[PATCH 1/1] mtd: cfi_flash: read device tree correctly

[Stefan Roese]

Hi Heinrich,

(added Simon to Cc)

On 24.07.20 18:34, Heinrich Schuchardt wrote:
> On 24.07.20 11:14, Rick Chen wrote:
>> Hi Heinrich
>>
>>> From: U-Boot [mailto:u-boot-bounces at lists.denx.de] On Behalf Of Heinrich Schuchardt
>>> Sent: Tuesday, July 21, 2020 10:51 AM
>>> To: Stefan Roese
>>> Cc: Simon Glass; u-boot at lists.denx.de; Heinrich Schuchardt
>>> Subject: [PATCH 1/1] mtd: cfi_flash: read device tree correctly
>>>
>>> dev_read_size_cells() and dev_read_addr_cells() do not walk up the device tree to find the number of cells. On error they return 1 and 2 respectively. On qemu_arm64_defconfig this leads to the incorrect detection of address of the second flash bank as 0x400000000000000 instead of 0x4000000.
>>>
>>> When running
>>>
>>>      qemu-system-aarch64 -machine virt -bios u-boot.bin \
>>>      -cpu cortex-a53 -nographic \
>>>      -drive if=pflash,format=raw,index=1,file=envstore.img
>>>
>>> the command 'saveenv' fails with
>>>
>>>      Saving Environment to Flash... Error: start and/or end address not on
>>>      sector boundary
>>>      Error: start and/or end address not on sector boundary
>>>      Failed (1)
>>>
>>> due to this incorrect address.
>>>
>>> Use function fdtdec_get_addr_size_auto_noparent() to read the array of flash banks from the device tree.
>>>
>>> Signed-off-by: Heinrich Schuchardt <xypron.glpk at gmx.de>
>>> ---
>>>   drivers/mtd/cfi_flash.c | 20 ++++++++------------
>>>   1 file changed, 8 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/drivers/mtd/cfi_flash.c b/drivers/mtd/cfi_flash.c index b7289ba539..dfa104bcf0 100644
>>> --- a/drivers/mtd/cfi_flash.c
>>> +++ b/drivers/mtd/cfi_flash.c
>>> @@ -2469,28 +2469,24 @@ unsigned long flash_init(void)  static int cfi_flash_probe(struct udevice *dev)  {
>>>          const fdt32_t *cell;
>>> -       int addrc, sizec;
>>> -       int len, idx;
>>> -
>>> -       addrc = dev_read_addr_cells(dev);
>>> -       sizec = dev_read_size_cells(dev);
>>> +       int len;
>>>
>>>          /* decode regs; there may be multiple reg tuples. */
>>>          cell = dev_read_prop(dev, "reg", &len);
>>>          if (!cell)
>>>                  return -ENOENT;
>>> -       idx = 0;
>>> -       len /= sizeof(fdt32_t);
>>> -       while (idx < len) {
>>> +
>>> +       for (cfi_flash_num_flash_banks = 0; ; ++cfi_flash_num_flash_banks) {
>>>                  phys_addr_t addr;
>>>
>>> -               addr = dev_translate_address(dev, cell + idx);
>>> +               addr = fdtdec_get_addr_size_auto_noparent(
>>> +                               gd->fdt_blob, dev_of_offset(dev), "reg",
>>> +                               cfi_flash_num_flash_banks, NULL, false);
>>> +               if (addr == FDT_ADDR_T_NONE)
>>> +                       break;
>>>
>>>                  flash_info[cfi_flash_num_flash_banks].dev = dev;
>>>                  flash_info[cfi_flash_num_flash_banks].base = addr;
>>> -               cfi_flash_num_flash_banks++;
>>> -
>>> -               idx += addrc + sizec;
>>>          }
>>>          gd->bd->bi_flashstart = flash_info[0].base;
>>>
>>> --
>>> 2.27.0
>>>
>>
>> This patch remind me that I have encounter flash bank detection
>> problem on AE350 platform a period time ago.
>> And have commit a patch to work around this problem as below:
>>
> 
>> commit cca8b1e5b20cdab7299a5ee7139e70783f73ccdf
>>
>>      riscv: dts: Add #address-cells and #size-cells in nor node
>>
>>      Those are required for cfi-flash driver to get correct address information.
>>      Also modify size description correctly.
>>
>> With this patch, there is unnecessary to re-declaration address-cells
>> and size-cells in nor node indeed.
>>
>> Tested-by: Rick Chen <rick at andestech.com>
>>
>> Thanks,
>> Rick
>>
> 
> Dear Stefan, dear Rick,
> 
> thanks for testing on different systems.
> 
> The reason for the different test results is the usage of CONFIG_OF_LIVE:
> 
> CONFIG_OF_LIVE=y
> * octeon_ebb7304_defconfig
> 
> CONFIG_OF_LIVE=n
> * ae350_rv32_defconfig
> * qemu_arm64_defconfig
> 
> dev_translate_address() behaves differently depending on the usage of a
> live tree.
> 
> I will send a revised patch that only changes the behavior only for the
> CONFIG_OF_LIVE=n case after testing qemu_arm64_defconfig both with and
> without live tree.

Thanks for looking into this. But I wonder, if it makes more sense to
change the OF_LIVE implementation so that it matches the "non-OF_LIVE"
version? We should strive to have both implementations achieving the
same results I think.

Or am I missing something?

Thanks,
Stefan