Skip to contentSkip to conversation

drm/i915: Increase WM memory latency values on SNB with high pixel clock

PATCH In review Jani Nikula 6 participants 18 messages Mar 21, 2014 – May 15, 2014

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

PATCH 1/1 drm/i915: Increase WM memory latency values on SNB with high pixel clock

From: Ville Syrjälä <[email protected]>

On SNB the BIOS provided WM memory latency values seem insufficient to handle high resolution displays.

In this particular case the display mode was a 2560x1440@60Hz, which makes the pixel clock 241.5 MHz. It was empirically found that a memory latency value if 1.2 usec is enough to avoid underruns, whereas the BIOS provided value of 0.7 usec was clearly too low. Incidentally 1.2 usec is what the typical BIOS provided values are on IVB systems.

Increase the WM memory latency values to 1.2 usec when encountering a display mode with pixel clock exceeding 225 MHz. 225 MHz was chosen as the threshold simply because that's the documented limit of SNB with HDMI, so one might surmise that the hardware may have been tested up to that frequency. In theory the latency shouldn't depend on the pixel clock at all. So it may be that we should just increase the latency values across the board for all SNB systems. But for now I'm inclined to limit this quirk to only those cases that are proven to need it, as doing otherwise might cause some increase in power consumption for everyone.

Cc: Robert N <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70254

---

I'm just relaying the patch from bugzilla to the list as it seems to fix the bug. Don't ask me anything about it.

BR,
Jani.

---
 drivers/gpu/drm/i915/intel_display.c |   10 ++++++++++
 drivers/gpu/drm/i915/intel_drv.h     |    1 +
 drivers/gpu/drm/i915/intel_pm.c      |   33 +++++++++++++++++++++++++++++++++
 3 files changed, 44 insertions(+)
drivers/gpu/drm/i915/intel_display.c +10 -0
index 0b19afdfbaa7..b0ac92d8f461 100644
@@ -9933,6 +9933,16 @@ static int __intel_set_mode(struct drm_crtc *crtc,
 		 */
 		drm_calc_timestamping_constants(crtc,
 						&pipe_config->adjusted_mode);
+
+
+		/*
+		 * The BIOS provided WM memory latency values are often
+		 * inadequate for high resolution displays. Adjust them.
+		 *
+		 * FIXME not sure 225MHz is a good threshold.
+		 */
+		if (IS_GEN6(dev) && pipe_config->adjusted_mode.crtc_clock > 225000)
+			snb_wm_latency_quirk(dev);
 	}
 
 	/* Only after disabling all output pipelines that will be changed can we
drivers/gpu/drm/i915/intel_drv.h +1 -0
index 60ffad376390..b1d631c9dfa5 100644
@@ -903,6 +903,7 @@ void intel_runtime_pm_put(struct drm_i915_private *dev_priv);
 void intel_init_runtime_pm(struct drm_i915_private *dev_priv);
 void intel_fini_runtime_pm(struct drm_i915_private *dev_priv);
 void ilk_wm_get_hw_state(struct drm_device *dev);
+void snb_wm_latency_quirk(struct drm_device *dev);
 
 
 /* intel_sdvo.c */
drivers/gpu/drm/i915/intel_pm.c +33 -0
index 39f3238bf1c3..226750452637 100644
@@ -2104,6 +2104,39 @@ static void ilk_setup_wm_latency(struct drm_device *dev)
 	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
 }
 
+static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
+				    uint16_t wm[5], uint16_t min)
+{
+	int level, max_level = ilk_wm_max_level(dev_priv->dev);
+
+	if (wm[0] >= min)
+		return false;
+
+	wm[0] = max(wm[0], min);
+	for (level = 1; level <= max_level; level++)
+		wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
+
+	return true;
+}
+
+void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
+
+	if (!changed)
+		return;
+
+	DRM_DEBUG_KMS("WM latency values increased due to high pixel clock\n");
+	intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
+	intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
+	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+}
+
 static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
 				      struct ilk_pipe_wm_parameters *p,
 				      struct intel_wm_config *config)
Signature · 6 lines
1.7.9.5

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 93 lines · Jani Nikula

From: Ville Syrjälä <[email protected]>

On SNB the BIOS provided WM memory latency values seem insufficient to handle high resolution displays.

