Re: [PATCH v5 1/2] rust: allocator: add KUnit tests for alignment guarantees

["Hui Zhu" <hui.zhu@linux.dev>]

2025年7月25日 18:02, "Danilo Krummrich" <dakr@kernel.org mailto:dakr@kernel.org?to=%22Danilo%20Krummrich%22%20%3Cdakr%40kernel.org%3E > 写到:


> 
> (Cc: Andrew)
> 
> On Fri Jul 25, 2025 at 11:50 AM CEST, Danilo Krummrich wrote:
> 
> > 
> > On Fri Jul 25, 2025 at 9:02 AM CEST, Hui Zhu wrote:
> > 
> > > 
> > > From: Hui Zhu <zhuhui@kylinos.cn>
> > > 
> > >  Add comprehensive tests to verify correct alignment handling in Rust
> > >  allocator wrappers. The tests validate:
> > > 
> > >  That kmalloc respects both standard (128-byte) and page-size
> > >  (8192-byte) alignments when allocating structs with explicit alignment
> > >  attributes.
> > > 
> > >  That vmalloc correctly handles standard alignments but intentionally
> > >  rejects allocations requiring alignments larger than its capabilities.
> > > 
> > >  That kvmalloc mirrors vmalloc's constraints, accepting standard
> > >  alignments but rejecting excessive alignment requirements.
> > > 
> > >  The test infrastructure uses specialized aligned structs (Blob and
> > >  LargeAlignBlob) and a test harness (TestAlign) to validate pointer
> > >  alignment through different allocation paths. This ensures our Rust
> > >  allocators correctly propagate kernel allocation constraints.
> > > 
> > >  Co-developed-by: Geliang Tang <geliang@kernel.org>
> > >  Signed-off-by: Geliang Tang <geliang@kernel.org>
> > >  Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
> > > 
> >  Thanks, this looks good. I think it would be good to rebase onto [1], since it
> >  will likely land in the same cycle. Additionally, two nits below.
> > 
> Please also Cc: Andrew for subsequent submissions, since this will, due to the
> interaction with [1] likely go through his tree.
> 
> > 
> > As a follow-up we could also test alignment in the context of
> >  Allocator::realloc(), i.e. when growing and shrinking buffers or requesting a
> >  different NUMA node.
> > 
> >  [1] https://lore.kernel.org/lkml/20250715135645.2230065-1-vitaly.wool@konsulko.se/
> > 
> > > 
> > > ---
> > >  rust/kernel/alloc/allocator.rs | 58 ++++++++++++++++++++++++++++++++++
> > >  1 file changed, 58 insertions(+)
> > > 
> > >  diff --git a/rust/kernel/alloc/allocator.rs b/rust/kernel/alloc/allocator.rs
> > >  index aa2dfa9dca4c..bcc916240f11 100644
> > >  --- a/rust/kernel/alloc/allocator.rs
> > >  +++ b/rust/kernel/alloc/allocator.rs
> > >  @@ -187,3 +187,61 @@ unsafe fn realloc(
> > >  unsafe { ReallocFunc::KVREALLOC.call(ptr, layout, old_layout, flags) }
> > >  }
> > >  }
> > >  +
> > >  +#[macros::kunit_tests(rust_allocator_kunit)]
> > >  +mod tests {
> > >  + use super::*;
> > >  + use core::mem::MaybeUninit;
> > >  + use kernel::prelude::*;
> > >  +
> > > 
> >  --8<--
> > 
> > > 
> > > + const TEST_SIZE: usize = 1024;
> > >  + const TEST_LARGE_ALIGN_SIZE: usize = kernel::page::PAGE_SIZE * 4;
> > >  +
> > >  + // These two structs are used to test allocating aligned memory.
> > >  + // they don't need to be accessed, so they're marked as dead_code.
> > >  + #[allow(dead_code)]
> > > 
> >  This should be #[expect(dead_code)].
> > 
> > > 
> > > + #[repr(align(128))]
> > >  + struct Blob([u8; TEST_SIZE]);
> > >  + #[allow(dead_code)]
> > >  + #[repr(align(8192))]
> > >  + struct LargeAlignBlob([u8; TEST_LARGE_ALIGN_SIZE]);
> > >  +
> > >  + struct TestAlign<T, A: Allocator>(Box<MaybeUninit<T>, A>);
> > >  + impl<T, A: Allocator> TestAlign<T, A> {
> > >  + fn new() -> Result<Self> {
> > >  + Ok(Self(Box::<_, A>::new_uninit(GFP_KERNEL)?))
> > >  + }
> > >  +
> > >  + fn alignment_valid(&self, align: usize) -> bool {
> > >  + assert!(align.is_power_of_two());
> > >  +
> > >  + let addr = self.0.as_ptr() as usize;
> > >  + if addr & (align - 1) != 0 {
> > >  + false
> > >  + } else {
> > >  + true
> > >  + }
> > > 
> >  This can just be
> > 
> >  addr & (align - 1) == 0
> > 
> >  instead of the conditional clause.
> > 
> > > 
> > > + }
> > >  + }
> > > 
> >  We could move all the above into test_alignment() given that it's likely only
> >  needed from there.
> > 
> > > 

Hi Danilo,

Thanks!

I sent v6 version that rebased on [1].

Best,
Hui

[1] https://lore.kernel.org/lkml/20250715135645.2230065-1-vitaly.wool@konsulko.se/

> > > +
> > >  + #[test]
> > >  + fn test_alignment() -> Result<()> {
> > >  + let ta = TestAlign::<Blob, Kmalloc>::new()?;
> > >  + assert!(ta.alignment_valid(128));
> > >  +
> > >  + let ta = TestAlign::<LargeAlignBlob, Kmalloc>::new()?;
> > >  + assert!(ta.alignment_valid(8192));
> > >  +
> > >  + let ta = TestAlign::<Blob, Vmalloc>::new()?;
> > >  + assert!(ta.alignment_valid(128));
> > >  +
> > >  + assert!(TestAlign::<LargeAlignBlob, Vmalloc>::new().is_err());
> > >  +
> > >  + let ta = TestAlign::<Blob, KVmalloc>::new()?;
> > >  + assert!(ta.alignment_valid(128));
> > >  +
> > >  + assert!(TestAlign::<LargeAlignBlob, KVmalloc>::new().is_err());
> > >  +
> > >  + Ok(())
> > >  + }
> > >  +}
> > >  -- 
> > >  2.43.0
> > >
> >
>