[PATCH] mm/hugetlb: suppress wrong warning info when alloc gigantic page

[PATCH] mm/hugetlb: suppress wrong warning info when alloc gigantic page

[Chen Wandun]

If hugetlb_cma is enabled, it will skip boot time allocation
when allocating gigantic page, that doesn't means allocation
failure, so suppress this warning info.

Fixes: cf11e85fc08c ("mm: hugetlb: optionally allocate gigantic hugepages using cma")
Signed-off-by: Chen Wandun <chenwandun@huawei.com>
---
 mm/hugetlb.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index b6992297aa16..98a49cb9250c 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2465,7 +2465,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 		if (hstate_is_gigantic(h)) {
 			if (hugetlb_cma_size) {
 				pr_warn_once("HugeTLB: hugetlb_cma is enabled, skip boot time allocation\n");
-				break;
+				goto free;
 			}
 			if (!alloc_bootmem_huge_page(h))
 				break;
@@ -2483,7 +2483,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 			h->max_huge_pages, buf, i);
 		h->max_huge_pages = i;
 	}
-
+free:
 	kfree(node_alloc_noretry);
 }
 
-- 
2.25.1

Re: [PATCH] mm/hugetlb: suppress wrong warning info when alloc gigantic page

[Mike Kravetz]

On 2/19/21 4:39 AM, Chen Wandun wrote:
> If hugetlb_cma is enabled, it will skip boot time allocation
> when allocating gigantic page, that doesn't means allocation
> failure, so suppress this warning info.
> 

Normally the addition of warning messages is discouraged.  However, in
this case the additional message provides value.  Why?

Prior to the commit cf11e85fc08c, one could have a kernel command line
that contains:

hugepagesz=1G hugepages=16

This would allocate 16 1G pages at boot time.

After the commit, someone could specify a command line containing:

hugepagesz=1G hugepages=16 hugetlb_cma=16G

In this case, 16G of CMA will be reserved for 1G huge page allocations
after boot time.  The parameter 'hugepages=16' is ignored, and the warning
message is logged.  The warning message should only be logged when the
kernel parameter 'hugepages=' is ignored.

IMO, it make sense to log a warning if ignoring a user specified parameter.
The user should not be attempting boot time allocation and CMA reservation
for 1G pages.

I do not think we should drop the warning as the it tells the user thay
have specified two incompatible allocation options.
-- 
Mike Kravetz


> Fixes: cf11e85fc08c ("mm: hugetlb: optionally allocate gigantic hugepages using cma")
> Signed-off-by: Chen Wandun <chenwandun@huawei.com>
> ---
>  mm/hugetlb.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index b6992297aa16..98a49cb9250c 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -2465,7 +2465,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
>  		if (hstate_is_gigantic(h)) {
>  			if (hugetlb_cma_size) {
>  				pr_warn_once("HugeTLB: hugetlb_cma is enabled, skip boot time allocation\n");
> -				break;
> +				goto free;
>  			}
>  			if (!alloc_bootmem_huge_page(h))
>  				break;
> @@ -2483,7 +2483,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
>  			h->max_huge_pages, buf, i);
>  		h->max_huge_pages = i;
>  	}
> -
> +free:
>  	kfree(node_alloc_noretry);
>  }
>  
> 

Re: [PATCH] mm/hugetlb: suppress wrong warning info when alloc gigantic page

[David Hildenbrand]

On 19.02.21 20:14, Mike Kravetz wrote:
> On 2/19/21 4:39 AM, Chen Wandun wrote:
>> If hugetlb_cma is enabled, it will skip boot time allocation
>> when allocating gigantic page, that doesn't means allocation
>> failure, so suppress this warning info.
>>
> 
> Normally the addition of warning messages is discouraged.  However, in
> this case the additional message provides value.  Why?
> 
> Prior to the commit cf11e85fc08c, one could have a kernel command line
> that contains:
> 
> hugepagesz=1G hugepages=16
> 
> This would allocate 16 1G pages at boot time.
> 
> After the commit, someone could specify a command line containing:
> 
> hugepagesz=1G hugepages=16 hugetlb_cma=16G
> 
> In this case, 16G of CMA will be reserved for 1G huge page allocations
> after boot time.  The parameter 'hugepages=16' is ignored, and the warning
> message is logged.  The warning message should only be logged when the
> kernel parameter 'hugepages=' is ignored.
> 
> IMO, it make sense to log a warning if ignoring a user specified parameter.
> The user should not be attempting boot time allocation and CMA reservation
> for 1G pages.
> 
> I do not think we should drop the warning as the it tells the user thay
> have specified two incompatible allocation options.
> 

I agree. It has value.

