3 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_dma.c +1 -0
index 336a835b..5eb37a3 100644
@@ -1268,6 +1268,7 @@ static void i915_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_
 static void i915_switcheroo_reprobe(struct pci_dev *pdev)
 {
 	struct drm_device *dev = pci_get_drvdata(pdev);
+	intel_lvds_get_edid(dev);
 	intel_fbdev_output_poll_changed(dev);
 }
 
drivers/gpu/drm/i915/intel_drv.h +1 -0
index 79f91f2..686954f 100644
@@ -769,6 +769,7 @@ bool intel_hdmi_compute_config(struct intel_encoder *encoder,
 
 
 /* intel_lvds.c */
+bool intel_lvds_get_edid(struct drm_device *dev);
 void intel_lvds_init(struct drm_device *dev);
 bool intel_is_dual_link_lvds(struct drm_device *dev);
 
drivers/gpu/drm/i915/intel_lvds.c +1 -1
index 8275551..f700355 100644
@@ -922,7 +922,7 @@ static bool intel_lvds_supported(struct drm_device *dev)
 	return false;
 }
 
-static bool intel_lvds_get_edid(struct drm_device *dev)
+bool intel_lvds_get_edid(struct drm_device *dev)
 {
 	struct edid *edid;
 	struct drm_i915_private *dev_priv = dev->dev_private;
Signature · 1 line
1.8.5.2.229.g4448466
#7 lore raw

Re: [PATCH 4/4] drm/i915: check LVDS for EDID on GPU switches

Quoted 3 lines · Ramkumar Ramachandra

If the LVDS panel wasn't connected at boot then we won't have an EDID for it. To fix this, call intel_lvds_get_edid() from the vga_switcheroo reprobe callback.

I would rather have an iterator over all our connectors (or perhaps encoders would be the right semantic, except we have connectors==encoders) and a reprobe callback. Saves exporting private functions and prevent me asking silly questions like what about eDP? -Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#8 lore raw

Re: [PATCH 4/4] drm/i915: check LVDS for EDID on GPU switches

Quoted 8 lines · Chris Wilson
Quoted 3 lines · Ramkumar Ramachandra

If the LVDS panel wasn't connected at boot then we won't have an EDID for it. To fix this, call intel_lvds_get_edid() from the vga_switcheroo reprobe callback.

I would rather have an iterator over all our connectors (or perhaps encoders would be the right semantic, except we have connectors==encoders) and a reprobe callback.

I don't follow; iterate over which connectors? There's one lvds_connector for which we get EDID.

Quoted 2 lines

Saves exporting private
functions and prevent me asking silly questions like what about eDP?

Avoid exporting intel_lvds_get_edid()? Why?

p.s- Excuse the stupidity of the questions; I know close to nothing about how i915 or switcheroo work.

#9 lore raw

Re: [PATCH 4/4] drm/i915: check LVDS for EDID on GPU switches

Quoted 6 lines · Ramkumar Ramachandra
Quoted 5 lines · Chris Wilson
Quoted 3 lines · Ramkumar Ramachandra

If the LVDS panel wasn't connected at boot then we won't have an EDID for it. To fix this, call intel_lvds_get_edid() from the vga_switcheroo reprobe callback.

Quoted 6 lines
Quoted 3 lines

I would rather have an iterator over all our connectors (or perhaps encoders would be the right semantic, except we have connectors==encoders) and a reprobe callback.

I don't follow; iterate over which connectors? There's one lvds_connector for which we get EDID.

Rather than special casing lvds (especially when there are other panel connectors that can also be muxed), extend the connector interface to support a reprobe and walk over all connectors associated with i915 after a switcheroo event.

Quoted 4 lines
Quoted 2 lines

Saves exporting private
functions and prevent me asking silly questions like what about eDP?

Avoid exporting intel_lvds_get_edid()? Why?

