Skip to contentSkip to conversation

lspci: Determine bridge window address size from bridge_flags

PATCH Applied David Matlack 3 participants 8 messages Aug 21 – Aug 24 PCI

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 lspci: Determine bridge window address size from bridge_flags

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

---
v2:
- Fix lspci.c to use bridge_flags[], per lib/pci.h, rather than changing the behavior of just the sysfs backend (Pali Rohár)

v1: https://lore.kernel.org/linux-pci/[email protected]/

 lspci.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
lspci.c +5 -7
index 2a14303e74f6..52fd9247aefb 100644
@@ -571,10 +571,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !io_disabled)
     {
-      io_base = p->bridge_base_addr[0] & PCI_IO_RANGE_MASK;
+      io_base = p->bridge_base_addr[0];
       io_limit = io_base + p->bridge_size[0] - 1;
-      io_type = p->bridge_base_addr[0] & PCI_IO_RANGE_TYPE_MASK;
-      io_bits = (io_type == PCI_IO_RANGE_TYPE_32) ? 32 : 16;
+      io_bits = (p->bridge_flags[0] & PCI_IORESOURCE_IO_16BIT_ADDR) ? 16 : 32;
       show_range("\tI/O behind bridge", io_base, io_limit, io_bits, io_disabled);
     }
   else if (io_type != (io_limit & PCI_IO_RANGE_TYPE_MASK) ||
@@ -600,7 +599,7 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !mem_disabled)
     {
-      mem_base = p->bridge_base_addr[1] & PCI_MEMORY_RANGE_MASK;
+      mem_base = p->bridge_base_addr[1];
       mem_limit = mem_base + p->bridge_size[1] - 1;
       show_range("\tMemory behind bridge", mem_base, mem_limit, 32, mem_disabled);
     }
@@ -616,10 +615,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !pref_disabled)
     {
-      u64 pref_base_64 = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_MASK;
+      u64 pref_base_64 = p->bridge_base_addr[2];
       u64 pref_limit_64 = pref_base_64 + p->bridge_size[2] - 1;
-      pref_type = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_TYPE_MASK;
-      pref_bits = (pref_type == PCI_PREF_RANGE_TYPE_64) ? 64 : 32;
+      pref_bits = (p->bridge_flags[2] & PCI_IORESOURCE_MEM_64) ? 64 : 32;
       show_range("\tPrefetchable memory behind bridge", pref_base_64, pref_limit_64, pref_bits, pref_disabled);
     }
   else if (pref_type != (pref_limit & PCI_PREF_RANGE_TYPE_MASK) ||

base-commit: b41ce14da749fb44ca7940ba9797027bfa62c23e

Signature · 1 line
2.55.0.766.g2966f0265a-goog
#1 lore raw

Hello! Thank you for taking this issue. I briefly checked the change and I think that this is the right way how to address that issue.

Anyway, it would be nice to provide some example of config space of PCI Bridge (lspci -x) which is affected by this issue and have it in pciutils/tests/ directory.

Quoted 73 lines · On Friday 21 August 2026 22:15:59 David Matlack wrote:

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

---
v2:
- Fix lspci.c to use bridge_flags[], per lib/pci.h, rather than changing the behavior of just the sysfs backend (Pali Rohár)

v1: https://lore.kernel.org/linux-pci/[email protected]/

 lspci.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
lspci.c +5 -7
index 2a14303e74f6..52fd9247aefb 100644
@@ -571,10 +571,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !io_disabled)
     {
-      io_base = p->bridge_base_addr[0] & PCI_IO_RANGE_MASK;
+      io_base = p->bridge_base_addr[0];
       io_limit = io_base + p->bridge_size[0] - 1;
-      io_type = p->bridge_base_addr[0] & PCI_IO_RANGE_TYPE_MASK;
-      io_bits = (io_type == PCI_IO_RANGE_TYPE_32) ? 32 : 16;
+      io_bits = (p->bridge_flags[0] & PCI_IORESOURCE_IO_16BIT_ADDR) ? 16 : 32;
       show_range("\tI/O behind bridge", io_base, io_limit, io_bits, io_disabled);
     }
   else if (io_type != (io_limit & PCI_IO_RANGE_TYPE_MASK) ||
@@ -600,7 +599,7 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !mem_disabled)
     {
-      mem_base = p->bridge_base_addr[1] & PCI_MEMORY_RANGE_MASK;
+      mem_base = p->bridge_base_addr[1];
       mem_limit = mem_base + p->bridge_size[1] - 1;
       show_range("\tMemory behind bridge", mem_base, mem_limit, 32, mem_disabled);
     }
@@ -616,10 +615,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !pref_disabled)
     {
-      u64 pref_base_64 = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_MASK;
+      u64 pref_base_64 = p->bridge_base_addr[2];
       u64 pref_limit_64 = pref_base_64 + p->bridge_size[2] - 1;
-      pref_type = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_TYPE_MASK;
-      pref_bits = (pref_type == PCI_PREF_RANGE_TYPE_64) ? 64 : 32;
+      pref_bits = (p->bridge_flags[2] & PCI_IORESOURCE_MEM_64) ? 64 : 32;
       show_range("\tPrefetchable memory behind bridge", pref_base_64, pref_limit_64, pref_bits, pref_disabled);
     }
   else if (pref_type != (pref_limit & PCI_PREF_RANGE_TYPE_MASK) ||

