Skip to contentSkip to conversation

drm/i915: Retry reading DPCD when bogus values are read

PATCH In review Takashi Iwai 4 participants 7 messages Nov 22, 2013 – Jan 23, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915: Retry reading DPCD when bogus values are read

I got kernel WARNINGs frequently on Haswell laptops complaining about invalid max DP link bw. With drm.debug=0x0e, it turned out that the obtained DPCD is utterly bogus when it happens:
[drm:intel_dp_get_dpcd], DPCD: 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d

This seems happening intermittently without plug state changes and even after the first few reads succeeded. As a workaround, add a sanity check for such bogus values and retry reading if it hits.

---

 drivers/gpu/drm/i915/intel_dp.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_dp.c +14 -1
index eb8139da9763..f596353da557 100644
@@ -2792,13 +2792,26 @@ intel_dp_get_dpcd(struct intel_dp *intel_dp)
 	struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
 	struct drm_device *dev = dig_port->base.base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-
+	int retry = 0;
 	char dpcd_hex_dump[sizeof(intel_dp->dpcd) * 3];
 
+ again:
 	if (intel_dp_aux_native_read_retry(intel_dp, 0x000, intel_dp->dpcd,
 					   sizeof(intel_dp->dpcd)) == 0)
 		return false; /* aux transfer failed */
 
+	/* sanity check: sometimes DPCD contains a series of same bogus value
+	 * even though the above returned success.  Retry reading in that case.
+	 */
+	if (intel_dp->dpcd[0]) {
+		int i;
+		for (i = 1; i < sizeof(intel_dp->dpcd); i++)
+			if (intel_dp->dpcd[i] != intel_dp->dpcd[0])
+				break;
+		if (i >= sizeof(intel_dp->dpcd) && retry++ < 2)
+			goto again;
+	}
+
 	hex_dump_to_buffer(intel_dp->dpcd, sizeof(intel_dp->dpcd),
 			   32, 1, dpcd_hex_dump, sizeof(dpcd_hex_dump), false);
 	DRM_DEBUG_KMS("DPCD: %s\n", dpcd_hex_dump);
Signature · 1 line
1.8.4.3
#1 lore raw
Quoted 10 lines · Takashi Iwai

I got kernel WARNINGs frequently on Haswell laptops complaining about invalid max DP link bw. With drm.debug=0x0e, it turned out that the obtained DPCD is utterly bogus when it happens:
[drm:intel_dp_get_dpcd], DPCD: 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d

This seems happening intermittently without plug state changes and even after the first few reads succeeded. As a workaround, add a sanity check for such bogus values and retry reading if it hits.

*insert lament about how hw never fails to disappoint*

A few questions:
- Has anyone checked whether the same panel (I guess this is about edp) also fails in the same way on other chips? Maybe with a dual gpu solution?

- Jesse has a patch somewhere, please test the diff embedded in this bz comment:

https://bugs.freedesktop.org/show_bug.cgi?id=67245#c29

Could be that we have an issue with the panel power timings.

- Shouldn't we instead clamp the values read from dpcd to a sane range? Worst case we'll do a few bogus link training rounds ...

Also we really should be extracting this stuff from drivers into common code. Somewhen next century maybe ;-)

Cheers, Daniel

Quoted 44 lines

---

 drivers/gpu/drm/i915/intel_dp.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_dp.c +14 -1
index eb8139da9763..f596353da557 100644
@@ -2792,13 +2792,26 @@ intel_dp_get_dpcd(struct intel_dp *intel_dp)
 	struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
 	struct drm_device *dev = dig_port->base.base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-
+	int retry = 0;
 	char dpcd_hex_dump[sizeof(intel_dp->dpcd) * 3];
 
+ again:
 	if (intel_dp_aux_native_read_retry(intel_dp, 0x000, intel_dp->dpcd,
 					   sizeof(intel_dp->dpcd)) == 0)
 		return false; /* aux transfer failed */
 
+	/* sanity check: sometimes DPCD contains a series of same bogus value
+	 * even though the above returned success.  Retry reading in that case.
+	 */
+	if (intel_dp->dpcd[0]) {
+		int i;
+		for (i = 1; i < sizeof(intel_dp->dpcd); i++)
+			if (intel_dp->dpcd[i] != intel_dp->dpcd[0])
+				break;
+		if (i >= sizeof(intel_dp->dpcd) && retry++ < 2)
+			goto again;
+	}
+
 	hex_dump_to_buffer(intel_dp->dpcd, sizeof(intel_dp->dpcd),
 			   32, 1, dpcd_hex_dump, sizeof(dpcd_hex_dump), false);
 	DRM_DEBUG_KMS("DPCD: %s\n", dpcd_hex_dump);
Signature · 6 lines
1.8.4.3

