Re: [PATCH v3 03/16] [gdb/gdbserver] refactor: Simplify SVE interface to read/write registers

[Luis Machado <luis.machado@arm.com>]

On 7/24/23 17:19, Thiago Jung Bauermann wrote:
> 
> Luis Machado via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
>> This is a patch in preparation to upcoming patches enabling SME support.  It
>> attempts to simplify the gdb/gdbserver shared interface used to read/write
>> SVE registers.
>>
>> Where the current code makes use of unique_ptr, allocating a new buffer by
>> hand and passing a buffer around, this patch makes that code use
>> gdb::byte_vector and passes a reference to this byte vector to the functions,
>> allowing the functions to have ready access to the size of the buffer.
> 
> Nice!
> 
>> It also shares a bit more code between gdb and gdbserver, in particular around
>> handling of ptrace get/set requests for SVE.
> 
> Also nice!
> 
>> diff --git a/gdb/nat/aarch64-scalable-linux-ptrace.c b/gdb/nat/aarch64-scalable-linux-ptrace.c
>> index cc43f510892..192eebcda19 100644
>> --- a/gdb/nat/aarch64-scalable-linux-ptrace.c
>> +++ b/gdb/nat/aarch64-scalable-linux-ptrace.c
>> @@ -120,28 +120,44 @@ aarch64_sve_set_vq (int tid, struct reg_buffer_common *reg_buf)
>>  
>>  /* See nat/aarch64-scalable-linux-ptrace.h.  */
>>  
>> -std::unique_ptr<gdb_byte[]>
>> -aarch64_sve_get_sveregs (int tid)
>> +gdb::byte_vector
>> +aarch64_fetch_sve_regset (int tid)
>>  {
>> -  struct iovec iovec;
>>    uint64_t vq = aarch64_sve_get_vq (tid);
>>  
>>    if (vq == 0)
>> -    perror_with_name (_("Unable to fetch SVE register header"));
>> +    perror_with_name (_("Unable to fetch SVE vector length"));
>>  
>>    /* A ptrace call with NT_ARM_SVE will return a header followed by either a
>>       dump of all the SVE and FP registers, or an fpsimd structure (identical to
>>       the one returned by NT_FPREGSET) if the kernel has not yet executed any
>>       SVE code.  Make sure we allocate enough space for a full SVE dump.  */
>>  
>> -  iovec.iov_len = SVE_PT_SIZE (vq, SVE_PT_REGS_SVE);
>> -  std::unique_ptr<gdb_byte[]> buf (new gdb_byte[iovec.iov_len]);
>> -  iovec.iov_base = buf.get ();
>> +  gdb::byte_vector sve_state (SVE_PT_SIZE (vq, SVE_PT_REGS_SVE), 0);
>> +
>> +  struct iovec iovec;
>> +  iovec.iov_base = sve_state.data ();
>> +  iovec.iov_len = sve_state.size ();
>>  
>>    if (ptrace (PTRACE_GETREGSET, tid, NT_ARM_SVE, &iovec) < 0)
>>      perror_with_name (_("Unable to fetch SVE registers"));
>>  
>> -  return buf;
>> +  return sve_state;
>> +}
>> +
>> +/* See nat/aarch64-scalable-linux-ptrace.h.  */
>> +
>> +bool
> 
> This function always returns true, and its only caller doesn't use the
> return value. Should it return void instead?
> 

I think it is reasonable for now. An earlier iteration of this patch used to let an error be handled by the callers, and I changed that.

I'll get it updated, and also change the same for the za regset.

>> +aarch64_store_sve_regset (int tid, const gdb::byte_vector &sve_state)
>> +{
>> +  struct iovec iovec;
>> +  iovec.iov_base = (void *) sve_state.data ();
> 
> Minor nit: is the cast necessary? The code looks cleaner without it.
> Also, it's not used in aarch64_fetch_sve_regset.
> 

I don't think it is needed. Probably an artifact of when the code used unique_ptr's.

>> +      else
>> +	{
>> +	  /* Otherwise, reformat the fpsimd structure into a full SVE set, by
>> +	     expanding the V registers (working backwards so we don't splat
>> +	     registers before they are copied) and using zero for everything
>> +	     else.
>> +	     Note that enough space for a full SVE dump was originally allocated
>> +	     for base.  */
>> +
>> +	  header->flags |= SVE_PT_REGS_SVE;
>> +	  header->size = SVE_PT_SIZE (vq, SVE_PT_REGS_SVE);
>> +
>> +	  memcpy (base + SVE_PT_SVE_FPSR_OFFSET (vq), &fpsimd->fpsr,
>> +		  sizeof (uint32_t));
>> +	  memcpy (base + SVE_PT_SVE_FPCR_OFFSET (vq), &fpsimd->fpcr,
>> +		  sizeof (uint32_t));
>> +
>> +	  for (int i = AARCH64_SVE_Z_REGS_NUM; i >= 0 ; i--)
> 
> Shouldn't i start from AARCH64_SVE_Z_REGS_NUM - 1? If this is correct, a
> comment explaining why would be helpful.
> 

Yes, it should. It seems this slipped through from the previous code. I'll fix it while at it.

Thanks for spotting it!