In this particular case the display mode was a 2560x1440@60Hz, which makes the pixel clock 241.5 MHz. It was empirically found that a memory latency value if 1.2 usec is enough to avoid underruns, whereas the BIOS provided value of 0.7 usec was clearly too low. Incidentally 1.2 usec is what the typical BIOS provided values are on IVB systems.

Increase the WM memory latency values to 1.2 usec when encountering a display mode with pixel clock exceeding 225 MHz. 225 MHz was chosen as the threshold simply because that's the documented limit of SNB with HDMI, so one might surmise that the hardware may have been tested up to that frequency. In theory the latency shouldn't depend on the pixel clock at all. So it may be that we should just increase the latency values across the board for all SNB systems. But for now I'm inclined to limit this quirk to only those cases that are proven to need it, as doing otherwise might cause some increase in power consumption for everyone.

Cc: Robert N <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70254

---

I'm just relaying the patch from bugzilla to the list as it seems to fix the bug. Don't ask me anything about it.

BR,
Jani.

---
 drivers/gpu/drm/i915/intel_display.c |   10 ++++++++++
 drivers/gpu/drm/i915/intel_drv.h     |    1 +
 drivers/gpu/drm/i915/intel_pm.c      |   33 +++++++++++++++++++++++++++++++++
 3 files changed, 44 insertions(+)
drivers/gpu/drm/i915/intel_display.c +10 -0
index 0b19afdfbaa7..b0ac92d8f461 100644
@@ -9933,6 +9933,16 @@ static int __intel_set_mode(struct drm_crtc *crtc,
 		 */
 		drm_calc_timestamping_constants(crtc,
 						&pipe_config->adjusted_mode);
+
+
+		/*
+		 * The BIOS provided WM memory latency values are often
+		 * inadequate for high resolution displays. Adjust them.
+		 *
+		 * FIXME not sure 225MHz is a good threshold.
+		 */
+		if (IS_GEN6(dev) && pipe_config->adjusted_mode.crtc_clock > 225000)
+			snb_wm_latency_quirk(dev);
 	}
 
 	/* Only after disabling all output pipelines that will be changed can we
drivers/gpu/drm/i915/intel_drv.h +1 -0
index 60ffad376390..b1d631c9dfa5 100644
@@ -903,6 +903,7 @@ void intel_runtime_pm_put(struct drm_i915_private *dev_priv);
 void intel_init_runtime_pm(struct drm_i915_private *dev_priv);
 void intel_fini_runtime_pm(struct drm_i915_private *dev_priv);
 void ilk_wm_get_hw_state(struct drm_device *dev);
+void snb_wm_latency_quirk(struct drm_device *dev);
 
 
 /* intel_sdvo.c */
drivers/gpu/drm/i915/intel_pm.c +13 -0
index 39f3238bf1c3..226750452637 100644
@@ -2104,6 +2104,39 @@ static void ilk_setup_wm_latency(struct drm_device *dev)
 	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
 }
 
+static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
+				    uint16_t wm[5], uint16_t min)
+{
+	int level, max_level = ilk_wm_max_level(dev_priv->dev);
+
+	if (wm[0] >= min)
+		return false;
+
+	wm[0] = max(wm[0], min);
+	for (level = 1; level <= max_level; level++)
+		wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
+
+	return true;

drivers/gpu/drm/i915/intel_pm.c static void ilk_setup_wm_latency(struct drm_device *dev) · line 2129

Quoted 10 lines
drivers/gpu/drm/i915/intel_pm.c +10 -0
+}
+
+void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

Unless I missed something this hereafter permanently bumps the latency for computations.

At the 225MHz pixel clock, what are the differences in final WM values? -Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#2 lore raw
In reply to Jani Nikula
Quoted 22 lines · Jani Nikula

From: Ville Syrjälä <[email protected]>

On SNB the BIOS provided WM memory latency values seem insufficient to handle high resolution displays.

In this particular case the display mode was a 2560x1440@60Hz, which makes the pixel clock 241.5 MHz. It was empirically found that a memory latency value if 1.2 usec is enough to avoid underruns, whereas the BIOS provided value of 0.7 usec was clearly too low. Incidentally 1.2 usec is what the typical BIOS provided values are on IVB systems.

