On 30/01/2020 13:53, Kamil Dudka wrote:
> tests/cp/proc-short-read.sh expects that a pair of subsequent reads from
> /proc/kallsyms will always return the same content.  This does not seem to
> be a safe assumption any more.  The test has started to fail in our build
> environment.  I am not sure how to fix the test.  We could probably make
> it use another file from /proc but most of them are much smaller than
> kallsyms and/or suffer from the same problem.  Output of the failing test
> follows.
> 
> Kamil
> 
> 
> FAIL: tests/cp/proc-short-read
> ==============================

> + compare_ 1 2
> + diff -u 1 2
> --- 1	2020-01-29 12:04:36.923963121 +0000
> +++ 2	2020-01-29 12:04:37.026963484 +0000
> @@ -114819,81 +114819,132 @@
>   0000000000000000 t nfs_file_direct_read.cold	[nfs]
>   0000000000000000 t nfs_file_direct_write.cold	[nfs]
>   0000000000000000 r .LC0	[nfs]
> -0000000000000000 r .LC2	[nfs]
> -0000000000000000 r __ksymtab_nfs_pgio_current_mirror	[nfs]
> -0000000000000000 r __kstrtab_nfs_pgio_current_mirror	[nfs]

> +0000000000000000 r __func__.87038	[nfs]
> +0000000000000000 t __nfs_revalidate_inode.cold	[nfs]
> +0000000000000000 t nfs_revalidate_mapping.cold	[nfs]
> +0000000000000000 d nfs_net_ops	[nfs]
> +0000000000000000 t exit_nfs_fs	[nfs]
> +0000000000000000 r __param_enable_ino64	[nfs]
> +0000000000000000 r __param_str_enable_ino64	[nfs]
> +0000000000000000 r .LC15	[nfs]
> +0000000000000000 r __ksymtab_nfs_fs_type	[nfs]
> +0000000000000000 r __kstrtab_nfs_fs_type	[nfs]

> + fail=1
> + md5sum /proc/kallsyms
> + md5sum 2
> + sed 's/ .*//' 3
> + sed 's/ .*//' 4
> + compare sum.proc sum.2
> + compare_dev_null_ sum.proc sum.2
> + test 2 = 2
> + test xsum.proc = x/dev/null
> + test xsum.2 = x/dev/null
> + return 2
> + case $? in
> + compare_ sum.proc sum.2
> + diff -u sum.proc sum.2
> --- sum.proc	2020-01-29 12:04:37.172963999 +0000
> +++ sum.2	2020-01-29 12:04:37.175964009 +0000
> @@ -1 +1 @@
> -226cd09830f68c56edda0b9272be66e4
> +37d7e78173b2a31d5f27cc66aa52e72a
> + fail=1

Interesting.
The attached changes to /proc/cpuinfo
which is a bit more awkward, but should be a valid test most of the time,
and is also the file for which the original bug report was against.

cheers,
Pádraig