-- 
Thanks,

David / dhildenb

Re: [PATCH] mm/hugetlb: suppress wrong warning info when alloc gigantic page

[Mike Kravetz]

On 3/4/21 1:35 AM, David Hildenbrand wrote:
> On 19.02.21 20:14, Mike Kravetz wrote:
>> On 2/19/21 4:39 AM, Chen Wandun wrote:
>>> If hugetlb_cma is enabled, it will skip boot time allocation
>>> when allocating gigantic page, that doesn't means allocation
>>> failure, so suppress this warning info.
>>>
>>
>> Normally the addition of warning messages is discouraged.  However, in
>> this case the additional message provides value.  Why?
>>
>> Prior to the commit cf11e85fc08c, one could have a kernel command line
>> that contains:
>>
>> hugepagesz=1G hugepages=16
>>
>> This would allocate 16 1G pages at boot time.
>>
>> After the commit, someone could specify a command line containing:
>>
>> hugepagesz=1G hugepages=16 hugetlb_cma=16G
>>
>> In this case, 16G of CMA will be reserved for 1G huge page allocations
>> after boot time.  The parameter 'hugepages=16' is ignored, and the warning
>> message is logged.  The warning message should only be logged when the
>> kernel parameter 'hugepages=' is ignored.
>>
>> IMO, it make sense to log a warning if ignoring a user specified parameter.
>> The user should not be attempting boot time allocation and CMA reservation
>> for 1G pages.
>>
>> I do not think we should drop the warning as the it tells the user thay
>> have specified two incompatible allocation options.
>>
> 
> I agree. It has value.
> 

Hi David,

Sorry my above reply was too quick as I did not take a close look at
the code/patch.  See,

https://lore.kernel.org/mm-commits/YDAbeDsG7GhV6s6B@carbon.dhcp.thefacebook.com/

This patch is actually in Andrew's tree.
-- 
Mike Kravetz

Re: [PATCH] mm/hugetlb: suppress wrong warning info when alloc gigantic page

[David Hildenbrand]

On 04.03.21 18:20, Mike Kravetz wrote:
> On 3/4/21 1:35 AM, David Hildenbrand wrote:
>> On 19.02.21 20:14, Mike Kravetz wrote:
>>> On 2/19/21 4:39 AM, Chen Wandun wrote:
>>>> If hugetlb_cma is enabled, it will skip boot time allocation
>>>> when allocating gigantic page, that doesn't means allocation
>>>> failure, so suppress this warning info.
>>>>
>>>
>>> Normally the addition of warning messages is discouraged.  However, in
>>> this case the additional message provides value.  Why?
>>>
>>> Prior to the commit cf11e85fc08c, one could have a kernel command line
>>> that contains:
>>>
>>> hugepagesz=1G hugepages=16
>>>
>>> This would allocate 16 1G pages at boot time.
>>>
>>> After the commit, someone could specify a command line containing:
>>>
>>> hugepagesz=1G hugepages=16 hugetlb_cma=16G
>>>
>>> In this case, 16G of CMA will be reserved for 1G huge page allocations
>>> after boot time.  The parameter 'hugepages=16' is ignored, and the warning
>>> message is logged.  The warning message should only be logged when the
>>> kernel parameter 'hugepages=' is ignored.
>>>
>>> IMO, it make sense to log a warning if ignoring a user specified parameter.
>>> The user should not be attempting boot time allocation and CMA reservation
>>> for 1G pages.
>>>
>>> I do not think we should drop the warning as the it tells the user thay
>>> have specified two incompatible allocation options.
>>>
>>
>> I agree. It has value.
>>
> 
> Hi David,
> 
> Sorry my above reply was too quick as I did not take a close look at
> the code/patch.  See,
> 
> https://lore.kernel.org/mm-commits/YDAbeDsG7GhV6s6B@carbon.dhcp.thefacebook.com/
> 
> This patch is actually in Andrew's tree.
> 

Oh, I missed that discussion - thanks for the pointer!

-- 
Thanks,

David / dhildenb