[Pdbg] [PATCH v2 22/39] gdbserver: check for attn using the SPATTN register
Joel Stanley
joel at jms.id.au
Tue May 3 17:10:47 AEST 2022
On Wed, 20 Apr 2022 at 06:51, Nicholas Piggin <npiggin at gmail.com> wrote:
>
> When POWER9/10 execute an attn instruction, they set a bit in the SPATTN
> register and quiesce the thread.
>
> This bit can be checked to confirm the thread hit an attn instruction,
> rather than assuming a thread was quiesced because of attn. This makes
> the gdb breakpoint code more robust in the presence of other direct
> controls (e.g., host-based direct controls from OPAL).
>
> This change also clears the SPATTN bit which clears the exception
> condition that raises the IPOLL global interrupt. That interrupt seems
> to be involved with the IPOLL interrupt storm and lock-up on POWER9
> (although this does not fix it).
>
> Signed-off-by: Nicholas Piggin <npiggin at gmail.com>
> ---
> src/pdbgproxy.c | 79 ++++++++++++++++++++++++++++++++++++++++++-------
> 1 file changed, 68 insertions(+), 11 deletions(-)
>
> diff --git a/src/pdbgproxy.c b/src/pdbgproxy.c
> index 23971c25..d7ceac8d 100644
> --- a/src/pdbgproxy.c
> +++ b/src/pdbgproxy.c
> @@ -457,6 +457,59 @@ static void v_contc(uint64_t *stack, void *priv)
> poll_interval = 1;
> }
>
> +#define P9_SPATTN_AND 0x20010A98
> +#define P9_SPATTN 0x20010A99
> +
> +#define P10_SPATTN_AND 0x20028498
> +#define P10_SPATTN 0x20028499
> +
> +static bool thread_check_attn(struct pdbg_target *target)
> +{
> + struct thread *thread = target_to_thread(target);
> + struct pdbg_target *core;
> + uint64_t spattn;
> +
> + if (pdbg_target_compatible(target, "ibm,power8-thread")) {
> + return true; /* XXX */
I assume this is ok?
Reviewed-by: Joel Stanley <joel at jms.id.au>
> + } else if (pdbg_target_compatible(target, "ibm,power9-thread")) {
> + core = pdbg_target_require_parent("core", target);
> + if (pib_read(core, P9_SPATTN, &spattn)) {
> + PR_ERROR("SPATTN read failed\n");
> + return false;
> + }
> +
> + if (spattn & PPC_BIT(1 + 4*thread->id)) {
> + uint64_t mask = ~PPC_BIT(1 + 4*thread->id);
> +
> + if (pib_write(core, P9_SPATTN_AND, mask)) {
> + PR_ERROR("SPATTN clear failed\n");
> + return false;
> + }
> +
> + return true;
> + }
> + } else if (pdbg_target_compatible(target, "ibm,power10-thread")) {
> + core = pdbg_target_require_parent("core", target);
> + if (pib_read(core, P10_SPATTN, &spattn)) {
> + PR_ERROR("SPATTN read failed\n");
> + return false;
> + }
> +
> + if (spattn & PPC_BIT(1 + 4*thread->id)) {
> + uint64_t mask = ~PPC_BIT(1 + 4*thread->id);
> +
> + if (pib_write(core, P10_SPATTN_AND, mask)) {
> + PR_ERROR("SPATTN clear failed\n");
> + return false;
> + }
> +
> + return true;
> + }
> + }
> +
> + return false;
> +}
> +
> static void interrupt(uint64_t *stack, void *priv)
> {
> struct thread_state status;
> @@ -477,7 +530,6 @@ static void interrupt(uint64_t *stack, void *priv)
>
> static void poll(void)
> {
> - uint64_t nia;
> struct thread_state status;
>
> thread_target->probe(thread_target);
> @@ -495,17 +547,22 @@ static void poll(void)
>
> state = IDLE;
> poll_interval = VCONT_POLL_DELAY;
> - if (!(status.active)) {
> - PR_ERROR("Thread inactive after trap\n");
> - send_response(fd, ERROR(EPERM));
> - return;
> - }
>
> - /* Restore NIA */
> - if (thread_getnia(thread_target, &nia))
> - PR_ERROR("Error during getnia\n");
> - if (thread_putnia(thread_target, nia - 4))
> - PR_ERROR("Error during putnia\n");
> + if (thread_check_attn(thread_target)) {
> + uint64_t nia;
> +
> + if (!(status.active)) {
> + PR_ERROR("Thread inactive after trap\n");
> + send_response(fd, ERROR(EPERM));
> + return;
> + }
> +
> + /* Restore NIA */
> + if (thread_getnia(thread_target, &nia))
> + PR_ERROR("Error during getnia\n");
> + if (thread_putnia(thread_target, nia - 4))
> + PR_ERROR("Error during putnia\n");
> + }
> send_response(fd, TRAP);
> break;
> }
> --
> 2.35.1
>
> --
> Pdbg mailing list
> Pdbg at lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/pdbg
More information about the Pdbg
mailing list