Re: 4.4-rt nilfs problem

[Ulrich Hecht <uli@fpond.eu>]

> On 12/11/2024 10:56 PM CET Pavel Machek <pavel@denx.de> wrote:
> This 4.4-rt release is proving more "fun" then average. Last problem I
> hit is this one:
> 
> https://gitlab.com/cip-project/cip-kernel/linux-cip/-/jobs/8612057735
> 
> fs/nilfs2/dir.c: In function 'nilfs_find_entry':
> 2705
> fs/nilfs2/dir.c:350:31: error: initialization of 'char *' from incompatible pointer type 'struct page *' [-Werror=incompatible-pointer-types]
> 2706
>   350 |                 char *kaddr = nilfs_get_page(dir, n);
> 2707
>       |                               ^~~~~~~~~~~~~~
> 2708
> 
> And I believe it is caused by 3ab089df2d67491785a5bdaab01cfab8eb35dab5
> :
> 
>     nilfs2: propagate directory read errors from nilfs_find_entry()
> 
> @@ -345,25 +347,26 @@ nilfs_find_entry(struct inode *dir, const struct qstr *qstr,
>                 start = 0;
>         n = start;
>         do {
> -               char *kaddr;
> -               page = nilfs_get_page(dir, n);
> -               if (!IS_ERR(page)) {
> -                       kaddr = page_address(page);
> ...
> +               char *kaddr = nilfs_get_page(dir, n);
> 
> Compiler does not like page * assigned into char *, and I don't blame
> him.

It seems this is missing a dependency, namely ea2ac9238d4919bdc6963a2b487a65ccdaa11c78 ("nilfs2: return the mapped address from nilfs_get_page()"), which in turn has more dependencies...

> I could probably hack something up -- like below -- but it raises
> questions such as
> 
> a) "should kaddr by checked for IS_ERR like that"?

AFAICT, no. page_address() returns NULL on error.

> b) "am I comfortable changing code we are not testing"?

Given that that is what this backport does and how well that worked, I guess the answer is no...

When maintaining file systems for 4.4, nilfs2 is the biggest effort by a mile. From what I can tell it's used in only one config (moxa_mxc_defconfig). Is there any way to find out if it is actually used or just happens to be enabled?

CU
Uli