diff options
Diffstat (limited to 'sysdeps/loongarch/dl-tlsdesc.S')
| -rw-r--r-- | sysdeps/loongarch/dl-tlsdesc.S | 398 |
1 files changed, 34 insertions, 364 deletions
diff --git a/sysdeps/loongarch/dl-tlsdesc.S b/sysdeps/loongarch/dl-tlsdesc.S index a6627cc754..be76c07cce 100644 --- a/sysdeps/loongarch/dl-tlsdesc.S +++ b/sysdeps/loongarch/dl-tlsdesc.S @@ -59,376 +59,46 @@ _dl_tlsdesc_undefweak: cfi_endproc .size _dl_tlsdesc_undefweak, .-_dl_tlsdesc_undefweak - #ifdef SHARED -#define FRAME_SIZE (-((-14 * SZREG) & ALMASK)) -#define FRAME_SIZE_LSX (-((-32 * SZVREG) & ALMASK)) -#define FRAME_SIZE_LASX (-((-32 * SZXREG) & ALMASK)) -#define FRAME_SIZE_FLOAT (-((-24 * SZFREG) & ALMASK)) - - /* Handler for dynamic TLS symbols. - Prototype: - _dl_tlsdesc_dynamic (tlsdesc *) ; - - The second word of the descriptor points to a - tlsdesc_dynamic_arg structure. - - Returns the offset between the thread pointer and the - object referenced by the argument. - - ptrdiff_t - _dl_tlsdesc_dynamic (struct tlsdesc *tdp) - { - struct tlsdesc_dynamic_arg *td = tdp->arg; - dtv_t *dtv = *(dtv_t **)((char *)__thread_pointer - SIZE_OF_TCB); - if (__glibc_likely (td->gen_count <= dtv[0].counter - && (dtv[td->tlsinfo.ti_module].pointer.val - != TLS_DTV_UNALLOCATED), - 1)) - return dtv[td->tlsinfo.ti_module].pointer.val - + td->tlsinfo.ti_offset - - __thread_pointer; - - return ___tls_get_addr (&td->tlsinfo) - __thread_pointer; - } */ - .hidden _dl_tlsdesc_dynamic - .global _dl_tlsdesc_dynamic - .type _dl_tlsdesc_dynamic,%function - cfi_startproc - .align 2 -_dl_tlsdesc_dynamic: - /* Save just enough registers to support fast path, if we fall - into slow path we will save additional registers. */ - ADDI sp, sp, -32 - REG_S t0, sp, 0 - REG_S t1, sp, 8 - REG_S t2, sp, 16 - -/* Runtime Storage Layout of Thread-Local Storage - TP point to the start of TLS block. - - dtv -Low address TCB ----------------> dtv0(counter) - TP --> static_block0 <----- dtv1 - static_block1 <----- dtv2 - static_block2 <----- dtv3 - dynamic_block0 <----- dtv4 -Hign address dynamic_block1 <----- dtv5 */ - - REG_L t0, tp, -SIZE_OF_TCB /* t0 = dtv */ - REG_L a0, a0, TLSDESC_ARG /* a0(td) = tdp->arg */ - REG_L t1, a0, TLSDESC_GEN_COUNT /* t1 = td->gen_count */ - REG_L t2, t0, DTV_COUNTER /* t2 = dtv[0].counter */ - /* If dtv[0].counter < td->gen_count, goto slow path. */ - bltu t2, t1, .Lslow - - REG_L t1, a0, TLSDESC_MODID /* t1 = td->tlsinfo.ti_module */ - /* t1 = t1 * sizeof(dtv_t) = t1 * (2 * sizeof(void*)) */ - slli.d t1, t1, 4 - add.d t1, t1, t0 /* t1 = dtv[td->tlsinfo.ti_module] */ - REG_L t1, t1, 0 /* t1 = dtv[td->tlsinfo.ti_module].pointer.val */ - li.d t2, TLS_DTV_UNALLOCATED - /* If dtv[td->tlsinfo.ti_module].pointer.val is TLS_DTV_UNALLOCATED, - goto slow path. */ - beq