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#2 lore raw

At Mon, 25 Nov 2013 16:47:25 +0100,

Quoted 11 lines · Daniel Vetter
Quoted 9 lines · Takashi Iwai

I got kernel WARNINGs frequently on Haswell laptops complaining about invalid max DP link bw. With drm.debug=0x0e, it turned out that the obtained DPCD is utterly bogus when it happens:
[drm:intel_dp_get_dpcd], DPCD: 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d

This seems happening intermittently without plug state changes and even after the first few reads succeeded. As a workaround, add a sanity check for such bogus values and retry reading if it hits.

Quoted 8 lines
Quoted 1 line

*insert lament about how hw never fails to disappoint*

A few questions:
- Has anyone checked whether the same panel (I guess this is about edp) also fails in the same way on other chips? Maybe with a dual gpu solution?

I should have mentioned more clearly that this was about the real DP monitor, not eDP. I haven't seen this problem with eDP.

Quoted 6 lines

- Jesse has a patch somewhere, please test the diff embedded in this bz comment:

https://bugs.freedesktop.org/show_bug.cgi?id=67245#c29

Could be that we have an issue with the panel power timings.

I guess this is only about eDP?

Takashi

Quoted 58 lines

- Shouldn't we instead clamp the values read from dpcd to a sane range? Worst case we'll do a few bogus link training rounds ...

Also we really should be extracting this stuff from drivers into common code. Somewhen next century maybe ;-)

Cheers, Daniel

Quoted 44 lines

---

 drivers/gpu/drm/i915/intel_dp.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_dp.c +14 -1
index eb8139da9763..f596353da557 100644
@@ -2792,13 +2792,26 @@ intel_dp_get_dpcd(struct intel_dp *intel_dp)
 	struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
 	struct drm_device *dev = dig_port->base.base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-
+	int retry = 0;
 	char dpcd_hex_dump[sizeof(intel_dp->dpcd) * 3];
 
+ again:
 	if (intel_dp_aux_native_read_retry(intel_dp, 0x000, intel_dp->dpcd,
 					   sizeof(intel_dp->dpcd)) == 0)
 		return false; /* aux transfer failed */
 
+	/* sanity check: sometimes DPCD contains a series of same bogus value
+	 * even though the above returned success.  Retry reading in that case.
+	 */
+	if (intel_dp->dpcd[0]) {
+		int i;
+		for (i = 1; i < sizeof(intel_dp->dpcd); i++)
+			if (intel_dp->dpcd[i] != intel_dp->dpcd[0])
+				break;
+		if (i >= sizeof(intel_dp->dpcd) && retry++ < 2)
+			goto again;
+	}
+
 	hex_dump_to_buffer(intel_dp->dpcd, sizeof(intel_dp->dpcd),
 			   32, 1, dpcd_hex_dump, sizeof(dpcd_hex_dump), false);
 	DRM_DEBUG_KMS("DPCD: %s\n", dpcd_hex_dump);
Signature · 6 lines
1.8.4.3

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#3 lore raw
Quoted 7 lines · Takashi Iwai
Quoted 4 lines

A few questions:
- Has anyone checked whether the same panel (I guess this is about edp) also fails in the same way on other chips? Maybe with a dual gpu solution?

I should have mentioned more clearly that this was about the real DP monitor, not eDP. I haven't seen this problem with eDP.

Then I'd really like to know what happens on other platforms, both intel and non-intel.

Quoted 8 lines
Quoted 6 lines

- Jesse has a patch somewhere, please test the diff embedded in this bz comment:

https://bugs.freedesktop.org/show_bug.cgi?id=67245#c29

Could be that we have an issue with the panel power timings.

I guess this is only about eDP?

Yeah, this is just an eDP-only issue that Jesse's patch addresses. -Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
maintainer
#4 lore raw
In reply to Takashi Iwai
Quoted 8 lines · Takashi Iwai

I got kernel WARNINGs frequently on Haswell laptops complaining about invalid max DP link bw. With drm.debug=0x0e, it turned out that the obtained DPCD is utterly bogus when it happens:
[drm:intel_dp_get_dpcd], DPCD: 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d

This seems happening intermittently without plug state changes and even after the first few reads succeeded. As a workaround, add a sanity check for such bogus values and retry reading if it hits.

