[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <f1c35067-9381-424b-b962-526b85d59ea9@igalia.com>
Date: Mon, 28 Oct 2024 08:34:08 -0300
From: Maíra Canal <mcanal@...lia.com>
To: Barry Song <21cnbao@...il.com>
Cc: Jonathan Corbet <corbet@....net>,
Andrew Morton <akpm@...ux-foundation.org>, Hugh Dickins <hughd@...gle.com>,
David Hildenbrand <david@...hat.com>, Ryan Roberts <ryan.roberts@....com>,
Baolin Wang <baolin.wang@...ux.alibaba.com>, Lance Yang
<ioworker0@...il.com>, linux-mm@...ck.org, linux-doc@...r.kernel.org,
linux-kernel@...r.kernel.org, kernel-dev@...lia.com
Subject: Re: [PATCH 3/3] mm: shmem: override mTHP shmem default with a kernel
parameter
Hi Barry,
On 28/10/24 08:09, Barry Song wrote:
> On Mon, Oct 28, 2024 at 6:10 PM Maíra Canal <mcanal@...lia.com> wrote:
>>
>> Hi Barry,
>>
>> On 27/10/24 18:54, Barry Song wrote:
>>> On Mon, Oct 28, 2024 at 6:58 AM Maíra Canal <mcanal@...lia.com> wrote:
>>>>
>>>> Add the ``thp_shmem=`` kernel command line to allow specifying the
>>>> default policy of each supported shmem hugepage size. The kernel parameter
>>>> accepts the following format:
>>>>
>>>> thp_shmem=<size>[KMG],<size>[KMG]:<policy>;<size>[KMG]-<size>[KMG]:<policy>
>>>>
>>>> For example,
>>>>
>>>> thp_shmem=16K-64K:always;128K,512K:inherit;256K:advise;1M-2M:never;4M-8M:within_size
>>>>
>>>> By configuring the default policy of several shmem huge pages, the user
>>>> can take advantage of mTHP before it's been configured through sysfs.
>>>>
>>>> Signed-off-by: Maíra Canal <mcanal@...lia.com>
>>>> ---
>>>> .../admin-guide/kernel-parameters.txt | 10 ++
>>>> Documentation/admin-guide/mm/transhuge.rst | 17 +++
>>>> mm/shmem.c | 109 +++++++++++++++++-
>>>> 3 files changed, 135 insertions(+), 1 deletion(-)
>>>>
>>>
>>> Hi Maíra,
>>>
>>>> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
>>>> index acabb04d0dd4..595fa096e28b 100644
>>>> --- a/Documentation/admin-guide/kernel-parameters.txt
>>>> +++ b/Documentation/admin-guide/kernel-parameters.txt
>>>> @@ -6700,6 +6700,16 @@
>>>> Force threading of all interrupt handlers except those
>>>> marked explicitly IRQF_NO_THREAD.
>>>>
>>>> + shmem_anon= [KNL]
>>>> + Format: <size>[KMG],<size>[KMG]:<policy>;<size>[KMG]-<size>[KMG]:<policy>
>>>> + Control the default policy of each hugepage size for the
>>>> + internal shmem mount. <policy> is one of policies available
>>>> + for the shmem mount ("always", "inherit", "never", "within_size",
>>>> + and "advise").
>>>> + It can be used multiple times for multiple shmem THP sizes.
>>>> + See Documentation/admin-guide/mm/transhuge.rst for more
>>>> + details.
>>>
>>> I'm not sure this is the right name. How about "thp_shmem"?
>>
>> Oops, sorry about that.
>>
>>>
>>>> +
>>>> topology= [S390,EARLY]
>>>> Format: {off | on}
>>>> Specify if the kernel should make use of the cpu
>>>> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
>>>> index 9b5b02c4d1ab..47e7fc30e22d 100644
>>>> --- a/Documentation/admin-guide/mm/transhuge.rst
>>>> +++ b/Documentation/admin-guide/mm/transhuge.rst
>>>> @@ -332,6 +332,23 @@ allocation policy for the internal shmem mount by using the kernel parameter
>>>> seven valid policies for shmem (``always``, ``within_size``, ``advise``,
>>>> ``never``, ``deny``, and ``force``).
>>>>
>>>> +In the same manner as ``thp_anon`` controls each supported anonymous THP
>>>> +size, ``thp_shmem`` controls each supported shmem THP size. ``thp_shmem``
>>>> +has the same format as ``thp_anon``, but also supports the policy
>>>> +``within_size``.
>>>> +
>>>> +``thp_shmem=`` may be specified multiple times to configure all THP sizes
>>>> +as required. If ``thp_shmem=`` is specified at least once, any shmem THP
>>>> +sizes not explicitly configured on the command line are implicitly set to
>>>> +``never``.
>>>> +
>>>> +``transparent_hugepage_shmem`` setting only affects the global toggle. If
>>>> +``thp_shmem`` is not specified, PMD_ORDER hugepage will default to
>>>> +``inherit``. However, if a valid ``thp_shmem`` setting is provided by the
>>>> +user, the PMD_ORDER hugepage policy will be overridden. If the policy for
>>>> +PMD_ORDER is not defined within a valid ``thp_shmem``, its policy will
>>>> +default to ``never``.
>>>> +
>>>> Hugepages in tmpfs/shmem
>>>> ========================
>>>>
>>>> diff --git a/mm/shmem.c b/mm/shmem.c
>>>> index 24cdeafd8260..0a7a7d04f725 100644
>>>> --- a/mm/shmem.c
>>>> +++ b/mm/shmem.c
[...]
