Re: [PATCH v2] dm: Fix OF_BAD_ADDR definition

[Jan Kiszka <jan.kiszka@siemens.com>]

On 15.02.22 14:49, Jan Kiszka wrote:
> On 15.02.22 14:34, Patrice CHOTARD wrote:
>> Hi Jan
>>
>> On 2/15/22 14:00, Jan Kiszka wrote:
>>> On 15.02.22 12:56, Patrice CHOTARD wrote:
>>>> Hi Jan
>>>>
>>>> On 2/14/22 16:21, Jan Kiszka wrote:
>>>>> On 04.01.22 08:42, Patrice Chotard wrote:
>>>>>> When OF_LIVE flag is enabled on a 64 bits platform, there is an
>>>>>> issue when dev_read_addr() is called and need to perform an address
>>>>>> translation using __of_translate_address().
>>>>>>
>>>>>> In case of error, __of_translate_address() return's value is OF_BAD_ADDR
>>>>>> (wich is defined in include/dm/of.h to ((u64)-1) = 0xffffffffffffffff).
>>>>>> The return value of dev_read_addr() is often compared to FDT_ADDR_T_NONE
>>>>>> which is defined as (-1U) = 0xffffffff.
>>>>>> In this case the comparison is always false.
>>>>>>
>>>>>> To fix this issue, define FDT_ADDR_T_NONE to (ulong)(-1) in case of
>>>>>> AARCH64. Update accordingly related tests.
>>>>>>
>>>>>> Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
>>>>>>
>>>>>> ---
>>>>>>
>>>>>> Changes in v2:
>>>>>>  - define FDT_ADDR_T_NONE as ((ulong)(-1)) and keep OF_BAD_ADDR unchanged
>>>>>>
>>>>>>  include/fdtdec.h   | 5 ++++-
>>>>>>  test/dm/ofnode.c   | 2 +-
>>>>>>  test/dm/pci.c      | 4 ++--
>>>>>>  test/dm/test-fdt.c | 2 +-
>>>>>>  4 files changed, 8 insertions(+), 5 deletions(-)
>>>>>>
>>>>>> diff --git a/include/fdtdec.h b/include/fdtdec.h
>>>>>> index 6c7ab887b2..e9e2916d6e 100644
>>>>>> --- a/include/fdtdec.h
>>>>>> +++ b/include/fdtdec.h
>>>>>> @@ -24,16 +24,19 @@
>>>>>>  typedef phys_addr_t fdt_addr_t;
>>>>>>  typedef phys_size_t fdt_size_t;
>>>>>>  
>>>>>> -#define FDT_ADDR_T_NONE (-1U)
>>>>>>  #define FDT_SIZE_T_NONE (-1U)
>>>>>>  
>>>>>>  #ifdef CONFIG_PHYS_64BIT
>>>>>> +#define FDT_ADDR_T_NONE ((ulong)(-1))
>>>>>> +
>>>>>>  #define fdt_addr_to_cpu(reg) be64_to_cpu(reg)
>>>>>>  #define fdt_size_to_cpu(reg) be64_to_cpu(reg)
>>>>>>  #define cpu_to_fdt_addr(reg) cpu_to_be64(reg)
>>>>>>  #define cpu_to_fdt_size(reg) cpu_to_be64(reg)
>>>>>>  typedef fdt64_t fdt_val_t;
>>>>>>  #else
>>>>>> +#define FDT_ADDR_T_NONE (-1U)
>>>>>> +
>>>>>>  #define fdt_addr_to_cpu(reg) be32_to_cpu(reg)
>>>>>>  #define fdt_size_to_cpu(reg) be32_to_cpu(reg)
>>>>>>  #define cpu_to_fdt_addr(reg) cpu_to_be32(reg)
>>>>>> diff --git a/test/dm/ofnode.c b/test/dm/ofnode.c
>>>>>> index cea0746bb3..e6c925eba6 100644
>>>>>> --- a/test/dm/ofnode.c
>>>>>> +++ b/test/dm/ofnode.c
>>>>>> @@ -286,7 +286,7 @@ static int dm_test_ofnode_get_reg(struct unit_test_state *uts)
>>>>>>  	ut_assert(ofnode_valid(node));
>>>>>>  	addr = ofnode_get_addr(node);
>>>>>>  	size = ofnode_get_size(node);
>>>>>> -	ut_asserteq(FDT_ADDR_T_NONE, addr);
>>>>>> +	ut_asserteq_64(FDT_ADDR_T_NONE, addr);
>>>>>>  	ut_asserteq(FDT_SIZE_T_NONE, size);
>>>>>>  
>>>>>>  	node = ofnode_path("/translation-test@8000/noxlatebus@3,300/dev@42");
>>>>>> diff --git a/test/dm/pci.c b/test/dm/pci.c
>>>>>> index fa2e4a8559..00e4440a9d 100644
>>>>>> --- a/test/dm/pci.c
>>>>>> +++ b/test/dm/pci.c
>>>>>> @@ -331,10 +331,10 @@ static int dm_test_pci_addr_live(struct unit_test_state *uts)
>>>>>>  	struct udevice *swap1f, *swap1;
