Source: mariadb-10.0 Version: 10.0.27-1 Severity: important Tags: patch Hi,
Currently mariadb-10.0 FTBFS on mips64el due to various testsuite failures. The two attached patches should resolve this. I am also looking at the failures on mips and mipsel, but haven't finished fixing those yet. mips-errno-enotempty.patch On mips* architectures, ENOTEMPTY == 93 which wasn't handled by two test cases. mips64-taocrypt-integer.patch The previous patch to fix mips64el made the package build, but was unfortunately completely wrong. I've replaced the code with a generic implementation using GCC's __int128. It could probably be generalized to other arches, but I didn't want to break anything. Thanks, James
Description: Permit 93 as a valid value of the ENOTEMPTY error in the testsuite On mips, ENOTEMPTY == 93 so permit that value in two test cases by replacing it with 39. Author: James Cowgill <[email protected]> --- This patch header follows DEP-3: http://dep.debian.net/deps/dep3/ --- a/mysql-test/extra/binlog_tests/database.test +++ b/mysql-test/extra/binlog_tests/database.test @@ -52,7 +52,7 @@ eval SELECT 'hello' INTO OUTFILE 'fake_f # Use '/' instead of '\' in the error message. On windows platform, dir is # formed with '\'. ---replace_regex /\\testing_1\\*/\/testing_1\// /66/39/ /17/39/ /File exists/Directory not empty/ +--replace_regex /\\testing_1\\*/\/testing_1\// /66/39/ /93/39/ /17/39/ /File exists/Directory not empty/ --error 1010 DROP DATABASE testing_1; let $wait_binlog_event= DROP TABLE IF EXIST; --- a/mysql-test/suite/rpl/t/rpl_drop_db.test +++ b/mysql-test/suite/rpl/t/rpl_drop_db.test @@ -13,7 +13,7 @@ insert into mysqltest1.t1 values (1); select * from mysqltest1.t1 into outfile 'mysqltest1/f1.txt'; create table mysqltest1.t2 (n int); create table mysqltest1.t3 (n int); ---replace_result \\ / 66 39 17 39 "File exists" "Directory not empty" +--replace_result \\ / 66 39 93 39 17 39 "File exists" "Directory not empty" --error 1010 drop database mysqltest1; use mysqltest1; @@ -30,7 +30,7 @@ while ($1) } --enable_query_log ---replace_result \\ / 66 39 17 39 "File exists" "Directory not empty" +--replace_result \\ / 66 39 93 39 17 39 "File exists" "Directory not empty" --error 1010 drop database mysqltest1; use mysqltest1;
Description: Correctly fix mips64 multiplication in taocrypt In Debian bugs #719196 and #798126, a patch was submitted to fix a FTBFS on mips64el of mysql-5.5 and mysql-5.6 respectively. This patch was accepted upstream and made its way into mariadb (MDEV-6528). Unfortunately the patch was incorrect and causes segfaults when executing the SSL parts of the testsuite. . The previous patch replaced the "h" asm constraint with the "d" constraint since the "h" is not supported in newer GCC versions. The "d" constraint was wrong as it tells GCC that the result can be found in _any_ general purpose register that it chooses. This resulted in GCC reading garbage into the high result bits. . This new patch replaces the assembly code with a generic 128-bit multiplication which works correctly on mips64. Author: James Cowgill <[email protected]> --- This patch header follows DEP-3: http://dep.debian.net/deps/dep3/ --- a/extra/yassl/taocrypt/src/integer.cpp +++ b/extra/yassl/taocrypt/src/integer.cpp @@ -193,8 +193,9 @@ DWord() {} "a" (a), "rm" (b) : "cc"); #elif defined(__mips64) - __asm__("dmultu %2,%3" : "=d" (r.halfs_.high), "=l" (r.halfs_.low) - : "r" (a), "r" (b)); + unsigned __int128 t = (unsigned __int128) a * b; + r.halfs_.high = t >> 64; + r.halfs_.low = (word) t; #elif defined(_M_IX86) // for testing
signature.asc
Description: OpenPGP digital signature
