Skip to contentSkip to conversation

mfd: intel-lpss: use devm_ioremap_uc for mmio

PATCH Superseded Tuowen Zhao 4 participants 5 messages Sep 27, 2019 – Oct 8, 2019

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mfd: intel-lpss: use devm_ioremap_uc for mmio

Write-combining BAR for intel-lpss-pci in MTRR causes system hangs during boot.

This patch adds devm_ioremap_uc as a new managed wrapper to ioremap_uc and with it forces the use of strongly uncachable mmio in intel-lpss.

This bahavior is seen on Dell XPS 13 7390 2-in-1:

[    0.001734]   5 base 4000000000 mask 6000000000 write-combining

4000000000-7fffffffff : PCI Bus 0000:00
4000000000-400fffffff : 0000:00:02.0 (i915) 4010000000-4010000fff : 0000:00:15.0 (intel-lpss-pci)

---
 drivers/mfd/intel-lpss.c |  2 +-
 include/linux/io.h       |  2 ++
 lib/devres.c             | 19 +++++++++++++++++++
 3 files changed, 22 insertions(+), 1 deletion(-)
drivers/mfd/intel-lpss.c +1 -1
index 277f48f1cc1c..06106c9320bb 100644
@@ -395,7 +395,7 @@ int intel_lpss_probe(struct device *dev,
 	if (!lpss)
 		return -ENOMEM;
 
-	lpss->priv = devm_ioremap(dev, info->mem->start + LPSS_PRIV_OFFSET,
+	lpss->priv = devm_ioremap_uc(dev, info->mem->start + LPSS_PRIV_OFFSET,
 				  LPSS_PRIV_SIZE);
 	if (!lpss->priv)
 		return -ENOMEM;
include/linux/io.h +2 -0
index accac822336a..a59834bc0a11 100644
@@ -64,6 +64,8 @@ static inline void devm_ioport_unmap(struct device *dev, void __iomem *addr)
 
 void __iomem *devm_ioremap(struct device *dev, resource_size_t offset,
 			   resource_size_t size);
+void __iomem *devm_ioremap_uc(struct device *dev, resource_size_t offset,
+				   resource_size_t size);
 void __iomem *devm_ioremap_nocache(struct device *dev, resource_size_t offset,
 				   resource_size_t size);
 void __iomem *devm_ioremap_wc(struct device *dev, resource_size_t offset,
lib/devres.c +19 -0
index 6a0e9bd6524a..beb0a064b891 100644
@@ -9,6 +9,7 @@
 enum devm_ioremap_type {
 	DEVM_IOREMAP = 0,
 	DEVM_IOREMAP_NC,
+	DEVM_IOREMAP_UC,
 	DEVM_IOREMAP_WC,
 };
 
@@ -39,6 +40,9 @@ static void __iomem *__devm_ioremap(struct device *dev, resource_size_t offset,
 	case DEVM_IOREMAP_NC:
 		addr = ioremap_nocache(offset, size);
 		break;
+	case DEVM_IOREMAP_UC:
+		addr = ioremap_uc(offset, size);
+		break;
 	case DEVM_IOREMAP_WC:
 		addr = ioremap_wc(offset, size);
 		break;
@@ -68,6 +72,21 @@ void __iomem *devm_ioremap(struct device *dev, resource_size_t offset,
 }
 EXPORT_SYMBOL(devm_ioremap);
 
+/**
+ * devm_ioremap_uc - Managed ioremap_uc()
+ * @dev: Generic device to remap IO address for
+ * @offset: Resource address to map
+ * @size: Size of map
+ *
+ * Managed ioremap_uc().  Map is automatically unmapped on driver detach.
+ */
+void __iomem *devm_ioremap_uc(struct device *dev, resource_size_t offset,
+			      resource_size_t size)
+{
+	return __devm_ioremap(dev, offset, size, DEVM_IOREMAP_UC);
+}
+EXPORT_SYMBOL(devm_ioremap_uc);
+
 /**
  * devm_ioremap_nocache - Managed ioremap_nocache()
  * @dev: Generic device to remap IO address for
Signature · 1 line
2.23.0
#1 lore raw

Hi,

In $subject, use MMIO instead of mmio.

Quoted 9 lines · Tuowen Zhao

Write-combining BAR for intel-lpss-pci in MTRR causes system hangs during boot.

This patch adds devm_ioremap_uc as a new managed wrapper to ioremap_uc and with it forces the use of strongly uncachable mmio in intel-lpss.