Increase the WM memory latency values to 1.2 usec when encountering a display mode with pixel clock exceeding 225 MHz. 225 MHz was chosen as the threshold simply because that's the documented limit of SNB with HDMI, so one might surmise that the hardware may have been tested up to that frequency. In theory the latency shouldn't depend on the pixel clock at all. So it may be that we should just increase the latency values across the board for all SNB systems. But for now I'm inclined to limit this quirk to only those cases that are proven to need it, as doing otherwise might cause some increase in power consumption for everyone.

Quoted 3 lines

Cc: Robert N <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70254

I think we should just unconditionally increase the memory latency. ivb has a rather similar memory subsystem compared to snb, so just increasing it always to this limits sounds ok. For lower resolutions we might simply get away because there's more excess bandwidth available.

Maybe Art can shed some light on this.
-Daniel

Quoted 98 lines

---

I'm just relaying the patch from bugzilla to the list as it seems to fix the bug. Don't ask me anything about it.

BR,
Jani.

---
 drivers/gpu/drm/i915/intel_display.c |   10 ++++++++++
 drivers/gpu/drm/i915/intel_drv.h     |    1 +
 drivers/gpu/drm/i915/intel_pm.c      |   33 +++++++++++++++++++++++++++++++++
 3 files changed, 44 insertions(+)
drivers/gpu/drm/i915/intel_display.c +10 -0
index 0b19afdfbaa7..b0ac92d8f461 100644
@@ -9933,6 +9933,16 @@ static int __intel_set_mode(struct drm_crtc *crtc,
  */
  drm_calc_timestamping_constants(crtc,
  &pipe_config->adjusted_mode);
+
+
+ /*
+ * The BIOS provided WM memory latency values are often
+ * inadequate for high resolution displays. Adjust them.
+ *
+ * FIXME not sure 225MHz is a good threshold.
+ */
+ if (IS_GEN6(dev) && pipe_config->adjusted_mode.crtc_clock > 225000)
+ snb_wm_latency_quirk(dev);
  }
 
  /* Only after disabling all output pipelines that will be changed can we
drivers/gpu/drm/i915/intel_drv.h +1 -0
index 60ffad376390..b1d631c9dfa5 100644
@@ -903,6 +903,7 @@ void intel_runtime_pm_put(struct drm_i915_private *dev_priv);
 void intel_init_runtime_pm(struct drm_i915_private *dev_priv);
 void intel_fini_runtime_pm(struct drm_i915_private *dev_priv);
 void ilk_wm_get_hw_state(struct drm_device *dev);
+void snb_wm_latency_quirk(struct drm_device *dev);
 
 
 /* intel_sdvo.c */
drivers/gpu/drm/i915/intel_pm.c +33 -1
index 39f3238bf1c3..226750452637 100644
@@ -2104,6 +2104,39 @@ static void ilk_setup_wm_latency(struct drm_device *dev)
  intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
 }
 
+static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
+    uint16_t wm[5], uint16_t min)
+{
+ int level, max_level = ilk_wm_max_level(dev_priv->dev);
+
+ if (wm[0] >= min)
+ return false;
+
+ wm[0] = max(wm[0], min);
+ for (level = 1; level <= max_level; level++)
+ wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
+
+ return true;
+}
+
+void snb_wm_latency_quirk(struct drm_device *dev)
+{
+ struct drm_i915_private *dev_priv = dev->dev_private;
+ bool changed;
+
+ changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+ ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+ ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
+
+ if (!changed)
+ return;
+
+ DRM_DEBUG_KMS("WM latency values increased due to high pixel clock\n");
+ intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
+ intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
+ intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+}
+
 static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
       struct ilk_pipe_wm_parameters *p,
       struct intel_wm_config *config)
--

1.7.9.5

_______________________________________________
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

Please check the DRAM configuration for the systems that fail. The higher latency is more likely with higher tRFC which is mainly found with 8 Gbit components.

Quoted 146 lines

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Friday, March 21, 2014 3:19 AM
To: Nikula, Jani; Runyan, Arthur J; Syrjala, Ville
Cc: intel-gfx
Subject: Re: [Intel-gfx] [PATCH] drm/i915: Increase WM memory latency values on SNB with high pixel clock

Quoted 22 lines · Jani Nikula

From: Ville Syrjälä <[email protected]>

On SNB the BIOS provided WM memory latency values seem insufficient to handle high resolution displays.

In this particular case the display mode was a 2560x1440@60Hz, which makes the pixel clock 241.5 MHz. It was empirically found that a memory latency value if 1.2 usec is enough to avoid underruns, whereas the BIOS provided value of 0.7 usec was clearly too low. Incidentally 1.2 usec is what the typical BIOS provided values are on IVB systems.

