On 10.01.23 13:25, Mirsad Todorovac wrote:
On 1/10/23 11:05, David Hildenbrand wrote:
On 09.01.23 22:41, Mirsad Goran Todorovac wrote:
On 1/9/2023 5:42 PM, David Hildenbrand wrote:
Hi, thank you for your reply.
I hope this is enough information for you to debug the issue.
I am standing by for any additional diagnostics needed.
Won't userfaultfd.c fail in a similar way?
Anyhow, khugepaged.c jas
#ifndef MADV_PAGEOUT #define MADV_PAGEOUT 21 #endif
So most probably we should do the same.
Actually, David, it turned out that userfaultfd.c compiled out-of-the-box, and side-by-side comparison showed that it also included "/home/marvin/linux/kernel/linux_torvalds/usr/include/asm-generic/mman-common.h"
The only remaining difference was including <linux/mman.h>, which fixed the issue w/o #ifdef ... #endif
Hope this helps.
Please find the following diff.
Regards, Mirsad
diff --git a/tools/testing/selftests/vm/cow.c b/tools/testing/selftests/vm/cow.c index 26f6ea3079e2..dd8cf12c6776 100644 --- a/tools/testing/selftests/vm/cow.c +++ b/tools/testing/selftests/vm/cow.c @@ -16,6 +16,7 @@ #include <fcntl.h> #include <dirent.h> #include <assert.h> +#include <linux/mman.h> #include <sys/mman.h> #include <sys/ioctl.h> #include <sys/wait.h>
I already sent a different fix [1]. I suspect when including linux/mman.h, it would still be problematic with older kernel headers that lack MADV_PAGEOUT (< v5.4).
I see your point.
But yeah, I saw that userfaultfd.c was fixed that way:
commit b773827e361952b3f53ac6fa4c4e39ccd632102e Author: Chengming Zhou zhouchengming@bytedance.com Date: Fri Mar 4 20:29:04 2022 -0800
kselftest/vm: fix tests build with old libc The error message when I build vm tests on debian10 (GLIBC 2.28): userfaultfd.c: In function `userfaultfd_pagemap_test': userfaultfd.c:1393:37: error: `MADV_PAGEOUT' undeclared (first use in this function); did you mean `MADV_RANDOM'? if (madvise(area_dst, test_pgsize, MADV_PAGEOUT)) ^~~~~~~~~~~~ MADV_RANDOM This patch includes these newer definitions from UAPI linux/mman.h, is useful to fix tests build on systems without these definitions in glibc sys/mman.h.
[1] https://lkml.kernel.org/r/20230109171255.488749-1-david@redhat.com
You're the boss :)
Heh, no I'm not :) I'm just raising that this turned out to be problematic unfortunately.
However, IMHO, having MADV_PAGEOUT defined in three or four places could make like miserable. OK, it is unlikely to change value, but something tells me that the right way to do it is to guarantee that the macro definition is unique.
I don't know what would be the right thing to do in pre-5.4 kernels w/o MADV_PAGEOUT defined.
Probably then the "(madvise(area_dst, test_pgsize, MADV_PAGEOUT)" gives EINVAL or is undefined?
Yes. The expectation is that it fails with ENOSYS or EINVAL if the kernel doesn't support it. The kernel might be different to the installed kernel headers.
In an ideal world, we'd be using the in-tree headers -- they are guaranteed to define what we need. So far, I failed to make it work (I thought it would work as expected before I had to do above mentioned fix). Maybe that can be made working?