This bahavior is seen on Dell XPS 13 7390 2-in-1:

[    0.001734]   5 base 4000000000 mask 6000000000 write-combining

I think it is worth mentioning that this is actually a BIOS bug and fixed by some vendors via BIOS upgrade.

Quoted 17 lines

4000000000-7fffffffff : PCI Bus 0000:00
4000000000-400fffffff : 0000:00:02.0 (i915) 4010000000-4010000fff : 0000:00:15.0 (intel-lpss-pci)

---
 drivers/mfd/intel-lpss.c |  2 +-
 include/linux/io.h       |  2 ++
 lib/devres.c             | 19 +++++++++++++++++++
 3 files changed, 22 insertions(+), 1 deletion(-)
drivers/mfd/intel-lpss.c +0 -0
index 277f48f1cc1c..06106c9320bb 100644

drivers/mfd/intel-lpss.c int intel_lpss_probe(struct device *dev, · line 399

Quoted 6 lines
drivers/mfd/intel-lpss.c +1 -1
@@ -395,7 +395,7 @@ int intel_lpss_probe(struct device *dev,
 	if (!lpss)
 		return -ENOMEM;
 
-	lpss->priv = devm_ioremap(dev, info->mem->start + LPSS_PRIV_OFFSET,
+	lpss->priv = devm_ioremap_uc(dev, info->mem->start + LPSS_PRIV_OFFSET,
 				  LPSS_PRIV_SIZE);

Can you add a comment here explaining why this particular driver needs to use _uc version? Normally drivers call the plain devm_ioremap() and expect to get non-cached memory, however ioremap() (which resolves to ioremap_nocache()) does not touch x86 PAT configuration which is the reason we need to call _uc variant here.

Otherwise looks good to me.

#2 lore raw
In reply to Tuowen Zhao
Quoted 6 lines · Tuowen Zhao

Write-combining BAR for intel-lpss-pci in MTRR causes system hangs during boot.

This patch adds devm_ioremap_uc as a new managed wrapper to ioremap_uc and with it forces the use of strongly uncachable mmio in intel-lpss.

Quoted 7 lines

This bahavior is seen on Dell XPS 13 7390 2-in-1:

[    0.001734]   5 base 4000000000 mask 6000000000 write-combining

4000000000-7fffffffff : PCI Bus 0000:00
4000000000-400fffffff : 0000:00:02.0 (i915) 4010000000-4010000fff : 0000:00:15.0 (intel-lpss-pci)

+Cc: Luis as author of UC flavour of ioremap.

Luis, some BIOSes in the wild have wrong MTRR setting for PCI resource window and thus when Linux tries to allocate 64-bit MMIO address space (and in opposite to Windows, which does this from the end of available space towards beginning, Linux do this from the beginning towards end). Ideally we have to push vendors to fix firmware.

This patch AFAIU overrides MTTR/PAT settings for those pages and makes it possible to workaround firmware bug.

What do you think is the best approach here?

Quoted 81 lines
---
 drivers/mfd/intel-lpss.c |  2 +-
 include/linux/io.h       |  2 ++
 lib/devres.c             | 19 +++++++++++++++++++
 3 files changed, 22 insertions(+), 1 deletion(-)
drivers/mfd/intel-lpss.c +1 -1
index 277f48f1cc1c..06106c9320bb 100644
@@ -395,7 +395,7 @@ int intel_lpss_probe(struct device *dev,
 	if (!lpss)
 		return -ENOMEM;
 
-	lpss->priv = devm_ioremap(dev, info->mem->start + LPSS_PRIV_OFFSET,
+	lpss->priv = devm_ioremap_uc(dev, info->mem->start + LPSS_PRIV_OFFSET,
 				  LPSS_PRIV_SIZE);
 	if (!lpss->priv)
 		return -ENOMEM;
include/linux/io.h +2 -0
index accac822336a..a59834bc0a11 100644
@@ -64,6 +64,8 @@ static inline void devm_ioport_unmap(struct device *dev, void __iomem *addr)
 
 void __iomem *devm_ioremap(struct device *dev, resource_size_t offset,
 			   resource_size_t size);
+void __iomem *devm_ioremap_uc(struct device *dev, resource_size_t offset,
+				   resource_size_t size);
 void __iomem *devm_ioremap_nocache(struct device *dev, resource_size_t offset,
 				   resource_size_t size);
 void __iomem *devm_ioremap_wc(struct device *dev, resource_size_t offset,
lib/devres.c +19 -0
index 6a0e9bd6524a..beb0a064b891 100644
@@ -9,6 +9,7 @@
 enum devm_ioremap_type {
 	DEVM_IOREMAP = 0,
 	DEVM_IOREMAP_NC,
+	DEVM_IOREMAP_UC,
 	DEVM_IOREMAP_WC,
 };
 
