Skip to contentSkip to conversation

x86:sysfb_efi:efifb_set_system: fix miss valid address range in later BARs

PATCH Applied Wang YanQing 3 participants 3 messages Apr 30, 2016 – May 3, 2016 x86

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 x86:sysfb_efi:efifb_set_system: fix miss valid address range in later BARs

We can't just break out when meet start is equal to zero, this will cause we miss valid address range in later BARs.

On the other hand, it isn't enough to test start only for below situation:
0(start) <= lfb_base < end

Note: this patch also add a trivial optimization,
break out after we find the address range
is valid without test later BARs.

---
Due to the BUG this patch fix, I can't use video=efifb:
boot parameter to get efifb on my new ThinkPad E550 for my old linux system hard disk with 3.10 kernel. In 3.10, efifb is the only choice due to DRM/I915 in it doesn't support the GPU.

Changes:
v1-v2:
1: Do a trivial code optimization.

 arch/x86/kernel/sysfb_efi.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
arch/x86/kernel/sysfb_efi.c +12 -2
index b285d4e..5da924b 100644
@@ -106,14 +106,24 @@ static int __init efifb_set_system(const struct dmi_system_id *id)
 					continue;
 				for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
 					resource_size_t start, end;
+					unsigned long flags;
+
+					flags = pci_resource_flags(dev, i);
+					if (!(flags & IORESOURCE_MEM))
+						continue;
+
+					if (flags & IORESOURCE_UNSET)
+						continue;
+
+					if (pci_resource_len(dev, i) == 0)
+						continue;
 
 					start = pci_resource_start(dev, i);
-					if (start == 0)
-						break;
 					end = pci_resource_end(dev, i);
 					if (screen_info.lfb_base >= start &&
 					    screen_info.lfb_base < end) {
 						found_bar = 1;
+						break;
 					}
 				}
 			}
Signature · 1 line
1.8.5.6.2.g3d8a54e.dirty
Reviewed-by on v2 1/1
#1 lore raw
Quoted 7 lines · Wang YanQing

We can't just break out when meet start is equal to zero, this will cause we miss valid address range in later BARs.

On the other hand, it isn't enough to test start only for below situation:
0(start) <= lfb_base < end

Quoted 5 lines

Note: this patch also add a trivial optimization,
break out after we find the address range
is valid without test later BARs.

This looks correct to me:

Cc'ing Matt Fleming, since this should probably go through the EFI tree.

Quoted 47 lines

---
Due to the BUG this patch fix, I can't use video=efifb:
boot parameter to get efifb on my new ThinkPad E550 for my old linux system hard disk with 3.10 kernel. In 3.10, efifb is the only choice due to DRM/I915 in it doesn't support the GPU.

Changes:
v1-v2:
1: Do a trivial code optimization.

 arch/x86/kernel/sysfb_efi.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
arch/x86/kernel/sysfb_efi.c +12 -2
index b285d4e..5da924b 100644
@@ -106,14 +106,24 @@ static int __init efifb_set_system(const struct dmi_system_id *id)
 					continue;
 				for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
 					resource_size_t start, end;
+					unsigned long flags;
+
+					flags = pci_resource_flags(dev, i);
+					if (!(flags & IORESOURCE_MEM))
+						continue;
+
+					if (flags & IORESOURCE_UNSET)
+						continue;
+
+					if (pci_resource_len(dev, i) == 0)
+						continue;
 
 					start = pci_resource_start(dev, i);
-					if (start == 0)
-						break;
 					end = pci_resource_end(dev, i);
 					if (screen_info.lfb_base >= start &&
 					    screen_info.lfb_base < end) {
 						found_bar = 1;
+						break;
 					}
 				}
 			}
Signature · 1 line
1.8.5.6.2.g3d8a54e.dirty
Signature · 1 line
  Peter
Applied
#2 lore raw
Quoted 12 lines · Peter Jones
Quoted 11 lines · Wang YanQing

We can't just break out when meet start is equal to zero, this will cause we miss valid address range in later BARs.

On the other hand, it isn't enough to test start only for below situation:
0(start) <= lfb_base < end

Note: this patch also add a trivial optimization,
break out after we find the address range
is valid without test later BARs.

Quoted 7 lines
Quoted 1 line

This looks correct to me:

Cc'ing Matt Fleming, since this should probably go through the EFI tree.

Applied, thanks.