In 10.0 there is not similar assertion. From error log in this case we can see:

2017-05-31 10:46:51 140227134180096 [Note] InnoDB: Space id 4667992 trying to reserve 3 extents actually reserved 256  reserve 258 free 260 size 1641472 rounds 66 total_reserved 16896

2017-05-31 12:14:51 7f5cfdff6700 InnoDB: Loading buffer pool(s) from .//ib_buffer_pool

2017-05-31 12:16:44 7f5cfdff6700  InnoDB: Assertion failure in thread 140037375092480 in file buf0rea.cc line 577

Not sure but are we trying to load buffer pool pages out of the tablespace current reserved size ?
Notable is also the fact that on assertion there is again buffer overflow message and stack but no symbols.

Based on code review it is possible that return value is DB_PAGE_CORRUPTED i.e. page read is corrupted. In that case we should not crash here as tablespace is marked as corrupted and that error is handled elsewhere. To prevent unknown error code assertions with no info, added logging of incorrect error code.

https://github.com/MariaDB/server/commit/13e7f74097ded0ab014d8a694d0f674e2cbe5419

Please write a simpler patch. My suggestion would be to replace the occurrences of

default:

	ut_error;

with

default:

	ib_logf(IB_LOG_LEVEL_FATAL, "Error %u in linear readahead", err);

and revert any other changes. This would also greatly simplify the merge, because it would avoid refactoring a lot of code that calls ib_logf(), which has been replaced in 10.2.

commit 109b8582583296c4bb4ee41bc7649f6cd2254124
Author: Jan Lindström <jan.lindstrom@mariadb.com>
Date: Thu Aug 17 07:19:12 2017 +0300

MDEV-13432: Assertion failure in buf0rea.cc line 577

Page read could return DB_PAGE_CORRUPTED error that should
be reported and passed to upper layer. In case of unknown
error code we should print both number and string.