UADK doesn't build under x86_64

[mikeBashStuff]

Sighting report

Recently, UADK was added to the pkgdep via 1360321. Since it was also pinned to install_all_dependencies() our imaging infrastructure picked it up. The build failed across all the distros with this (example taken from fedora38, though the same failure is seen under ubuntu, centos, rocky):

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: wd_alg.c: In function 'wd_check_ce_support':

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: wd_alg.c:103:58: error: 'HWCAP_CE_SM3' undeclared (first use in this function)

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38:   103 |         if (!strcmp("isa_ce_sm3", dev_name) && (hwcaps & HWCAP_CE_SM3))

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38:       |                                                          ^~~~~~~~~~~~

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: wd_alg.c:103:58: note: each undeclared identifier is reported only once for each function it appears in

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: wd_alg.c:106:58: error: 'HWCAP_CE_SM4' undeclared (first use in this function)

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38:   106 |         if (!strcmp("isa_ce_sm4", dev_name) && (hwcaps & HWCAP_CE_SM4))

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38:       |                                                          ^~~~~~~~~~~~

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: wd_alg.c: In function 'wd_check_sve_support':

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: wd_alg.c:119:22: error: 'HWCAP_SVE' undeclared (first use in this function)

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38:   119 |         if (hwcaps & HWCAP_SVE)

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38:       |                      ^~~~~~~~~

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: make[1]: *** [Makefile:1011: wd_alg.lo] Error 1

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00:     fedora38: make[1]: Leaving directory '/usr/local/src/uadk'

[2024-05-13T14:56:14.920Z] 2024-05-13T16:56:14+02:00: ==> fedora38: make: *** [Makefile:1105: install-recursive] Error 1

[2024-05-13T14:56:15.180Z] 2024-05-13T16:56:14+02:00: ==> fedora38: retry limit reached.

[2024-05-13T14:56:15.180Z] 2024-05-13T16:56:14+02:00: ==> fedora38: Provisioning step had errors: Running the cleanup provisioner, if present...

[2024-05-13T14:56:15.181Z] 2024-05-13T16:56:14+02:00: ==> fedora38: Deleting output directory...

After a closer inspection it seems that UADK is quite ARM-centric since the above macros the compiler is complaining about are declared only for ARM-based systems (see ./include/wd_alg.h):

#if defined(__arm__) || defined(__arm)
# define HWCAP                  AT_HWCAP
# define HWCAP_NEON             (1 << 12)

# define HWCAP_CE               AT_HWCAP2
# define HWCAP_CE_AES           (1 << 0)
# define HWCAP_CE_PMULL         (1 << 1)
# define HWCAP_CE_SHA1          (1 << 2)
# define HWCAP_CE_SHA256        (1 << 3)
#elif defined(__aarch64__)
# define HWCAP                  AT_HWCAP
# define HWCAP_NEON             (1 << 1)

# define HWCAP_CE               HWCAP
# define HWCAP_CE_AES           (1 << 3)
# define HWCAP_CE_PMULL         (1 << 4)
# define HWCAP_CE_SHA1          (1 << 5)
# define HWCAP_CE_SHA256        (1 << 6)
# define HWCAP_CPUID            (1 << 11)
# define HWCAP_SHA3             (1 << 17)
# define HWCAP_CE_SM3           (1 << 18)
# define HWCAP_CE_SM4           (1 << 19)
# define HWCAP_CE_SHA512        (1 << 21)
# define HWCAP_SVE              (1 << 22)
/* AT_HWCAP2 */
# define HWCAP2                 26
# define HWCAP2_SVE2            (1 << 1)
# define HWCAP2_RNG             (1 << 16)
#endif

include/drv/arm_arch_ce.h also seems to look only for ARM.

Considering all that, UADK should be, at minimum, removed from install_all_dependencies() to make sure the --all does not suddenly break.

Also, just to note, under FreeBSD the build fails already at the autoconf stage:

[2024-05-13T15:36:47.134Z] 2024-05-13T17:36:47+02:00: ==> freebsd14: configure.ac:30: error: possibly undefined macro: AC_DEFINE

[2024-05-13T15:36:47.134Z] 2024-05-13T17:36:47+02:00: ==> freebsd14:       If this token and others are legitimate, please use m4_pattern_allow.

[2024-05-13T15:36:47.134Z] 2024-05-13T17:36:47+02:00: ==> freebsd14:       See the Autoconf documentation.

[2024-05-13T15:36:47.135Z] 2024-05-13T17:36:47+02:00: ==> freebsd14: autoreconf2.72: error: /usr/local/bin/autoconf2.72 failed with exit status: 1

[mikeBashStuff]

https://review.spdk.io/gerrit/c/spdk/spdk/+/23089. The alternative is to drive the $INSTALL_UADK setup with extra arch checks but this seems to be the simplest of solutions.

[mikeBashStuff]

Patch got merged, closing.