RE: [PATCH] net/ipv4: fix type mismatch in inet_ehash_locks_alloc() causing build failure

From: Kuniyuki Iwashima <redacted>
Date: Sun,  8 Jun 2025 13:11:51 -0700

quoted

From: Eliav Farber <redacted>
Date: Sun, 8 Jun 2025 06:07:26 +0000

quoted

Fix compilation warning:

In file included from ./include/linux/kernel.h:15,
                 from ./include/linux/list.h:9,
                 from ./include/linux/module.h:12,
                 from net/ipv4/inet_hashtables.c:12:
net/ipv4/inet_hashtables.c: In function ‘inet_ehash_locks_alloc’:
./include/linux/minmax.h:20:35: warning: comparison of distinct pointer types lacks a cast
   20 |         (!!(sizeof((typeof(x) *)1 == (typeof(y) *)1)))
      |                                   ^~
./include/linux/minmax.h:26:18: note: in expansion of macro ‘__typecheck’
   26 |                 (__typecheck(x, y) && __no_side_effects(x, y))
      |                  ^~~~~~~~~~~
./include/linux/minmax.h:36:31: note: in expansion of macro ‘__safe_cmp’
   36 |         __builtin_choose_expr(__safe_cmp(x, y), \
      |                               ^~~~~~~~~~
./include/linux/minmax.h:52:25: note: in expansion of macro ‘__careful_cmp’
   52 | #define max(x, y)       __careful_cmp(x, y, >)
      |                         ^~~~~~~~~~~~~
net/ipv4/inet_hashtables.c:946:19: note: in expansion of macro ‘max’
  946 |         nblocks = max(nblocks, num_online_nodes() * PAGE_SIZE / locksz);
      |                   ^~~
  CC      block/badblocks.o

When warnings are treated as errors, this causes the build to fail.

The issue is a type mismatch between the operands passed to the max()
macro. Here, nblocks is an unsigned int, while the expression
num_online_nodes() * PAGE_SIZE / locksz is promoted to unsigned long.

This happens because:
 - num_online_nodes() returns int
 - PAGE_SIZE is typically defined as an unsigned long (depending on the
   architecture)
 - locksz is unsigned int

The resulting arithmetic expression is promoted to unsigned long.

Thus, the max() macro compares values of different types: unsigned int
vs unsigned long.

This issue was introduced in commit b53d6e9525af ("tcp: bring back NUMA
dispersion in inet_ehash_locks_alloc()") during the update from kernel
v5.10.237 to v5.10.238.

Please use the upstream SHA1, f8ece40786c9.

Fixed in V2

quoted

quoted

It does not exist in newer kernel branches (e.g., v5.15.185 and all 6.x
branches), because they include commit d53b5d862acd ("minmax: allow

Same here, d03eba99f5bf.

Fixed in V2

quoted

But why not backport it to stable instead ?

I just checked the 5.10.238 thread.
https://lore.kernel.org/stable/2025060412-cursor-navigate-126d@gregkh/ (local)

---8<---

quoted

quoted

For both of these, I'll just let them be as they are ok, it's just the
mess of our min/max macro unwinding causes these issues.

Unless they really bother someone, and in that case, a patch to add the
correct type to the backport to make the noise go away would be greatly
appreciated.

Yeah that's a reasonable resolution, I will try to track down the missing
patches for minmax.h so we are warning free for the stable kernels.

I tried in the past, it's non-trivial.  What would be easier is to just
properly cast the variables in the places where this warning is showing
up to get rid of that warning.  We've done that in some backports in the
past as well.
---8<---

So this should be fixed up in the backport, and I guess this patch
targeted the stable trees ?

This patch is for v5.10.238. Newer kernels don't need it.

If so, please clarify that by specifying the stable version in the
subject and CCing the stable mainling list:

  Subject: [PATCH 5.10.y] tcp: ...
  Cc: stable@vger.kernel.org, ...

Done in v2