Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Nathan Lynch <nathanl@linux.ibm.com>
To:     Leonardo Bras <leonardo@linux.ibm.com>
Cc:     linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org,
        Michael Ellerman <mpe@ellerman.id.au>,
        Nicholas Piggin <npiggin@gmail.com>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Allison Randal <allison@lohutok.net>,
        "Gautham R. Shenoy" <ego@linux.vnet.ibm.com>,
        Nadav Amit <namit@vmware.com>
Subject: Re: [PATCH 1/1] powerpc/rtas: Implement reentrant rtas call
In-Reply-To: <20200408223901.760733-1-leonardo@linux.ibm.com>
References: <20200408223901.760733-1-leonardo@linux.ibm.com>
Date:   Fri, 10 Apr 2020 14:28:20 -0500
Message-ID: <87ftdb87jf.fsf@linux.ibm.com>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Leonardo Bras <leonardo@linux.ibm.com> writes:
> Implement rtas_call_reentrant() for reentrant rtas-calls:
> "ibm,int-on", "ibm,int-off",ibm,get-xive" and  "ibm,set-xive".
>
> On LoPAPR Version 1.1 (March 24, 2016), from 7.3.10.1 to 7.3.10.4,
> items 2 and 3 say:
>
> 2 - For the PowerPC External Interrupt option: The * call must be
> reentrant to the number of processors on the platform.
> 3 - For the PowerPC External Interrupt option: The * argument call
> buffer for each simultaneous call must be physically unique.
>
> So, these rtas-calls can be called in a lockless way, if using
> a different buffer for each call.

From the language in the spec it's clear that these calls are intended
to be reentrant with respect to themselves, but it's less clear to me
that they are safe to call simultaneously with respect to each other or
arbitrary other RTAS methods.


> This can be useful to avoid deadlocks in crashing, where rtas-calls are
> needed, but some other thread crashed holding the rtas.lock.

Are these calls commonly used in the crash-handling path? Is this
addressing a real issue you've seen?


> +/*
> + * Used for reentrant rtas calls.
> + * According to LoPAR documentation, only "ibm,int-on", "ibm,int-off",
> + * "ibm,get-xive" and "ibm,set-xive" are currently reentrant.
> + * Reentrant calls need their own rtas_args buffer, so not using rtas.args.
> + */

Please use kernel-doc format in new code.


> +int rtas_call_reentrant(int token, int nargs, int nret, int *outputs, ...)
> +{
> +	va_list list;
> +	struct rtas_args rtas_args;
> +
> +	if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
> +		return -1;
> +
> +	va_start(list, outputs);
> +	va_rtas_call_unlocked(&rtas_args, token, nargs, nret, list);
> +	va_end(list);

No, I don't think you can place the RTAS argument buffer on the stack:

  7.2.7, Software Implementation Note:
  | The OS must be aware that the effective address range for RTAS is 4
  | GB when instantiated in 32-bit mode and the OS should not pass RTAS
  | addresses or blocks of data which might fall outside of this range.