@@ -39,6 +40,9 @@ static void __iomem *__devm_ioremap(struct device *dev, resource_size_t offset,
 	case DEVM_IOREMAP_NC:
 		addr = ioremap_nocache(offset, size);
 		break;
+	case DEVM_IOREMAP_UC:
+		addr = ioremap_uc(offset, size);
+		break;
 	case DEVM_IOREMAP_WC:
 		addr = ioremap_wc(offset, size);
 		break;
@@ -68,6 +72,21 @@ void __iomem *devm_ioremap(struct device *dev, resource_size_t offset,
 }
 EXPORT_SYMBOL(devm_ioremap);
 
+/**
+ * devm_ioremap_uc - Managed ioremap_uc()
+ * @dev: Generic device to remap IO address for
+ * @offset: Resource address to map
+ * @size: Size of map
+ *
+ * Managed ioremap_uc().  Map is automatically unmapped on driver detach.
+ */
+void __iomem *devm_ioremap_uc(struct device *dev, resource_size_t offset,
+			      resource_size_t size)
+{
+	return __devm_ioremap(dev, offset, size, DEVM_IOREMAP_UC);
+}
+EXPORT_SYMBOL(devm_ioremap_uc);
+
 /**
  * devm_ioremap_nocache - Managed ioremap_nocache()
  * @dev: Generic device to remap IO address for
Signature · 1 line
2.23.0
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#3 lore raw
Quoted 23 lines · Andy Shevchenko
Quoted 6 lines · Tuowen Zhao

Write-combining BAR for intel-lpss-pci in MTRR causes system hangs during boot.

This patch adds devm_ioremap_uc as a new managed wrapper to ioremap_uc and with it forces the use of strongly uncachable mmio in intel-lpss.

Quoted 7 lines

This bahavior is seen on Dell XPS 13 7390 2-in-1:

[    0.001734]   5 base 4000000000 mask 6000000000 write-combining

4000000000-7fffffffff : PCI Bus 0000:00
4000000000-400fffffff : 0000:00:02.0 (i915) 4010000000-4010000fff : 0000:00:15.0 (intel-lpss-pci)

+Cc: Luis as author of UC flavour of ioremap.

Luis, some BIOSes in the wild have wrong MTRR setting for PCI resource window and thus when Linux tries to allocate 64-bit MMIO address space (and in opposite to Windows, which does this from the end of available space towards beginning, Linux do this from the beginning towards end). Ideally we have to push vendors to fix firmware.

Quoted 4 lines

This patch AFAIU overrides MTTR/PAT settings for those pages and makes it possible to workaround firmware bug.

What do you think is the best approach here?

Tuowen,

since Luis didn't respond, I think we may proceed with v2 after addressing Mika's comments.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#4 lore raw
In reply to Andy Shevchenko
Quoted 23 lines · Andy Shevchenko
Quoted 6 lines · Tuowen Zhao

Write-combining BAR for intel-lpss-pci in MTRR causes system hangs during boot.

This patch adds devm_ioremap_uc as a new managed wrapper to ioremap_uc and with it forces the use of strongly uncachable mmio in intel-lpss.

Quoted 7 lines

This bahavior is seen on Dell XPS 13 7390 2-in-1:

[    0.001734]   5 base 4000000000 mask 6000000000 write-combining

4000000000-7fffffffff : PCI Bus 0000:00
4000000000-400fffffff : 0000:00:02.0 (i915) 4010000000-4010000fff : 0000:00:15.0 (intel-lpss-pci)

+Cc: Luis as author of UC flavour of ioremap.

Luis, some BIOSes in the wild have wrong MTRR setting for PCI resource window and thus when Linux tries to allocate 64-bit MMIO address space (and in opposite to Windows, which does this from the end of available space towards beginning, Linux do this from the beginning towards end). Ideally we have to push vendors to fix firmware.

Quoted 4 lines

This patch AFAIU overrides MTTR/PAT settings for those pages and makes it possible to workaround firmware bug.

What do you think is the best approach here?

Indeed, such cases can come up, and yes _uc can be a workaround for such cases.

Quoted 1 line
Quoted 1 line

+EXPORT_SYMBOL(devm_ioremap_uc);

EXPORT_SYMBOL_GPL() would be my preference. But other than that, this makes sense.

Luis