[Pdbg] [PATCH 13/14] gdbserver: use read-modify-write for put_mem that is not 8-byte aligned

[Nicholas Piggin]

Excerpts from Joel Stanley's message of March 16, 2022 9:46 am:
> On Mon, 14 Mar 2022 at 04:18, Nicholas Piggin <npiggin at gmail.com> wrote:
>>
>> Not all targets have memory access that can support arbitrary byte
>> writes. Use RMW for put_mem commands that are not 8-byte aligned.
> 
> Do we know which targets can/can't support unaligned byte writes?

sbefifo chip ops at least can only do 8 byte aligned / sized writes.

>> @@ -318,17 +318,27 @@ static void put_mem(uint64_t *stack, void *priv)
>>                         err = 2;
>>                         goto out;
>>                 }
>> -       } else {
>> -               stack[2] = __builtin_bswap64(stack[2]) >> 32;
>>         }
>>
>> -       PR_INFO("put_mem 0x%016" PRIx64 " = 0x%016" PRIx64 "\n", addr, stack[2]);
>> +       if (len == 4 && littleendian) {
>> +               insn = (uint32_t *)data;
>> +               *insn = bswap_32(*insn);
>> +       }
>>
>> -       if (mem_write(adu_target, addr, data, len, 0, false)) {
>> +       start_addr = addr & ~(align - 1);
>> +       end_addr = (addr + len + (align - 1)) & ~(align - 1);
> 
> Would some ALIGN_UP / ALIGN_DOWN macros make sense here?

Yeah that could improve things.

> 
>> +       if (addr != start_addr || (addr + len) != end_addr) {
>> +               buf = malloc(end_addr - start_addr);
>> +               mem_read(adu_target, start_addr, buf, end_addr - start_addr, 0, false);
>> +               memcpy(buf + (addr - start_addr), data, len);
>> +       } else {
>> +               buf = data;
>> +       }
>> +
>> +       if (mem_write(adu_target, start_addr, buf, end_addr - start_addr, 0, false)) {
>>                 PR_ERROR("Unable to write memory\n");
>>                 err = 3;
>>         }
> 
> Needs to free buf if it was malloc'd.

Good catch.

Thanks,
Nick