base-commit: b41ce14da749fb44ca7940ba9797027bfa62c23e

Signature · 1 line
2.55.0.766.g2966f0265a-goog
#2 lore raw
Quoted 7 lines · Pali Rohár

Hello! Thank you for taking this issue. I briefly checked the change and I think that this is the right way how to address that issue.

Anyway, it would be nice to provide some example of config space of PCI Bridge (lspci -x) which is affected by this issue and have it in pciutils/tests/ directory.

Ack, I will incorporate that into a v3.

Quoted 75 lines · On Friday 21 August 2026 22:15:59 David Matlack wrote:
Quoted 73 lines

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

---
v2:
- Fix lspci.c to use bridge_flags[], per lib/pci.h, rather than changing the behavior of just the sysfs backend (Pali Rohár)

v1: https://lore.kernel.org/linux-pci/[email protected]/

 lspci.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
lspci.c +5 -7
index 2a14303e74f6..52fd9247aefb 100644
@@ -571,10 +571,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !io_disabled)
     {
-      io_base = p->bridge_base_addr[0] & PCI_IO_RANGE_MASK;
+      io_base = p->bridge_base_addr[0];
       io_limit = io_base + p->bridge_size[0] - 1;
-      io_type = p->bridge_base_addr[0] & PCI_IO_RANGE_TYPE_MASK;
-      io_bits = (io_type == PCI_IO_RANGE_TYPE_32) ? 32 : 16;
+      io_bits = (p->bridge_flags[0] & PCI_IORESOURCE_IO_16BIT_ADDR) ? 16 : 32;
       show_range("\tI/O behind bridge", io_base, io_limit, io_bits, io_disabled);
     }
   else if (io_type != (io_limit & PCI_IO_RANGE_TYPE_MASK) ||
@@ -600,7 +599,7 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !mem_disabled)
     {
-      mem_base = p->bridge_base_addr[1] & PCI_MEMORY_RANGE_MASK;
+      mem_base = p->bridge_base_addr[1];
       mem_limit = mem_base + p->bridge_size[1] - 1;
       show_range("\tMemory behind bridge", mem_base, mem_limit, 32, mem_disabled);
     }
@@ -616,10 +615,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !pref_disabled)
     {
-      u64 pref_base_64 = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_MASK;
+      u64 pref_base_64 = p->bridge_base_addr[2];
       u64 pref_limit_64 = pref_base_64 + p->bridge_size[2] - 1;
-      pref_type = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_TYPE_MASK;
-      pref_bits = (pref_type == PCI_PREF_RANGE_TYPE_64) ? 64 : 32;
+      pref_bits = (p->bridge_flags[2] & PCI_IORESOURCE_MEM_64) ? 64 : 32;
       show_range("\tPrefetchable memory behind bridge", pref_base_64, pref_limit_64, pref_bits, pref_disabled);
     }
   else if (pref_type != (pref_limit & PCI_PREF_RANGE_TYPE_MASK) ||

base-commit: b41ce14da749fb44ca7940ba9797027bfa62c23e --
2.55.0.766.g2966f0265a-goog

#3 lore raw
Quoted 10 lines · On Monday 24 August 2026 09:24:33 David Matlack wrote:
Quoted 7 lines · Pali Rohár

Hello! Thank you for taking this issue. I briefly checked the change and I think that this is the right way how to address that issue.

