Commit 9f52ccf371 for openssl.org
commit 9f52ccf3713a137898313f427dc76c2e415d243a
Author: Samaresh Kumar Singh <ssam3003@gmail.com>
Date: Thu Mar 26 09:19:22 2026 -0500
chacha/asm: fix ELFv2 ABI violation in ChaCha20_ctr32_vsx_8x
The 8-block POWER10 ChaCha20 path uses vxxlor to spill VMX values into
VSR0-VSR26, which aliases FPR0-FPR26. FPR14-FPR31 are callee-saved per
the ELFv2 ABI, but the function was never saving or restoring them,
silently corrupting 11 FPRs (12 on big-endian) across any call with a
buffer larger than 255 bytes. VMX registers v20-v23, also
callee-saved, had the same problem.
Fix by increasing the frame size to accommodate save slots for
FPR14-FPR25 (and FPR26 on BE) and VMX v20-v23, and adding the
corresponding stfd/lfd and stvx/lvx pairs in the prologue and
epilogue. The VRSAVE save offset is updated to a fixed expression so
it stays clear of the new save area.
Fix for the bug #30584
Reviewed-by: Tomas Mraz <tomas@openssl.foundation>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
MergeDate: Sat Apr 11 20:06:02 2026
(Merged from https://github.com/openssl/openssl/pull/30587)
diff --git a/crypto/chacha/asm/chachap10-ppc.pl b/crypto/chacha/asm/chachap10-ppc.pl
index ef43a117ae..60992d6714 100755
--- a/crypto/chacha/asm/chachap10-ppc.pl
+++ b/crypto/chacha/asm/chachap10-ppc.pl
@@ -501,7 +501,7 @@ my ($xv8,$xv9,$xv10,$xv11,$xv12,$xv13,$xv14,$xv15,$xv16,$xv17) = map("v$_",(8..1
my ($xv18,$xv19,$xv20,$xv21) = map("v$_",(18..21));
my ($xv22,$xv23,$xv24,$xv25,$xv26) = map("v$_",(22..26));
-my $FRAME=$LOCALS+64+9*16; # 8*16 is for v24-v31 offload
+my $FRAME=$LOCALS+64+9*16+12*8+4*16; # 8*16 for v24-v31 offload, 12*8 for f14-f26, 4*16 for v20-v23
sub VSX_lane_ROUND_8x {
my ($a0,$b0,$c0,$d0,$a4,$b4,$c4,$d4)=@_;
@@ -665,7 +665,27 @@ $code.=<<___;
addi r11,r11,32
stvx v30,r10,$sp
stvx v31,r11,$sp
- stw r12,`$FRAME-4`($sp) # save vrsave
+ stfd f14,`$LOCALS+64+9*16+0*8`($sp) # save FPR14-FPR25 (callee-saved per ELFv2 ABI)
+ stfd f15,`$LOCALS+64+9*16+1*8`($sp)
+ stfd f16,`$LOCALS+64+9*16+2*8`($sp)
+ stfd f18,`$LOCALS+64+9*16+3*8`($sp)
+ stfd f19,`$LOCALS+64+9*16+4*8`($sp)
+ stfd f20,`$LOCALS+64+9*16+5*8`($sp)
+ stfd f21,`$LOCALS+64+9*16+6*8`($sp)
+ stfd f22,`$LOCALS+64+9*16+7*8`($sp)
+ stfd f23,`$LOCALS+64+9*16+8*8`($sp)
+ stfd f24,`$LOCALS+64+9*16+9*8`($sp)
+ stfd f25,`$LOCALS+64+9*16+10*8`($sp)
+ be?stfd f26,`$LOCALS+64+9*16+11*8`($sp) # BE only
+ li r10,`$LOCALS+64+9*16+12*8+15`
+ li r11,`$LOCALS+64+9*16+12*8+31`
+ stvx v20,r10,$sp # save VMX v20-v23 (callee-saved per ELFv2 ABI)
+ addi r10,r10,32
+ stvx v21,r11,$sp
+ addi r11,r11,32
+ stvx v22,r10,$sp
+ stvx v23,r11,$sp
+ stw r12,`$LOCALS+64+9*16-4`($sp) # save vrsave
li r12,-4096+63
$PUSH r0, `$FRAME+$LRSAVE`($sp)
mtspr 256,r12 # preserve 29 AltiVec registers
@@ -1159,7 +1179,27 @@ $code.=<<___;
bne Loop_outer_vsx_8x
Ldone_vsx_8x:
- lwz r12,`$FRAME-4`($sp) # pull vrsave
+ lwz r12,`$LOCALS+64+9*16-4`($sp) # pull vrsave
+ lfd f14,`$LOCALS+64+9*16+0*8`($sp) # restore FPR14-FPR25 (callee-saved per ELFv2 ABI)
+ lfd f15,`$LOCALS+64+9*16+1*8`($sp)
+ lfd f16,`$LOCALS+64+9*16+2*8`($sp)
+ lfd f18,`$LOCALS+64+9*16+3*8`($sp)
+ lfd f19,`$LOCALS+64+9*16+4*8`($sp)
+ lfd f20,`$LOCALS+64+9*16+5*8`($sp)
+ lfd f21,`$LOCALS+64+9*16+6*8`($sp)
+ lfd f22,`$LOCALS+64+9*16+7*8`($sp)
+ lfd f23,`$LOCALS+64+9*16+8*8`($sp)
+ lfd f24,`$LOCALS+64+9*16+9*8`($sp)
+ lfd f25,`$LOCALS+64+9*16+10*8`($sp)
+ be?lfd f26,`$LOCALS+64+9*16+11*8`($sp) # BE only
+ li r10,`$LOCALS+64+9*16+12*8+15`
+ li r11,`$LOCALS+64+9*16+12*8+31`
+ lvx v20,r10,$sp # restore VMX v20-v23 (callee-saved per ELFv2 ABI)
+ addi r10,r10,32
+ lvx v21,r11,$sp
+ addi r11,r11,32
+ lvx v22,r10,$sp
+ lvx v23,r11,$sp
li r10,`15+$LOCALS+64`
li r11,`31+$LOCALS+64`
$POP r0, `$FRAME+$LRSAVE`($sp)