Increase the WM memory latency values to 1.2 usec when encountering a display mode with pixel clock exceeding 225 MHz. 225 MHz was chosen as the threshold simply because that's the documented limit of SNB with HDMI, so one might surmise that the hardware may have been tested up to that frequency. In theory the latency shouldn't depend on the pixel clock at all. So it may be that we should just increase the latency values across the board for all SNB systems. But for now I'm inclined to limit this quirk to only those cases that are proven to need it, as doing otherwise might cause some increase in power consumption for everyone.

Quoted 3 lines

Cc: Robert N <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70254

I think we should just unconditionally increase the memory latency. ivb has a rather similar memory subsystem compared to snb, so just increasing it always to this limits sounds ok. For lower resolutions we might simply get away because there's more excess bandwidth available.

Maybe Art can shed some light on this.
-Daniel

Quoted 94 lines

---

I'm just relaying the patch from bugzilla to the list as it seems to fix the bug. Don't ask me anything about it.

BR,
Jani.

---
 drivers/gpu/drm/i915/intel_display.c |   10 ++++++++++
 drivers/gpu/drm/i915/intel_drv.h     |    1 +
 drivers/gpu/drm/i915/intel_pm.c      |   33 +++++++++++++++++++++++++++++++++
 3 files changed, 44 insertions(+)
drivers/gpu/drm/i915/intel_display.c +10 -0
index 0b19afdfbaa7..b0ac92d8f461 100644
@@ -9933,6 +9933,16 @@ static int __intel_set_mode(struct drm_crtc *crtc,
  */
  drm_calc_timestamping_constants(crtc,
  &pipe_config->adjusted_mode);
+
+
+ /*
+ * The BIOS provided WM memory latency values are often
+ * inadequate for high resolution displays. Adjust them.
+ *
+ * FIXME not sure 225MHz is a good threshold.
+ */
+ if (IS_GEN6(dev) && pipe_config->adjusted_mode.crtc_clock > 225000)
+ snb_wm_latency_quirk(dev);
  }
 
  /* Only after disabling all output pipelines that will be changed can we
drivers/gpu/drm/i915/intel_drv.h +1 -0
index 60ffad376390..b1d631c9dfa5 100644
@@ -903,6 +903,7 @@ void intel_runtime_pm_put(struct drm_i915_private *dev_priv);
 void intel_init_runtime_pm(struct drm_i915_private *dev_priv);
 void intel_fini_runtime_pm(struct drm_i915_private *dev_priv);
 void ilk_wm_get_hw_state(struct drm_device *dev);
+void snb_wm_latency_quirk(struct drm_device *dev);
 
 
 /* intel_sdvo.c */
drivers/gpu/drm/i915/intel_pm.c +33 -1
index 39f3238bf1c3..226750452637 100644
@@ -2104,6 +2104,39 @@ static void ilk_setup_wm_latency(struct drm_device *dev)
  intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
 }
 
+static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
+    uint16_t wm[5], uint16_t min)
+{
+ int level, max_level = ilk_wm_max_level(dev_priv->dev);
+
+ if (wm[0] >= min)
+ return false;
+
+ wm[0] = max(wm[0], min);
+ for (level = 1; level <= max_level; level++)
+ wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
+
+ return true;
+}
+
+void snb_wm_latency_quirk(struct drm_device *dev)
+{
+ struct drm_i915_private *dev_priv = dev->dev_private;
+ bool changed;
+
+ changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+ ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+ ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
+
+ if (!changed)
+ return;
+
+ DRM_DEBUG_KMS("WM latency values increased due to high pixel clock\n");
+ intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
+ intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
+ intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+}
+
 static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
       struct ilk_pipe_wm_parameters *p,
       struct intel_wm_config *config)
--

1.7.9.5

Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

#4 lore raw
Quoted 2 lines · Runyan, Arthur J <arthur.j.runyan <at> intel.com> writes:

Please check the DRAM configuration for the systems that fail. The higher

latency is more likely with

Quoted 2 lines

higher tRFC which is mainly found with 8 Gbit components.

What other information do we need to get this included?

The DRAM config, is this something that I have to/should check? How do I get this information to you?

