Changes in libcfa/src/concurrency/preemption.cfa [8fc652e0:82a2fed]

File:

• libcfa/src/concurrency/preemption.cfa (modified) (11 diffs)

libcfa/src/concurrency/preemption.cfa

@@ -10 +10 @@
 // Created On       : Mon Jun 5 14:20:42 2017
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Wed Aug 26 16:46:03 2020
-// Update Count     : 53
+// Last Modified On : Fri Nov  6 07:42:13 2020
+// Update Count     : 54
 //
 
@@ -163 +163 @@
 // Kernel Signal Tools
 //=============================================================================================
+// In a user-level threading system, there are handful of thread-local variables where this problem occurs on the ARM.
+//
+// For each kernel thread running user-level threads, there is a flag variable to indicate if interrupts are
+// enabled/disabled for that kernel thread. Therefore, this variable is made thread local.
+//
+// For example, this code fragment sets the state of the "interrupt" variable in thread-local memory.
+//
+// _Thread_local volatile int interrupts;
+// int main() {
+//     interrupts = 0; // disable interrupts }
+//
+// which generates the following code on the ARM
+//
+// (gdb) disassemble main
+// Dump of assembler code for function main:
+//    0x0000000000000610 <+0>:  mrs     x1, tpidr_el0
+//    0x0000000000000614 <+4>:  mov     w0, #0x0                        // #0
+//    0x0000000000000618 <+8>:  add     x1, x1, #0x0, lsl #12
+//    0x000000000000061c <+12>: add     x1, x1, #0x10
+//    0x0000000000000620 <+16>: str     wzr, [x1]
+//    0x0000000000000624 <+20>: ret
+//
+// The mrs moves a pointer from coprocessor register tpidr_el0 into register x1.  Register w0 is set to 0. The two adds
+// increase the TLS pointer with the displacement (offset) 0x10, which is the location in the TSL of variable
+// "interrupts".  Finally, 0 is stored into "interrupts" through the pointer in register x1 that points into the
+// TSL. Now once x1 has the pointer to the location of the TSL for kernel thread N, it can be be preempted at a
+// user-level and the user thread is put on the user-level ready-queue. When the preempted thread gets to the front of
+// the user-level ready-queue it is run on kernel thread M. It now stores 0 into "interrupts" back on kernel thread N,
+// turning off interrupt on the wrong kernel thread.
+//
+// On the x86, the following code is generated for the same code fragment.
+//
+// (gdb) disassemble main
+// Dump of assembler code for function main:
+//    0x0000000000400420 <+0>:  movl   $0x0,%fs:0xfffffffffffffffc
+//    0x000000000040042c <+12>: xor    %eax,%eax
+//    0x000000000040042e <+14>: retq
+//
+// and there is base-displacement addressing used to atomically reset variable "interrupts" off of the TSL pointer in
+// register "fs".
+//
+// Hence, the ARM has base-displacement address for the general purpose registers, BUT not to the coprocessor
+// registers. As a result, generating the address for the write into variable "interrupts" is no longer atomic.
+//
+// Note this problem does NOT occur when just using multiple kernel threads because the preemption ALWAYS restarts the
+// thread on the same kernel thread.
+//
+// The obvious question is why does ARM use a coprocessor register to store the TSL pointer given that coprocessor
+// registers are second-class registers with respect to the instruction set. One possible answer is that they did not
+// want to dedicate one of the general registers to hold the TLS pointer and there was a free coprocessor register
+// available.
 