If it's not related to crtc disable/enable state changes, then I assume this might happen with any native aux reads at any time, and your patch would just paper over one instance of it.

Does this occur with several different displays? With the displays this occurs with, does it also occur on non-Haswell?

Have you observed that you'd ever need to try again more than once? I.e. does the first retry sometimes return garbage too?

Instinctively I'd say we need to resist the urge to apply this workaround, and try to find the root cause.

BR,
Jani.

Quoted 46 lines

---

 drivers/gpu/drm/i915/intel_dp.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_dp.c +14 -1
index eb8139da9763..f596353da557 100644
@@ -2792,13 +2792,26 @@ intel_dp_get_dpcd(struct intel_dp *intel_dp)
 	struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
 	struct drm_device *dev = dig_port->base.base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-
+	int retry = 0;
 	char dpcd_hex_dump[sizeof(intel_dp->dpcd) * 3];
 
+ again:
 	if (intel_dp_aux_native_read_retry(intel_dp, 0x000, intel_dp->dpcd,
 					   sizeof(intel_dp->dpcd)) == 0)
 		return false; /* aux transfer failed */
 
+	/* sanity check: sometimes DPCD contains a series of same bogus value
+	 * even though the above returned success.  Retry reading in that case.
+	 */
+	if (intel_dp->dpcd[0]) {
+		int i;
+		for (i = 1; i < sizeof(intel_dp->dpcd); i++)
+			if (intel_dp->dpcd[i] != intel_dp->dpcd[0])
+				break;
+		if (i >= sizeof(intel_dp->dpcd) && retry++ < 2)
+			goto again;
+	}
+
 	hex_dump_to_buffer(intel_dp->dpcd, sizeof(intel_dp->dpcd),
 			   32, 1, dpcd_hex_dump, sizeof(dpcd_hex_dump), false);
 	DRM_DEBUG_KMS("DPCD: %s\n", dpcd_hex_dump);
Signature · 6 lines
1.8.4.3

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 1 line
Jani Nikula, Intel Open Source Technology Center
#5 lore raw

At Wed, 04 Dec 2013 12:17:03 +0200,

Quoted 7 lines · Jani Nikula
Quoted 5 lines · Takashi Iwai

I got kernel WARNINGs frequently on Haswell laptops complaining about invalid max DP link bw. With drm.debug=0x0e, it turned out that the obtained DPCD is utterly bogus when it happens:
[drm:intel_dp_get_dpcd], DPCD: 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d

Quoted 7 lines
Quoted 3 lines

This seems happening intermittently without plug state changes and even after the first few reads succeeded. As a workaround, add a sanity check for such bogus values and retry reading if it hits.

If it's not related to crtc disable/enable state changes, then I assume this might happen with any native aux reads at any time, and your patch would just paper over one instance of it.

Yes, exactly, that's the purpose :)

Quoted 2 lines

Does this occur with several different displays? With the displays this occurs with, does it also occur on non-Haswell?

I haven't seen on IvyBridge machines, at least. And I have no other DP monitor, so cannot test right now.

Quoted 2 lines

Have you observed that you'd ever need to try again more than once? I.e. does the first retry sometimes return garbage too?

I don't remember whether more than once were needed.

Takashi

Quoted 56 lines

Instinctively I'd say we need to resist the urge to apply this workaround, and try to find the root cause.

BR,
Jani.

Quoted 46 lines

---

 drivers/gpu/drm/i915/intel_dp.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_dp.c +14 -1
index eb8139da9763..f596353da557 100644
@@ -2792,13 +2792,26 @@ intel_dp_get_dpcd(struct intel_dp *intel_dp)
 	struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
 	struct drm_device *dev = dig_port->base.base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-
+	int retry = 0;
 	char dpcd_hex_dump[sizeof(intel_dp->dpcd) * 3];
 
+ again:
 	if (intel_dp_aux_native_read_retry(intel_dp, 0x000, intel_dp->dpcd,
 					   sizeof(intel_dp->dpcd)) == 0)
 		return false; /* aux transfer failed */
 
+	/* sanity check: sometimes DPCD contains a series of same bogus value
+	 * even though the above returned success.  Retry reading in that case.
+	 */
+	if (intel_dp->dpcd[0]) {
+		int i;
+		for (i = 1; i < sizeof(intel_dp->dpcd); i++)
+			if (intel_dp->dpcd[i] != intel_dp->dpcd[0])
+				break;
+		if (i >= sizeof(intel_dp->dpcd) && retry++ < 2)
+			goto again;
+	}
+
 	hex_dump_to_buffer(intel_dp->dpcd, sizeof(intel_dp->dpcd),
 			   32, 1, dpcd_hex_dump, sizeof(dpcd_hex_dump), false);
 	DRM_DEBUG_KMS("DPCD: %s\n", dpcd_hex_dump);
Signature · 6 lines
1.8.4.3

