v3.18-rc5 build failure on ppc64 in Documentation/mic/mpssd/mpssd.o
One or more in the following commit set in the 3.18 merge window is causing build failure on ppc64 with the default config:
7b345771ba921361b318e95bf21b257c65ac141c Documentation: update include path for mpssd 8c2b0dc83d9840da4d993a5dbb15c5974ad5a188 Documentation: support glibc versions without htole macros 6ab0e475f1f38b6be90aff4ef3ebf928c4a73dc8 Documentation: fix misc. warnings adb19fb66eeebac07fe37d968725bb8906dadb8e Documentation: add makefiles for more targets
Building v3.18-rc5 and I have glibc 2.17.
The problem is that for big endian, the htole* macros are not simple compile time transforms by the precompiler, but rather block code that must be executed in function context.
Build errors:
===
HOSTCC Documentation/mic/mpssd/mpssd.o
In file included from /usr/include/bits/byteswap.h:34:0,
from /usr/include/endian.h:60,
from /usr/include/bits/waitstatus.h:64,
from /usr/include/stdlib.h:42,
from Documentation/mic/mpssd/mpssd.c:23:
Documentation/mic/mpssd/mpssd.c:93:10: error: braced-group within expression allowed only inside a function
.num = htole16(MIC_VRING_ENTRIES),
^
Documentation/mic/mpssd/mpssd.c:96:10: error: braced-group within expression allowed only inside a function
.num = htole16(MIC_VRING_ENTRIES),
^
Documentation/mic/mpssd/mpssd.c:113:10: error: braced-group within expression allowed only inside a function
.num = htole16(MIC_VRING_ENTRIES),
^
Documentation/mic/mpssd/mpssd.c:116:10: error: braced-group within expression allowed only inside a function
.num = htole16(MIC_VRING_ENTRIES),
^Documentation/mic/mpssd/mpssd.c:119:3: error: initializer element is not constant
.host_features = htole32(
^Documentation/mic/mpssd/mpssd.c:119:3: error: (near initialization for ‘virtnet_dev_page.host_features’) In file included from /usr/include/bits/byteswap.h:34:0,
from /usr/include/endian.h:60,
from /usr/include/bits/waitstatus.h:64,
from /usr/include/stdlib.h:42,
from Documentation/mic/mpssd/mpssd.c:23:
Documentation/mic/mpssd/mpssd.c:146:10: error: braced-group within expression allowed only inside a function
.num = htole16(MIC_VRING_ENTRIES),
^Documentation/mic/mpssd/mpssd.c:149:3: error: initializer element is not constant
htole32(1<<VIRTIO_BLK_F_SEG_MAX),
^Documentation/mic/mpssd/mpssd.c:149:3: error: (near initialization for ‘virtblk_dev_page.host_features’) Documentation/mic/mpssd/mpssd.c:151:3: error: initializer element is not constant
.seg_max = htole32(MIC_VRING_ENTRIES - 2),
^Documentation/mic/mpssd/mpssd.c:151:3: error: (near initialization for ‘virtblk_dev_page.blk_config.seg_max’) Documentation/mic/mpssd/mpssd.c:152:3: error: initializer element is not constant
.capacity = htole64(0),
^Documentation/mic/mpssd/mpssd.c:152:3: error: (near initialization for ‘virtblk_dev_page.blk_config.capacity’) make[3]: *** [Documentation/mic/mpssd/mpssd.o] Error 1
===
Reverting the commits fixes the issue, however, a more targeted approach could just skip make on Documentation/mic/mpssd/ until the endianness issue is resolved. Or something else...
Thanks,
Seth
[ adding John to CC ]
Quoted 68 lines · Seth Jennings
One or more in the following commit set in the 3.18 merge window is causing build failure on ppc64 with the default config:
7b345771ba921361b318e95bf21b257c65ac141cDocumentation: update include path for mpssd8c2b0dc83d9840da4d993a5dbb15c5974ad5a188Documentation: support glibc versions without htole macros6ab0e475f1f38b6be90aff4ef3ebf928c4a73dc8Documentation: fix misc. warningsadb19fb66eeebac07fe37d968725bb8906dadb8eDocumentation: add makefiles for more targetsBuilding v3.18-rc5 and I have glibc 2.17.
The problem is that for big endian, the htole* macros are not simple compile time transforms by the precompiler, but rather block code that must be executed in function context.
