2021-01-30 09:28:45

by Miaohe Lin

Subject: [PATCH] hugetlbfs: show pagesize in unit of GB if possible

Hugepage size in unit of GB is supported. We could show pagesize in unit of
GB to make it more friendly to read. Also rework the calculation code of
page size unit to make it more readable.

Signed-off-by: Miaohe Lin <[email protected]>
---
fs/hugetlbfs/inode.c | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 3a08fbae3b53..40a9795f250a 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -1014,11 +1014,15 @@ static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
if (sbinfo->max_inodes != -1)
seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);

- hpage_size /= 1024;
- mod = 'K';
- if (hpage_size >= 1024) {
- hpage_size /= 1024;
+ if (hpage_size >= SZ_1G) {
+ hpage_size /= SZ_1G;
+ mod = 'G';
+ } else if (hpage_size >= SZ_1M) {
+ hpage_size /= SZ_1M;
mod = 'M';
+ } else {
+ hpage_size /= SZ_1K;
+ mod = 'K';
}
seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
if (spool) {
--
2.19.1

2021-01-30 22:14:36

by David Rientjes

Subject: Re: [PATCH] hugetlbfs: show pagesize in unit of GB if possible

On Sat, 30 Jan 2021, Miaohe Lin wrote:

> Hugepage size in unit of GB is supported. We could show pagesize in unit of
> GB to make it more friendly to read. Also rework the calculation code of
> page size unit to make it more readable.
>
> Signed-off-by: Miaohe Lin <[email protected]>
> ---
> fs/hugetlbfs/inode.c | 12 ++++++++----
> 1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> index 3a08fbae3b53..40a9795f250a 100644
> --- a/fs/hugetlbfs/inode.c
> +++ b/fs/hugetlbfs/inode.c
> @@ -1014,11 +1014,15 @@ static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
> if (sbinfo->max_inodes != -1)
> seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
>
> - hpage_size /= 1024;
> - mod = 'K';
> - if (hpage_size >= 1024) {
> - hpage_size /= 1024;
> + if (hpage_size >= SZ_1G) {
> + hpage_size /= SZ_1G;
> + mod = 'G';
> + } else if (hpage_size >= SZ_1M) {
> + hpage_size /= SZ_1M;
> mod = 'M';
> + } else {
> + hpage_size /= SZ_1K;
> + mod = 'K';
> }
> seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
> if (spool) {

NACK, this can break existing userspace parsing.

2021-02-01 01:26:10

by Miaohe Lin

Subject: Re: [PATCH] hugetlbfs: show pagesize in unit of GB if possible

Hi:
On 2021/1/31 6:07, David Rientjes wrote:
> On Sat, 30 Jan 2021, Miaohe Lin wrote:
>
>> Hugepage size in unit of GB is supported. We could show pagesize in unit of
>> GB to make it more friendly to read. Also rework the calculation code of
>> page size unit to make it more readable.
>>
>> Signed-off-by: Miaohe Lin <[email protected]>
>> ---
>> fs/hugetlbfs/inode.c | 12 ++++++++----
>> 1 file changed, 8 insertions(+), 4 deletions(-)
>>
>> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
>> index 3a08fbae3b53..40a9795f250a 100644
>> --- a/fs/hugetlbfs/inode.c
>> +++ b/fs/hugetlbfs/inode.c
>> @@ -1014,11 +1014,15 @@ static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
>> if (sbinfo->max_inodes != -1)
>> seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
>>
>> - hpage_size /= 1024;
>> - mod = 'K';
>> - if (hpage_size >= 1024) {
>> - hpage_size /= 1024;
>> + if (hpage_size >= SZ_1G) {
>> + hpage_size /= SZ_1G;
>> + mod = 'G';
>> + } else if (hpage_size >= SZ_1M) {
>> + hpage_size /= SZ_1M;
>> mod = 'M';
>> + } else {
>> + hpage_size /= SZ_1K;
>> + mod = 'K';
>> }
>> seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
>> if (spool) {
>
> NACK, this can break existing userspace parsing.
> .
>

I see. I should take care of this. Many thanks.

