arch/intel64: optimise context switch

We save interrupted TCB state on tcb->xcp.regs not interrupt stack now
which allows us to remove x86_64_savestate() from up_switch_context()
and other places.

Signed-off-by: p-szafonimateusz <[email protected]>
Signed-off-by: hujun5 <[email protected]>

arch/x86_64/include/intel64/irq.h

@@ -461,6 +461,10 @@
 
 #define XMMAREA_REGS     (25)
 
+/* Aux register used by implementation */
+
+#define REG_AUX          (26 + XMMAREA_REG_OFFSET)
+
 /* NOTE 2: This is not really state data.  Rather, this is just a convenient
  *   way to pass parameters from the interrupt handler to C code.
  */
@@ -478,6 +482,10 @@
 #define XCP_ALIGN_DOWN(a) ((a) & ~XCP_ALIGN_MASK)
 #define XCP_ALIGN_UP(a)   (((a) + XCP_ALIGN_MASK) & ~XCP_ALIGN_MASK)
 
+/* Aux register flags */
+
+#define REG_AUX_FULLCONTEXT (1 << 0) /* Force full context switch */
+
 /****************************************************************************
  * Public Types
  ****************************************************************************/

arch/x86_64/src/common/x86_64_internal.h

@@ -220,7 +220,6 @@ void x86_64_boardinitialize(void);
 /* Defined in files with the same name as the function */
 
 void x86_64_copystate(uint64_t *dest, uint64_t *src);
-void x86_64_savestate(uint64_t *regs);
 void x86_64_decodeirq(uint64_t *regs);
 #ifdef CONFIG_ARCH_DMA
 void weak_function x86_64_dmainitialize(void);

arch/x86_64/src/common/x86_64_switchcontext.c

@@ -63,17 +63,11 @@ void up_switch_context(struct tcb_s *tcb, struct tcb_s *rtcb)
 
   if (up_interrupt_context())
     {
-      /* Yes, then we have to do things differently.
-       * Just copy the g_current_regs into the OLD rtcb.
-       */
-
-      x86_64_savestate(rtcb->xcp.regs);
+      /* Restore addition x86_64 state */
 
       x86_64_restore_auxstate(tcb);
 
-      /* Then switch contexts.  Any necessary address environment
-       * changes will be made when the interrupt returns.
-       */
+      /* Update current regs to signal that we need context switch */
 
       x86_64_restorestate(tcb->xcp.regs);
     }

arch/x86_64/src/intel64/CMakeLists.txt

@@ -31,7 +31,6 @@ set(SRCS
     intel64_releasestack.c
     intel64_rtc.c
     intel64_restore_auxstate.c
-    intel64_savestate.c
     intel64_stackframe.c
     intel64_schedulesigaction.c
     intel64_sigdeliver.c

arch/x86_64/src/intel64/Make.defs

@@ -22,7 +22,7 @@ include common/Make.defs
 
 CMN_CSRCS += intel64_createstack.c intel64_initialstate.c intel64_irq.c
 CMN_CSRCS += intel64_map_region.c intel64_regdump.c intel64_releasestack.c
-CMN_CSRCS += intel64_rtc.c intel64_restore_auxstate.c intel64_savestate.c
+CMN_CSRCS += intel64_rtc.c intel64_restore_auxstate.c
 CMN_CSRCS += intel64_stackframe.c intel64_schedulesigaction.c
 CMN_CSRCS += intel64_sigdeliver.c intel64_usestack.c x86_64_tcbinfo.c
 CMN_CSRCS += intel64_systemreset.c intel64_freq.c intel64_cache.c

arch/x86_64/src/intel64/intel64_handlers.c

@@ -80,9 +80,8 @@ static uint64_t *common_handler(int irq, uint64_t *regs)
 
