bsp/pc386: Turn start16.S into a startAP.S

start16.S is now only used for SMP configurations to start the
application processors.

This commit removes all unnecessary parts for this job,
i.e. video conssole initalisation, A20 gate activation
and all non-AP related code.

Update #3335

4 files changed, 159 insertions, 258 deletions

diff --git a/bsps/i386/pc386/start/smp-imps.c b/bsps/i386/pc386/start/smp-imps.c
index 0543b17ec5..58d9178f90 100644
--- a/bsps/i386/pc386/start/smp-imps.c
+++ b/bsps/i386/pc386/start/smp-imps.c
@@ -309,6 +309,11 @@ boot_cpu(imps_processor *proc)
   }
 
   /*
+   *  Wait until AP is in protected mode before starting the next AP
+   */
+  while (reset[2] != 0);
+
+  /*
    *  Generic CPU startup sequence ends here, the rest is cleanup.
    */
 
@@ -342,12 +347,17 @@ add_processor(imps_processor *proc)
     printk("#0  BootStrap Processor (BSP)\n");
     return;
   }
+  /* Setup the apic/cpu maps before booting the APs
+   * otherwise calls to _Get_current_processor can deliver
+   * wrong values if the BSP gets interrupted
+   */
+  imps_cpu_apic_map[imps_num_cpus] = apicid;
+  imps_apic_cpu_map[apicid] = imps_num_cpus;
   if (boot_cpu(proc)) {
 
