linux-arm-kernel.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
From: Cristian Marussi <cristian.marussi@arm.com>
To: linux-kernel@vger.kernel.org
Cc: linux-arch@vger.kernel.org, mark.rutland@arm.com,
	sparclinux@vger.kernel.org, dzickus@redhat.com,
	ehabkost@redhat.com, peterz@infradead.org,
	catalin.marinas@arm.com, x86@kernel.org, linux@armlinux.org.uk,
	davem@davemloft.net, hch@infradead.org,
	takahiro.akashi@linaro.org, mingo@redhat.com,
	james.morse@arm.com, hidehiro.kawai.ez@hitachi.com,
	tglx@linutronix.de, will@kernel.org, dave.martin@arm.com,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [RFC PATCH v2 01/12] smp: add generic SMP-stop support to common code
Date: Mon, 23 Sep 2019 15:20:12 +0100	[thread overview]
Message-ID: <1d7e90b7-71d8-3c12-9c1b-7049a1d7e32f@arm.com> (raw)
In-Reply-To: <20190913181953.45748-2-cristian.marussi@arm.com>

On 13/09/2019 19:19, Cristian Marussi wrote:
> There was a lot of code duplication across architectures regarding the
> SMP stop procedures' logic; moreover some of this duplicated code logic
> happened to be similarly faulty across a few architectures: while fixing
> such logic, move such generic logic as much as possible inside common
> code.
> 
> Collect all the common logic related to SMP stop operations into the
> common SMP code; any architecture willing to use such centralized logic
> can select CONFIG_ARCH_USE_COMMON_STOP=y and provide the related
> arch-specific helpers: in such a scenario, those architectures will
> transparently start using the common code provided by smp_send_stop()
> common function.
> 
> On the other side, Architectures not willing to use common code SMP stop
> logic will simply leave CONFIG_ARCH_USE_COMMON_STOP undefined and carry
> on executing their local arch-specific smp_send_stop() as before.
> 
> Suggested-by: Dave Martin <Dave.Martin@arm.com>
> Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>
> ---
> v1 --> v2
> - moved related Kconfig to common code inside arch/Kconfig
> - introduced additional CONFIG_USE_COMMON_STOP selected by
>   CONFIG_ARCH_USE_COMMON_STOP
> - introduced helpers to let architectures optionally alter
>   the default common code behaviour while waiting for CPUs:
>   change timeout or wait for ever. (will be needed by x86)
> ---
>  arch/Kconfig        |  7 +++++
>  include/linux/smp.h | 55 +++++++++++++++++++++++++++++++++++++
>  kernel/smp.c        | 67 +++++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 129 insertions(+)
> 
> diff --git a/arch/Kconfig b/arch/Kconfig
> index a7b57dd42c26..89766e6c0ac8 100644
> --- a/arch/Kconfig
> +++ b/arch/Kconfig
> @@ -166,6 +166,13 @@ config ARCH_USE_BUILTIN_BSWAP
>  	 instructions should set this. And it shouldn't hurt to set it
>  	 on architectures that don't have such instructions.
>  
> +config ARCH_USE_COMMON_SMP_STOP
> +       def_bool n
> +
> +config USE_COMMON_SMP_STOP
> +       depends on SMP && ARCH_USE_COMMON_SMP_STOP
> +       def_bool y
> +
>  config KRETPROBES
>  	def_bool y
>  	depends on KPROBES && HAVE_KRETPROBES
> diff --git a/include/linux/smp.h b/include/linux/smp.h
> index 6fc856c9eda5..381a14bfcd96 100644
> --- a/include/linux/smp.h
> +++ b/include/linux/smp.h
> @@ -77,6 +77,61 @@ int smp_call_function_single_async(int cpu, call_single_data_t *csd);
>   */
>  extern void smp_send_stop(void);
>  
> +#ifdef CONFIG_USE_COMMON_SMP_STOP
> +static atomic_t wait_forever;
> +static atomic_t wait_timeout = ATOMIC_INIT(USEC_PER_SEC);
> +
> +/*
> + * An Architecture can optionally decide to use this helper to change the
> + * waiting behaviour of common STOP logic, forcing to wait forever for
> + * all CPUs to be stopped.
> + */
> +static inline void smp_stop_set_wait_forever(int wait)
> +{
> +	atomic_set(&wait_forever, wait);
> +	/* ensure wait atomic-op is visible */
> +	smp_mb__after_atomic();
> +}
> +