#5 lore raw
Quoted 4 lines · Robert Navarro
Quoted 2 lines · Runyan, Arthur J <arthur.j.runyan <at> intel.com> writes:

Please check the DRAM configuration for the systems that fail. The higher

Quoted 8 lines

latency is more likely with

Quoted 2 lines

higher tRFC which is mainly found with 8 Gbit components.

What other information do we need to get this included?

The DRAM config, is this something that I have to/should check? How do I get this information to you?

If decode-dimms works on your machine it might print out potentially interesting stuff. It doesn't seem to work on most of my machines. It does work on the x220 I have here though, so I'm assuming it should work on your x220 too. decode-dimms comes from the i2c-tools package.

Here's what I get on mine:

# modprobe i2c-i801
# modprobe eeprom
# decode-dimms 
# decode-dimms version 6231 (2014-02-20 10:54:34 +0100)

Memory Serial Presence Detect Decoder
By Philip Edelbrock, Christian Zuckschwerdt, Burkart Lingner, Jean Delvare, Trent Piepho and others

Decoding EEPROM: /sys/bus/i2c/drivers/eeprom/0-0050
Guessing DIMM is in bank 1

---=== SPD EEPROM Information ===---

EEPROM CRC of bytes 0-116                       OK (0xFC7B)
# of bytes written to SDRAM EEPROM              176
Total number of bytes in EEPROM                 256
Fundamental Memory type                         DDR3 SDRAM
Module Type                                     SO-DIMM

---=== Memory Characteristics ===---

Fine time base                                  1.000 ps
Medium time base                                0.125 ns
Maximum module speed                            1333 MHz (PC3-10600)
Size                                            4096 MB
Banks x Rows x Columns x Bits                   8 x 15 x 10 x 64
Ranks                                           2
SDRAM Device Width                              8 bits
Bus Width Extension                             0 bits
tCL-tRCD-tRP-tRAS                               9-9-9-24
Supported CAS Latencies (tCL)                   9T, 8T, 7T, 6T, 5T

---=== Timing Parameters ===---

Minimum Write Recovery time (tWR)               15.000 ns
Minimum Row Active to Row Active Delay (tRRD)   6.000 ns
Minimum Active to Auto-Refresh Delay (tRC)      49.125 ns
Minimum Recovery Delay (tRFC)                   160.000 ns
Minimum Write to Read CMD Delay (tWTR)          7.500 ns
Minimum Read to Pre-charge CMD Delay (tRTP)     7.500 ns
Minimum Four Activate Window Delay (tFAW)       30.000 ns

---=== Optional Features ===---

Operable voltages                               1.5V
RZQ/6 supported?                                Yes
RZQ/7 supported?                                Yes
DLL-Off Mode supported?                         Yes
Operating temperature range                     0-95 degrees C
Refresh Rate in extended temp range             1X
Auto Self-Refresh?                              No
On-Die Thermal Sensor readout?                  No
Partial Array Self-Refresh?                     No
Thermal Sensor Accuracy                         Not implemented
SDRAM Device Type                               Standard Monolithic

---=== Physical Characteristics ===---

Module Height (mm)                              30
Module Thickness (mm)                           2 front, 2 back
Module Width (mm)                               67.6
Module Reference Card                           F

---=== Manufacturer Data ===---

Module Manufacturer                             Samsung
DRAM Manufacturer                               Samsung
Manufacturing Location Code                     0x03
Manufacturing Date                              2012-W04
Assembly Serial Number                          0x009B48F1
Part Number                                     M471B5273DH0-CH9  

Number of SDRAM DIMMs detected and decoded: 1

Just for figuring out what tRFC is I think the following would do it:

# intel_reg_read 0x144298 0x144698 0x144a98 0x145d10 0x145e04

These are from the machines I have lying around on my desk:

// dell xps (snb)

# ./intel_reg_read 0x144298 0x144698 0x144a98 0x145d10 0x145e04
0x144298 : 0x5A6B1450
0x144698 : 0x5A6B1450
0x144A98 : 0x0
0x145D10 : 0x16040307
0x145E04 : 0x5

// x220 (snb)

# ./intel_reg_read 0x144298 0x144698 0x144a98 0x145d10 0x145e04
0x144298 : 0x5A6B1450
0x144698 : 0x46B41004
0x144A98 : 0x0
0x145D10 : 0x16040307
0x145E04 : 0x5

// some ivb desktop

# ./intel_reg_read 0x144298 0x144698 0x144a98 0x145d10 0x145e04
0x144298 : 0x5A6B1450
0x144698 : 0x5A6B1450
0x144A98 : 0x0
0x145D10 : 0x2010040C
0x145E04 : 0x5

So on for most things I get an answer of tRFC=107, which means 160ns if my math is any good. And that matches the SPD info from decode-dimms on the x220. The second channel for the x220 has something different but it actually looks like the reset value for the register which makes sense since decode-dimms says it only has one DIMM.