     /*  XXXXX  add OS-specific setup for secondary CPUs here */
 
-    imps_cpu_apic_map[imps_num_cpus] = apicid;
-    imps_apic_cpu_map[apicid] = imps_num_cpus;
+    /* AP booted successfully, increase number of available cores */
     imps_num_cpus++;
   }
 }
diff --git a/bsps/i386/pc386/start/start16.S b/bsps/i386/pc386/start/start16.S
deleted file mode 100644
index 3d46f40ed6..0000000000
--- a/bsps/i386/pc386/start/start16.S
+++ /dev/null
@@ -1,254 +0,0 @@
-/*--------------------------------------------------------------------------+
- * start16.s v1.0 - PC386 BSP - 1998/04/13
- *--------------------------------------------------------------------------+
- * This file contains the entry point for the application.
- * The name of this entry point is compiler dependent.
- * It jumps to the BSP which is responsible for performing all initialization.
- *--------------------------------------------------------------------------+
- * (C) Copyright 1997 -
- * - NavIST Group - Real-Time Distributed Systems and Industrial Automation
- *
- * http://pandora.ist.utl.pt
- *
- * Instituto Superior Tecnico * Lisboa * PORTUGAL
- *--------------------------------------------------------------------------+
- * Disclaimer:
- *
- * This file is provided "AS IS" without warranty of any kind, either
- * expressed or implied.
- *--------------------------------------------------------------------------+
- */
-
-/*
- *  COPYRIGHT (c) 2011.
- *  On-Line Applications Research Corporation (OAR).
- *
- *  The license and distribution terms for this file may be
- *  found in the file LICENSE in this distribution or at
- *  http://www.rtems.org/license/LICENSE.
- */
-
-
-#include <bspopts.h>
-
-/*---------------------------------------------------------------------------+
-| Constants
-+----------------------------------------------------------------------------*/
-
-#if defined(SMP_SECONDARY_CORE)
-.set PROT_CODE_SEG, 0x08        # offset of code segment descriptor into GDT
-#else
-.set PROT_CODE_SEG, 0x0         # offset of code segment descriptor into GDT
-#endif
-
-.set PROT_DATA_SEG, 0x10        # offset of code segment descriptor into GDT
-.set CR0_PE,        1           # protected mode flag on CR0 register
-.set HDRSTART,      HEADERADDR  # address of start of bin2boot header
-.set HDROFF,        0x24        # offset into bin2boot header of start32 addr
-.set STACKOFF,      0x200-0x10  # offset to load into %esp, from start of image
-
-/* #define NEW_GAS */
-#ifdef NEW_GAS
-  #define LJMPL ljmpl
-#else
-  #define LJMPL ljmp
-#endif
-
-/*----------------------------------------------------------------------------+
-| CODE section
-+----------------------------------------------------------------------------*/
-
-.text
-#if defined(SMP_SECONDARY_CORE)
-        .globl app_processor_start                # entry point
-app_processor_start:
-#else
-        .globl _start16         # entry point
-        .globl start16
-start16:
-_start16:
-#endif
-
-.code16
-        cli                     # DISABLE INTERRUPTS!!!
-#if defined(SMP_SECONDARY_CORE)
-        jmp     1f
-	.align 4
-app_cpu_start:
-	.long   0
-app_cpu_stack:
-	.long   0
-1:
-#endif
-        movw    %cs, %ax                   # Initialize the rest of 
-        movw    %ax, %ds                   #   segment registers
-        movw    %ax, %es
-        movw    %ax, %ss
-
-#if !defined(SMP_SECONDARY_CODE) && (RTEMS_VIDEO_80x50 == 1)
-        movl    $0x0040,%eax    # use 32 bit constant to ensure 16 MSB=0
-        mov     %ax,%es
-        movw    %es:0x4a, %ax   # get 16 bit number of columns
-        cmpw    $0, %ax         # or 0 if no video adapter
-        je      1f              # if no video, skip touching it
-        /*---------------------------------------------------------------------+
-        | Switch VGA video to 80 lines x 50 columns mode. Has to be done before
-        | turning protected mode on since it uses BIOS int 10h (video) services.
-        +---------------------------------------------------------------------*/
-
-        movw    $0x0003, %ax    # forced set
-        int     $0x10
-        movw    $0x1112, %ax    # use 8x8 font
-        xorb    %bl, %bl
-        int     $0x10
-        movw    $0x1201, %ax    # turn off cursor emulation
-        movb    $0x34, %bl
-        int     $0x10
-        movb    $0x01, %ah      # define cursor (scan lines 0 to 7)
-        movw    $0x0007, %cx
-        int     $0x10
-1:
-#endif /* !SMP_SECONDARY_CODE and RTEMS_VIDEO_80x50 */
-
-        /*---------------------------------------------------------------------+
-        | Bare PC machines boot in real mode! We have to turn protected mode on.
-        +---------------------------------------------------------------------*/
-
-#if defined(SMP_SECONDARY_CORE)
