[Pdbg] [PATCH 13/19] pdbg/gdbserver: Add in basic skeleton for a gdbserver on p8

[Rashmica]

On 30/08/18 21:25, Nicholas Piggin wrote:

> On Thu, 30 Aug 2018 10:18:30 +1000
> Rashmica <rashmica.g at gmail.com> wrote:
>> On 29/08/18 18:36, Nicholas Piggin wrote:
>>> On Wed, 29 Aug 2018 11:50:41 +1000
>>> Rashmica Gupta <rashmica.g at gmail.com> wrote:
>>>> From: Alistair Popple <alistair at popple.id.au>
>>>>
>>>> I have changed a few bits here and there but this patch is largely
>>>> authored by Alistair Popple.  
>>> So what's missing for P9? enable_attn? Anything else?  
>> There's a couple of things that I'm trying to work out regarding P9.
>>
>> - we can't mtnia from an arbitrary register in ramming mode (can only do it
>>  from the link register).
> There's a few bits that seem to make P9 work. No guarantee I didn't
> break P8 MTNIA because I haven't tested it.
>
> As we can see, the need for target specific .ram_spr function is
> increasing. P9 still isn't quite right with all its MTSPRs either.

When I tried doing value -> LR -> NIA like below, the last two bits of NIA on p9
were always set to 0. I'll try on some other machines to see if it happens there.

> --
>
> P9 MTNIA and MTMSR fixes
>
> ---
>  libpdbg/chip.c   | 19 +++++++++++++++----
>  libpdbg/p9chip.c |  6 +++---
>  src/reg.c        |  2 +-
>  3 files changed, 19 insertions(+), 8 deletions(-)
>
> diff --git a/libpdbg/chip.c b/libpdbg/chip.c
> index 80a2261..1d41d87 100644
> --- a/libpdbg/chip.c
> +++ b/libpdbg/chip.c
> @@ -145,8 +145,8 @@ int ram_sreset_thread(struct pdbg_target *thread_target)
>   * into *results. *results must point to an array the same size as
>   * *opcodes. Each entry from *results is put into SCR0 prior to
>   * executing an opcode so that it may also be used to pass in
> - * data. Note that only register r0 is saved and restored so opcodes
> - * must not touch other registers.
> + * data. Note that only registers r0 and r1 are saved and restored so
> + * opcode sequences must preserve other registers.
>   */
>  static int ram_instructions(struct pdbg_target *thread_target, uint64_t *opcodes,
>  			    uint64_t *results, int len, unsigned int lpar)
> @@ -242,10 +242,21 @@ int ram_getnia(struct pdbg_target *thread, uint64_t *value)
>  	return 0;
>  }
>  
> +/*
> + * P9 must MTNIA from LR, P8 can MTNIA from R0. So we set both LR and R0
> + * to value. LR must be saved and restored.
> + *
> + * This is a hack and should be made much cleaner once we have target
> + * specific putspr commands.
> + */
>  int ram_putnia(struct pdbg_target *thread, uint64_t value)
>  {
> -	uint64_t opcodes[] = {mfspr(0, 277), mtnia(0)};
> -	uint64_t results[] = {value, 0};
> +	uint64_t opcodes[] = {	mfspr(1, 8),	/* mflr r1 */
> +				mfspr(0, 277),	/* value -> r0 */
> +				mtspr(8, 0),	/* mtlr r0 */
> +				mtnia(0),
> +				mtspr(8, 1), };	/* mtlr r1 */
> +	uint64_t results[] = {0, value, 0, 0, 0};
>  
>  	CHECK_ERR(ram_instructions(thread, opcodes, results, ARRAY_SIZE(opcodes), 0));
>  	return 0;
> diff --git a/libpdbg/p9chip.c b/libpdbg/p9chip.c
> index 189d80a..9094c24 100644
> --- a/libpdbg/p9chip.c
> +++ b/libpdbg/p9chip.c
> @@ -291,10 +291,9 @@ static int __p9_ram_instruction(struct thread *thread, uint64_t opcode, uint64_t
>  
>  	switch(opcode & OPCODE_MASK) {
>  	case MTNIA_OPCODE:
> +		opcode = 0x4c0000a4;
> +		opcode |= 0x001E0000;
>  		predecode = 8;
> -
> -		/* Not currently supported as we can only MTNIA from LR */
> -		PR_ERROR("MTNIA is not currently supported\n");
>  		break;
>  
>  	case MFNIA_OPCODE:
> @@ -304,6 +303,7 @@ static int __p9_ram_instruction(struct thread *thread, uint64_t opcode, uint64_t
>  
>  	case MTMSR_OPCODE:
>  		predecode = 8;
> +		opcode |= 0x001E0000;
>  		break;
>  
>  	case MFSPR_OPCODE:
> diff --git a/src/reg.c b/src/reg.c
> index aa77a8a..c63f6da 100644
> --- a/src/reg.c
> +++ b/src/reg.c
> @@ -116,7 +116,7 @@ OPTCMD_DEFINE_CMD(getnia, getnia);
>  static int putnia(uint64_t nia)
>  {
>  	uint64_t reg = REG_NIA;
> -	return for_each_target("thread", getprocreg, &reg, &nia);
> +	return for_each_target("thread", putprocreg, &reg, &nia);
>  }
>  OPTCMD_DEFINE_CMD_WITH_ARGS(putnia, putnia, (DATA));
>  
>