[U-Boot] [PATCH v3] drivers: core: Add translation in live tree case
The function dev_read_addr calls ofnode_get_addr_index in the live tree case, which does not apply bus translations to the address read from the device tree. This results in illegal addresses on boards that rely on bus translations being applied.
Fix this situation by applying bus translations in the live tree case as well.
Signed-off-by: Mario Six mario.six@gdsys.cc
---
v2 -> v3: * Fixed endianness issue on little-endian platforms
v1 -> v2: * Added IS_ENABLED(CONFIG_OF_TRANSLATE) case distinction
drivers/core/ofnode.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/drivers/core/ofnode.c b/drivers/core/ofnode.c index 0030ab962e..09f0aeab4f 100644 --- a/drivers/core/ofnode.c +++ b/drivers/core/ofnode.c @@ -200,13 +200,22 @@ fdt_addr_t ofnode_get_addr_index(ofnode node, int index) uint flags; u64 size; int na; + __be32 addr;
prop_val = of_get_address(ofnode_to_np(node), index, &size, &flags); if (!prop_val) return FDT_ADDR_T_NONE; na = of_n_addr_cells(ofnode_to_np(node)); - return of_read_number(prop_val, na); + addr = of_read_number(prop_val, na); + + if (IS_ENABLED(CONFIG_OF_TRANSLATE)) { + u64 paddr = of_translate_address(ofnode_to_np(node), &addr); + + return be32_to_cpu((fdt_addr_t)paddr); + } else { + return addr; + } } else { return fdt_get_base_address(gd->fdt_blob, ofnode_to_offset(node)); -- 2.13.6
On 12/18/2017 01:34 AM, Mario Six wrote:
The function dev_read_addr calls ofnode_get_addr_index in the live tree case, which does not apply bus translations to the address read from the device tree. This results in illegal addresses on boards that rely on bus translations being applied.
Fix this situation by applying bus translations in the live tree case as well.
Tested-by: Stephen Warren swarren@nvidia.com
On 12/18/2017 03:29 PM, Stephen Warren wrote:
On 12/18/2017 01:34 AM, Mario Six wrote:
The function dev_read_addr calls ofnode_get_addr_index in the live tree case, which does not apply bus translations to the address read from the device tree. This results in illegal addresses on boards that rely on bus translations being applied.
Fix this situation by applying bus translations in the live tree case as well.
Tested-by: Stephen Warren swarren@nvidia.com
Uggh. Sorry, I take that back. This seems to break NVIDIA Jetson TX1 (p2371-2180 board), even though it did solve this issue that was present on other boards in the previous patch version. I'll try and see what's up (something to do with I2C accesses early during boot)...
On 12/18/2017 03:33 PM, Stephen Warren wrote:
On 12/18/2017 03:29 PM, Stephen Warren wrote:
On 12/18/2017 01:34 AM, Mario Six wrote:
The function dev_read_addr calls ofnode_get_addr_index in the live tree case, which does not apply bus translations to the address read from the device tree. This results in illegal addresses on boards that rely on bus translations being applied.
Fix this situation by applying bus translations in the live tree case as well.
Tested-by: Stephen Warren swarren@nvidia.com
Uggh. Sorry, I take that back. This seems to break NVIDIA Jetson TX1 (p2371-2180 board), even though it did solve this issue that was present on other boards in the previous patch version. I'll try and see what's up (something to do with I2C accesses early during boot)...
I guess this is something to do with:
if (IS_ENABLED(CONFIG_OF_TRANSLATE)) {u64 paddr = of_translate_address(ofnode_to_np(node), &addr);return be32_to_cpu((fdt_addr_t)paddr);
My tests passed on a 32-bit platform but failed on a 64-bit platform. I expect you need be64_to_cpu() there on 64-bit platforms? Actually, it should really base the bit-width on #address-cells, but we only support 32- or 64-bit at the moment so switching between the two is probably fine.
On Tue, Dec 19, 2017 at 12:03 AM, Stephen Warren swarren@wwwdotorg.org wrote:
On 12/18/2017 03:33 PM, Stephen Warren wrote:
On 12/18/2017 03:29 PM, Stephen Warren wrote:
On 12/18/2017 01:34 AM, Mario Six wrote:
The function dev_read_addr calls ofnode_get_addr_index in the live tree case, which does not apply bus translations to the address read from the device tree. This results in illegal addresses on boards that rely on bus translations being applied.
Fix this situation by applying bus translations in the live tree case as well.
Tested-by: Stephen Warren swarren@nvidia.com
Uggh. Sorry, I take that back. This seems to break NVIDIA Jetson TX1 (p2371-2180 board), even though it did solve this issue that was present on other boards in the previous patch version. I'll try and see what's up (something to do with I2C accesses early during boot)...
I guess this is something to do with:
if (IS_ENABLED(CONFIG_OF_TRANSLATE)) {u64 paddr =of_translate_address(ofnode_to_np(node), &addr);
return be32_to_cpu((fdt_addr_t)paddr);My tests passed on a 32-bit platform but failed on a 64-bit platform. I expect you need be64_to_cpu() there on 64-bit platforms? Actually, it should really base the bit-width on #address-cells, but we only support 32- or 64-bit at the moment so switching between the two is probably fine.
Just sent a v4, which will hopefully fix this for all platforms. I tested with 32-bit and 64-bit sandbox, and I now get the same result with and without translation for both.
@Simon: I also sent a RFC patch for a sandbox with 64-bit addresses, since it might be useful for testing.
Best regards,
Mario
participants (2)
-
Mario Six -
Stephen Warren