   /* Check for a context switch.  If a context switch occurred, then
    * g_current_regs will have a different value than it did on entry.  If an
-   * interrupt level context switch has occurred, then restore the floating
-   * point state and the establish the correct address environment before
-   * returning from the interrupt.
+   * interrupt level context switch has occurred, then the establish the
+   * correct address environment before returning from the interrupt.
    */
 
   if (regs != up_current_regs())
@@ -238,3 +237,27 @@ uint64_t *irq_handler(uint64_t *regs, uint64_t irq_no)
   return ret;
 #endif
 }
+
+/****************************************************************************
+ * Name: irq_xcp_regs
+ *
+ * Description:
+ *   Return current task XCP registers area.
+ *
+ * ASSUMPTION:
+ *   Interrupts are disabled.
+ *
+ *   This function should be called only form intel64_vector.S file !
+ *   Any other use must be carefully considered.
+ *
+ ****************************************************************************/
+
+uint64_t *irq_xcp_regs(void)
+{
+  /* This must be the simplest as possible, so we not use too much registers.
+   * Now this function corrupts only RAX and RDX registers regardless of
+   * the compiler optimization.
+   */
+
+  return (current_task(this_cpu()))->xcp.regs;
+}

arch/x86_64/src/intel64/intel64_saveusercontext.S

@@ -80,13 +80,13 @@ up_saveusercontext:
 	/* Save the value of SP as will be at the time of the IRET that will
 	 * appear to be the return from this function.
 	 *
-	 *         CURRENT STACK           IRET STACK
-	 *        ------------------------------        -----------------
+	 * CURRENT STACK                IRET STACK
+	 * ---------------------        -----------------
 	 *                              RIP
 	 *                              CS
 	 *                              RFLAGS
 	 *                              RSP
-	 * ESP->Return address          SS
+	 * RSP->Return address          SS
 	 * Argument                     Alignment (16bytes)
 	 *
 	 */

arch/x86_64/src/intel64/intel64_schedulesigaction.c

@@ -126,12 +126,6 @@ void up_schedule_sigaction(struct tcb_s *tcb)
           up_current_regs()[REG_RIP]    = (uint64_t)x86_64_sigdeliver;
           up_current_regs()[REG_RSP]    = up_current_regs()[REG_RSP] - 8;
           up_current_regs()[REG_RFLAGS] = 0;
-
-          /* And make sure that the saved context in the TCB
-           * is the same as the interrupt return context.
-           */
-
-          x86_64_savestate(tcb->xcp.regs);
         }
     }
 
@@ -222,11 +216,14 @@ void up_schedule_sigaction(struct tcb_s *tcb)
           up_current_regs()[REG_RSP]    = up_current_regs()[REG_RSP] - 8;
           up_current_regs()[REG_RFLAGS] = 0;
 
-          /* And make sure that the saved context in the TCB
-           * is the same as the interrupt return context.
+          /* Mark that full context switch is necessary when we
+           * return from interrupt handler.
+           * In that case RIP, RSP and RFLAGS are changed, but
+           * register area pointer remains the same, so we need an
+           * additional variable to signal the need for full context switch
            */
 
-          x86_64_savestate(tcb->xcp.regs);
+          tcb->xcp.regs[REG_AUX] = REG_AUX_FULLCONTEXT;
         }
     }
 

arch/x86_64/src/intel64/intel64_smpcall.c

@@ -62,7 +62,6 @@ int x86_64_smp_call_handler(int irq, void *c, void *arg)
   int cpu = this_cpu();
 
   tcb = current_task(cpu);
-  x86_64_savestate(tcb->xcp.regs);
   nxsched_smp_call_handler(irq, c, arg);
   tcb = current_task(cpu);
   x86_64_restorestate(tcb->xcp.regs);
@@ -97,7 +96,6 @@ int x86_64_smp_sched_handler(int irq, void *c, void *arg)
 
   tcb = current_task(cpu);
   nxsched_suspend_scheduler(tcb);
-  x86_64_savestate(tcb->xcp.regs);
   nxsched_process_delivered(cpu);
   tcb = current_task(cpu);
   nxsched_resume_scheduler(tcb);