>>>>>>  
>>>>>>  	ut_assertok(dm_pci_bus_find_bdf(PCI_BDF(0, 0x1f, 0), &swap1f));
>>>>>> -	ut_asserteq(FDT_ADDR_T_NONE, dev_read_addr_pci(swap1f));
>>>>>> +	ut_asserteq_64(FDT_ADDR_T_NONE, dev_read_addr_pci(swap1f));
>>>>>>  
>>>>>>  	ut_assertok(dm_pci_bus_find_bdf(PCI_BDF(0, 0x1, 0), &swap1));
>>>>>> -	ut_asserteq(FDT_ADDR_T_NONE, dev_read_addr_pci(swap1));
>>>>>> +	ut_asserteq_64(FDT_ADDR_T_NONE, dev_read_addr_pci(swap1));
>>>>>>  
>>>>>>  	return 0;
>>>>>>  }
>>>>>> diff --git a/test/dm/test-fdt.c b/test/dm/test-fdt.c
>>>>>> index 8866d4d959..e1de066226 100644
>>>>>> --- a/test/dm/test-fdt.c
>>>>>> +++ b/test/dm/test-fdt.c
>>>>>> @@ -768,7 +768,7 @@ static int dm_test_fdt_livetree_writing(struct unit_test_state *uts)
>>>>>>  	/* Test setting generic properties */
>>>>>>  
>>>>>>  	/* Non-existent in DTB */
>>>>>> -	ut_asserteq(FDT_ADDR_T_NONE, dev_read_addr(dev));
>>>>>> +	ut_asserteq_64(FDT_ADDR_T_NONE, dev_read_addr(dev));
>>>>>>  	/* reg = 0x42, size = 0x100 */
>>>>>>  	ut_assertok(ofnode_write_prop(node, "reg", 8,
>>>>>>  				      "\x00\x00\x00\x42\x00\x00\x01\x00"));
>>>>>
>>>>> Since this commit, I'm getting this dev_get_priv warning:
>>>>>
>>>>> [...]
>>>>> U-Boot 2022.01-00766-g9876ae7db6d-dirty (Feb 14 2022 - 16:15:21 +0100)
>>>>>
>>>>> Model: SIMATIC IOT2050 Basic
>>>>> DRAM:  1 GiB
>>>>> Core:  94 devices, 28 uclasses, devicetree: separate
>>>>> WDT:   Not starting watchdog@40610000
>>>>> MMC:   mmc@4fa0000: 0
>>>>> Loading Environment from SPIFlash... SF: Detected w25q128 with page size
>>>>> 256 Bytes, erase size 64 KiB, total 16 MiB
>>>>> drivers/dma/ti/k3-udma.c:1744 __dev_get_priv: null device
>>>>> drivers/dma/ti/k3-udma.c:1744 __dev_get_priv: null device
>>>>> OK
>>>>> In:    serial
>>>>> Out:   serial
>>>>> Err:   serial
>>>>> Net:   No ethernet found.
>>>>> Hit any key to stop autoboot:  0
>>>>>
>>>>> (I've instrumented dev_get_priv to tell me file:line)
>>>>>
>>>>> Is that a sleeping problem surfaced by the commit or a real regression?
>>>>> I can boot, though.
>>>>>
>>>>> Jan
>>>>>
>>>>
>>>> It should be interesting to understand why uclass_get_device_by_phandle() return tmp = NULL.
>>>
>>> Yep.
>>>
>>>> What's the return value of uclass_get_device_by_phandle() ?
>>>>
>>>
>>> -22, EINVAL.
>>
>> As EINVAL is one of the more "common" error value, i think you have to go deeper 
>> to see where the uclass_get_device_by_phandle() is failing.
>>
> 
> Sigh, I was hoping to get around debugging this myself.
> 
> In any case: With this patch revert, the related code path is still
> taken, just successfully:
> 
> ti-udma dma-controller@285c0000: ret=0 tmp=00000000bdf10750
> 

To conclude this thread: The patch does what it is supposed to do, i.e.
define the right FDT_ADDR_T_NONE so that comparisons finally work
correctly on 64-bit archs.

As they work correctly now, in this case in dev_remap_addr_name, they
reveal that k3_nav_ringacc_init() tries to get a non-existent
configuration register "cfg". So far it got -1LL as result, != NULL, and
likely used that happily. The missing register came from a missing
u-boot specific fragment in our board dts, compare to the TI reference
board. Working on a fix.

Jan

-- 
Siemens AG, Technology
Competence Center Embedded Linux