diff options
| author | Samuel Thibault <samuel.thibault@ens-lyon.org> | 2025-04-19 20:35:01 +0200 |
|---|---|---|
| committer | Samuel Thibault <samuel.thibault@ens-lyon.org> | 2025-04-19 20:35:01 +0200 |
| commit | 2e03784967a89b36942e498820db7b3639d74548 (patch) | |
| tree | a778719d71beaf39c0c5b4817f885a9f66e5f02c | |
| parent | 883420f4201f50dc91b3d2fefe5cc2837565dc7b (diff) | |
fpu: Fix saving/restoring SSE MXCSR state
| -rw-r--r-- | i386/i386/fpu.c | 6 | ||||
| -rw-r--r-- | tests/test-thread-state-fp.c | 13 |
2 files changed, 17 insertions, 2 deletions
diff --git a/i386/i386/fpu.c b/i386/i386/fpu.c index 967ba91d..93762086 100644 --- a/i386/i386/fpu.c +++ b/i386/i386/fpu.c @@ -491,6 +491,9 @@ ASSERT_IPL(SPL0); ifps->xfp_save_state.fp_opcode = user_fp_state->fp_opcode; ifps->xfp_save_state.fp_dp = user_fp_state->fp_dp; ifps->xfp_save_state.fp_ds = user_fp_state->fp_ds; + ifps->xfp_save_state.fp_dp3 = user_fp_state->fp_dp3; + ifps->xfp_save_state.fp_mxcsr = user_fp_state->fp_mxcsr & mxcsr_feature_mask; + ifps->xfp_save_state.fp_mxcsr_mask = user_fp_state->fp_mxcsr_mask & mxcsr_feature_mask;; for (i=0; i<8; i++) memcpy(&ifps->xfp_save_state.fp_reg_word[i], &user_fp_state->fp_reg_word[i], sizeof(user_fp_state->fp_reg_word[i])); for (i=0; i<16; i++) @@ -619,6 +622,9 @@ ASSERT_IPL(SPL0); user_fp_state->fp_opcode = ifps->xfp_save_state.fp_opcode; user_fp_state->fp_dp = ifps->xfp_save_state.fp_dp; user_fp_state->fp_ds = ifps->xfp_save_state.fp_ds; + user_fp_state->fp_dp3 = ifps->xfp_save_state.fp_dp3; + user_fp_state->fp_mxcsr = ifps->xfp_save_state.fp_mxcsr; + user_fp_state->fp_mxcsr_mask = ifps->xfp_save_state.fp_mxcsr_mask; for (i=0; i<8; i++) memcpy(&user_fp_state->fp_reg_word[i], &ifps->xfp_save_state.fp_reg_word[i], sizeof(user_fp_state->fp_reg_word[i])); for (i=0; i<16; i++) diff --git a/tests/test-thread-state-fp.c b/tests/test-thread-state-fp.c index 2aab8ca9..1a7cbdaa 100644 --- a/tests/test-thread-state-fp.c +++ b/tests/test-thread-state-fp.c @@ -30,9 +30,9 @@ static void printx(struct i386_xfloat_state *state, int size) printf("xfloat init %d fp %d exc %d\n", state->initialized, state->fpkind, state->exc_status); struct i386_xfp_save *xfp = (struct i386_xfp_save *) &state->hw_state[0]; - printf("xfp %x %x %x %x %x %x %x %x\n", + printf("xfp %x %x %x %x %x %x %x %x %x\n", xfp->fp_control, xfp->fp_status, xfp->fp_tag, xfp->fp_eip, - xfp->fp_cs, xfp->fp_opcode, xfp->fp_dp, xfp->fp_ds); + xfp->fp_cs, xfp->fp_opcode, xfp->fp_dp, xfp->fp_ds, xfp->fp_mxcsr); for (int i=0; i<8; i++) { printf("fp%d", i); @@ -185,8 +185,10 @@ static void thread_xfp_getset(void *arg) for (int j=0; j<16; j++) ASSERT(xfp->fp_xreg_word[3][j] == 0x33, "register xmm3 wasn't correctly retrieved from the getset thread"); + ASSERT(xfp->fp_mxcsr == 0x1f80, "mxcsr wasn't correctly retrieved from the getset thread"); memset(xfp->fp_xreg_word[7], 0x77, 16); + xfp->fp_mxcsr = 0x1fa0; err = thread_set_state(th, i386_XFLOAT_STATE, (thread_state_t) state, state_count); @@ -208,6 +210,10 @@ static void test_xfp_state_getset() memset(buf3, 0x33, 16); asm volatile ("movups (%0),%%xmm3" :: "r" (buf3) :); + /* And to MXCSR */ + unsigned int mxcsr = 0x1f80; + asm volatile ("ldmxcsr %0" :: "m" (mxcsr)); + /* then switch to the get/set test thread, and wait to be resumed */ test_thread_start(mach_task_self(), thread_xfp_getset, &th); err = thread_suspend(th); @@ -225,6 +231,9 @@ static void test_xfp_state_getset() for (int j=0; j<16; j++) ASSERT(buf7[j] == 0x77, "register xmm7 wasn't correctly set by the getset thread"); + + asm volatile ("stmxcsr %0" :: "m" (mxcsr)); + ASSERT(mxcsr == 0x1fa0, "mxcsr wasn't updated"); } #endif |