These new helpers I added in V2 to let x86 configure wait/timeout SMP common stop behavior
are in fact deadly broken as of now since based on underlying static header-globals.
I'll fix in V3.

Cheers

Cristian


> +/*
> + * An Architecture can optionally decide to use this helper to change the
> + * waiting timeout of common STOP logic. A ZERO timeout means no timeout
> + * at all as long as wait_forever was not previously set.
> + *
> + * Note that wait_forever and timeout must remain individually selectable:
> + * so you can temporarily request wait_forever while keeping the same timeout
> + * settings.
> + */
> +static inline void smp_stop_set_wait_timeout_us(unsigned long timeout)
> +{
> +	atomic_set(&wait_timeout, timeout);
> +	/* ensure timeout atomic-op is visible */
> +	smp_mb__after_atomic();
> +}
> +
> +/* Retrieve the current wait settings. */
> +static inline bool smp_stop_get_wait_timeout_us(unsigned long *timeout)
> +{
> +	if (timeout)
> +		*timeout = atomic_read(&wait_timeout);
> +	return atomic_read(&wait_forever);
> +}
> +
> +/*
> + * Any Architecture willing to use STOP common logic implementation
> + * MUST at least provide the arch_smp_stop_call() helper which is in
> + * charge of its own arch-specific CPU-stop mechanism.
> + */
> +extern void arch_smp_stop_call(cpumask_t *cpus);
> +
> +/*
> + * An Architecture CAN also provide the arch_smp_cpus_stop_complete()
> + * dedicated helper, to perform any final arch-specific operation on
> + * the local CPU once the other CPUs have been successfully stopped.
> + */
> +void arch_smp_cpus_stop_complete(void);
> +#endif
> +
>  /*
>   * sends a 'reschedule' event to another CPU:
>   */
> diff --git a/kernel/smp.c b/kernel/smp.c
> index 7dbcb402c2fc..72f99bf13fd0 100644
> --- a/kernel/smp.c
> +++ b/kernel/smp.c
> @@ -20,6 +20,7 @@
>  #include <linux/sched.h>
>  #include <linux/sched/idle.h>
>  #include <linux/hypervisor.h>
> +#include <linux/delay.h>
>  
>  #include "smpboot.h"
>  
> @@ -817,3 +818,69 @@ int smp_call_on_cpu(unsigned int cpu, int (*func)(void *), void *par, bool phys)
>  	return sscs.ret;
>  }
>  EXPORT_SYMBOL_GPL(smp_call_on_cpu);
> +
> +#ifdef CONFIG_USE_COMMON_SMP_STOP
> +void __weak arch_smp_cpus_stop_complete(void) { }
> +
> +static inline bool any_other_cpus_online(cpumask_t *mask,
> +					 unsigned int this_cpu_id)
> +{
> +	cpumask_copy(mask, cpu_online_mask);
> +	cpumask_clear_cpu(this_cpu_id, mask);
> +
> +	return !cpumask_empty(mask);
> +}
> +
> +/*
> + * This centralizes the common logic to:
> + *
> + *  - evaluate which CPUs are online and needs to be notified for stop,
> + *    while considering properly the status of the calling CPU
> + *
> + *  - call the arch-specific helpers to request the effective stop
> + *
> + *  - wait for the stop operation to be completed across all involved CPUs
> + *    monitoring the cpu_online_mask
> + */
> +void smp_send_stop(void)
> +{
> +	unsigned int this_cpu_id;
> +	cpumask_t mask;
> +
> +	this_cpu_id = smp_processor_id();
> +	if (any_other_cpus_online(&mask, this_cpu_id)) {
> +		bool wait;
> +		unsigned long timeout;
> +		unsigned int this_cpu_online = cpu_online(this_cpu_id);
> +
> +		if (system_state <= SYSTEM_RUNNING)
> +			pr_crit("stopping secondary CPUs\n");
> +		arch_smp_stop_call(&mask);
> +
> +		/*
> +		 * Defaults to wait up to one second for other CPUs to stop;
> +		 * architectures can modify the default timeout or request
> +		 * to wait forever.
> +		 *
> +		 * Here we rely simply on cpu_online_mask to sync with
> +		 * arch-specific stop code without bloating the code with an
> +		 * additional atomic_t ad-hoc counter.
> +		 *
> +		 * As a consequence we'll need proper explicit memory barriers
> +		 * in case the other CPUs running the arch-specific stop-code
> +		 * would need to commit to memory some data (like saved_regs).
> +		 */
> +		wait = smp_stop_get_wait_timeout_us(&timeout);
> +		while (num_online_cpus() > this_cpu_online &&
> +		       (wait || timeout--))
> +			udelay(1);
> +		/* ensure any stopping-CPUs memory access is made visible */
> +		smp_rmb();
> +		if (num_online_cpus() > this_cpu_online)
> +			pr_warn("failed to stop secondary CPUs %*pbl\n",
> +				cpumask_pr_args(cpu_online_mask));
> +	}
> +	/* Perform final (possibly arch-specific) work on this CPU */
> +	arch_smp_cpus_stop_complete();
> +}
> +#endif
> 