 //----------
@@ -180 +231 @@
 }
 
+struct asm_region {
+        void * before;
+        void * after;
+};
+
+static inline bool __cfaasm_in( void * ip, struct asm_region & region ) {
+        return ip >= region.before && ip <= region.after;
+}
+
+
 //----------
 // Get data from the TLS block
+// struct asm_region __cfaasm_get;
 uintptr_t __cfatls_get( unsigned long int offset ) __attribute__((__noinline__)); //no inline to avoid problems
 uintptr_t __cfatls_get( unsigned long int offset ) {
+        // __cfaasm_get.before = ({ void * value; asm("movq $__cfaasm_get_before, %[v]\n\t" : [v]"=r"(value) ); value; });
+        // __cfaasm_get.after  = ({ void * value; asm("movq $__cfaasm_get_after , %[v]\n\t" : [v]"=r"(value) ); value; });
         // create a assembler label before
         // marked as clobber all to avoid movement
@@ -196 +260 @@
         return val;
 }
-
-// //----------
-// // Write data to the TLS block
-// // sadly it looses the type information and can only write 1 word at a time
-// // use with __builtin_offsetof
-// void __cfatls_set(uintptr_t offset, void * value) __attribute__((__noinline__));
-// void __cfatls_set(uintptr_t offset, void * value) {
-//     // create a assembler label before
-//     // marked as clobber all to avoid movement
-//     asm volatile("__cfaasm_set_before:":::"memory");
-
-//     // access tls as normal (except for type information)
-//     *(void**)(offset + (uintptr_t)&my_tls) = value;
-
-//     // create a assembler label after
-//     // marked as clobber all to avoid movement
-//     asm volatile("__cfaasm_set_after:":::"memory");
-// }
-
-// //----------
-// #include <stdio.h>
-// int main() {
-//     // Get the information
-//     // Must use inline assembly to get access to label
-//     // C is annoying here because this could easily be a static const but "initializer element is not a compile-time constant"
-//     // The big advantage of this approach is that there is 0 overhead for the read and writes function
-//     void * __cfaasm_addr_get_before = ({ void * value; asm("movq $__cfaasm_get_before, %[v]\n\t" : [v]"=r"(value) ); value; });
-//     void * __cfaasm_addr_get_after  = ({ void * value; asm("movq $__cfaasm_get_after , %[v]\n\t" : [v]"=r"(value) ); value; });
-//     void * __cfaasm_addr_set_before = ({ void * value; asm("movq $__cfaasm_set_before, %[v]\n\t" : [v]"=r"(value) ); value; });
-//     void * __cfaasm_addr_set_after  = ({ void * value; asm("movq $__cfaasm_set_after , %[v]\n\t" : [v]"=r"(value) ); value; });
-
-//     printf("%p to %p\n", __cfaasm_addr_get_before, __cfaasm_addr_get_after);
-//     printf("%p to %p\n", __cfaasm_addr_set_before, __cfaasm_addr_set_after);
-//     return 0;
-// }
-
-__cfaabi_dbg_debug_do( static thread_local void * last_interrupt = 0; )
 
 extern "C" {
@@ -239 +266 @@
                 // create a assembler label before
                 // marked as clobber all to avoid movement
-                asm volatile("__cfaasm_disable_before:":::"memory");
+                asm volatile("__cfaasm_dsable_before:":::"memory");
 
                 with( __cfaabi_tls.preemption_state ) {
@@ -261 +288 @@
                 // create a assembler label after
                 // marked as clobber all to avoid movement
-                asm volatile("__cfaasm_disable_after:":::"memory");
+                asm volatile("__cfaasm_dsable_after:":::"memory");
         }
 
@@ -269 +296 @@
                 // create a assembler label before
                 // marked as clobber all to avoid movement
-                asm volatile("__cfaasm_enable_before:":::"memory");
+                asm volatile("__cfaasm_enble_before:":::"memory");
 
                 processor   * proc = __cfaabi_tls.this_processor; // Cache the processor now since interrupts can start happening after the atomic store
@@ -304 +331 @@
                 // create a assembler label after
                 // marked as clobber all to avoid movement
-                asm volatile("__cfaasm_enable_after:":::"memory");
+                asm volatile("__cfaasm_enble_after:":::"memory");
         }
 