Anyway, it would be nice to provide some example of config space of PCI Bridge (lspci -x) which is affected by this issue and have it in pciutils/tests/ directory.

Ack, I will incorporate that into a v3.

Feel free to send it separately. No need to have it in this one change.

Quoted 75 lines
Quoted 75 lines · On Friday 21 August 2026 22:15:59 David Matlack wrote:
Quoted 73 lines

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

---
v2:
- Fix lspci.c to use bridge_flags[], per lib/pci.h, rather than changing the behavior of just the sysfs backend (Pali Rohár)

v1: https://lore.kernel.org/linux-pci/[email protected]/

 lspci.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
lspci.c +5 -7
index 2a14303e74f6..52fd9247aefb 100644
@@ -571,10 +571,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !io_disabled)
     {
-      io_base = p->bridge_base_addr[0] & PCI_IO_RANGE_MASK;
+      io_base = p->bridge_base_addr[0];
       io_limit = io_base + p->bridge_size[0] - 1;
-      io_type = p->bridge_base_addr[0] & PCI_IO_RANGE_TYPE_MASK;
-      io_bits = (io_type == PCI_IO_RANGE_TYPE_32) ? 32 : 16;
+      io_bits = (p->bridge_flags[0] & PCI_IORESOURCE_IO_16BIT_ADDR) ? 16 : 32;
       show_range("\tI/O behind bridge", io_base, io_limit, io_bits, io_disabled);
     }
   else if (io_type != (io_limit & PCI_IO_RANGE_TYPE_MASK) ||
@@ -600,7 +599,7 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !mem_disabled)
     {
-      mem_base = p->bridge_base_addr[1] & PCI_MEMORY_RANGE_MASK;
+      mem_base = p->bridge_base_addr[1];
       mem_limit = mem_base + p->bridge_size[1] - 1;
       show_range("\tMemory behind bridge", mem_base, mem_limit, 32, mem_disabled);
     }
@@ -616,10 +615,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !pref_disabled)
     {
-      u64 pref_base_64 = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_MASK;
+      u64 pref_base_64 = p->bridge_base_addr[2];
       u64 pref_limit_64 = pref_base_64 + p->bridge_size[2] - 1;
-      pref_type = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_TYPE_MASK;
-      pref_bits = (pref_type == PCI_PREF_RANGE_TYPE_64) ? 64 : 32;
+      pref_bits = (p->bridge_flags[2] & PCI_IORESOURCE_MEM_64) ? 64 : 32;
       show_range("\tPrefetchable memory behind bridge", pref_base_64, pref_limit_64, pref_bits, pref_disabled);
     }
   else if (pref_type != (pref_limit & PCI_PREF_RANGE_TYPE_MASK) ||

base-commit: b41ce14da749fb44ca7940ba9797027bfa62c23e --
2.55.0.766.g2966f0265a-goog

#4 lore raw
Quoted 7 lines · Pali Rohár
Quoted 5 lines · On Monday 24 August 2026 09:24:33 David Matlack wrote:
Quoted 4 lines · Pali Rohár

Hello! Thank you for taking this issue. I briefly checked the change and I think that this is the right way how to address that issue.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Anyway, it would be nice to provide some example of config space of PCI Bridge (lspci -x) which is affected by this issue and have it in pciutils/tests/ directory.

Ack, I will incorporate that into a v3.

Feel free to send it separately. No need to have it in this one change.

Sure, I can send a 2 patch series in v3 if that's what you mean?

By the way, since lspci parsing of the pciutils/tests/ directory would use the dump backend, which does not use PCI_FILL_BRIDGE_BASES, it will not exercise the bug fixed by this commit. Is it still worth adding the test case? I am happy to send it, but I'm curious how it will be used.

Quoted 76 lines
Quoted 76 lines
Quoted 76 lines · On Friday 21 August 2026 22:15:59 David Matlack wrote:
Quoted 73 lines

