Skip to contentSkip to conversation

Question about device tree usage and ranges

U-Boot Aaron Williams 2 participants 4 messages Sep 19, 2019

lore
Share & tools
lore
#0 lore raw

Hi,

In my device tree I need to translate some address via ranges but I'm running into some issues. If I use ofnode_get_addr_size() it is not using the #address-cells and #size-cells nor is it performing the ranges translation. It always assumes #address-cells and #size-cells is 2 and doesn't translate. Is this by design?

If, on the other hand, I use ofnode_get_addr_size_index, it is having problems on ARM64. One problem I see is in of_translate_one in fdt_support.c. Shouldn't there be a fdt32_to_cpu or fdt64_to_cpu for the address after the memcpy in of_translate_one? I'm seeing the wrong endian address and size show up.

OF: ** translation for device console at 0 ** __of_translate_address: bus: default, address: 00000007fff46d7d
OF: bus is default (na=1, ns=1) on pci-console at 0x03000000
OF: translating address: 00000000
OF: parent bus is default (na=2, ns=2) on soc at 0
OF: walking ranges...
OF: default map, cp=0, s=1, da=0
OF: parent translation for: 80003000 80400000 <======= WRONG ENDIAN!
OF: with offset: 0
OF: one level translation: 80003000 80400000 <======= WRONG ENDIAN!
OF: parent bus is default (na=2, ns=2) on
OF: no ranges, 1:1 translation

-Aaron

#1 lore raw

Hi Aaron,

Quoted 13 lines · Aaron Williams

Hi,

In my device tree I need to translate some address via ranges but I'm running into some issues. If I use ofnode_get_addr_size() it is not using the #address-cells and #size-cells nor is it performing the ranges translation. It always assumes #address-cells and #size-cells is 2 and doesn't translate. Is this by design?

If, on the other hand, I use ofnode_get_addr_size_index, it is having problems on ARM64. One problem I see is in of_translate_one in fdt_support.c. Shouldn't there be a fdt32_to_cpu or fdt64_to_cpu for the address after the memcpy in of_translate_one? I'm seeing the wrong endian address and size show up.

Quoted 12 lines

OF: ** translation for device console at 0 ** __of_translate_address: bus: default, address: 00000007fff46d7d
OF: bus is default (na=1, ns=1) on pci-console at 0x03000000
OF: translating address: 00000000
OF: parent bus is default (na=2, ns=2) on soc at 0
OF: walking ranges...
OF: default map, cp=0, s=1, da=0
OF: parent translation for: 80003000 80400000 <======= WRONG ENDIAN!
OF: with offset: 0
OF: one level translation: 80003000 80400000 <======= WRONG ENDIAN!
OF: parent bus is default (na=2, ns=2) on
OF: no ranges, 1:1 translation

Without looking deeper into your mail, let me ask if you enabled CONFIG_(SPL_)OF_TRANSLATE?

Thanks,
Stefan

#2 lore raw

Hi Stefan,

That doesn't explain the wrong endian. The code should be using fdt32_to_cpu for little-endian CPUs.

-Aaron

Quoted 37 lines · PDT Stefan Roese

----------------------------------------------------------------------
Hi Aaron,

Quoted 14 lines · Aaron Williams

Hi,

In my device tree I need to translate some address via ranges but I'm running into some issues. If I use ofnode_get_addr_size() it is not using the #address-cells and #size-cells nor is it performing the ranges translation. It always assumes #address-cells and #size-cells is 2 and doesn't translate. Is this by design?

If, on the other hand, I use ofnode_get_addr_size_index, it is having problems on ARM64. One problem I see is in of_translate_one in fdt_support.c. Shouldn't there be a fdt32_to_cpu or fdt64_to_cpu for the address after the memcpy in of_translate_one? I'm seeing the wrong endian address and size show up.

Quoted 12 lines

OF: ** translation for device console at 0 ** __of_translate_address: bus: default, address: 00000007fff46d7d
OF: bus is default (na=1, ns=1) on pci-console at 0x03000000
OF: translating address: 00000000
OF: parent bus is default (na=2, ns=2) on soc at 0
OF: walking ranges...
OF: default map, cp=0, s=1, da=0
OF: parent translation for: 80003000 80400000 <======= WRONG ENDIAN!
OF: with offset: 0
OF: one level translation: 80003000 80400000 <======= WRONG ENDIAN!
OF: parent bus is default (na=2, ns=2) on
OF: no ranges, 1:1 translation

Without looking deeper into your mail, let me ask if you enabled CONFIG_(SPL_)OF_TRANSLATE?

Thanks,
Stefan

#3 lore raw
In reply to Stefan Roese

On Thursday, September 19, 2019 4:43:27 AM PDT Stefan Roese wrote:
Hi Stefan,

Quoted 32 lines

----------------------------------------------------------------------
Hi Aaron,

Quoted 14 lines · Aaron Williams

Hi,

In my device tree I need to translate some address via ranges but I'm running into some issues. If I use ofnode_get_addr_size() it is not using the #address-cells and #size-cells nor is it performing the ranges translation. It always assumes #address-cells and #size-cells is 2 and doesn't translate. Is this by design?

If, on the other hand, I use ofnode_get_addr_size_index, it is having problems on ARM64. One problem I see is in of_translate_one in fdt_support.c. Shouldn't there be a fdt32_to_cpu or fdt64_to_cpu for the address after the memcpy in of_translate_one? I'm seeing the wrong endian address and size show up.

Quoted 12 lines

OF: ** translation for device console at 0 ** __of_translate_address: bus: default, address: 00000007fff46d7d
OF: bus is default (na=1, ns=1) on pci-console at 0x03000000
OF: translating address: 00000000
OF: parent bus is default (na=2, ns=2) on soc at 0
OF: walking ranges...
OF: default map, cp=0, s=1, da=0
OF: parent translation for: 80003000 80400000 <======= WRONG ENDIAN!
OF: with offset: 0
OF: one level translation: 80003000 80400000 <======= WRONG ENDIAN!
OF: parent bus is default (na=2, ns=2) on
OF: no ranges, 1:1 translation

Quoted 5 lines

Without looking deeper into your mail, let me ask if you enabled CONFIG_(SPL_)OF_TRANSLATE?

Thanks,
Stefan

Yes, CONFIG_OF_TRANSLATE is enabled.

In looking at the code, it does not come into the picture when ofnode_get_addr_size() is called. It is used in ofnode_get_addr_size_index(), however. Also, ofnode_get_addr_size() does not honor the #address-cells and #size-cells whereas the other function does. Furthermore, the code does not translate the endianess in of_translate_one() which is used in ofnode_get_addr_size_index().

-Aaron