[PATCH v4 00/15] mm: jit/text allocator

[Kent Overstreet]

On Thu, Apr 11, 2024 at 07:00:36PM +0300, Mike Rapoport wrote:
> From: "Mike Rapoport (IBM)" <rppt at kernel.org>
> 
> Hi,
> 
> Since v3 I looked into making execmem more of an utility toolbox, as we
> discussed at LPC with Mark Rutland, but it was getting more hairier than
> having a struct describing architecture constraints and a type identifying
> the consumer of execmem.
> 
> And I do think that having the description of architecture constraints for
> allocations of executable memory in a single place is better that having it
> spread all over the place.
> 
> The patches available via git:
> https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git/log/?h=execmem/v4
> 
> v4 changes:
> * rebase on v6.9-rc2
> * rename execmem_params to execmem_info and execmem_arch_params() to
>   execmem_arch_setup()
> * use single execmem_alloc() API instead of execmem_{text,data}_alloc() (Song)
> * avoid extra copy of execmem parameters (Rick)
> * run execmem_init() as core_initcall() except for the architectures that
>   may allocated text really early (currently only x86) (Will)
> * add acks for some of arm64 and riscv changes, thanks Will and Alexandre
> * new commits:
>   - drop call to kasan_alloc_module_shadow() on arm64 because it's not
>     needed anymore
>   - rename MODULE_START to MODULES_VADDR on MIPS
>   - use CONFIG_EXECMEM instead of CONFIG_MODULES on powerpc as per Christophe:
>     https://lore.kernel.org/all/79062fa3-3402-47b3-8920-9231ad05e964@csgroup.eu/
> 
> v3: https://lore.kernel.org/all/20230918072955.2507221-1-rppt@kernel.org
> * add type parameter to execmem allocation APIs
> * remove BPF dependency on modules
> 
> v2: https://lore.kernel.org/all/20230616085038.4121892-1-rppt@kernel.org
> * Separate "module" and "others" allocations with execmem_text_alloc()
> and jit_text_alloc()
> * Drop ROX entailment on x86
> * Add ack for nios2 changes, thanks Dinh Nguyen
> 
> v1: https://lore.kernel.org/all/20230601101257.530867-1-rppt@kernel.org
> 
> = Cover letter from v1 (sligtly updated) =
> 
> module_alloc() is used everywhere as a mean to allocate memory for code.
> 
> Beside being semantically wrong, this unnecessarily ties all subsystmes
> that need to allocate code, such as ftrace, kprobes and BPF to modules and
> puts the burden of code allocation to the modules code.
> 
> Several architectures override module_alloc() because of various
> constraints where the executable memory can be located and this causes
> additional obstacles for improvements of code allocation.
> 
> A centralized infrastructure for code allocation allows allocations of
> executable memory as ROX, and future optimizations such as caching large
> pages for better iTLB performance and providing sub-page allocations for
> users that only need small jit code snippets.
> 
> Rick Edgecombe proposed perm_alloc extension to vmalloc [1] and Song Liu
> proposed execmem_alloc [2], but both these approaches were targeting BPF
> allocations and lacked the ground work to abstract executable allocations
> and split them from the modules core.
> 
> Thomas Gleixner suggested to express module allocation restrictions and
> requirements as struct mod_alloc_type_params [3] that would define ranges,
> protections and other parameters for different types of allocations used by
> modules and following that suggestion Song separated allocations of
> different types in modules (commit ac3b43283923 ("module: replace
> module_layout with module_memory")) and posted "Type aware module
> allocator" set [4].
> 
> I liked the idea of parametrising code allocation requirements as a
> structure, but I believe the original proposal and Song's module allocator
> was too module centric, so I came up with these patches.
> 
> This set splits code allocation from modules by introducing execmem_alloc()
> and and execmem_free(), APIs, replaces call sites of module_alloc() and
> module_memfree() with the new APIs and implements core text and related
> allocations in a central place.
> 
> Instead of architecture specific overrides for module_alloc(), the
> architectures that require non-default behaviour for text allocation must
> fill execmem_info structure and implement execmem_arch_setup() that returns
> a pointer to that structure. If an architecture does not implement
> execmem_arch_setup(), the defaults compatible with the current
> modules::module_alloc() are used.
> 
> Since architectures define different restrictions on placement,
> permissions, alignment and other parameters for memory that can be used by
> different subsystems that allocate executable memory, execmem APIs
> take a type argument, that will be used to identify the calling subsystem
> and to allow architectures to define parameters for ranges suitable for that
> subsystem.
> 
> The new infrastructure allows decoupling of BPF, kprobes and ftrace from
> modules, and most importantly it paves the way for ROX allocations for
> executable memory.

It looks like you're just doing API cleanup first, then improving the
implementation later?

Patch set looks nice and clean; previous versions did seem to leak too
much arch/module details (or perhaps we were just bikeshedding too much
;) - but the API first approach is nice.

Looking forward to seeing this merged.