Build errors:
===
HOSTCCDocumentation/mic/mpssd/mpssd.o
In file included from/usr/include/bits/byteswap.h:34:0,from /usr/include/endian.h:60, from /usr/include/bits/waitstatus.h:64, from /usr/include/stdlib.h:42, from Documentation/mic/mpssd/mpssd.c:23: Documentation/mic/mpssd/mpssd.c:93:10: error: braced-group within expression allowed only inside a function .num = htole16(MIC_VRING_ENTRIES), ^ Documentation/mic/mpssd/mpssd.c:96:10: error: braced-group within expression allowed only inside a function .num = htole16(MIC_VRING_ENTRIES), ^ Documentation/mic/mpssd/mpssd.c:113:10: error: braced-group within expression allowed only inside a function .num = htole16(MIC_VRING_ENTRIES), ^ Documentation/mic/mpssd/mpssd.c:116:10: error: braced-group within expression allowed only inside a function .num = htole16(MIC_VRING_ENTRIES), ^
Documentation/mic/mpssd/mpssd.c:119:3: error: initializer element is not constant
.host_features = htole32(
^Documentation/mic/mpssd/mpssd.c:119:3: error: (near initialization for ‘virtnet_dev_page.host_features’) In file included from/usr/include/bits/byteswap.h:34:0,from /usr/include/endian.h:60, from /usr/include/bits/waitstatus.h:64, from /usr/include/stdlib.h:42, from Documentation/mic/mpssd/mpssd.c:23: Documentation/mic/mpssd/mpssd.c:146:10: error: braced-group within expression allowed only inside a function .num = htole16(MIC_VRING_ENTRIES), ^
Documentation/mic/mpssd/mpssd.c:149:3: error: initializer element is not constant
htole32(1<<VIRTIO_BLK_F_SEG_MAX),
^Documentation/mic/mpssd/mpssd.c:149:3: error: (near initialization for ‘virtblk_dev_page.host_features’)Documentation/mic/mpssd/mpssd.c:151:3: error: initializer element is not constant
.seg_max =htole32(MIC_VRING_ENTRIES - 2),
^Documentation/mic/mpssd/mpssd.c:151:3: error: (near initialization for ‘virtblk_dev_page.blk_config.seg_max’)Documentation/mic/mpssd/mpssd.c:152:3: error: initializer element is not constant
.capacity = htole64(0),
^Documentation/mic/mpssd/mpssd.c:152:3: error: (near initialization for ‘virtblk_dev_page.blk_config.capacity’) make[3]: *** [Documentation/mic/mpssd/mpssd.o] Error 1
===Reverting the commits fixes the issue, however, a more targeted approach could just skip make on
Documentation/mic/mpssd/until the endianness issue is resolved. Or something else...Thanks,
Seth
Signature · 2 lines
Jiri Kosina SUSE Labs
PATCH 1/1 Documentation/mic/mpssd: fix compile error on BE platforms
Change any use of htole* macro in static context to instead manually check for endianness and use __bswap_constant_* macro instead.
The current glibc definitions of the htole* macros don't allow their use in a static context; on big endian systems the build fails with:
HOSTCC Documentation/mic/mpssd/mpssd.o
In file included from /usr/include/bits/byteswap.h:34:0,
from /usr/include/endian.h:60,
from /usr/include/bits/waitstatus.h:64,
from /usr/include/stdlib.h:42,
from .../Documentation/mic/mpssd/mpssd.c:23:.../Documentation/mic/mpssd/mpssd.c:93:10:
error: braced-group within expression allowed only inside a function
.num = htole16(MIC_VRING_ENTRIES),
...and...
.../Documentation/mic/mpssd/mpssd.c:119:3:
error: initializer element is not constant
.host_features = htole32(
I also opened a glibc bug:
https://sourceware.org/bugzilla/show_bug.cgi?id=17679
But this patch is still needed to prevent build failures with the current glibc htole* macro definitions.