t1, t2, .Lslow - - REG_L t2, a0, TLSDESC_MODOFF /* t2 = td->tlsinfo.ti_offset */ - /* dtv[td->tlsinfo.ti_module].pointer.val + td->tlsinfo.ti_offset */ - add.d a0, t1, t2 -.Lret: - sub.d a0, a0, tp - REG_L t0, sp, 0 - REG_L t1, sp, 8 - REG_L t2, sp, 16 - ADDI sp, sp, 32 - RET - -.Lslow: - /* This is the slow path. We need to call __tls_get_addr() which - means we need to save and restore all the register that the - callee will trash. */ - - /* Save the remaining registers that we must treat as caller save. */ - ADDI sp, sp, -FRAME_SIZE - REG_S ra, sp, 0 * SZREG - REG_S a1, sp, 1 * SZREG - REG_S a2, sp, 2 * SZREG - REG_S a3, sp, 3 * SZREG - REG_S a4, sp, 4 * SZREG - REG_S a5, sp, 5 * SZREG - REG_S a6, sp, 6 * SZREG - REG_S a7, sp, 7 * SZREG - REG_S t3, sp, 8 * SZREG - REG_S t4, sp, 9 * SZREG - REG_S t5, sp, 10 * SZREG - REG_S t6, sp, 11 * SZREG - REG_S t7, sp, 12 * SZREG - REG_S t8, sp, 13 * SZREG - #ifndef __loongarch_soft_float - /* Save fcsr0 register. - Only one physical fcsr0 register, fcsr1-fcsr3 are aliases - of some fields in fcsr0. */ - movfcsr2gr t0, fcsr0 - st.w t0, sp, FRAME_SIZE + 24 /* Use the spare slot above t2 */ - - /* Whether support LASX. */ - la.global t0, _rtld_global_ro - REG_L t0, t0, GLRO_DL_HWCAP_OFFSET - andi t1, t0, HWCAP_LOONGARCH_LASX - beqz t1, .Llsx - - /* Save 256-bit vector registers. - FIXME: Without vector ABI, save all vector registers. */ - ADDI sp, sp, -FRAME_SIZE_LASX - xvst xr0, sp, 0*SZXREG - xvst xr1, sp, 1*SZXREG - xvst xr2, sp, 2*SZXREG - xvst xr3, sp, 3*SZXREG - xvst xr4, sp, 4*SZXREG - xvst xr5, sp, 5*SZXREG - xvst xr6, sp, 6*SZXREG - xvst xr7, sp, 7*SZXREG - xvst xr8, sp, 8*SZXREG - xvst xr9, sp, 9*SZXREG - xvst xr10, sp, 10*SZXREG - xvst xr11, sp, 11*SZXREG - xvst xr12, sp, 12*SZXREG - xvst xr13, sp, 13*SZXREG - xvst xr14, sp, 14*SZXREG - xvst xr15, sp, 15*SZXREG - xvst xr16, sp, 16*SZXREG - xvst xr17, sp, 17*SZXREG - xvst xr18, sp, 18*SZXREG - xvst xr19, sp, 19*SZXREG - xvst xr20, sp, 20*SZXREG - xvst xr21, sp, 21*SZXREG - xvst xr22, sp, 22*SZXREG - xvst xr23, sp, 23*SZXREG - xvst xr24, sp, 24*SZXREG - xvst xr25, sp, 25*SZXREG - xvst xr26, sp, 26*SZXREG - xvst xr27, sp, 27*SZXREG - xvst xr28, sp, 28*SZXREG - xvst xr29, sp, 29*SZXREG - xvst xr30, sp, 30*SZXREG - xvst xr31, sp, 31*SZXREG - b .Ltga - -.Llsx: - /* Whether support LSX. */ - andi t1, t0, HWCAP_LOONGARCH_LSX - beqz t1, .Lfloat - - /* Save 128-bit vector registers. */ - ADDI sp, sp, -FRAME_SIZE_LSX - vst vr0, sp, 0*SZVREG - vst vr1, sp, 1*SZVREG - vst vr2, sp, 2*SZVREG - vst vr3, sp, 3*SZVREG - vst vr4, sp, 4*SZVREG - vst vr5, sp, 5*SZVREG - vst vr6, sp, 6*SZVREG - vst vr7, sp, 7*SZVREG - vst vr8, sp, 8*SZVREG - vst vr9, sp, 9*SZVREG - vst vr10, sp, 10*SZVREG - vst vr11, sp, 11*SZVREG - vst vr12, sp, 12*SZVREG - vst vr13, sp, 