Hello, I know there are currently enough TIF_ flags, but some time ago I promised to post a patch that extends their very limited number to 8. The main reason for having special work-to-be-done bits is that they can be checked by a more efficient CPU instruction. The reason for only having 7 such bits on IA64 is that the "and" instruction sign-extends its immediate argument. Since the assembler code never cares about the high bits, it is possible to load only the first byte from the thread flags. The 8-bit value is zero-extended, so the sign extension of the most significant bit in the "and immediate" instruction does not matter (it gets masked off by the register content itself). This allows to use 8 TIF_ flags for pending work to be done. Signed-off-by: Petr Tesarik <ptesarik@xxxxxxx> diff --git a/arch/ia64/include/asm/thread_info.h b/arch/ia64/include/asm/thread_info.h index 7c60fcd..e30c470 100644 --- a/arch/ia64/include/asm/thread_info.h +++ b/arch/ia64/include/asm/thread_info.h @@ -95,8 +95,8 @@ extern void tsk_clear_notify_resume(struct task_struct *tsk); /* * thread information flags * - these are process state flags that various assembly files may need to access - * - pending work-to-be-done flags are in least-significant 16 bits, other flags - * in top 16 bits + * - pending work-to-be-done flags are in least-significant 8 bits, other flags + * in top 24 bits */ #define TIF_SIGPENDING 0 /* signal pending */ #define TIF_NEED_RESCHED 1 /* rescheduling necessary */ diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S index 0dd6c14..3bc2119 100644 --- a/arch/ia64/kernel/entry.S +++ b/arch/ia64/kernel/entry.S @@ -626,7 +626,7 @@ GLOBAL_ENTRY(ia64_ret_from_clone) .ret8: adds r2=TI_FLAGS+IA64_TASK_SIZE,r13 ;; - ld4 r2=[r2] + ld1 r2=[r2] ;; mov r8=0 and r2=_TIF_SYSCALL_TRACEAUDIT,r2 @@ -732,7 +732,7 @@ __paravirt_work_processed_syscall: (pUStk) mov.m r22=ar.itc // fetch time at leave adds r18=TI_FLAGS+IA64_TASK_SIZE,r13 ;; -(p6) ld4 r31=[r18] // load current_thread_info()->flags +(p6) ld1 r31=[r18] // r31 = current_thread_info()->flags & 0xff ld8 r19=[r2],PT(B6)-PT(LOADRS) // load ar.rsc value for "loadrs" adds r3=PT(AR_BSPSTORE)+16,r12 // deferred ;; @@ -741,7 +741,7 @@ __paravirt_work_processed_syscall: adds r3=PT(AR_BSPSTORE)+16,r12 adds r18=TI_FLAGS+IA64_TASK_SIZE,r13 ;; -(p6) ld4 r31=[r18] // load current_thread_info()->flags +(p6) ld1 r31=[r18] // r31 = current_thread_info()->flags & 0xff ld8 r19=[r2],PT(B6)-PT(LOADRS) // load ar.rsc value for "loadrs" nop.i 0 ;; @@ -896,7 +896,7 @@ GLOBAL_ENTRY(__paravirt_leave_kernel) .work_processed_kernel: adds r17=TI_FLAGS+IA64_TASK_SIZE,r13 ;; -(p6) ld4 r31=[r17] // load current_thread_info()->flags +(p6) ld1 r31=[r17] // r31 = current_thread_info()->flags & 0xff adds r21=PT(PR)+16,r12 ;; diff --git a/arch/ia64/kernel/fsys.S b/arch/ia64/kernel/fsys.S index c1625c7..d9feb9c 100644 --- a/arch/ia64/kernel/fsys.S +++ b/arch/ia64/kernel/fsys.S @@ -66,7 +66,7 @@ ENTRY(fsys_getpid) ld8 r17=[r17] // r17 = current->group_leader add r9=TI_FLAGS+IA64_TASK_SIZE,r16 ;; - ld4 r9=[r9] + ld1 r9=[r9] add r17=IA64_TASK_TGIDLINK_OFFSET,r17 ;; and r9=TIF_ALLWORK_MASK,r9 @@ -100,7 +100,7 @@ ENTRY(fsys_getppid) add r9=TI_FLAGS+IA64_TASK_SIZE,r16 ;; - ld4 r9=[r9] + ld1 r9=[r9] add r17=IA64_TASK_REAL_PARENT_OFFSET,r17 // r17 = ¤t->group_leader->real_parent ;; and r9=TIF_ALLWORK_MASK,r9 @@ -144,7 +144,7 @@ ENTRY(fsys_set_tid_address) add r9=TI_FLAGS+IA64_TASK_SIZE,r16 add r17=IA64_TASK_TGIDLINK_OFFSET,r16 ;; - ld4 r9=[r9] + ld1 r9=[r9] tnat.z p6,p7=r32 // check argument register for being NaT ld8 r17=[r17] // r17 = current->pids[PIDTYPE_PID].pid ;; @@ -242,7 +242,7 @@ ENTRY(fsys_gettimeofday) (p6) br.cond.spnt.few .fail_einval movl r20 = fsyscall_gtod_data // load fsyscall gettimeofday data address ;; - ld4 r2 = [r2] // process work pending flags + ld1 r2 = [r2] // process work pending flags movl r29 = itc_jitter_data // itc_jitter add r22 = IA64_GTOD_WALL_TIME_OFFSET,r20 // wall_time add r21 = IA64_CLKSRC_MMIO_OFFSET,r20 @@ -391,7 +391,7 @@ ENTRY(fsys_rt_sigprocmask) add r31=IA64_TASK_SIGHAND_OFFSET,r16 ;; ld8 r3=[r2] // read/prefetch current->blocked - ld4 r9=[r9] + ld1 r9=[r9] tnat.nz.or p6,p0=r35 cmp.ne.or p6,p0=_NSIG_WORDS*8,r35 @@ -476,14 +476,14 @@ EX(.fail_efault, ld8 r14=[r33]) // r14 <- *set (p6) br.cond.dpnt.many .sig_pending ;; -1: ld4 r17=[r9] // r17 <- current->thread_info->flags +1: ld1 r17=[r9] // r17 <- current->thread_info->flags & 0xff ;; mov ar.ccv=r17 and r18=~_TIF_SIGPENDING,r17 // r18 <- r17 & ~(1 << TIF_SIGPENDING) ;; st8 [r2]=r14 // update current->blocked with new mask - cmpxchg4.acq r8=[r9],r18,ar.ccv // current->thread_info->flags <- r18 + cmpxchg1.acq r8=[r9],r18,ar.ccv // current->thread_info->flags <- r18 ;; cmp.ne p6,p0=r17,r8 // update failed? (p6) br.cond.spnt.few 1b // yes -> retry @@ -542,7 +542,7 @@ ENTRY(fsys_getcpu) add r3=TI_CPU+IA64_TASK_SIZE,r16 ;; ld4 r3=[r3] // M r3 = thread_info->cpu - ld4 r2=[r2] // M r2 = thread_info->flags + ld1 r2=[r2] // M r2 = thread_info->flags & 0xff (p6) br.cond.spnt.few .fail_einval // B ;; tnat.nz p7,p0 = r33 // I guard against NaT argument @@ -677,7 +677,7 @@ GLOBAL_ENTRY(fsys_bubble_down) addl r22=IA64_RBS_OFFSET,r2 // A compute base of RBS add r3=TI_FLAGS+IA64_TASK_SIZE,r2 // A ;; - ld4 r3=[r3] // M0|1 r3 = current_thread_info()->flags + ld1 r3=[r3] // M0|1 r3 = current_thread_info()->flags & 0xff lfetch.fault.excl.nt1 [r22] // M0|1 prefetch register backing-store nop.i 0 ;; diff --git a/arch/ia64/kernel/ivt.S b/arch/ia64/kernel/ivt.S index 416a952..4ab0253 100644 --- a/arch/ia64/kernel/ivt.S +++ b/arch/ia64/kernel/ivt.S @@ -765,7 +765,7 @@ ENTRY(break_fault) mov r3=NR_syscalls - 1 ;; ld1.bias r17=[r16] // M0|1 r17 = current->thread.on_ustack flag - ld4 r9=[r9] // M0|1 r9 = current_thread_info()->flags + ld1 r9=[r9] // M0|1 r9 = current_thread_info()->flags extr.u r8=r29,41,2 // I0 extract ei field from cr.ipsr shladd r30=r15,3,r30 // A r30 = sys_call_table + 8*(syscall-1024) @@ -1775,7 +1775,7 @@ ENTRY(dispatch_to_ia32_handler) movl r16=ia32_syscall_table ;; (p6) shladd r16=r8,3,r16 // force ni_syscall if not valid syscall number - ld4 r2=[r2] // r2 = current_thread_info()->flags + ld1 r2=[r2] // r2 = current_thread_info()->flags ;; ld8 r16=[r16] and r2=_TIF_SYSCALL_TRACEAUDIT,r2 // mask trace or audit -- To unsubscribe from this list: send the line "unsubscribe linux-ia64" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
|