[PATCH 0/3] x86: "brk" allocator

Jan Beulich posted 3 patches 5 days, 13 hours ago
Failed in applying to current master (apply log)
[PATCH 0/3] x86: "brk" allocator
Posted by Jan Beulich 5 days, 13 hours ago
Derive for our purposes something similar to what Linux has been having for
a long while.

1: x86: introduce "brk" allocator
2: x86/EFI: replace ebmalloc()
3: xhci-dbc: use brk_alloc()

To reduce padding holes, .bss.page_aligned and per-CPU data may want moving
immediately ahead of __brk_start[]. Albeit then the tail of per-CPU data will
all be padding space; sadly the TSS wants/needs page-aligning for XPTI
purposes.

Jan
Re: [PATCH 0/3] x86: "brk" allocator
Posted by Andrew Cooper 5 days, 9 hours ago
On 13/11/2025 11:06 am, Jan Beulich wrote:
> Derive for our purposes something similar to what Linux has been having for
> a long while.
>
> 1: x86: introduce "brk" allocator
> 2: x86/EFI: replace ebmalloc()
> 3: xhci-dbc: use brk_alloc()
>
> To reduce padding holes, .bss.page_aligned and per-CPU data may want moving
> immediately ahead of __brk_start[]. Albeit then the tail of per-CPU data will
> all be padding space; sadly the TSS wants/needs page-aligning for XPTI
> purposes.

I've not done this yet, but I was intending to make the TSSes be
dynamically allocated, so they (along with the IDTs) can be omitted
entirely under FRED.

The VT-x tr limit bug/misfeature complicates this somewhat, hence why
I've left this out of the FRED basic support, but it can be addressed by
pointing HOST_TR_BASE at a singleton TSS.

~Andrew