[PATCH v7] eal: fix rte_memcpy strict aliasing/alignment bugs

[Ananyev, Konstantin]

> Calls to rte_memcpy for 1 < n < 16 could result in unaligned
> loads/stores, which is undefined behaviour according to the C
> standard, and strict aliasing violations.
> 
> The code was changed to use a packed structure that allows aliasing
> (using the __may_alias__ attribute) to perform the load/store
> operations. This results in code that has the same performance as the
> original code and that is also C standards-compliant.
> 
> Fixes: d35cc1fe6a7a ("eal/x86: revert select optimized memcpy at run-time")
> Cc: Xiaoyun Li <xiaoyun.li at intel.com>
> Cc: stable at dpdk.org
> 
> Signed-off-by: Luc Pelletier <lucp.at.work at gmail.com>
> ---
> v7:
> * Fix coding style issue by adding new __rte_may_alias macro rather
> than directly use __attribute__
> 
> v6:
> * Refocus to fix strict aliasing problems discovered following
> discussions in this thread.
> * Modified the code to use __may_alias__ and packed structure. This fixes
> both the undefined behaviour of unaligned access (which is not the main
> concern), and also fixes the strict aliasing violations (which can cause
> major bugs, as demonstrated in a previous message in this thread).
> * Renamed new function from rte_mov15_or_less_unaligned to
> rte_mov15_or_less.
> * Modified code that copies <= 15 bytes to call rte_mov15_or_less.
> 
> v5:
> * Replaced assembly with pure C code that uses a packed structure to make
> unaligned loads conform to C standard.
> 
> v4:
> * Added volatile to asm statements, which is required under gcc.
> 
> v3:
> * Removed for loop and went back to using assembly.
> 
> v2:
> * Replaced assembly with a regular for loop that copies bytes one by
> one.
> 
> v1:
> * Fix undefined behaviour of unaligned stores/loads by using assembly
> to perform stores/loads.

LGTM, thanks for highlighting and fixing the problem.
Acked-by: Konstantin Ananyev <konstantin.ananyev at intel.com>
Tested-by: Konstantin Ananyev <konstantin.ananyev at intel.com>

As a side note, we probably need to check other similar places in DPDK code.