[PATCH v10] elf: Support THP segment load with madvise enabled THP

[H.J. Lu]

On Thu, May 28, 2026 at 1:00 AM Adhemerval Zanella Netto
<adhemerval.zanella@linaro.org> wrote:
>
>
>
> On 27/05/26 11:21, Andreas Schwab wrote:
> > On Mai 26 2026, Adhemerval Zanella Netto wrote:
> >
> >> It seems RISCV default linker scripts sets p_vaddr to 0, different than
> >> x86_64/aarch64, and /proc/sys/vm/mmap_min_addr is a non-zero value:
> >
> > The difference is that x86-64 and aarch64 put executables at 0x400000
> > by default, whereas riscv64 uses 0x100000.  You get the same effect on
> > x86-64 and aarch64 if you increase the page size beyond 0x400000.

I tried

THP-PAGE-SIZE = 0x800000

on x86-64.  It works:

[hjl@gnu-tgl-3 build-x86_64-linux]$ readelf -lW elf/tst-thp-1-static

Elf file type is DYN (Position-Independent Executable file)
Entry point 0x802070
There are 12 program headers, starting at offset 64

Program Headers:
  Type           Offset   VirtAddr           PhysAddr
FileSiz  MemSiz   Flg Align
  LOAD           0x000000 0x0000000000000000 0x0000000000000000
0x030d38 0x030d38 R   0x800000
  LOAD           0x800000 0x0000000000800000 0x0000000000800000
0x20b4fb5 0x20b4fb5 R E 0x800000
  LOAD           0x2ffb2c0 0x00000000037fb2c0 0x00000000037fb2c0
0x006778 0x00c9a8 RW  0x800000
  DYNAMIC        0x2fffcc8 0x00000000037ffcc8 0x00000000037ffcc8
0x000200 0x000200 RW  0x8
  NOTE           0x0002e0 0x00000000000002e0 0x00000000000002e0
0x000024 0x000024 R   0x4
  NOTE           0x030cc8 0x0000000000030cc8 0x0000000000030cc8
0x000050 0x000050 R   0x8
  NOTE           0x030d18 0x0000000000030d18 0x0000000000030d18
0x000020 0x000020 R   0x4
  TLS            0x2ffb2c0 0x00000000037fb2c0 0x00000000037fb2c0
0x000038 0x000050 R   0x8
  GNU_PROPERTY   0x030cc8 0x0000000000030cc8 0x0000000000030cc8
0x000050 0x000050 R   0x8
  GNU_EH_FRAME   0x01ee44 0x000000000001ee44 0x000000000001ee44
0x002b24 0x002b24 R   0x4
  GNU_STACK      0x000000 0x0000000000000000 0x0000000000000000
0x000000 0x000000 RW  0x10
  GNU_RELRO      0x2ffb2c0 0x00000000037fb2c0 0x00000000037fb2c0
0x004d40 0x004d40 R   0x1

 Section to Segment mapping:
  Segment Sections...
   00     .note.gnu.build-id .gnu.hash .dynsym .dynstr .rela.dyn
.rela.plt .relr.dyn .rodata rodata.cst32 .stapsdt.base .eh_frame_hdr
.eh_frame .gcc_except_table .note.gnu.property .note.ABI-tag
   01     .init .plt .plt.got .plt.sec .text .fini
   02     .tdata .init_array .fini_array .data.rel.ro .dynamic .got
.got.plt .data .bss
   03     .dynamic
   04     .note.gnu.build-id
   05     .note.gnu.property
   06     .note.ABI-tag
   07     .tdata .tbss
   08     .note.gnu.property
   09     .eh_frame_hdr
   10
   11     .tdata .init_array .fini_array .data.rel.ro .dynamic .got
[hjl@gnu-tgl-3 build-x86_64-linux]$

> >
>
> So should we use  -Wl,-Ttext-segment=0x200000 or --image-base on riscv or
> define a different large page size on this patch?

Since riscv also uses 0x200000 for THP page size, we can't use a lower value.
Does it mean that riscv can't have p_align >= 0x200000?  It sounds a
riscv linker or kernel bug.  I opened:

https://sourceware.org/bugzilla/show_bug.cgi?id=34184

--
H.J.