@@ -372 +399 @@
 }
 
+//-----------------------------------------------------------------------------
+// Some assembly required
+#if defined( __i386 )
+        #ifdef __PIC__
+                #define RELOC_PRELUDE( label ) \
+                        "calll   .Lcfaasm_prelude_" #label "$pb\n\t" \
+                        ".Lcfaasm_prelude_" #label "$pb:\n\t" \
+                        "popl    %%eax\n\t" \
+                        ".Lcfaasm_prelude_" #label "_end:\n\t" \
+                        "addl    $_GLOBAL_OFFSET_TABLE_+(.Lcfaasm_prelude_" #label "_end-.Lcfaasm_prelude_" #label "$pb), %%eax\n\t"
+                #define RELOC_PREFIX ""
+                #define RELOC_SUFFIX "@GOT(%%eax)"
+        #else
+                #define RELOC_PREFIX "$"
+                #define RELOC_SUFFIX ""
+        #endif
+        #define __cfaasm_label( label ) static struct asm_region label = \
+                ({ \
+                        struct asm_region region; \
+                        asm( \
+                                RELOC_PRELUDE( label ) \
+                                "movl " RELOC_PREFIX "__cfaasm_" #label "_before" RELOC_SUFFIX ", %[vb]\n\t" \
+                                "movl " RELOC_PREFIX "__cfaasm_" #label "_after"  RELOC_SUFFIX ", %[va]\n\t" \
+                                 : [vb]"=r"(region.before), [va]"=r"(region.after) \
+                        ); \
+                        region; \
+                });
+#elif defined( __x86_64 )
+        #ifdef __PIC__
+                #define RELOC_PREFIX ""
+                #define RELOC_SUFFIX "@GOTPCREL(%%rip)"
+        #else
+                #define RELOC_PREFIX "$"
+                #define RELOC_SUFFIX ""
+        #endif
+        #define __cfaasm_label( label ) static struct asm_region label = \
+                ({ \
+                        struct asm_region region; \
+                        asm( \
+                                "movq " RELOC_PREFIX "__cfaasm_" #label "_before" RELOC_SUFFIX ", %[vb]\n\t" \
+                                "movq " RELOC_PREFIX "__cfaasm_" #label "_after"  RELOC_SUFFIX ", %[va]\n\t" \
+                                 : [vb]"=r"(region.before), [va]"=r"(region.after) \
+                        ); \
+                        region; \
+                });
+#elif defined( __aarch64__ )
+        #ifdef __PIC__
+                #define RELOC_TAG "@PLT"
+        #else
+                #define RELOC_TAG ""
+        #endif
+        #define __cfaasm_label( label ) \
+                ({ \
+                        struct asm_region region; \
+                        asm( \
+                                "mov %[vb], __cfaasm_" #label "_before@GOTPCREL(%%rip)"  "\n\t" \
+                                "mov %[va], __cfaasm_" #label "_after@GOTPCREL(%%rip)"   "\n\t" \
+                                 : [vb]"=r"(region.before), [va]"=r"(region.after) \
+                        ); \
+                        region; \
+                });
+#else
+        #error unknown hardware architecture
+#endif
+
 // KERNEL ONLY
 // Check if a __cfactx_switch signal handler shoud defer
 // If true  : preemption is safe
 // If false : preemption is unsafe and marked as pending
-static inline bool preemption_ready() {
+static inline bool preemption_ready( void * ip ) {
+        // Get all the region for which it is not safe to preempt
+        __cfaasm_label( get    );
+        __cfaasm_label( check  );
+        __cfaasm_label( dsable );
+        __cfaasm_label( enble );
+        __cfaasm_label( nopoll );
+
         // Check if preemption is safe
-        bool ready = __cfaabi_tls.preemption_state.enabled && ! __cfaabi_tls.preemption_state.in_progress;
-
+        bool ready = true;
+        if( __cfaasm_in( ip, get    ) ) { ready = false; goto EXIT; };
+        if( __cfaasm_in( ip, check  ) ) { ready = false; goto EXIT; };
+        if( __cfaasm_in( ip, dsable ) ) { ready = false; goto EXIT; };
+        if( __cfaasm_in( ip, enble  ) ) { ready = false; goto EXIT; };
+        if( __cfaasm_in( ip, nopoll ) ) { ready = false; goto EXIT; };
+        if( !__cfaabi_tls.preemption_state.enabled) { ready = false; goto EXIT; };
+        if( __cfaabi_tls.preemption_state.in_progress ) { ready = false; goto EXIT; };
+
+EXIT:
         // Adjust the pending flag accordingly
         __cfaabi_tls.this_processor->pending_preemption = !ready;
@@ -454 +561 @@
 // Kernel Signal Handlers
 //=============================================================================================
-struct asm_region {
-        void * before;
-        void * after;
-};
-
-//-----------------------------------------------------------------------------
-// Some assembly required
-#if defined( __i386 )
-        #define __cfaasm_label( label ) \
-                ({ \
-                        struct asm_region region; \
-                        asm( \
-                                "movl $__cfaasm_" #label "_before, %[vb]\n\t" \
-                                "movl $__cfaasm_" #label "_after , %[va]\n\t" \
-                                 : [vb]"=r"(region.before), [vb]"=r"(region.before) \
-                        ); \
-                        region; \
-                });
-#elif defined( __x86_64 )
-        #ifdef __PIC__
-                #define PLT "@PLT"
-        #else
-                #define PLT ""
-        #endif
-        #define __cfaasm_label( label ) \
-                ({ \
-                        struct asm_region region; \
-                        asm( \
-                                "movq $__cfaasm_" #label "_before" PLT ", %[vb]\n\t" \
-                                "movq $__cfaasm_" #label "_after"  PLT ", %[va]\n\t" \
-                                 : [vb]"=r"(region.before), [va]"=r"(region.after) \
-                        ); \
-                        region; \
-                });
-#elif defined( __aarch64__ )
-        #error __cfaasm_label undefined for arm
-#else
-        #error unknown hardware architecture
-#endif
+__cfaabi_dbg_debug_do( static thread_local void * last_interrupt = 0; )
 
 // Context switch signal handler
@@ -514 +583 @@
 
         // Check if it is safe to preempt here
-        if( !preemption_ready() ) { return; }
-
-        struct asm_region region;
-        region = __cfaasm_label( get     ); if( ip >= region.before && ip <= region.after ) return;
-        region = __cfaasm_label( check   ); if( ip >= region.before && ip <= region.after ) return;
-        region = __cfaasm_label( disable ); if( ip >= region.before && ip <= region.after ) return;
-        region = __cfaasm_label( enable  ); if( ip >= region.before && ip <= region.after ) return;
-        region = __cfaasm_label( nopoll  ); if( ip >= region.before && ip <= region.after ) return;
+        if( !preemption_ready( ip ) ) { return; }
 
         __cfaabi_dbg_print_buffer_decl( " KERNEL: preempting core %p (%p @ %p).\n", __cfaabi_tls.this_processor, __cfaabi_tls.this_thread, (void *)(cxt->uc_mcontext.CFA_REG_IP) );