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 1 line
Jani Nikula, Intel Open Source Technology Center
#6 lore raw
In reply to Takashi Iwai
Quoted 37 lines · Takashi Iwai

I got kernel WARNINGs frequently on Haswell laptops complaining about invalid max DP link bw. With drm.debug=0x0e, it turned out that the obtained DPCD is utterly bogus when it happens:

   [drm:intel_dp_get_dpcd], DPCD: 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d 4d

This seems happening intermittently without plug state changes and even after the first few reads succeeded. As a workaround, add a sanity check for such bogus values and retry reading if it hits.

---

  drivers/gpu/drm/i915/intel_dp.c | 15 ++++++++++++++-
  1 file changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_dp.c +7 -1
index eb8139da9763..f596353da557 100644
@@ -2792,13 +2792,26 @@ intel_dp_get_dpcd(struct intel_dp *intel_dp)
  	struct intel_digital_port *dig_port = dp_to_dig_port(intel_dp);
  	struct drm_device *dev = dig_port->base.base.dev;
  	struct drm_i915_private *dev_priv = dev->dev_private;
-
+	int retry = 0;
  	char dpcd_hex_dump[sizeof(intel_dp->dpcd) * 3];
 
+ again:
  	if (intel_dp_aux_native_read_retry(intel_dp, 0x000, intel_dp->dpcd,
  					   sizeof(intel_dp->dpcd)) == 0)
  		return false; /* aux transfer failed */
 
+	/* sanity check: sometimes DPCD contains a series of same bogus value
+	 * even though the above returned success.  Retry reading in that case.
+	 */
+	if (intel_dp->dpcd[0]) {
+		int i;

drivers/gpu/drm/i915/intel_dp.c intel_dp_get_dpcd(struct intel_dp *intel_dp) · line 2817

Quoted 10 lines
drivers/gpu/drm/i915/intel_dp.c +7 -0
+		for (i = 1; i < sizeof(intel_dp->dpcd); i++)
+			if (intel_dp->dpcd[i] != intel_dp->dpcd[0])
+				break;
+		if (i >= sizeof(intel_dp->dpcd) && retry++ < 2)
+			goto again;
+	}
+
  	hex_dump_to_buffer(intel_dp->dpcd, sizeof(intel_dp->dpcd),
  			   32, 1, dpcd_hex_dump, sizeof(dpcd_hex_dump), false);
  	DRM_DEBUG_KMS("DPCD: %s\n", dpcd_hex_dump);

I'm in agreement with Jani's assessment. While it does address the particular problem at hand, it just looks for duplicate data and issues a retry if that specific condition is detected. It seems more appropriate that instead of checking for specific failure modes, we check to ensure that we have useable values for the basic, critical fields required for Displayport.

I would suggest that we validate the first 3 bytes of the DPCD. Those fields are as follows:

    0x00 - DPCD revision level
        Currently can only be 0x10, 0x11 or 0x12
    0x01 - Link rate
        Also only 3 values currently - 0x06, 0x0A and 0x14
    0x02 - Max lane count
        For validation purposes, we just need to look at 4:0 and verify 

it's 0x01, 0x02 or 0x04
While bits 6 and 7 are used, they're just feature bits and we
check them later anyways.

This is kind of what I was thinking:

    #define MAX_DPCD_REVISION 0x12
    u8 lane_count = DPCD[2] & 0x1F;

    // Verify DPCD revision is in a valid range. MAX_DPCD_REVISION 
should be updated as necessasry
    // to reflect the current supported DPCD version.
    if (DPCD[0] < 0x10 || DPCD > MAX_DPCD_REVISION)
        goto retry;
    // These are the only valid values for link rate. More will surely 
be added as Displayport moves forward.
    if  !(DPCD[1] == 0x06 || DPCD[1] == 0x0A || DPCD[1] == 0x14)
        goto retry;
    // Lane count must be 1, 2 or 4. 3 shalt thou not count, and 5 is 
right out...
    if !(lane_count == 0x01 || lane_count == 0x02 || lane_count == 0x04)
        goto retry;

retry:
retry_count++;
if retry_count > max_retries
goto error
else
goto read_dpcd

-T