--- Documentation/mic/mpssd/mpssd.c | 54 ++++++++++++++++++++++++++++------------- 1 file changed, 37 insertions(+), 17 deletions(-)
Documentation/mic/mpssd/mpssd.c +37 -17
@@ -77,6 +77,33 @@ static struct mic_info mic_list;#define VIRTIO_NET_HDR_F_DATA_VALID 2/* Csum is valid */ #endif +#define VIRTNET_HOST_FEATURES \ + (1 << VIRTIO_NET_F_CSUM | \ + 1 << VIRTIO_NET_F_GSO | \ + 1 << VIRTIO_NET_F_GUEST_TSO4 | \ + 1 << VIRTIO_NET_F_GUEST_TSO6 | \ + 1 << VIRTIO_NET_F_GUEST_ECN | \ + 1 << VIRTIO_NET_F_GUEST_UFO) + +/* Currently, glibc htole* macros don't allow use in static context, +* so we check and use __bswap_constant_* instead +*/ +#if __BYTE_ORDER == __LITTLE_ENDIAN +# define MIC_VRING_ENTRIES_LE MIC_VRING_ENTRIES +# define VIRTNET_HOST_FEATURES_LE VIRTNET_HOST_FEATURES +# define VIRTBLK_HOST_FEATURES_LE (1<<VIRTIO_BLK_F_SEG_MAX) +# define VIRTBLK_CONFIG_SEG_MAX_LE (MIC_VRING_ENTRIES - 2) +#else /* __BIG_ENDIAN */ +# define MIC_VRING_ENTRIES_LE \ + __bswap_constant_16(MIC_VRING_ENTRIES) +# define VIRTNET_HOST_FEATURES_LE \ + __bswap_constant_32(VIRTNET_HOST_FEATURES) +# define VIRTBLK_HOST_FEATURES_LE \ + __bswap_constant_32(1<<VIRTIO_BLK_F_SEG_MAX) +# define VIRTBLK_CONFIG_SEG_MAX_LE \ + __bswap_constant_32(MIC_VRING_ENTRIES - 2) +#endif + static struct { struct mic_device_desc dd; struct mic_vqconfig vqconfig[2];
@@ -90,10 +117,10 @@ static struct { .config_len = sizeof(virtcons_dev_page.cons_config), }, .vqconfig[0] = { - .num = htole16(MIC_VRING_ENTRIES), + .num = MIC_VRING_ENTRIES_LE, }, .vqconfig[1] = { - .num = htole16(MIC_VRING_ENTRIES), + .num = MIC_VRING_ENTRIES_LE, }, };
@@ -110,21 +137,15 @@ static struct { .config_len = sizeof(virtnet_dev_page.net_config), }, .vqconfig[0] = { - .num = htole16(MIC_VRING_ENTRIES), + .num = MIC_VRING_ENTRIES_LE, }, .vqconfig[1] = { - .num = htole16(MIC_VRING_ENTRIES), + .num = MIC_VRING_ENTRIES_LE, },#if GSO_ENABLED - .host_features = htole32( -1 <<VIRTIO_NET_F_CSUM | -1 <<VIRTIO_NET_F_GSO | -1 <<VIRTIO_NET_F_GUEST_TSO4 | -1 <<VIRTIO_NET_F_GUEST_TSO6 | -1 <<VIRTIO_NET_F_GUEST_ECN | - 1 << VIRTIO_NET_F_GUEST_UFO), + .host_features = VIRTNET_HOST_FEATURES_LE, #else - .host_features = 0, + .host_features = 0,#endif };
@@ -143,13 +164,12 @@ static struct { .config_len = sizeof(virtblk_dev_page.blk_config), }, .vqconfig[0] = { - .num = htole16(MIC_VRING_ENTRIES), + .num = MIC_VRING_ENTRIES_LE, }, - .host_features = - htole32(1<<VIRTIO_BLK_F_SEG_MAX), + .host_features = VIRTBLK_HOST_FEATURES_LE, .blk_config = { - .seg_max = htole32(MIC_VRING_ENTRIES - 2), - .capacity = htole64(0), + .seg_max = VIRTBLK_CONFIG_SEG_MAX_LE, + .capacity = 0, } };
Signature · 1 line
1.8.3.1
Re: [PATCH] Documentation/mic/mpssd: fix compile error on BE platforms
Quoted 5 lines · Dan Streetman
Change any use of htole* macro in static context to instead manually check for endianness and use __bswap_constant_* macro instead.
The current glibc definitions of the htole* macros don't allow their use in a static context; on big endian systems the build fails with:
Instead of building Documentation/mic/mpssd for all architectures, I have submitted a patch which would build it only for x86_64 since that is the only place where it will ever be used. This would resolve this and future builds breaks and also make this patch unnecessary. The maintainers can decide to accept either or these patches, or even both.