Re: Performance regressions in "boot_time" tests in Linux 5.8 Kernel

["bhe@redhat.com" <bhe@redhat.com>]

On 12/11/20 at 04:16pm, Rahul Gopakumar wrote:
> Hi Baoquan,
> 
> We re-evaluated your last patch and it seems to be fixing the
> initial performance bug reported. During our previous testing,
> we did not apply the patch rightly hence it was reporting
> some issues. 
> 
> Here is the dmesg log confirming no delay in the draft patch.
> 
> Vanilla (5.10 rc3)
> ------------------
> 
> [    0.024011] On node 2 totalpages: 89391104
> [    0.024012]   Normal zone: 1445888 pages used for memmap
> [    0.024012]   Normal zone: 89391104 pages, LIFO batch:63
> [    2.054646] ACPI: PM-Timer IO Port: 0x448 --------------> 2 secs delay
> 
> Patch
> ------
> 
> [    0.024166] On node 2 totalpages: 89391104
> [    0.024167]   Normal zone: 1445888 pages used for memmap
> [    0.024167]   Normal zone: 89391104 pages, LIFO batch:63
> [    0.026694] ACPI: PM-Timer IO Port: 0x448 --------------> No delay
> 
> Attached dmesg logs. Let me know if anything is needed from our end.

I posted formal patchset to fix this issue. The patch 1 is doing the
fix, and almost the same as the draft v2 patch I attached in this thread.
Please feel free to help test and add your Tested-by: tag in the patch
thread if possible.

> 
> 
> 
> From: Rahul Gopakumar <gopakumarr@vmware.com>
> Sent: 24 November 2020 8:33 PM
> To: bhe@redhat.com <bhe@redhat.com>
> Cc: linux-mm@kvack.org <linux-mm@kvack.org>; linux-kernel@vger.kernel.org <linux-kernel@vger.kernel.org>; akpm@linux-foundation.org <akpm@linux-foundation.org>; natechancellor@gmail.com <natechancellor@gmail.com>; ndesaulniers@google.com <ndesaulniers@google.com>; clang-built-linux@googlegroups.com <clang-built-linux@googlegroups.com>; rostedt@goodmis.org <rostedt@goodmis.org>; Rajender M <manir@vmware.com>; Yiu Cho Lau <lauyiuch@vmware.com>; Peter Jonasson <pjonasson@vmware.com>; Venkatesh Rajaram <rajaramv@vmware.com>
> Subject: Re: Performance regressions in "boot_time" tests in Linux 5.8 Kernel 
>  
> Hi Baoquan,
> 
> We applied the new patch to 5.10 rc3 and tested it. We are still
> observing the same page corruption issue which we saw with the
> old patch. This is causing 3 secs delay in boot time.
> 
> Attached dmesg log from the new patch and also from vanilla
> 5.10 rc3 kernel.
> 
> There are multiple lines like below in the dmesg log of the
> new patch.
> 
> "BUG: Bad page state in process swapper  pfn:ab08001"
> 
> ________________________________________
> From: bhe@redhat.com <bhe@redhat.com>
> Sent: 22 November 2020 6:38 AM
> To: Rahul Gopakumar
> Cc: linux-mm@kvack.org; linux-kernel@vger.kernel.org; akpm@linux-foundation.org; natechancellor@gmail.com; ndesaulniers@google.com; clang-built-linux@googlegroups.com; rostedt@goodmis.org; Rajender M; Yiu Cho Lau; Peter Jonasson; Venkatesh Rajaram
> Subject: Re: Performance regressions in "boot_time" tests in Linux 5.8 Kernel
> 
> On 11/20/20 at 03:11am, Rahul Gopakumar wrote:
> > Hi Baoquan,
> >
> > To which commit should we apply the draft patch. We tried applying
> > the patch to the commit 3e4fb4346c781068610d03c12b16c0cfb0fd24a3
> > (the one we used for applying the previous patch) but it fails.
> 
> I tested on 5.10-rc3+. You can append below change to the old patch in
> your testing kernel.
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index fa6076e1a840..5e5b74e88d69 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -448,6 +448,8 @@ defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
>         if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
>                 return false;
> 
> +       if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
> +               return true;
>         /*
>          * We start only with one section of pages, more pages are added as
>          * needed until the rest of deferred pages are initialized.