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

---
v2:
- Fix lspci.c to use bridge_flags[], per lib/pci.h, rather than changing the behavior of just the sysfs backend (Pali Rohár)

v1: https://lore.kernel.org/linux-pci/[email protected]/

 lspci.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
lspci.c +5 -7
index 2a14303e74f6..52fd9247aefb 100644
@@ -571,10 +571,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !io_disabled)
     {
-      io_base = p->bridge_base_addr[0] & PCI_IO_RANGE_MASK;
+      io_base = p->bridge_base_addr[0];
       io_limit = io_base + p->bridge_size[0] - 1;
-      io_type = p->bridge_base_addr[0] & PCI_IO_RANGE_TYPE_MASK;
-      io_bits = (io_type == PCI_IO_RANGE_TYPE_32) ? 32 : 16;
+      io_bits = (p->bridge_flags[0] & PCI_IORESOURCE_IO_16BIT_ADDR) ? 16 : 32;
       show_range("\tI/O behind bridge", io_base, io_limit, io_bits, io_disabled);
     }
   else if (io_type != (io_limit & PCI_IO_RANGE_TYPE_MASK) ||
@@ -600,7 +599,7 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !mem_disabled)
     {
-      mem_base = p->bridge_base_addr[1] & PCI_MEMORY_RANGE_MASK;
+      mem_base = p->bridge_base_addr[1];
       mem_limit = mem_base + p->bridge_size[1] - 1;
       show_range("\tMemory behind bridge", mem_base, mem_limit, 32, mem_disabled);
     }
@@ -616,10 +615,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !pref_disabled)
     {
-      u64 pref_base_64 = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_MASK;
+      u64 pref_base_64 = p->bridge_base_addr[2];
       u64 pref_limit_64 = pref_base_64 + p->bridge_size[2] - 1;
-      pref_type = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_TYPE_MASK;
-      pref_bits = (pref_type == PCI_PREF_RANGE_TYPE_64) ? 64 : 32;
+      pref_bits = (p->bridge_flags[2] & PCI_IORESOURCE_MEM_64) ? 64 : 32;
       show_range("\tPrefetchable memory behind bridge", pref_base_64, pref_limit_64, pref_bits, pref_disabled);
     }
   else if (pref_type != (pref_limit & PCI_PREF_RANGE_TYPE_MASK) ||

base-commit: b41ce14da749fb44ca7940ba9797027bfa62c23e --
2.55.0.766.g2966f0265a-goog

#5 lore raw
Quoted 16 lines · On Monday 24 August 2026 13:22:32 David Matlack wrote:
Quoted 7 lines · Pali Rohár
Quoted 5 lines · On Monday 24 August 2026 09:24:33 David Matlack wrote:
Quoted 4 lines · Pali Rohár

Hello! Thank you for taking this issue. I briefly checked the change and I think that this is the right way how to address that issue.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Anyway, it would be nice to provide some example of config space of PCI Bridge (lspci -x) which is affected by this issue and have it in pciutils/tests/ directory.

Ack, I will incorporate that into a v3.

Feel free to send it separately. No need to have it in this one change.

Quoted 7 lines

Sure, I can send a 2 patch series in v3 if that's what you mean?

By the way, since lspci parsing of the pciutils/tests/ directory would use the dump backend, which does not use PCI_FILL_BRIDGE_BASES, it will not exercise the bug fixed by this commit. Is it still worth adding the test case? I am happy to send it, but I'm curious how it will be used.

I see. Those flags cannot be read from the PCI config space and they need to be provided by OS when it is doing enumeration / setup of PCI. So lspci -x dump outputs will not provide them.

Quoted 76 lines
Quoted 76 lines
Quoted 76 lines
Quoted 76 lines · On Friday 21 August 2026 22:15:59 David Matlack wrote:
Quoted 73 lines

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

---
v2:
- Fix lspci.c to use bridge_flags[], per lib/pci.h, rather than changing the behavior of just the sysfs backend (Pali Rohár)

v1: https://lore.kernel.org/linux-pci/[email protected]/

 lspci.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
lspci.c +5 -7
index 2a14303e74f6..52fd9247aefb 100644
@@ -571,10 +571,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !io_disabled)
     {
-      io_base = p->bridge_base_addr[0] & PCI_IO_RANGE_MASK;
+      io_base = p->bridge_base_addr[0];
       io_limit = io_base + p->bridge_size[0] - 1;
-      io_type = p->bridge_base_addr[0] & PCI_IO_RANGE_TYPE_MASK;
-      io_bits = (io_type == PCI_IO_RANGE_TYPE_32) ? 32 : 16;
+      io_bits = (p->bridge_flags[0] & PCI_IORESOURCE_IO_16BIT_ADDR) ? 16 : 32;
       show_range("\tI/O behind bridge", io_base, io_limit, io_bits, io_disabled);
     }
   else if (io_type != (io_limit & PCI_IO_RANGE_TYPE_MASK) ||
@@ -600,7 +599,7 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !mem_disabled)
     {
-      mem_base = p->bridge_base_addr[1] & PCI_MEMORY_RANGE_MASK;
+      mem_base = p->bridge_base_addr[1];
       mem_limit = mem_base + p->bridge_size[1] - 1;
       show_range("\tMemory behind bridge", mem_base, mem_limit, 32, mem_disabled);
     }
@@ -616,10 +615,9 @@ show_htype1(struct device *d)
 