No idea if my tRFC is particularly low or high. The latency values are quite different between the ivb and snb machines however. Also I must admit that I've never tried plugging in a 25x14 monitor to these machines, and in fact with the dell xps I can't since it doesn't even have a DP connector.

Signature · 2 lines
Ville Syrjälä
Intel OTC
#6 lore raw

PATCH v2 1/1 drm/i915: Increase WM memory latency values on SNB

In reply to Daniel Vetter

From: Ville Syrjälä <[email protected]>

On SNB the BIOS provided WM memory latency values seem insufficient to handle high resolution displays.

In this particular case the display mode was a 2560x1440@60Hz, which makes the pixel clock 241.5 MHz. It was empirically found that a memory latency value if 1.2 usec is enough to avoid underruns, whereas the BIOS provided value of 0.7 usec was clearly too low. Incidentally 1.2 usec is what the typical BIOS provided values are on IVB systems.

Increase the WM memory latency values to at least 1.2 usec on SNB. Hopefully this won't have a significant effect on power consumption.

v2: Increase the latency values regardless of the pixel clock

Cc: Robert N <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70254

---
 drivers/gpu/drm/i915/intel_pm.c | 40 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)
drivers/gpu/drm/i915/intel_pm.c +40 -0
index 834c49c..2809365 100644
@@ -2113,6 +2113,43 @@ static void intel_print_wm_latency(struct drm_device *dev,
 	}
 }
 
+static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
+				    uint16_t wm[5], uint16_t min)
+{
+	int level, max_level = ilk_wm_max_level(dev_priv->dev);
+
+	if (wm[0] >= min)
+		return false;
+
+	wm[0] = max(wm[0], min);
+	for (level = 1; level <= max_level; level++)
+		wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
+
+	return true;
+}
+
+static void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	/*
+	 * The BIOS provided WM memory latency values are often
+	 * inadequate for high resolution displays. Adjust them.
+	 */
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
+
+	if (!changed)
+		return;
+
+	DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
+	intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
+	intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
+	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+}
+
 static void ilk_setup_wm_latency(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
@@ -2130,6 +2167,9 @@ static void ilk_setup_wm_latency(struct drm_device *dev)
 	intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
 	intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
 	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+
+	if (IS_GEN6(dev))
+		snb_wm_latency_quirk(dev);
 }
 
 static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
Signature · 6 lines
1.8.3.2

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#7 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values on SNB

Thanks for this Ville.

Should this apply to 3.14 and 3.15?

I'll try it on 3.15 first and report back.

#8 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values?on SNB

Quoted 5 lines · Robert Navarro

Thanks for this Ville.

Should this apply to 3.14 and 3.15?

I'll try it on 3.15 first and report back.

I think it should apply to 3.13+. If not directly then with a bit of manual frobbery. Which reminds me that we should perhaps slap a cc stable on it to get it included in 3.13+. For older kernels the patch would have to look totally different, so I'm not going to bother about those.

Signature · 2 lines
Ville Syrjälä
Intel OTC
#9 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values?on SNB

Quoted 6 lines · Ville Syrjälä <ville.syrjala <at> linux.intel.com> writes:

I think it should apply to 3.13+. If not directly then with a bit of manual frobbery. Which reminds me that we should perhaps slap a cc stable on it to get it included in 3.13+. For older kernels the patch would have to look totally different, so I'm not going to bother about those.

Sounds good, one more question. Does this replace the previous patch or are they used together?

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#10 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values?on SNB

Quoted 11 lines · Robert Navarro
Quoted 6 lines · Ville Syrjälä <ville.syrjala <at> linux.intel.com> writes:

I think it should apply to 3.13+. If not directly then with a bit of manual frobbery. Which reminds me that we should perhaps slap a cc stable on it to get it included in 3.13+. For older kernels the patch would have to look totally different, so I'm not going to bother about those.