>>>> static int __init setup_transparent_hugepage_shmem(char *str)
>>>> {
>>>> int huge, ret = 0;
>>>> @@ -5206,6 +5228,91 @@ static int __init setup_transparent_hugepage_shmem(char *str)
>>>> }
>>>> __setup("transparent_hugepage_shmem=", setup_transparent_hugepage_shmem);
>>>>
>>>> +static char str_dup[PAGE_SIZE] __initdata;
>>>> +static int __init setup_thp_shmem(char *str)
>>>> +{
>>>> + char *token, *range, *policy, *subtoken;
>>>> + unsigned long always, inherit, madvise, within_size;
>>>> + char *start_size, *end_size;
>>>> + int start, end, nr;
>>>> + char *p;
>>>> +
>>>> + if (!str || strlen(str) + 1 > PAGE_SIZE)
>>>> + goto err;
>>>> + strcpy(str_dup, str);
>>>> +
>>>> + always = huge_shmem_orders_always;
>>>> + inherit = huge_shmem_orders_inherit;
>>>> + madvise = huge_shmem_orders_madvise;
>>>> + within_size = huge_shmem_orders_within_size;
>>>> + p = str_dup;
>>>> + while ((token = strsep(&p, ";")) != NULL) {
>>>> + range = strsep(&token, ":");
>>>> + policy = token;
>>>> +
>>>> + if (!policy)
>>>> + goto err;
>>>> +
>>>> + while ((subtoken = strsep(&range, ",")) != NULL) {
>>>> + if (strchr(subtoken, '-')) {
>>>> + start_size = strsep(&subtoken, "-");
>>>> + end_size = subtoken;
>>>> +
>>>> + start = get_order_from_str(start_size);
>>>> + end = get_order_from_str(end_size);
>>>> + } else {
>>>> + start = end = get_order_from_str(subtoken);
>>>> + }
>>>> +
>>>> + if (start < 0 || end < 0 || start > end)
>>>> + goto err;
>>>> +
>>>> + nr = end - start + 1;
>>>> + if (!strcmp(policy, "always")) {
>>>> + bitmap_set(&always, start, nr);
>>>> + bitmap_clear(&inherit, start, nr);
>>>> + bitmap_clear(&madvise, start, nr);
>>>> + bitmap_clear(&within_size, start, nr);
>>>> + } else if (!strcmp(policy, "advise")) {
>>>> + bitmap_set(&madvise, start, nr);
>>>> + bitmap_clear(&inherit, start, nr);
>>>> + bitmap_clear(&always, start, nr);
>>>> + bitmap_clear(&within_size, start, nr);
>>>> + } else if (!strcmp(policy, "inherit")) {
>>>> + bitmap_set(&inherit, start, nr);
>>>> + bitmap_clear(&madvise, start, nr);
>>>> + bitmap_clear(&always, start, nr);
>>>> + bitmap_clear(&within_size, start, nr);
>>>> + } else if (!strcmp(policy, "within_size")) {
>>>> + bitmap_set(&within_size, start, nr);
>>>> + bitmap_clear(&inherit, start, nr);
>>>> + bitmap_clear(&madvise, start, nr);
>>>> + bitmap_clear(&always, start, nr);
>>>> + } else if (!strcmp(policy, "never")) {
>>>> + bitmap_clear(&inherit, start, nr);
>>>> + bitmap_clear(&madvise, start, nr);
>>>> + bitmap_clear(&always, start, nr);
>>>> + bitmap_clear(&within_size, start, nr);
>>>> + } else {
>>>> + pr_err("invalid policy %s in thp_shmem boot parameter\n", policy);
>>>> + goto err;
>>>> + }
>>>> + }
>>>> + }
>>>> +
>>>> + huge_shmem_orders_always = always;
>>>> + huge_shmem_orders_madvise = madvise;
>>>> + huge_shmem_orders_inherit = inherit;
>>>> + huge_shmem_orders_within_size = within_size;
>>>> + shmem_orders_configured = true;
>>>> + return 1;
>>>> +
>>>> +err:
>>>> + pr_warn("thp_shmem=%s: error parsing string, ignoring setting\n", str);
>>>> + return 0;
>>>> +}
>>>
>>> Can we share source code with thp_anon since there's a lot of duplication?
>>
>> I'm not a regular mm contributor and I'm most usually around drivers, so
>> I don't know exactly here I could add shared code. Should I add the
>> headers to "internal.h"?
>
> My comment isn't related to drivers or memory management. It's solely about
> avoiding code duplication. For example, we could create a shared function to
> handle both controls, reducing redundant code :-)
Let me rephrase it.
I completely agree that we should avoid code duplication. I'm asking
where is the best place to add the headers of the shared functions.
"linux/shmem_fs.h" doesn't look appropriate to me, so I believe the
remaining options would be "linux/huge_mm.h" or "internal.h".
I would like to know your opinion about those two options.
Best Regards,
- Maíra
>
>>
>> Best Regards,
>> - Maíra
>>
>>>
>>>> +__setup("thp_shmem=", setup_thp_shmem);
>>>> +
>>>> #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
>>>>
>>>> #else /* !CONFIG_SHMEM */
>>>> --
>>>> 2.46.2
>>>>
>>>
>>> Thanks
>>> barry
>>
Powered by blists - more mailing lists