   if ((p->known_fields & PCI_FILL_BRIDGE_BASES) && !pref_disabled)
     {
-      u64 pref_base_64 = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_MASK;
+      u64 pref_base_64 = p->bridge_base_addr[2];
       u64 pref_limit_64 = pref_base_64 + p->bridge_size[2] - 1;
-      pref_type = p->bridge_base_addr[2] & PCI_MEMORY_RANGE_TYPE_MASK;
-      pref_bits = (pref_type == PCI_PREF_RANGE_TYPE_64) ? 64 : 32;
+      pref_bits = (p->bridge_flags[2] & PCI_IORESOURCE_MEM_64) ? 64 : 32;
       show_range("\tPrefetchable memory behind bridge", pref_base_64, pref_limit_64, pref_bits, pref_disabled);
     }
   else if (pref_type != (pref_limit & PCI_PREF_RANGE_TYPE_MASK) ||

base-commit: b41ce14da749fb44ca7940ba9797027bfa62c23e --
2.55.0.766.g2966f0265a-goog

#6 lore raw
Quoted 26 lines · Pali Rohár
Quoted 16 lines · On Monday 24 August 2026 13:22:32 David Matlack wrote:
Quoted 7 lines · Pali Rohár
Quoted 5 lines · On Monday 24 August 2026 09:24:33 David Matlack wrote:
Quoted 4 lines · Pali Rohár

Hello! Thank you for taking this issue. I briefly checked the change and I think that this is the right way how to address that issue.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Anyway, it would be nice to provide some example of config space of PCI Bridge (lspci -x) which is affected by this issue and have it in pciutils/tests/ directory.

Ack, I will incorporate that into a v3.

Feel free to send it separately. No need to have it in this one change.

Quoted 7 lines

Sure, I can send a 2 patch series in v3 if that's what you mean?

By the way, since lspci parsing of the pciutils/tests/ directory would use the dump backend, which does not use PCI_FILL_BRIDGE_BASES, it will not exercise the bug fixed by this commit. Is it still worth adding the test case? I am happy to send it, but I'm curious how it will be used.

Quoted 3 lines

I see. Those flags cannot be read from the PCI config space and they need to be provided by OS when it is doing enumeration / setup of PCI. So lspci -x dump outputs will not provide them.

lspci -x will provide enough information for lspci to determine the bit-size of bridge windows. But when PCI_FILL_BRIDGE_BASES is set, show_hypte1() ignores the flags in the config space and uses the OS-provided flags instead, and that's where the bug was. So when testing (./lspci -v -F tests/bridge-window-sizes), it outputted the correct bridge windows with and without the fix.

#7 lore raw
In reply to David Matlack

Hello!

Quoted 12 lines

Use bridge_flags to determine address size bits (16- vs 32-bit for I/O, 32- vs 64-bit for prefetchable memory) when displaying bridge ranges populated via PCI_FILL_BRIDGE_BASES.

Commit ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported") introduced code in show_htype1() that reads PCI_IO_RANGE_TYPE_MASK and PCI_MEMORY_RANGE_TYPE_MASK from bridge_base_addr[]. However, bridge_base_addr[] holds base addresses without flags, as documented in lib/pci.h and implemented across backends like sysfs.c and win32-cfgmgr32.c.

Quoted 7 lines

Because bridge_base_addr[] does not contain configuration register flag bits, masking the lower bits evaluates to zero. As a result, 32-bit I/O windows are mislabeled as "[16-bit]" and 64-bit prefetchable memory windows are mislabeled as "[32-bit]" when using PCI_FILL_BRIDGE_BASES.

Fixes ccf68033a452 ("lspci: Use PCI_FILL_BRIDGE_BASES to detect if range behind bridge is disabled or unsupported")

Thanks, applied.

Unfortunately, the current system of test cases cannot handle this type of errors, because it is based solely on parsing dumps. Maybe we should collect sysfs subtrees instead?

Martin