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

  reply	other threads:[~2019-09-23 14:20 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-09-13 18:19 [RFC PATCH v2 00/12] Unify SMP stop generic logic to common code Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 01/12] smp: add generic SMP-stop support " Cristian Marussi
2019-09-23 14:20   ` Cristian Marussi [this message]
2019-09-13 18:19 ` [RFC PATCH v2 02/12] smp: unify crash_ and smp_send_stop() logic Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 03/12] smp: coordinate concurrent crash/smp stop calls Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 04/12] smp: address races of starting CPUs while stopping Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 05/12] arm64: smp: use generic SMP stop common code Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 06/12] arm64: smp: use SMP crash-stop " Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 07/12] arm64: smp: add arch specific cpu parking helper Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 08/12] x86: smp: use generic SMP stop common code Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 09/12] x86: smp: use SMP crash-stop " Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 10/12] arm: smp: use generic SMP stop " Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 11/12] arm: smp: use SMP crash-stop " Cristian Marussi
2019-09-13 18:19 ` [RFC PATCH v2 12/12] sparc64: smp: use generic SMP stop " Cristian Marussi
2019-09-13 18:27 ` [RFC PATCH v2 00/12] Unify SMP stop generic logic to " Russell King - ARM Linux admin
2019-09-16  9:38   ` Cristian Marussi

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1d7e90b7-71d8-3c12-9c1b-7049a1d7e32f@arm.com \
    --to=cristian.marussi@arm.com \
    --cc=catalin.marinas@arm.com \
    --cc=dave.martin@arm.com \
    --cc=davem@davemloft.net \
    --cc=dzickus@redhat.com \
    --cc=ehabkost@redhat.com \
    --cc=hch@infradead.org \
    --cc=hidehiro.kawai.ez@hitachi.com \
    --cc=james.morse@arm.com \
    --cc=linux-arch@vger.kernel.org \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux@armlinux.org.uk \
    --cc=mark.rutland@arm.com \
    --cc=mingo@redhat.com \
    --cc=peterz@infradead.org \
    --cc=sparclinux@vger.kernel.org \
    --cc=takahiro.akashi@linaro.org \
    --cc=tglx@linutronix.de \
    --cc=will@kernel.org \
    --cc=x86@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).