upstream commit 277c8cb3e8ac ("MIPS: fix local_{add,sub}_return on MIPS64")
was backported to v5.15.27 as
commit f98371d2ac83 ("MIPS: fix local_{add,sub}_return on MIPS64")
but breaks MIPS build:
In file included from ./arch/mips/include/asm/local.h:8:0, from ./include/linux/genhd.h:20, from ./include/linux/blkdev.h:8, from ./include/linux/blk-cgroup.h:23, from ./include/linux/writeback.h:14, from ./include/linux/memcontrol.h:22, from ./include/net/sock.h:53, from ./include/linux/tcp.h:19, from drivers/net/slip/slip.c:91: ./arch/mips/include/asm/asm.h:68:0: warning: "END" redefined #define END(function) \
In file included from drivers/net/slip/slip.c:88:0: drivers/net/slip/slip.h:44:0: note: this is the location of the previous definition #define END 0300 /* indicates end of frame */
Analyses reveals that with the backported MIPS fix there is a new #include <asm/asm.h> introduced by ./arch/mips/include/asm/local.h which already defines some END macro.
But why does v5.16.x compile fine where
commit a0ecfd10d669c ("MIPS: fix local_{add,sub}_return on MIPS64")
is also present since v5.16.3?
Deeper analyses shows that there is another patch introduced in v5.16-rc1 which removed one #include in the above chain and therefore does not define END by asm/asm.h:
commit 348332e000697 ("mm: don't include <linux/blk-cgroup.h> in <linux/writeback.h>")
Hence, the MIPS fix should only be applied to branches where the mm fix is already present. Or the mm fix should be backported as well (if it has no side-effects).
Note: the MIPS fix was apparently not (yet?) applied to v5.10.y or earlier even tough the Fixes: 7232311ef14c ("local_t: mips extension") would be true.
BR and thanks, Nikolaus Schaller
Please fix <asm/asm.h>. The most trivial fix might be to only defined END() under __ASSEMBLY__, although in the long run it probably wants a better name as well.
Am 10.03.2022 um 17:28 schrieb Christoph Hellwig hch@lst.de:
Please fix <asm/asm.h>. The most trivial fix might be to only defined END() under __ASSEMBLY__, although in the long run it probably wants a better name as well.
Well
a) the bug (name conflict) does not occur in v5.16 or later b) I am in no way responsible for either of the patches or subsystems
So someone else has to either (partially?) backport
commit 348332e000697 ("mm: don't include <linux/blk-cgroup.h> in <linux/writeback.h>")
to 5.17.y
or write your proposed fix just for 5.17.y because it is not required elsewhere.
BR and thanks, Nikolaus
Am 10.03.2022 um 17:34 schrieb H. Nikolaus Schaller hns@goldelico.com:
Am 10.03.2022 um 17:28 schrieb Christoph Hellwig hch@lst.de:
Please fix <asm/asm.h>. The most trivial fix might be to only defined END() under __ASSEMBLY__, although in the long run it probably wants a better name as well.
Well
a) the bug (name conflict) does not occur in v5.16 or later b) I am in no way responsible for either of the patches or subsystems
So someone else has to either (partially?) backport
commit 348332e000697 ("mm: don't include <linux/blk-cgroup.h> in <linux/writeback.h>")
to 5.17.y
or write your proposed fix just for 5.17.y because it is not required elsewhere.
sorry, typo: just for v5.15.y
BR and thanks, Nikolaus
On Thu, 10 Mar 2022, Christoph Hellwig wrote:
Please fix <asm/asm.h>. The most trivial fix might be to only defined END() under __ASSEMBLY__, although in the long run it probably wants a better name as well.
FWIW the END macro has its roots in IRIX if not MIPSCO. It's been around for some 30 years if not more. MIPS/ULTRIX may have had it too, as it's originally all ECOFF stuff, i.e. `.end', `.ent', `.frame', etc. assembly pseudo-ops for symbol/debug information.
Maciej
On Thu, Mar 10, 2022 at 05:25:07PM +0100, H. Nikolaus Schaller wrote:
upstream commit 277c8cb3e8ac ("MIPS: fix local_{add,sub}_return on MIPS64")
was backported to v5.15.27 as
commit f98371d2ac83 ("MIPS: fix local_{add,sub}_return on MIPS64")
but breaks MIPS build:
In file included from ./arch/mips/include/asm/local.h:8:0, from ./include/linux/genhd.h:20, from ./include/linux/blkdev.h:8, from ./include/linux/blk-cgroup.h:23, from ./include/linux/writeback.h:14, from ./include/linux/memcontrol.h:22, from ./include/net/sock.h:53, from ./include/linux/tcp.h:19, from drivers/net/slip/slip.c:91: ./arch/mips/include/asm/asm.h:68:0: warning: "END" redefined #define END(function) \ In file included from drivers/net/slip/slip.c:88:0: drivers/net/slip/slip.h:44:0: note: this is the location of the previous definition #define END 0300 /* indicates end of frame */
Analyses reveals that with the backported MIPS fix there is a new #include <asm/asm.h> introduced by ./arch/mips/include/asm/local.h which already defines some END macro.
But why does v5.16.x compile fine where
commit a0ecfd10d669c ("MIPS: fix local_{add,sub}_return on MIPS64")
is also present since v5.16.3?
Deeper analyses shows that there is another patch introduced in v5.16-rc1 which removed one #include in the above chain and therefore does not define END by asm/asm.h:
commit 348332e000697 ("mm: don't include <linux/blk-cgroup.h> in <linux/writeback.h>")
Hence, the MIPS fix should only be applied to branches where the mm fix is already present. Or the mm fix should be backported as well (if it has no side-effects).
Note: the MIPS fix was apparently not (yet?) applied to v5.10.y or earlier even tough the Fixes: 7232311ef14c ("local_t: mips extension") would be true.
Thanks for the report. I'll work on resolving this for the next round of stable releases _after_ the ones that are currently out for review are released.
thanks,
greg k-h
On Thu, Mar 10, 2022 at 08:05:04PM +0100, Greg KH wrote:
Thanks for the report. I'll work on resolving this for the next round of stable releases _after_ the ones that are currently out for review are released.
If you think the genhd.h include cleanup is the best fix feel free to pick it up. But I really think it is just hiding the problem, and that problem will resurface.
linux-stable-mirror@lists.linaro.org
-
Christoph Hellwig -
Greg KH -
H. Nikolaus Schaller -
Maciej W. Rozycki