> On Feb 11, 2021, at 4:12 AM, Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Feb 10, 2021 at 10:58:37PM +0000, Chuck Lever wrote:
>>> Not in the short term due to bug load and other obligations.
>>>
>>> The original series had "mm, page_allocator: Only use per-cpu allocator
>>> for irq-safe requests" but that was ultimately rejected because softirqs
>>> were affected so it would have to be done without that patch.
>>>
>>> The last patch can be rebased easily enough but it only batch allocates
>>> order-0 pages. It's also only build tested and could be completely
>>> miserable in practice and as I didn't even try boot test let, let alone
>>> actually test it, it could be a giant pile of crap. To make high orders
>>> work, it would need significant reworking but if the API showed even
>>> partial benefit, it might motiviate someone to reimplement the bulk
>>> interfaces to perform better.
>>>
>>> Rebased diff, build tested only, might not even work
>>
>> Thanks, Mel, for kicking off a forward port.
>>
>> It compiles. I've added a patch to replace the page allocation loop
>> in svc_alloc_arg() with a call to alloc_pages_bulk().
>>
>> The server system deadlocks pretty quickly with any NFS traffic. Based
>> on some initial debugging, it appears that a pcplist is getting corrupted
>> and this causes the list_del() in __rmqueue_pcplist() to fail during a
>> a call to alloc_pages_bulk().
>>
>
> Parameters to __rmqueue_pcplist are garbage as the parameter order changed.
> I'm surprised it didn't blow up in a spectacular fashion. Again, this
> hasn't been near any testing and passing a list with high orders to
> free_pages_bulk() will corrupt lists too. Mostly it's a curiousity to see
> if there is justification for reworking the allocator to fundamentally
> deal in batches and then feed batches to pcp lists and the bulk allocator
> while leaving the normal GFP API as single page "batches". While that
> would be ideal, it's relatively high risk for regressions. There is still
> some scope for adding a basic bulk allocator before considering a major
> refactoring effort.
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index f8353ea7b977..8f3fe7de2cf7 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5892,7 +5892,7 @@ __alloc_pages_bulk_nodemask(gfp_t gfp_mask, unsigned int order,
> pcp_list = &pcp->lists[migratetype];
>
> while (nr_pages) {
> - page = __rmqueue_pcplist(zone, gfp_mask, migratetype,
> + page = __rmqueue_pcplist(zone, migratetype, alloc_flags,
> pcp, pcp_list);
> if (!page)
> break;
The NFS server is considerably more stable now. Thank you!
I confirmed that my patch is requesting and getting multiple pages.
The new NFSD code and the API seem to be working as expected.
The results are stunning. Each svc_alloc_arg() call here allocates
65 pages to satisfy a 256KB NFS READ request.
Before:
nfsd-972 [000] 584.513817: funcgraph_entry: + 35.385 us | svc_alloc_arg();
nfsd-979 [002] 584.513870: funcgraph_entry: + 29.051 us | svc_alloc_arg();
nfsd-980 [001] 584.513951: funcgraph_entry: + 29.178 us | svc_alloc_arg();
nfsd-983 [000] 584.514014: funcgraph_entry: + 29.211 us | svc_alloc_arg();
nfsd-976 [002] 584.514059: funcgraph_entry: + 29.315 us | svc_alloc_arg();
nfsd-974 [001] 584.514127: funcgraph_entry: + 29.237 us | svc_alloc_arg();
After:
nfsd-977 [002] 87.049425: funcgraph_entry: 4.293 us | svc_alloc_arg();
nfsd-981 [000] 87.049478: funcgraph_entry: 4.059 us | svc_alloc_arg();
nfsd-988 [001] 87.049549: funcgraph_entry: 4.474 us | svc_alloc_arg();
nfsd-983 [003] 87.049612: funcgraph_entry: 3.819 us | svc_alloc_arg();
nfsd-976 [000] 87.049619: funcgraph_entry: 3.869 us | svc_alloc_arg();
nfsd-980 [002] 87.049738: funcgraph_entry: 4.124 us | svc_alloc_arg();
nfsd-975 [000] 87.049769: funcgraph_entry: 3.734 us | svc_alloc_arg();
There appears to be little cost change for single-page allocations
using the bulk allocator (nr_pages=1):
Before:
nfsd-985 [003] 572.324517: funcgraph_entry: 0.332 us | svc_alloc_arg();
nfsd-986 [001] 572.324531: funcgraph_entry: 0.311 us | svc_alloc_arg();
nfsd-985 [003] 572.324701: funcgraph_entry: 0.311 us | svc_alloc_arg();
nfsd-986 [001] 572.324727: funcgraph_entry: 0.424 us | svc_alloc_arg();
nfsd-985 [003] 572.324760: funcgraph_entry: 0.332 us | svc_alloc_arg();
nfsd-986 [001] 572.324786: funcgraph_entry: 0.390 us | svc_alloc_arg();
After:
nfsd-989 [002] 75.043226: funcgraph_entry: 0.322 us | svc_alloc_arg();
nfsd-988 [001] 75.043436: funcgraph_entry: 0.368 us | svc_alloc_arg();
nfsd-989 [002] 75.043464: funcgraph_entry: 0.424 us | svc_alloc_arg();
nfsd-988 [001] 75.043490: funcgraph_entry: 0.317 us | svc_alloc_arg();
nfsd-989 [002] 75.043517: funcgraph_entry: 0.425 us | svc_alloc_arg();
nfsd-988 [001] 75.050025: funcgraph_entry: 0.407 us | svc_alloc_arg();
--
Chuck Lever
