From: Baolin Wang <baolin.wang@linux.alibaba.com>
To: "Maíra Canal" <mcanal@igalia.com>,
"Jonathan Corbet" <corbet@lwn.net>,
"Andrew Morton" <akpm@linux-foundation.org>,
"Hugh Dickins" <hughd@google.com>,
"Barry Song" <baohua@kernel.org>,
"David Hildenbrand" <david@redhat.com>,
"Ryan Roberts" <ryan.roberts@arm.com>,
"Lance Yang" <ioworker0@gmail.com>
Cc: linux-mm@kvack.org, linux-doc@vger.kernel.org,
linux-kernel@vger.kernel.org, kernel-dev@igalia.com
Subject: Re: [PATCH v2 2/4] mm: shmem: control THP support through the kernel command line
Date: Tue, 29 Oct 2024 10:03:58 +0800 [thread overview]
Message-ID: <6044e56a-7747-45d4-82c9-7a070950ddd2@linux.alibaba.com> (raw)
In-Reply-To: <20241029002324.1062723-3-mcanal@igalia.com>
On 2024/10/29 08:13, Maíra Canal wrote:
> Add a new kernel command line to control the hugepage allocation policy
> for the internal shmem mount, ``transparent_hugepage_shmem``. The
> parameter is similar to ``transparent_hugepage`` and has the following
> format:
>
> transparent_hugepage_shmem=<policy>
>
> where ``<policy>`` is one of the seven valid policies available for
> shmem.
>
> By configuring the default hugepage allocation policy for the internal
> shmem mount, applications that use shmem, such as the DRM GEM objects,
> can take advantage of mTHP before it's been configured through sysfs.
Just out of curiosity, do you have any performance benefit data when
using mTHP for DRM GEM objects?
> Signed-off-by: Maíra Canal <mcanal@igalia.com>
> ---
> .../admin-guide/kernel-parameters.txt | 7 +++
> Documentation/admin-guide/mm/transhuge.rst | 6 +++
> mm/shmem.c | 53 +++++++++++++++----
> 3 files changed, 57 insertions(+), 9 deletions(-)
>
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index 1666576acc0e..acabb04d0dd4 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -6926,6 +6926,13 @@
> See Documentation/admin-guide/mm/transhuge.rst
> for more details.
>
> + transparent_hugepage_shmem= [KNL]
> + Format: [always|within_size|advise|never|deny|force]
> + Can be used to control the hugepage allocation policy for
> + the internal shmem mount.
> + See Documentation/admin-guide/mm/transhuge.rst
> + for more details.
> +
> trusted.source= [KEYS]
> Format: <string>
> This parameter identifies the trust source as a backend
> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
> index 745055c3dc09..9b5b02c4d1ab 100644
> --- a/Documentation/admin-guide/mm/transhuge.rst
> +++ b/Documentation/admin-guide/mm/transhuge.rst
> @@ -326,6 +326,12 @@ PMD_ORDER THP policy will be overridden. If the policy for PMD_ORDER
> is not defined within a valid ``thp_anon``, its policy will default to
> ``never``.
>
> +Similarly to ``transparent_hugepage``, you can control the hugepage
> +allocation policy for the internal shmem mount by using the kernel parameter
> +``transparent_hugepage_shmem=<policy>``, where ``<policy>`` is one of the
> +seven valid policies for shmem (``always``, ``within_size``, ``advise``,
> +``never``, ``deny``, and ``force``).
> +
> Hugepages in tmpfs/shmem
> ========================
>
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 44282a296c33..26c1eb1b4b1d 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -582,7 +582,6 @@ static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
> }
> }
>
> -#if defined(CONFIG_SYSFS)
> static int shmem_parse_huge(const char *str)
> {
> if (!strcmp(str, "never"))
> @@ -599,7 +598,6 @@ static int shmem_parse_huge(const char *str)
> return SHMEM_HUGE_FORCE;
> return -EINVAL;
> }
> -#endif
>
> #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
> static const char *shmem_format_huge(int huge)
> @@ -624,6 +622,20 @@ static const char *shmem_format_huge(int huge)
> }
> #endif
>
> +static int shmem_valid_huge(int huge)
> +{
> + if (!has_transparent_hugepage() &&
> + huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
> + return -EINVAL;
> +
> + /* Do not override huge allocation policy with non-PMD sized mTHP */
> + if (huge == SHMEM_HUGE_FORCE &&
> + huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
> + return -EINVAL;
> +
> + return 0;
> +}
> +
> static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
> struct shrink_control *sc, unsigned long nr_to_free)
> {
> @@ -5070,14 +5082,10 @@ static ssize_t shmem_enabled_store(struct kobject *kobj,
> huge = shmem_parse_huge(tmp);
> if (huge == -EINVAL)
> return -EINVAL;
> - if (!has_transparent_hugepage() &&
> - huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
> - return -EINVAL;
>
> - /* Do not override huge allocation policy with non-PMD sized mTHP */
> - if (huge == SHMEM_HUGE_FORCE &&
> - huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
> - return -EINVAL;