-        lgdt    gdtptr - app_processor_start  # load Global Descriptor Table
-#else
-        lgdt    gdtptr - start16        # load Global Descriptor Table
-#endif /* SMP_SECONDARY_CORE */
-
-        movl    %cr0, %eax
-        orl     $CR0_PE, %eax
-        movl    %eax, %cr0              # turn on protected mode
-#if defined(SMP_SECONDARY_CORE)
-        LJMPL   $PROT_CODE_SEG, $2f     # flush prefetch queue, and reload %cs
-#else
-        LJMPL   $PROT_CODE_SEG, $2f     # flush prefetch queue, and reload %cs
-#endif
-.code32
-2:
-
-        /*---------------------------------------------------------------------+
-        | load the other segment registers
-        +---------------------------------------------------------------------*/
-        movl    $PROT_DATA_SEG, %eax
-        movw    %ax, %ds
-        movw    %ax, %es
-        movw    %ax, %ss
-#if defined(SMP_SECONDARY_CORE)
-        movl    app_cpu_stack, %esp            # stack pointer
-        movl    app_cpu_stack, %ebp            # base pointer
- #else
-        movl    $start16 + STACKOFF, %esp       # set up stack pointer
-        addl    $start16 + STACKOFF, %ebp       # set up stack pointer
-#endif  /* SMP_SECONDARY_CORE */
-
-        /*---------------------------------------------------------------------+
-        | we have to enable A20 in order to access memory above 1MByte
-        +---------------------------------------------------------------------*/
-        call    empty_8042
-        movb    $0xD1, %al                      # command write
-        outb    %al, $0x64
-        call    empty_8042
-        movb    $0xDF, %al                      # A20 on
-        outb    %al, $0x60
-        call    empty_8042
-
-        call    pc386_delay
-        call    pc386_delay
-        call    pc386_delay
-
-#if defined(SMP_SECONDARY_CORE)
-        movl    app_cpu_start, %eax             # jump to app CPU start
-#else
-        movl    %cs:HDRSTART + HDROFF, %eax     # jump to start of 32 bit code
-#endif  /* SMP_SECONDARY_CORE */
-        pushl   %eax                            
-        ret                                    
- 
-
-/*----------------------------------------------------------------------------+
-| pc386_delay
-+------------------------------------------------------------------------------
-| Delay is needed after doing I/O.
-|
-| The outb version is OK on most machines BUT the loop version ...
-|
-| will delay for 1us on 1Gz machine, it will take a little bit
-| longer on slower machines, however, it does not matter because we
-| are going to call this function only a few times
-!
-| NOTE: Saving the content of the EAX register just in case. - Rosimildo.
-+----------------------------------------------------------------------------*/
-        .p2align 4
-        .globl _pc386_delay
-        .globl pc386_delay
-pc386_delay:
-_pc386_delay:
-        pushl   %eax
-#if defined(USE_OUTB_FOR_DELAY)
-        outb    %al, $0x80       # about 1uS delay on most machines
-
-#else
-
-        movl    $0x200, %eax
-pc386_delay1:
-        dec     %eax
-        jnz     pc386_delay1
-#endif
-        popl    %eax
-        ret
-
-/*----------------------------------------------------------------------------+
-| empty_8042
-+------------------------------------------------------------------------------
-| This routine checks that the keyboard command queue is empty (after emptying
-| the output buffers).
-| No timeout is used - if this hangs there is something wrong with the machine,
-| and we probably couldn't proceed anyway.
-+----------------------------------------------------------------------------*/
-        .p2align 4
-        .globl _empty_8042
-        .globl empty_8042
-empty_8042:
-_empty_8042:
-        call    pc386_delay
-        inb     $0x64, %al      # 8042 status port
-        testb   $0x01, %al      # output buffer?
-        jz      no_output
-        call    pc386_delay
-        in      $0x60, %al      # read it
-        jmp     empty_8042
-no_output:
-        test    $0x02, %al      # is input buffer full?
-        jnz     empty_8042      # yes - loop
-        ret
-
-/*----------------------------------------------------------------------------+
-| DATA section
-+----------------------------------------------------------------------------*/
-
-/**************************
-* GLOBAL DESCRIPTOR TABLE *
-**************************/
-
-        .p2align 4
-gdtptr:
-        /* we use the NULL descriptor to store the GDT pointer - a trick quite
-           nifty due to: Robert Collins (rcollins@x86.org) */
-        .word   gdtlen - 1
-        .long   gdtptr
-        .word   0x0000
-
-        /* code segment */
-        .word   0xffff, 0
-        .byte   0, 0x9f, 0xcf, 0
-
-        /* data segment */
-        .word   0xffff, 0
