fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix for 1st-gen Ryzen crashes

Browse source
This commit is contained in:
SChernykh 2019-12-27 12:40:38 +02:00
parent 99826a6b51
commit 3a2941b719
10 changed files with 102 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

1
src/crypto/randomx/asm/program_epilogue_store.inc
View file

@ -1,4 +1,5 @@
;# save VM register values
add rsp, 24
pop rcx
mov qword ptr [rcx+0], r8
mov qword ptr [rcx+8], r9

4
src/crypto/randomx/asm/program_loop_load.inc
View file

@ -1,5 +1,5 @@
lea rcx, [rsi+rax]
push rcx
mov [rsp+8], rcx
xor r8, qword ptr [rcx+0]
xor r9, qword ptr [rcx+8]
xor r10, qword ptr [rcx+16]
@ -9,7 +9,7 @@
xor r14, qword ptr [rcx+48]
xor r15, qword ptr [rcx+56]
lea rcx, [rsi+rdx]
push rcx
mov [rsp+16], rcx
cvtdq2pd xmm0, qword ptr [rcx+0]
cvtdq2pd xmm1, qword ptr [rcx+8]
cvtdq2pd xmm2, qword ptr [rcx+16]

4
src/crypto/randomx/asm/program_loop_store.inc
View file

@ -1,4 +1,4 @@
pop rcx
mov rcx, [rsp+16]
mov qword ptr [rcx+0], r8
mov qword ptr [rcx+8], r9
mov qword ptr [rcx+16], r10
@ -7,7 +7,7 @@
mov qword ptr [rcx+40], r13
mov qword ptr [rcx+48], r14
mov qword ptr [rcx+56], r15
pop rcx
mov rcx, [rsp+8]
xorpd xmm0, xmm4
xorpd xmm1, xmm5
xorpd xmm2, xmm6

9
src/crypto/randomx/jit_compiler_x86.cpp
View file

@ -36,6 +36,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "crypto/randomx/program.hpp"
#include "crypto/randomx/reciprocal.h"
#include "crypto/randomx/virtual_memory.hpp"
#include "crypto/rx/Rx.h"
#ifdef _MSC_VER
# include <intrin.h>
@ -168,8 +169,8 @@ namespace randomx {
static const uint8_t REX_MAXPD[] = { 0x66, 0x41, 0x0f, 0x5f };
static const uint8_t REX_DIVPD[] = { 0x66, 0x41, 0x0f, 0x5e };
static const uint8_t SQRTPD[] = { 0x66, 0x0f, 0x51 };
static const uint8_t AND_OR_MOV_LDMXCSR[] = { 0x25, 0x00, 0x60, 0x00, 0x00, 0x0D, 0xC0, 0x9F, 0x00, 0x00, 0x89, 0x44, 0x24, 0xFC, 0x0F, 0xAE, 0x54, 0x24, 0xFC };
static const uint8_t AND_OR_MOV_LDMXCSR_RYZEN[] = { 0x25, 0x00, 0x60, 0x00, 0x00, 0x0D, 0xC0, 0x9F, 0x00, 0x00, 0x3B, 0x44, 0x24, 0xFC, 0x74, 0x09, 0x89, 0x44, 0x24, 0xFC, 0x0F, 0xAE, 0x54, 0x24, 0xFC };
static const uint8_t AND_OR_MOV_LDMXCSR[] = { 0x25, 0x00, 0x60, 0x00, 0x00, 0x0D, 0xC0, 0x9F, 0x00, 0x00, 0x89, 0x04, 0x24, 0x0F, 0xAE, 0x14, 0x24 };
static const uint8_t AND_OR_MOV_LDMXCSR_RYZEN[] = { 0x25, 0x00, 0x60, 0x00, 0x00, 0x0D, 0xC0, 0x9F, 0x00, 0x00, 0x3B, 0x04, 0x24, 0x74, 0x07, 0x89, 0x04, 0x24, 0x0F, 0xAE, 0x14, 0x24 };
static const uint8_t ROL_RAX[] = { 0x48, 0xc1, 0xc0 };
static const uint8_t XOR_ECX_ECX[] = { 0x33, 0xC9 };
static const uint8_t REX_CMP_R32I[] = { 0x41, 0x81 };
@ -380,10 +381,12 @@ namespace randomx {
*(uint32_t*)(code + codePos + 10) = RandomX_CurrentConfig.ScratchpadL3Mask64_Calculated;
*(uint32_t*)(code + codePos + 20) = RandomX_CurrentConfig.ScratchpadL3Mask64_Calculated;
if (hasAVX) {
uint32_t* p = (uint32_t*)(code + codePos + 29);
uint32_t* p = (uint32_t*)(code + codePos + 32);
*p = (*p & 0xFF000000U) | 0x0077F8C5U;
}
xmrig::Rx::setMainLoopBounds(code + prologueSize, code + epilogueOffset);
codePos = prologueSize;
memcpy(code + codePos - 48, &pcfg.eMask, sizeof(pcfg.eMask));
memcpy(code + codePos, codeLoopLoad, loopLoadSize);

3
src/crypto/randomx/jit_compiler_x86_static.S
View file

@ -93,7 +93,8 @@ DECL(randomx_program_prologue_first_load):
and eax, RANDOMX_SCRATCHPAD_MASK
ror rdx, 32
and edx, RANDOMX_SCRATCHPAD_MASK
stmxcsr dword ptr [rsp-20]
sub rsp, 24
stmxcsr dword ptr [rsp]
nop
nop
nop

3
src/crypto/randomx/jit_compiler_x86_static.asm
View file

@ -81,7 +81,8 @@ randomx_program_prologue_first_load PROC
and eax, RANDOMX_SCRATCHPAD_MASK
ror rdx, 32
and edx, RANDOMX_SCRATCHPAD_MASK
stmxcsr dword ptr [rsp-20]
sub rsp, 24
stmxcsr dword ptr [rsp]
nop
nop
nop