2021-02-01 21:34:19

by David Rientjes

Subject: Re: [PATCH] hugetlbfs: show pagesize in unit of GB if possible

On Mon, 1 Feb 2021, Miaohe Lin wrote:

> >> Hugepage size in unit of GB is supported. We could show pagesize in unit of
> >> GB to make it more friendly to read. Also rework the calculation code of
> >> page size unit to make it more readable.
> >>
> >> Signed-off-by: Miaohe Lin <[email protected]>
> >> ---
> >> fs/hugetlbfs/inode.c | 12 ++++++++----
> >> 1 file changed, 8 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> >> index 3a08fbae3b53..40a9795f250a 100644
> >> --- a/fs/hugetlbfs/inode.c
> >> +++ b/fs/hugetlbfs/inode.c
> >> @@ -1014,11 +1014,15 @@ static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
> >> if (sbinfo->max_inodes != -1)
> >> seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
> >>
> >> - hpage_size /= 1024;
> >> - mod = 'K';
> >> - if (hpage_size >= 1024) {
> >> - hpage_size /= 1024;
> >> + if (hpage_size >= SZ_1G) {
> >> + hpage_size /= SZ_1G;
> >> + mod = 'G';
> >> + } else if (hpage_size >= SZ_1M) {
> >> + hpage_size /= SZ_1M;
> >> mod = 'M';
> >> + } else {
> >> + hpage_size /= SZ_1K;
> >> + mod = 'K';
> >> }
> >> seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
> >> if (spool) {
> >
> > NACK, this can break existing userspace parsing.
> > .
> >
>
> I see. I should take care of this. Many thanks.
>

Thanks. It's an important point to emphasize because it shows how
important user-facing interfaces are. Once the hugetlbfs mount options
are printed in the way they are, it becomes very difficult to change them
because there can be userspace in the wild that would break as a result.
This is why it's crucial to be very careful when specifying user-facing
interfaces the first time so they are extensible.

2021-02-02 01:28:16

by Miaohe Lin

Subject: Re: [PATCH] hugetlbfs: show pagesize in unit of GB if possible

Hi:
On 2021/2/2 5:31, David Rientjes wrote:
> On Mon, 1 Feb 2021, Miaohe Lin wrote:
>
>>>> Hugepage size in unit of GB is supported. We could show pagesize in unit of
>>>> GB to make it more friendly to read. Also rework the calculation code of
>>>> page size unit to make it more readable.
>>>>
>>>> Signed-off-by: Miaohe Lin <[email protected]>
>>>> ---
>>>> fs/hugetlbfs/inode.c | 12 ++++++++----
>>>> 1 file changed, 8 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
>>>> index 3a08fbae3b53..40a9795f250a 100644
>>>> --- a/fs/hugetlbfs/inode.c
>>>> +++ b/fs/hugetlbfs/inode.c
>>>> @@ -1014,11 +1014,15 @@ static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
>>>> if (sbinfo->max_inodes != -1)
>>>> seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
>>>>
>>>> - hpage_size /= 1024;
>>>> - mod = 'K';
>>>> - if (hpage_size >= 1024) {
>>>> - hpage_size /= 1024;
>>>> + if (hpage_size >= SZ_1G) {
>>>> + hpage_size /= SZ_1G;
>>>> + mod = 'G';
>>>> + } else if (hpage_size >= SZ_1M) {
>>>> + hpage_size /= SZ_1M;
>>>> mod = 'M';
>>>> + } else {
>>>> + hpage_size /= SZ_1K;
>>>> + mod = 'K';
>>>> }
>>>> seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
>>>> if (spool) {
>>>
>>> NACK, this can break existing userspace parsing.
>>> .
>>>
>>
>> I see. I should take care of this. Many thanks.
>>
>
> Thanks. It's an important point to emphasize because it shows how
> important user-facing interfaces are. Once the hugetlbfs mount options
> are printed in the way they are, it becomes very difficult to change them
> because there can be userspace in the wild that would break as a result.
> This is why it's crucial to be very careful when specifying user-facing
> interfaces the first time so they are extensible.
> .
>

Many thanks for detailed explanation! I will keep this in mind! Thanks again.