Re: [PATCH] arm64: debug: Make 'btc' and similar work in kdb

[Will Deacon <will@kernel.org>]

Hi Doug,

On Tue, Jul 30, 2019 at 03:18:00PM -0700, Douglas Anderson wrote:
> diff --git a/arch/arm64/kernel/kgdb.c b/arch/arm64/kernel/kgdb.c
> index 43119922341f..b666210fbc75 100644
> --- a/arch/arm64/kernel/kgdb.c
> +++ b/arch/arm64/kernel/kgdb.c
> @@ -148,6 +148,45 @@ sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task)
>  	gdb_regs[32] = cpu_context->pc;
>  }
>  
> +void kgdb_call_nmi_hook(void *ignored)
> +{
> +	struct pt_regs *regs;
> +
> +	/*
> +	 * NOTE: get_irq_regs() is supposed to get the registers from
> +	 * before the IPI interrupt happened and so is supposed to
> +	 * show where the processor was.  In some situations it's
> +	 * possible we might be called without an IPI, so it might be
> +	 * safer to figure out how to make kgdb_breakpoint() work
> +	 * properly here.
> +	 */
> +	regs = get_irq_regs();
> +
> +	/*
> +	 * Some commands (like 'btc') assume that they can find info about
> +	 * a task in the 'cpu_context'.  Unfortunately that's only valid
> +	 * for sleeping tasks.  ...but let's make it work anyway by just
> +	 * writing the registers to the right place.  This is safe because
> +	 * nobody else is using the 'cpu_context' for a running task.
> +	 */
> +	current->thread.cpu_context.x19 = regs->regs[19];
> +	current->thread.cpu_context.x20 = regs->regs[20];
> +	current->thread.cpu_context.x21 = regs->regs[21];
> +	current->thread.cpu_context.x22 = regs->regs[22];
> +	current->thread.cpu_context.x23 = regs->regs[23];
> +	current->thread.cpu_context.x24 = regs->regs[24];
> +	current->thread.cpu_context.x25 = regs->regs[25];
> +	current->thread.cpu_context.x26 = regs->regs[26];
> +	current->thread.cpu_context.x27 = regs->regs[27];
> +	current->thread.cpu_context.x28 = regs->regs[28];
> +	current->thread.cpu_context.fp = regs->regs[29];
> +
> +	current->thread.cpu_context.sp = regs->sp;
> +	current->thread.cpu_context.pc = regs->pc;
> +
> +	kgdb_nmicallback(raw_smp_processor_id(), regs);
> +}

This is really gross... :/

Can you IPI the other CPUs instead and have them backtrace locally, like we
do for things like magic sysrq (sysrq_handle_showallcpus())?

Will