Sounds good, one more question. Does this replace the previous patch or are they used together?

Replace.

Signature · 2 lines
Ville Syrjälä
Intel OTC
#11 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values?on SNB

Quoted 3 lines · Ville Syrjälä <ville.syrjala <at> linux.intel.com> writes:

Replace.

Finally got around to compiling this for my system, there were a few issues with the build scripts on the latest Ubuntu.

Currently running 3.15.0-rc3-custom-drm-intel-nightly-bug70254+ with no issues thus far.

I'll give a few more days just to make sure.

Thanks for all your hard work so far!

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#12 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values?on SNB

Quoted 7 lines · Robert Navarro <crshman <at> gmail.com> writes:

Finally got around to compiling this for my system, there were a few issues with the build scripts on the latest Ubuntu.

Currently running 3.15.0-rc3-custom-drm-intel-nightly-bug70254+ with no issues thus far.

Quoted 9 lines

I'll give a few more days just to make sure.

Thanks for all your hard work so far!

_______________________________________________
Intel-gfx mailing list
Intel-gfx <at> lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Just reporting back, I see you've included the fix, thanks!

A few days (and really hot temperature days for the laptop) not a single flicker or anything.

Thanks again for everyone's hard work in making this happen!

#13 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values on SNB

In reply to [email protected]
Quoted 32 lines · On Thu, 08 May 2014, [email protected] wrote:

From: Ville Syrjälä <[email protected]>

On SNB the BIOS provided WM memory latency values seem insufficient to handle high resolution displays.

In this particular case the display mode was a 2560x1440@60Hz, which makes the pixel clock 241.5 MHz. It was empirically found that a memory latency value if 1.2 usec is enough to avoid underruns, whereas the BIOS provided value of 0.7 usec was clearly too low. Incidentally 1.2 usec is what the typical BIOS provided values are on IVB systems.

Increase the WM memory latency values to at least 1.2 usec on SNB. Hopefully this won't have a significant effect on power consumption.

v2: Increase the latency values regardless of the pixel clock

Cc: Robert N <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70254

---
 drivers/gpu/drm/i915/intel_pm.c | 40 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)
drivers/gpu/drm/i915/intel_pm.c +1 -0
index 834c49c..2809365 100644
@@ -2113,6 +2113,43 @@ static void intel_print_wm_latency(struct drm_device *dev,
 	}
 }
 
+static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,

drivers/gpu/drm/i915/intel_pm.c static void intel_print_wm_latency(struct drm_device *dev, · line 2126

Quoted 10 lines
drivers/gpu/drm/i915/intel_pm.c +10 -0
+				    uint16_t wm[5], uint16_t min)
+{
+	int level, max_level = ilk_wm_max_level(dev_priv->dev);
+
+	if (wm[0] >= min)
+		return false;
+
+	wm[0] = max(wm[0], min);
+	for (level = 1; level <= max_level; level++)
+		wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));

DIV_ROUND_UP(min, 5)?

I don't get the units, and the code vs. comments in the wm code aren't helping. Please clarify.

Quoted 6 lines
drivers/gpu/drm/i915/intel_pm.c +6 -0
+
+	return true;
+}
+
+static void snb_wm_latency_quirk(struct drm_device *dev)
+{

drivers/gpu/drm/i915/intel_pm.c static void intel_print_wm_latency(struct drm_device *dev, · line 2142

Quoted 10 lines
drivers/gpu/drm/i915/intel_pm.c +10 -0
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	/*
+	 * The BIOS provided WM memory latency values are often
+	 * inadequate for high resolution displays. Adjust them.
+	 */
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

Nitpick, s/|/||/g for bools.

BR,
Jani.

Quoted 26 lines
drivers/gpu/drm/i915/intel_pm.c +13 -0
+
+	if (!changed)
+		return;
+
+	DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
+	intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
+	intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
+	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+}
+
 static void ilk_setup_wm_latency(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
@@ -2130,6 +2167,9 @@ static void ilk_setup_wm_latency(struct drm_device *dev)
 	intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
 	intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
 	intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
+
+	if (IS_GEN6(dev))
+		snb_wm_latency_quirk(dev);
 }
 
 static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
Signature · 1 line
1.8.3.2
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#14 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values on SNB

Quoted 16 lines · Jani Nikula

drivers/gpu/drm/i915/intel_pm.c static void intel_print_wm_latency(struct drm_device *dev, · line 2142

Quoted 13 lines · On Thu, 08 May 2014, [email protected] wrote:

From: Ville Syrjälä <[email protected]>

drivers/gpu/drm/i915/intel_pm.c +12 -0
+static void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	/*
+	 * The BIOS provided WM memory latency values are often
+	 * inadequate for high resolution displays. Adjust them.
+	 */
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

Nitpick, s/|/||/g for bools.

