[PATCH] make new y-modem (PATCH v2) work on big-endian CPU

[Antony Pavlov]

On 8 November 2012 03:41, Robert Jarzmik <robert.jarzmik at free.fr> wrote:
> Antony Pavlov <antonynpavlov at gmail.com> writes:
>
>> apply this commit after that one:
>>
>>     Author: Robert Jarzmik <robert.jarzmik at free.fr>
>>     Date:   Sun Nov 4 18:55:23 2012 +0100
>>
>>         commands: change Y-Modem implementation
>> ---
>>  lib/xymodem.c |   13 +++++++------
>>  1 file changed, 7 insertions(+), 6 deletions(-)
>>
>> diff --git a/lib/xymodem.c b/lib/xymodem.c
>> index 1469a9a..0e82ce2 100644
>> --- a/lib/xymodem.c
>> +++ b/lib/xymodem.c
>> @@ -257,8 +257,8 @@ static ssize_t xy_read_block(struct xyz_ctxt *proto, struct xy_block *blk,
>>       uint64_t timeout)
>>  {
>>       ssize_t rc, data_len = 0;
>> -     unsigned char hdr, seqs[2], crcs[2];
>> -     int crc = 0, hdr_found = 0;
>> +     unsigned char hdr, seqs[2];
>> +     uint16_t crc = 0, hdr_found = 0;
>>       uint64_t start = get_time_ns();
>>
>>       while (!hdr_found) {
>> @@ -308,12 +308,13 @@ static ssize_t xy_read_block(struct xyz_ctxt *proto, struct xy_block *blk,
>>
>>       switch (proto->crc_mode) {
>>       case CRC_ADD8:
>> -             rc = xy_gets(proto->cdev, proto->fifo, crcs, 1, timeout);
>> -             crc = crcs[0];
>> +             rc = xy_gets(proto->cdev, proto->fifo,
>> +                             (unsigned char *)&crc, 1, timeout);
> This doesn't look good to me.
> In big-endian arch, suppose you read 0xab as the crc.
> In that case, won't crc be equal to 0xab00 instead of 0x00ab in this code ? If
> that's the case, the previous code was right ...

To make the patch I got your old working version and your PATCH v2
branches. I got 'git diff'. I found the key difference and tested the
patch. That is all.

May be the 'proto->crc_mode==CRC_ADD8' condition is always false
during my tests?

>>               break;
>>       case CRC_CRC16:
>> -             rc = xy_gets(proto->cdev, proto->fifo, crcs, 2, timeout);
>> -             crc = be16_to_cpu(*(uint16_t *)crcs);
>> +             rc = xy_gets(proto->cdev, proto->fifo,
>> +                             (unsigned char *)&crc, 2, timeout);
>> +             crc = be16_to_cpu(crc);
>
> Does that mean that the code up there doesn't work on big endian ? I'm really
> puzzled, as the sender sends in big endian, the xy_gets() receives the bytes
> into crcs[2], and the be16_to_cpu() should be a no-op ...

I have no idea just now. I'll try to disassemble generated code.

> I think that we should do this simpler :
>   crc = crcs[0] << 8 + crcs[1]
>
> I'll put that into my github branch for testing.

-- 
Best regards,
  Antony Pavlov