• Improved size checking

Changed to reflect the correct behavior of OF_getencprop().
A patch to fix ofw_fdt_getprop() was submitted as D15680.

• Removed extra printf param

• Improved linux,initrd-{start,end} length checking

In D15667#331145, @jhibbits wrote:

In D15667#331142, @lffpires_ruabrasil.org wrote:

After further investigation, linux,initrd-start/end are always 32-bit values

That's odd, because they're 64-bit values in the device tree:

Node 0xe38: chosen

linux,initrd-end:
  00 00 00 00 00 00 00 00
linux,initrd-start:
  00 00 00 00 00 00 00 00

Rework after comments

Removed dummy comment

After further investigation, linux,initrd-start/end are always 32-bit values

• Added PT_GETVSRREGS/PT_SETVSRREGS to ptrace man pages

Fixing diff in revision

• memcpy optimization for powerpc64
• moved vsx detection to C code (#3)
• added license
• set bcopy_has_vsx to 0 during vsx test and use elf_aux_info()

2 follow up revisions created:
D15340 (Removed -fstack-protector-all restriction from libc)
D15341 (Updated rtld to use libc_nossp_pic.a)

Made changes according to review comments.

• Separated libc restriction- and rtld-related changes
• Removed unnecessary MK_TOOLCHAIN check

In D15208#320453, @kib wrote:

I think this is more involved than just avoiding memcpy(). If you look at the libexec/rtld-elf/Makefile near the start, you would see MK_SSP=no. In other words, we know that rtld must be compiled without stack protector, but you effectively point out that libc_pic.a is compiled with it, and this causes more troubles.

I suggest to consider changing the build for libc_pic.a to also avoid -fstack-protector instead of fixing the single place.

The fix for the stack protection issue was merged to main (D15173), so now this diff is ready to be considered for merging as well.

@kib, I added them just because this showed up while working on an optimization for powerpc64, and at the time I didn't know who I should add as reviewer. Please feel free to commit this.

• addressed review comments

Use elf_aux_info() for VSX check and set bcopy_has_vsx during the check to avoid infinite loop
For this to work, I had to make changes to the stack guard setup code - I'll post those in a separate review.

Moved VSX detection to C code - now it should be simpler to migrate to ifunc's in the future.
Changed to /* */ - style comments.
Removed unnecessary references to %cr0.