NAK: [SRU][F][PATCH 1/1] s390/qeth: use memory reserves to back RX buffers

[Krzysztof Kozlowski]

On 04/02/2022 14:49, Thadeu Lima de Souza Cascardo wrote:
> On Fri, Feb 04, 2022 at 10:29:21AM +0100, Krzysztof Kozlowski wrote:
>> On 04/02/2022 07:47, frank.heimes at canonical.com wrote:
>>> From: Julian Wiedmann <jwi at linux.ibm.com>
>>>
>>> BugLink: https://bugs.launchpad.net/bugs/1959529
>>>
>>> Use dev_alloc_page() for backing the RX buffers with pages. This way we
>>> pick up __GFP_MEMALLOC.
>>>
>>> Signed-off-by: Julian Wiedmann <jwi at linux.ibm.com>
>>> Signed-off-by: David S. Miller <davem at davemloft.net>
>>> (backported from commit 714c9108851743bb718fbc1bfb81290f12a53854)
>>> Signed-off-by: Alexandra Winter <wintera at linux.ibm.com>
>>> Signed-off-by: Frank Heimes <frank.heimes at canonical.com>
>>> ---
>>>  drivers/s390/net/qeth_core_main.c | 4 ++--
>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
>>> index ec8c7a640d9e..61372e5c279b 100644
>>> --- a/drivers/s390/net/qeth_core_main.c
>>> +++ b/drivers/s390/net/qeth_core_main.c
>>> @@ -227,7 +227,7 @@ static int qeth_alloc_buffer_pool(struct qeth_card *card)
>>>  			return -ENOMEM;
>>>  		}
>>>  		for (j = 0; j < QETH_MAX_BUFFER_ELEMENTS(card); ++j) {
>>> -			ptr = (void *) __get_free_page(GFP_KERNEL);
>>> +			ptr = (void *) __dev_alloc_page(GFP_KERNEL);
>>
> 
> Or since the whole point is to use __GPF_MEMALLOC, simply do:
> 

It is not exact equivalent of __dev_alloc_page (__get_free_page() picks
closest NUMA node, __dev_alloc_page() loads entire NUMA policy and
chooses node according to it) , but I guess the differences here do not
matter.

The MEMALLOC, which allows using emergency memory pools and should not
be used outside of memory management subsystem or outside paths freeing
memory, seems a weird solution to reported problem of lack of memory.
The Bug report is saying that memory allocation failed, without too many
details so probably because of lack of memory in the system. :) This is
a RX buffer. Allocating a generic network RX buffer in case of
out-of-memory situation from emergency pools does not fit into MEMALLOC
purpose...

However if IBM feels it is proper approach for their problem, let's go
with simply MEMALLOC flag, like Thadeu suggested.

Best regards,
Krzysztof