On Tue, Jun 18, 2024 at 2:58 AM Sami Tolvanen [off-list ref] wrote:

Hi folks,

This series implements CONFIG_MODVERSIONS for Rust, an important
feature for distributions like Android that want to ship Rust
kernel modules, and depend on modversions to help ensure module ABI
compatibility.

There have been earlier proposals [1][2] that would allow Rust
modules to coexist with modversions, but none that actually implement
symbol versioning. Unlike C, Rust source code doesn't have sufficient
information about the final ABI, as the compiler has considerable
freedom in adjusting structure layout for improved performance [3],
for example, which makes using a source code parser like genksyms
a non-starter. Based on Matt's suggestion and previous feedback
from maintainers, this series uses DWARF debugging information for
computing versions. DWARF is an established and relatively stable
format, which includes all the necessary ABI details, and adding a
CONFIG_DEBUG_INFO dependency for Rust symbol versioning seems like a
reasonable trade-off.

The first 12 patches of this series add a small tool for computing
symbol versions from DWARF, called gendwarfksyms. When passed a list
of exported symbols, the tool generates an expanded type string
for each symbol, and computes symbol CRCs similarly to genksyms.
gendwarfksyms is written in C and uses libdw to process DWARF, mainly
because of the existing support for C host tools that use elfutils
(e.g., objtool).

Another compatibility issue is fitting the extremely long mangled
Rust symbol names into struct modversion_info, which can only hold
55-character names (on 64-bit systems). Previous proposals suggested
changing the structure to support longer names, but the conclusion was
that we cannot break userspace tools that parse the version table.

The next two patches of the series implement support for hashed
symbol names in struct modversion_info, where names longer than 55
characters are hashed, and the hash is stored in the name field. To
avoid breaking userspace tools, the binary hash is prefixed with a
null-terminated string containing the name of the hash function. While
userspace tools can later be updated to potentially produce more
useful information about the long symbols, this allows them to
continue working in the meantime.

The final patch allows CONFIG_MODVERSIONS to be selected with Rust,
provided that debugging information is also available.

I am surprised at someone who attempts to add another variant of genksyms.


I am also surprised at the tool being added under the tools/ directory.

At least, we can avoid the latter misfortune, though.


A patch attached (on top of your patch set).

Such a tool can be compiled in scripts/, or even better
in rust/Makefile if it is only used in rust/Makefile.





--
Best Regards
Masahiro Yamada