Consider side effects.
-Chris

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

Re: [PATCH v2] drm/i915: Increase WM memory latency values on SNB

Quoted 18 lines · Chris Wilson
Quoted 16 lines · Jani Nikula

drivers/gpu/drm/i915/intel_pm.c static void intel_print_wm_latency(struct drm_device *dev, · line 2142

Quoted 13 lines · On Thu, 08 May 2014, [email protected] wrote:

From: Ville Syrjälä <[email protected]>

drivers/gpu/drm/i915/intel_pm.c +12 -0
+static void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	/*
+	 * The BIOS provided WM memory latency values are often
+	 * inadequate for high resolution displays. Adjust them.
+	 */
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

Nitpick, s/|/||/g for bools.

Quoted 1 line

Consider side effects.

Ugh I'm slow today. Some might claim business as usual. I'll hide somewhere.

Before I head under the rock, may I say bitops on bools are still ugly?

BR,
Jani.

Quoted 4 lines

-Chris

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

Re: [PATCH v2] drm/i915: Increase WM memory latency values on SNB

Quoted 19 lines · Jani Nikula
Quoted 18 lines · Chris Wilson
Quoted 16 lines · Jani Nikula

drivers/gpu/drm/i915/intel_pm.c static void intel_print_wm_latency(struct drm_device *dev, · line 2142

Quoted 13 lines · On Thu, 08 May 2014, [email protected] wrote:

From: Ville Syrjälä <[email protected]>

drivers/gpu/drm/i915/intel_pm.c +12 -0
+static void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	/*
+	 * The BIOS provided WM memory latency values are often
+	 * inadequate for high resolution displays. Adjust them.
+	 */
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

Nitpick, s/|/||/g for bools.

Quoted 6 lines
Quoted 1 line

Consider side effects.

Ugh I'm slow today. Some might claim business as usual. I'll hide somewhere.

Before I head under the rock, may I say bitops on bools are still ugly?

I tend to use them but Paulo was also confused by them somewhere else in the watermark code, so maybe I should stop using them?

I can rewrite as:

changed |= ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12);
changed |= ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12);
changed |= ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

or just 3x
if (ilk_increase_wm_latency(...))

	changed = true;

if that helps.

Signature · 2 lines
Ville Syrjälä
Intel OTC
#17 lore raw

Re: [PATCH v2] drm/i915: Increase WM memory latency values on SNB

Quoted 35 lines · Ville Syrjälä
Quoted 19 lines · Jani Nikula
Quoted 18 lines · Chris Wilson
Quoted 16 lines · Jani Nikula

drivers/gpu/drm/i915/intel_pm.c static void intel_print_wm_latency(struct drm_device *dev, · line 2142

Quoted 13 lines · On Thu, 08 May 2014, [email protected] wrote:

From: Ville Syrjälä <[email protected]>

drivers/gpu/drm/i915/intel_pm.c +12 -0
+static void snb_wm_latency_quirk(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	bool changed;
+
+	/*
+	 * The BIOS provided WM memory latency values are often
+	 * inadequate for high resolution displays. Adjust them.
+	 */
+	changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
+		ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);

Nitpick, s/|/||/g for bools.

Quoted 6 lines
Quoted 1 line

Consider side effects.

Ugh I'm slow today. Some might claim business as usual. I'll hide somewhere.

Before I head under the rock, may I say bitops on bools are still ugly?

I tend to use them but Paulo was also confused by them somewhere else in the watermark code, so maybe I should stop using them?

I can rewrite as:

changed |= ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12);
changed |= ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12);
changed |= ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
Quoted 5 lines

or just 3x
if (ilk_increase_wm_latency(...))

	changed = true;

if that helps.

Too late, patch pushed to -fixes, thanks for the patch and ridicule. ;) You can send cleanups later for -next if you like.

BR,
Jani.

Quoted 4 lines
Signature · 3 lines
-- 
Ville Syrjälä
Intel OTC
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx