| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <yzgn3nbaeftrthqwddwt3gap4uni4api2r2uik2gxoimnpdiy7@hty77udv6un2>
Date: Tue, 19 Aug 2025 20:03:08 +1000
From: Alistair Popple <apopple@...dia.com>
To: Thomas Hellström <thomas.hellstrom@...ux.intel.com>
Cc: intel-xe@...ts.freedesktop.org, Jason Gunthorpe <jgg@...pe.ca>,
Andrew Morton <akpm@...ux-foundation.org>, Simona Vetter <simona.vetter@...ll.ch>,
Dave Airlie <airlied@...il.com>, dri-devel@...ts.freedesktop.org, linux-mm@...ck.org,
linux-kernel@...r.kernel.org, Matthew Brost <matthew.brost@...el.com>,
Christian König <christian.koenig@....com>
Subject: Re: [RFC PATCH 1/6] mm/mmu_notifier: Allow multiple struct
mmu_interval_notifier passes
On Sat, Aug 09, 2025 at 03:51:32PM +0200, Thomas Hellström wrote:
> GPU use-cases for mmu_interval_notifiers with hmm often involve
> starting a gpu operation and then waiting for it to complete.
> These operations are typically context preemption or TLB flushing.
>
> With single-pass notifiers per GPU this doesn't scale in
> multi-gpu scenarios. In those scenarios we'd want to first start
> preemption- or TLB flushing on all GPUs and as a second pass wait
> for them to complete on all gpus.
>
> One can do this on per-driver basis multiplexing per-driver
> notifiers but that would mean sharing the notifier "user" lock
> across all GPUs and that doesn't scale well either, so adding support
> for multi-pass in the core appears like the right choice.
>
> Implement multi-pass capability in the mmu_interval_notifier. Use a
> linked list for the additional passes to minimize the impact for
> use-cases that don't need the multi-pass functionality.
>
> Cc: Jason Gunthorpe <jgg@...pe.ca>
> Cc: Andrew Morton <akpm@...ux-foundation.org>
> Cc: Simona Vetter <simona.vetter@...ll.ch>
> Cc: Dave Airlie <airlied@...il.com>
> Cc: <dri-devel@...ts.freedesktop.org>
> Cc: <linux-mm@...ck.org>
> Cc: <linux-kernel@...r.kernel.org>
>
> Signed-off-by: Thomas Hellström <thomas.hellstrom@...ux.intel.com>
> ---
> include/linux/mmu_notifier.h | 30 ++++++++++++++++
> mm/mmu_notifier.c | 67 +++++++++++++++++++++++++++++++-----
> 2 files changed, 88 insertions(+), 9 deletions(-)
>
> diff --git a/include/linux/mmu_notifier.h b/include/linux/mmu_notifier.h
> index d1094c2d5fb6..1107a8eafd8a 100644
> --- a/include/linux/mmu_notifier.h
> +++ b/include/linux/mmu_notifier.h
> @@ -233,6 +233,32 @@ struct mmu_notifier {
> unsigned int users;
> };
>
> +/**
> + * struct mmu_interval_notifier_pass - mmu_interval_notifier multi-pass abstraction
> + * @link: List link for the notifiers pending pass list
> + *
> + * Allocate, typically using GFP_NOWAIT in the interval notifier's first pass.
> + * If allocation fails (which is not unlikely under memory pressure), fall back
> + * to single-pass operation.
> + */
> +struct mmu_interval_notifier_pass {
If we limit the number of passes to two maybe call this
`mmu_interval_notifier_finish()`? ...
> + struct list_head link;
> + /**
> + * @pass: Driver callback for additionall pass.
> + * @additional_pass: Pointer to the mmu_interval_notifier_pass structure.
> + * @range: The mmu_notifier_range.
> + * @cur_seq: The current sequence set by the first pass.
> + *
> + * Return: Either a pointer to a valid mmu_interval_notifier_pass for
> + * another pass to be called, or %NULL if processing is complete for this
> + * notifier. There is no error reporting mechanism for additional passes.
> + */
> + struct mmu_interval_notifier_pass *
> + (*pass) (struct mmu_interval_notifier_pass *additional_pass,
... and call this `finish()` ...
> + const struct mmu_notifier_range *range,
> + unsigned long cur_seq);
> +};
> +
> /**
> * struct mmu_interval_notifier_ops
> * @invalidate: Upon return the caller must stop using any SPTEs within this
> @@ -243,6 +269,10 @@ struct mmu_interval_notifier_ops {
> bool (*invalidate)(struct mmu_interval_notifier *interval_sub,
> const struct mmu_notifier_range *range,
> unsigned long cur_seq);
> + bool (*invalidate_multipass)(struct mmu_interval_notifier *interval_sub,
... and then this could be called `invalidate_start()`. That might address some
of the concerns with naming.
> + const struct mmu_notifier_range *range,
> + unsigned long cur_seq,
> + struct mmu_interval_notifier_pass **pass);
> };
>
> struct mmu_interval_notifier {
> diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c
> index 8e0125dc0522..dd6af87db103 100644
> --- a/mm/mmu_notifier.c
> +++ b/mm/mmu_notifier.c
> @@ -260,6 +260,22 @@ mmu_interval_read_begin(struct mmu_interval_notifier *interval_sub)
> }
> EXPORT_SYMBOL_GPL(mmu_interval_read_begin);
>
> +static void mn_itree_additional_passes(struct list_head *additional_passes,
> + const struct mmu_notifier_range *range,
> + unsigned long cur_seq)
> +{
> + struct mmu_interval_notifier_pass *p, *next;
> +
> + while (!list_empty(additional_passes)) {
> + list_for_each_entry_safe(p, next, additional_passes, link) {
> + list_del_init(&p->link);
> + p = p->pass(p, range, cur_seq);
> + if (p)
> + list_add_tail(&p->link, additional_passes);
> + }
> + }
> +}
> +
> static void mn_itree_release(struct mmu_notifier_subscriptions *subscriptions,
> struct mm_struct *mm)
> {
> @@ -272,17 +288,32 @@ static void mn_itree_release(struct mmu_notifier_subscriptions *subscriptions,
> };
> struct mmu_interval_notifier *interval_sub;
> unsigned long cur_seq;
> + LIST_HEAD(additional_passes);
> bool ret;
>
> for (interval_sub =
> mn_itree_inv_start_range(subscriptions, &range, &cur_seq);
> interval_sub;
> interval_sub = mn_itree_inv_next(interval_sub, &range)) {
> - ret = interval_sub->ops->invalidate(interval_sub, &range,
> - cur_seq);
> + if (interval_sub->ops->invalidate_multipass) {
> + struct mmu_interval_notifier_pass *second = NULL;
> +
> + ret = interval_sub->ops->invalidate_multipass(interval_sub,
> + &range,
> + cur_seq,
> + &second);
> + if (ret && second)
> + list_add_tail(&second->link, &additional_passes);
> +
> + } else {
> + ret = interval_sub->ops->invalidate(interval_sub,
> + &range,
> + cur_seq);
> + }
> WARN_ON(!ret);
> }
>
> + mn_itree_additional_passes(&additional_passes, &range, cur_seq);
> mn_itree_inv_end(subscriptions);
> }
>
> @@ -431,6 +462,8 @@ static int mn_itree_invalidate(struct mmu_notifier_subscriptions *subscriptions,
> {
> struct mmu_interval_notifier *interval_sub;
> unsigned long cur_seq;
> + LIST_HEAD(additional_passes);
> + int err = 0;
>
> for (interval_sub =
> mn_itree_inv_start_range(subscriptions, range, &cur_seq);
> @@ -438,23 +471,39 @@ static int mn_itree_invalidate(struct mmu_notifier_subscriptions *subscriptions,
> interval_sub = mn_itree_inv_next(interval_sub, range)) {
> bool ret;
>
> - ret = interval_sub->ops->invalidate(interval_sub, range,
> - cur_seq);
> + if (interval_sub->ops->invalidate_multipass) {
> + struct mmu_interval_notifier_pass *second = NULL;
> +
> + ret = interval_sub->ops->invalidate_multipass(interval_sub,
> + range,
> + cur_seq,
> + &second);
> + if (ret && second)
> + list_add_tail(&second->link, &additional_passes);
> +
> + } else {
> + ret = interval_sub->ops->invalidate(interval_sub,
> + range,
> + cur_seq);
> + }
> if (!ret) {
> if (WARN_ON(mmu_notifier_range_blockable(range)))
> continue;
> - goto out_would_block;
> + err = -EAGAIN;
> + break;
> }
> }
> - return 0;
>
> -out_would_block:
> + mn_itree_additional_passes(&additional_passes, range, cur_seq);
> +
> /*
> * On -EAGAIN the non-blocking caller is not allowed to call
> * invalidate_range_end()
> */
> - mn_itree_inv_end(subscriptions);
> - return -EAGAIN;
> + if (err)
> + mn_itree_inv_end(subscriptions);
> +
> + return err;
> }
>
> static int mn_hlist_invalidate_range_start(
> --
> 2.50.1
>
>
Powered by blists - more mailing lists