13*SZVREG - vst vr14, sp, 14*SZVREG - vst vr15, sp, 15*SZVREG - vst vr16, sp, 16*SZVREG - vst vr17, sp, 17*SZVREG - vst vr18, sp, 18*SZVREG - vst vr19, sp, 19*SZVREG - vst vr20, sp, 20*SZVREG - vst vr21, sp, 21*SZVREG - vst vr22, sp, 22*SZVREG - vst vr23, sp, 23*SZVREG - vst vr24, sp, 24*SZVREG - vst vr25, sp, 25*SZVREG - vst vr26, sp, 26*SZVREG - vst vr27, sp, 27*SZVREG - vst vr28, sp, 28*SZVREG - vst vr29, sp, 29*SZVREG - vst vr30, sp, 30*SZVREG - vst vr31, sp, 31*SZVREG - b .Ltga - -.Lfloat: - /* Save float registers. */ - ADDI sp, sp, -FRAME_SIZE_FLOAT - FREG_S fa0, sp, 0*SZFREG - FREG_S fa1, sp, 1*SZFREG - FREG_S fa2, sp, 2*SZFREG - FREG_S fa3, sp, 3*SZFREG - FREG_S fa4, sp, 4*SZFREG - FREG_S fa5, sp, 5*SZFREG - FREG_S fa6, sp, 6*SZFREG - FREG_S fa7, sp, 7*SZFREG - FREG_S ft0, sp, 8*SZFREG - FREG_S ft1, sp, 9*SZFREG - FREG_S ft2, sp, 10*SZFREG - FREG_S ft3, sp, 11*SZFREG - FREG_S ft4, sp, 12*SZFREG - FREG_S ft5, sp, 13*SZFREG - FREG_S ft6, sp, 14*SZFREG - FREG_S ft7, sp, 15*SZFREG - FREG_S ft8, sp, 16*SZFREG - FREG_S ft9, sp, 17*SZFREG - FREG_S ft10, sp, 18*SZFREG - FREG_S ft11, sp, 19*SZFREG - FREG_S ft12, sp, 20*SZFREG - FREG_S ft13, sp, 21*SZFREG - FREG_S ft14, sp, 22*SZFREG - FREG_S ft15, sp, 23*SZFREG - -#endif /* #ifndef __loongarch_soft_float */ - -.Ltga: - bl HIDDEN_JUMPTARGET(__tls_get_addr) - ADDI a0, a0, -TLS_DTV_OFFSET - -#ifndef __loongarch_soft_float - - la.global t0, _rtld_global_ro - REG_L t0, t0, GLRO_DL_HWCAP_OFFSET - andi t1, t0, HWCAP_LOONGARCH_LASX - beqz t1, .Llsx1 - - /* Restore 256-bit vector registers. */ - xvld xr0, sp, 0*SZXREG - xvld xr1, sp, 1*SZXREG - xvld xr2, sp, 2*SZXREG - xvld xr3, sp, 3*SZXREG - xvld xr4, sp, 4*SZXREG - xvld xr5, sp, 5*SZXREG - xvld xr6, sp, 6*SZXREG - xvld xr7, sp, 7*SZXREG - xvld xr8, sp, 8*SZXREG - xvld xr9, sp, 9*SZXREG - xvld xr10, sp, 10*SZXREG - xvld xr11, sp, 11*SZXREG - xvld xr12, sp, 12*SZXREG - xvld xr13, sp, 13*SZXREG - xvld xr14, sp, 14*SZXREG - xvld xr15, sp, 15*SZXREG - xvld xr16, sp, 16*SZXREG - xvld xr17, sp, 17*SZXREG - xvld xr18, sp, 18*SZXREG - xvld xr19, sp, 19*SZXREG - xvld xr20, sp, 20*SZXREG - xvld xr21, sp, 21*SZXREG - xvld xr22, sp, 22*SZXREG - xvld xr23, sp, 23*SZXREG - xvld xr24, sp, 24*SZXREG - xvld xr25, sp, 25*SZXREG - xvld xr26, sp, 26*SZXREG - xvld xr27, sp, 27*SZXREG - xvld xr28, sp, 28*SZXREG - xvld xr29, sp, 29*SZXREG - xvld xr30, sp, 30*SZXREG - xvld xr31, sp, 31*SZXREG - ADDI sp, sp, FRAME_SIZE_LASX - b .Lfcsr - -.Llsx1: - andi t1, t0, HWCAP_LOONGARCH_LSX - beqz t1, .Lfloat1 - - /* Restore 128-bit vector registers. */ - vld vr0, sp, 0*SZVREG - vld vr1, sp, 1*SZVREG - vld vr2, sp, 2*SZVREG - vld vr3, sp, 3*SZVREG - vld vr4, sp, 4*SZVREG - vld vr5, sp, 5*SZVREG - vld vr6, sp, 6*SZVREG - vld vr7, sp, 7*SZVREG - vld vr8, sp, 8*SZVREG - vld vr9, sp, 9*SZVREG - vld vr10, sp, 10*SZVREG - vld vr11, sp, 11*SZVREG - vld vr12, sp, 12*SZVREG - vld vr13, sp, 13*SZVREG - vld vr14, sp, 14*SZVREG - vld vr15, sp, 15*SZVREG - vld vr16, sp, 16*SZVREG - vld vr17, sp, 17*SZVREG - vld vr18, sp, 18*SZVREG - vld vr19, sp, 19*SZVREG - vld vr20, sp, 20*SZVREG - vld vr21, sp, 21*SZVREG - vld vr22, sp, 22*SZVREG - vld vr23, sp, 23*SZVREG - vld vr24, sp, 24*SZVREG - vld vr25, sp, 25*SZVREG - vld vr26, sp, 26*SZVREG - vld vr27, sp, 27*SZVREG - vld vr28, sp, 28*SZVREG - vld vr29, sp, 29*SZVREG - vld vr30, sp, 30*SZVREG - vld vr31, sp, 31*SZVREG - ADDI sp, sp, FRAME_SIZE_LSX - b .Lfcsr - -.Lfloat1: - /* Restore float registers. */ - FREG_L fa0, sp, 0*SZFREG - FREG_L fa1, sp, 1*SZFREG - FREG_L fa2, sp, 2*SZFREG - FREG_L fa3, sp, 3*SZFREG - FREG_L fa4, sp, 4*SZFREG - FREG_L fa5, sp, 5*SZFREG - FREG_L fa6, sp, 6*SZFREG - FREG_L fa7, sp, 7*SZFREG - FREG_L ft0, sp, 8*SZFREG - FREG_L ft1, sp, 9*SZFREG - FREG_L ft2, sp, 10*SZFREG - FREG_L ft3, sp, 11*SZFREG - FREG_L ft4, sp, 12*SZFREG - FREG_L ft5, sp, 13*SZFREG - FREG_L ft6, sp, 14*SZFREG - FREG_L ft7, sp, 15*SZFREG - FREG_L ft8, sp, 16*SZFREG - FREG_L ft9, sp, 17*SZFREG - FREG_L ft10, sp, 18*SZFREG - FREG_L ft11, sp, 19*SZFREG - FREG_L ft12, sp, 20*SZFREG - FREG_L ft13, sp, 21*SZFREG - FREG_L ft14, sp, 22*SZFREG - FREG_L ft15, sp, 23*SZFREG - ADDI sp, sp, FRAME_SIZE_FLOAT - -.Lfcsr: - /* Restore fcsr0 register. */ - ld.w t0, sp, FRAME_SIZE + 24 - movgr2fcsr fcsr0, t0 +#define USE_LASX +#define _dl_tlsdesc_dynamic _dl_tlsdesc_dynamic_lasx +#define Lret Lret_lasx +#define Lslow Lslow_lasx +#include "dl-tlsdesc-dynamic.h" +#undef FRAME_SIZE +#undef V_REG_S +#undef V_REG_L +#undef V_SPACE +#undef V_REG +#undef V_REGS +#undef V_REGSZ +#undef USE_LASX +#undef _dl_tlsdesc_dynamic +#undef Lret +#undef Lslow + +#define USE_LSX +#define _dl_tlsdesc_dynamic _dl_tlsdesc_dynamic_lsx +#define Lret Lret_lsx +#define Lslow Lslow_lsx +#include "dl-tlsdesc-dynamic.h" +#undef FRAME_SIZE +#undef V_REG_S +#undef V_REG_L +#undef V_SPACE +#undef V_REG +#undef V_REGS +#undef V_REGSZ +#undef USE_LSX +#undef _dl_tlsdesc_dynamic +#undef Lret +#undef Lslow #endif /* #ifndef __loongarch_soft_float */ - REG_L ra, sp, 0 * SZREG - REG_L a1, sp, 1 * SZREG - REG_L a2, sp, 2 * SZREG - REG_L a3, sp, 3 * SZREG - REG_L a4, sp, 4 * SZREG - REG_L a5, sp, 5 * SZREG - REG_L a6, sp, 6 * SZREG - REG_L a7, sp, 7 * SZREG - REG_L t3, sp, 8 * SZREG - REG_L t4, sp, 9 * SZREG - REG_L t5, sp, 10 * SZREG - REG_L t6, sp, 11 * SZREG - REG_L t7, sp, 12 * SZREG - REG_L t8, sp, 13 * SZREG - ADDI sp, sp, FRAME_SIZE - - b .Lret - cfi_endproc - .size _dl_tlsdesc_dynamic, .-_dl_tlsdesc_dynamic - .hidden HIDDEN_JUMPTARGET(__tls_get_addr) +#include "dl-tlsdesc-dynamic.h" #endif /* #ifdef SHARED */ |