> + err = shmem_valid_huge(huge);
> + if (err)
> + return err;
>
> shmem_huge = huge;
> if (shmem_huge > SHMEM_HUGE_DENY)
> @@ -5174,6 +5182,33 @@ struct kobj_attribute thpsize_shmem_enabled_attr =
> __ATTR(shmem_enabled, 0644, thpsize_shmem_enabled_show, thpsize_shmem_enabled_store);
> #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
>
> +#if defined(CONFIG_TRANSPARENT_HUGEPAGE)
> +
> +static int __init setup_transparent_hugepage_shmem(char *str)
> +{
> + int huge, ret = 0;
> +
> + if (!str)
> + goto out;
> +
> + huge = shmem_parse_huge(str);
> + if (huge == -EINVAL)
> + goto out;
Looks better. But shmem_parse_huge() is also a common part, and what I
am thinking is below:
diff --git a/mm/shmem.c b/mm/shmem.c
index f8b8b1ad2631..646d8943950a 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -629,24 +629,39 @@ static unsigned int
shmem_huge_global_enabled(struct inode *inode, pgoff_t index
}
}
-#if defined(CONFIG_SYSFS)
static int shmem_parse_huge(const char *str)
{
+ int huge;
+
+ if (!str)
+ return -EINVAL;
+
if (!strcmp(str, "never"))
- return SHMEM_HUGE_NEVER;
- if (!strcmp(str, "always"))
- return SHMEM_HUGE_ALWAYS;
- if (!strcmp(str, "within_size"))
- return SHMEM_HUGE_WITHIN_SIZE;
- if (!strcmp(str, "advise"))
- return SHMEM_HUGE_ADVISE;
- if (!strcmp(str, "deny"))
- return SHMEM_HUGE_DENY;
- if (!strcmp(str, "force"))
- return SHMEM_HUGE_FORCE;
- return -EINVAL;
+ huge = SHMEM_HUGE_NEVER;
+ else if (!strcmp(str, "always"))
+ huge = SHMEM_HUGE_ALWAYS;
+ else if (!strcmp(str, "within_size"))
+ huge = SHMEM_HUGE_WITHIN_SIZE;
+ else if (!strcmp(str, "advise"))
+ huge = SHMEM_HUGE_ADVISE;
+ else if (!strcmp(str, "deny"))
+ huge = SHMEM_HUGE_DENY;
+ else if (!strcmp(str, "force"))
+ huge = SHMEM_HUGE_FORCE;
+ else
+ return -EINVAL;
+
+ if (!has_transparent_hugepage() &&
+ huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
+ return -EINVAL;
+
+ /* Do not override huge allocation policy with non-PMD sized mTHP */
+ if (huge == SHMEM_HUGE_FORCE &&
+ huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
+ return -EINVAL;
+
+ return huge;
}
-#endif
#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
static const char *shmem_format_huge(int huge)
@@ -5104,16 +5119,8 @@ static ssize_t shmem_enabled_store(struct kobject
*kobj,
tmp[count - 1] = '\0';
huge = shmem_parse_huge(tmp);
- if (huge == -EINVAL)
- return -EINVAL;
- if (!has_transparent_hugepage() &&
- huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
- return -EINVAL;
-
- /* Do not override huge allocation policy with non-PMD sized mTHP */
- if (huge == SHMEM_HUGE_FORCE &&
- huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
- return -EINVAL;
+ if (huge < 0)
+ return huge;
shmem_huge = huge;
if (shmem_huge > SHMEM_HUGE_DENY)
@@ -5210,6 +5217,25 @@ struct kobj_attribute thpsize_shmem_enabled_attr =
__ATTR(shmem_enabled, 0644, thpsize_shmem_enabled_show,
thpsize_shmem_enabled_store);
#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
+#if defined(CONFIG_TRANSPARENT_HUGEPAGE)
+
+static int __init setup_transparent_hugepage_shmem(char *str)
+{
+ int huge;
+
+ huge = shmem_parse_huge(str);
+ if (huge < 0) {
+ pr_warn("transparent_hugepage_shmem= cannot parse,
ignored\n");
+ return 0;
+ }
+
+ shmem_huge = huge;
+ return 1;
+}
+__setup("transparent_hugepage_shmem=", setup_transparent_hugepage_shmem);
+
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+
#else /* !CONFIG_SHMEM */
/*
next prev parent reply other threads:[~2024-10-29 2:04 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-10-29 0:13 [PATCH v2 0/4] mm: add more kernel parameters to control mTHP Maíra Canal
2024-10-29 0:13 ` [PATCH v2 1/4] mm: fix docs for the kernel parameter ``thp_anon=`` Maíra Canal
2024-10-29 1:03 ` Barry Song
2024-10-29 11:03 ` Maíra Canal
2024-10-29 0:13 ` [PATCH v2 2/4] mm: shmem: control THP support through the kernel command line Maíra Canal
2024-10-29 2:03 ` Baolin Wang [this message]
2024-10-29 11:13 ` Maíra Canal
2024-10-29 0:13 ` [PATCH v2 3/4] mm: generalize the implementation of ``thp_anon=`` Maíra Canal
2024-10-29 1:07 ` Barry Song
2024-10-29 0:13 ` [PATCH v2 4/4] mm: shmem: override mTHP shmem default with a kernel parameter Maíra Canal
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=6044e56a-7747-45d4-82c9-7a070950ddd2@linux.alibaba.com \
--to=baolin.wang@linux.alibaba.com \
--cc=akpm@linux-foundation.org \
--cc=baohua@kernel.org \
--cc=corbet@lwn.net \
--cc=david@redhat.com \
--cc=hughd@google.com \
--cc=ioworker0@gmail.com \
--cc=kernel-dev@igalia.com \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mcanal@igalia.com \
--cc=ryan.roberts@arm.com \
/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