Because it's a special case and misses others.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#10 lore raw

Re: [PATCH 4/4] drm/i915: check LVDS for EDID on GPU switches

Quoted 4 lines · Chris Wilson

Rather than special casing lvds (especially when there are other panel connectors that can also be muxed), extend the connector interface to support a reprobe and walk over all connectors associated with i915 after a switcheroo event.

Okay, so I can see the following get_edid functions:

intel_lvds_get_edid()
intel_crt_get_edid()
intel_sdvo_get_edid()
intel_dp_get_edid()

In the hdmi case, drm_get_edid() is called in two places:
intel_hdmi_detect() and intel_hdmi_detect_audio(); should I factor out something into a corresponding intel_hdmi_get_edid() function?

Now, in i915_switcheroo_reprobe(), I need to call all these functions, right? They all accept a a generic drm_connector and an specific i2c_adapter. But how do I do that without exporting each of these functions and their adapters?

Thanks.

#11 lore raw

Re: [PATCH 4/4] drm/i915: check LVDS for EDID on GPU switches

Quoted 13 lines · Ramkumar Ramachandra
Quoted 4 lines · Chris Wilson

Rather than special casing lvds (especially when there are other panel connectors that can also be muxed), extend the connector interface to support a reprobe and walk over all connectors associated with i915 after a switcheroo event.

Okay, so I can see the following get_edid functions:

intel_lvds_get_edid()
intel_crt_get_edid()
intel_sdvo_get_edid()
intel_dp_get_edid()

Quoted 8 lines

In the hdmi case, drm_get_edid() is called in two places:
intel_hdmi_detect() and intel_hdmi_detect_audio(); should I factor out something into a corresponding intel_hdmi_get_edid() function?

Now, in i915_switcheroo_reprobe(), I need to call all these functions, right? They all accept a a generic drm_connector and an specific i2c_adapter. But how do I do that without exporting each of these functions and their adapters?

If you look at struct intel_connector, you will see space for adding a callback that each of the LVDS/CRT/SDVO/DP connectors can hook into if they want to reprobe after a switcheroo.

struct intel_connector *connector;
list_for_each_entry(connector,

                    &dev->mode_config.connector_list,
		    base.head) {
	connector->reprobe(connector);

}

(or something similar) would do in the switcheroo callback.

-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#12 lore raw
In reply to Ramkumar Ramachandra
Quoted 9 lines · Ramkumar Ramachandra

Hi,

VGA switcheroo doesn't work on my 2013 MBP, and I'm trying to fix it. From what I've gathered from previous patches, it seems that the EDID is not computed at boottime, because LVDS isn't connected to the i915 card (and is connected to the nouveau card instead). So, here's a series to get switcheroo-reprobe to call get-edid. I think it's a step in the right direction, although I think more stuff is required to fix the issue.

It's been a while since I've thought about this, so my memory is suspect. It's also a bit dated, and I haven't kept up to know what's changed in since then. But I don't think just reading the mode on reprobe gets you very far.

These are the main problems as I remember them:

1. On many Macbook models the firmware isn't putting LVDS information in opregion. Reading the EDID during device initialization also fails since the DDC is muxed to the DGPU, so i915 concludes there's no LVDS and doesn't register an LVDS connector at all. I also think it was impossible to register a new connector at a later time, though I'm not sure about that.

2. If i915 is forced to register the LVDS connector (say via a quirk) it still may not know the panel mode and guesses something rather conservative. This mode it assumes is used to allocate framebuffer memory, which is likely to end up being too small, and I don't think there was any support for resizing it later. This resulted in VTs not working because there isn't enough framebuffer memory for the mode they're trying to use.

3. So what is really wanted is to be able to mux just the DDC over to the IGPU when the device is initialized to read the panel mode. One possible problem is that the hw might not support muxing the DDC separately, but Apple hw does so we'll ignore that for now. The basic code for muxing only the DDC isn't too difficult, and I had some patches for this at one point. The problem is that the handler must register with vga_switcheroo before this is possible, so there's a module initialization ordering issue that wasn't easily solved. Maybe now it would be possible to do something with EPROBE_DEFER, though I'm not sure how i915 would know whether or not it should wait for a switcheroo handler.

I just didn't have enough time to spend on this issue to work through all of this. I also never looked at doing this with eDP, which is what I thought all the Macbooks have been using for the last couple of years. Are you sure your machine uses LVDS and not eDP?

Seth

#13 lore raw
Quoted 6 lines · Seth Forshee

1. On many Macbook models the firmware isn't putting LVDS information in opregion. Reading the EDID during device initialization also fails since the DDC is muxed to the DGPU, so i915 concludes there's no LVDS and doesn't register an LVDS connector at all. I also think it was impossible to register a new connector at a later time, though I'm not sure about that.

On a different thread, Chris is helping me write a series to fill in the switcheroo-reprobe callback with corresponding get-edid functions. Do you think it'll be useful?

Quoted 7 lines

2. If i915 is forced to register the LVDS connector (say via a quirk) it still may not know the panel mode and guesses something rather conservative. This mode it assumes is used to allocate framebuffer memory, which is likely to end up being too small, and I don't think there was any support for resizing it later. This resulted in VTs not working because there isn't enough framebuffer memory for the mode they're trying to use.

I saw your quirk-patch (https://lkml.org/lkml/2012/8/4/128). Seems like a rather inelegant approach to register the LVDS connector unconditionally during initialization; I'll try to avoid going down this road.

Quoted 11 lines

3. So what is really wanted is to be able to mux just the DDC over to the IGPU when the device is initialized to read the panel mode. One possible problem is that the hw might not support muxing the DDC separately, but Apple hw does so we'll ignore that for now. The basic code for muxing only the DDC isn't too difficult, and I had some patches for this at one point. The problem is that the handler must register with vga_switcheroo before this is possible, so there's a module initialization ordering issue that wasn't easily solved. Maybe now it would be possible to do something with EPROBE_DEFER, though I'm not sure how i915 would know whether or not it should wait for a switcheroo handler.

I didn't understand much of the above; I suppose it'll start making sense when I'm deeper into the issue.

Quoted 4 lines

I just didn't have enough time to spend on this issue to work through all of this. I also never looked at doing this with eDP, which is what I thought all the Macbooks have been using for the last couple of years. Are you sure your machine uses LVDS and not eDP?

I just assumed that it was LVDS; turns out it's eDP.

I have time to work on the issue now, so I'd appreciate any help I can get.

#14 lore raw

Am 2014-01-07 15:30, schrieb Ramkumar Ramachandra:

Quoted 17 lines
Quoted 6 lines · Seth Forshee

3. So what is really wanted is to be able to mux just the DDC over to the IGPU when the device is initialized to read the panel mode.
One
possible problem is that the hw might not support muxing the DDC separately, but Apple hw does so we'll ignore that for now. The basic code for muxing only the DDC isn't too difficult, and I had

Quoted 8 lines

some patches for this at one point. The problem is that the
handler
must register with vga_switcheroo before this is possible, so there's a module initialization ordering issue that wasn't easily solved. Maybe now it would be possible to do something with EPROBE_DEFER, though I'm not sure how i915 would know whether or
not
it should wait for a switcheroo handler.

Quoted 3 lines

I didn't understand much of the above; I suppose it'll start making sense when I'm deeper into the issue.

You can switch the line used to detect the display separately from the line which is used to display the picture, so you can just switch DDC to the intel card when you know it will try to detect the display, without having to introduce flickering. There were a few different versions of patches that did this, I think this one is the latest: http://lists.freedesktop.org/archives/dri-devel/2012-September/027528.html

Sorry that I forgot about this yesterday, it has been quite some time since I worked on this.

#15 lore raw
Quoted 6 lines · [email protected] wrote:

You can switch the line used to detect the display separately from the line which is used to display the picture, so you can just switch DDC to the intel card when you know it will try to detect the display, without having to introduce flickering. There were a few different versions of patches that did this, I think this one is the latest:
http://lists.freedesktop.org/archives/dri-devel/2012-September/027528.html

Any idea why this series didn't get merged?