-        .byte   0, 0x93, 0xcf, 0
-
-        .set    gdtlen, . - gdtptr      # length of GDT
diff --git a/bsps/i386/pc386/start/startAP.S b/bsps/i386/pc386/start/startAP.S
new file mode 100644
index 0000000000..0f81c03144
--- /dev/null
+++ b/bsps/i386/pc386/start/startAP.S
@@ -0,0 +1,145 @@
+/*--------------------------------------------------------------------------+
+ * start16.s v1.0 - PC386 BSP - 1998/04/13
+ * startAP.s 05/2019
+ *--------------------------------------------------------------------------+
+ * This file contains the initialization code for application processors (AP)
+ * for i386 based board support packages in SMP configuration.
+ * The APs start in 16 bit real mode. The goal is to:
+ * 1. Initialize the CPU registers
+ * 2. Load the global descriptor table
+ * 3. Switch to protected mode
+ * 4. Setup the stack pointers
+ * 5. Switch to the higher level initialization routine
+ *
+ *--------------------------------------------------------------------------+
+ * (C) Copyright 1997 -
+ * - NavIST Group - Real-Time Distributed Systems and Industrial Automation
+ *
+ * http://pandora.ist.utl.pt
+ *
+ * Instituto Superior Tecnico * Lisboa * PORTUGAL
+ *--------------------------------------------------------------------------+
+ * Disclaimer:
+ *
+ * This file is provided "AS IS" without warranty of any kind, either
+ * expressed or implied.
+ *--------------------------------------------------------------------------+
+ */
+
+/*
+ *  COPYRIGHT (c) 2011.
+ *  On-Line Applications Research Corporation (OAR).
+ *
+ *  The license and distribution terms for this file may be
+ *  found in the file LICENSE in this distribution or at
+ *  http://www.rtems.org/license/LICENSE.
+ */
+
+
+#include <bspopts.h>
+
+/*---------------------------------------------------------------------------+
+| Constants
++----------------------------------------------------------------------------*/
+
+.set PROT_CODE_SEG, 0x08        # offset of code segment descriptor into GDT
+.set PROT_DATA_SEG, 0x10        # offset of code segment descriptor into GDT
+.set CR0_PE,        1           # protected mode flag on CR0 register
+.set HDRSTART,      HEADERADDR  # address of start of bin2boot header
+.set HDROFF,        0x24        # offset into bin2boot header of start32 addr
+.set STACKOFF,      0x200-0x10  # offset to load into %esp, from start of image
+
+/* #define NEW_GAS */
+#ifdef NEW_GAS
+  #define LJMPL ljmpl
+#else
+  #define LJMPL ljmp
+#endif
+
+/*----------------------------------------------------------------------------+
+| CODE section
++----------------------------------------------------------------------------*/
+
+.text
+        .globl app_processor_start                # entry point
+app_processor_start:
+
+.code16
+        cli                     # DISABLE INTERRUPTS!!!
+        jmp     setup_processor
+/*
+ * Placeholder to copy information from boot_cpu()
+ * Do NOT move or add asm instruction before
+ */
+.align 4
+app_cpu_start:
+	.long   0
+app_cpu_stack:
+	.long   0
+
+setup_processor:
+        movw    %cs, %ax                   # Initialize the rest of
+        movw    %ax, %ds                   #   segment registers
+        movw    %ax, %es
+        movw    %ax, %ss
+
+        /*---------------------------------------------------------------------+
+        | Bare PC machines boot in real mode! We have to turn protected mode on.
+        +---------------------------------------------------------------------*/
+
+        lgdt    gdtptr - app_processor_start  # load Global Descriptor Table
+
+        movl    %cr0, %eax
+        orl     $CR0_PE, %eax
+        movl    %eax, %cr0              # turn on protected mode
+        LJMPL   $PROT_CODE_SEG, $start_32bit     # flush prefetch queue, and reload %cs
+
+.code32
+start_32bit:
+
+        /*---------------------------------------------------------------------+
+        | load the other segment registers
+        +---------------------------------------------------------------------*/
+        movl    $PROT_DATA_SEG, %eax
+        movw    %ax, %ds
+        movw    %ax, %es
+        movw    %ax, %ss
+        /* Prepare stack pointers */
+        movl    app_cpu_stack, %esp            # stack pointer
+        movl    app_cpu_stack, %ebp            # base pointer
+        movl    app_cpu_start, %eax             # jump to app CPU start
+        pushl   %eax
+        /* Clear stack pointer to signal that the we jump to the kernel */
+        movl    $0, app_cpu_stack
+        /* Switch to the higher level initialization routines */
+        ret
+
+/*----------------------------------------------------------------------------+
+| DATA section
++----------------------------------------------------------------------------*/
+
+/**************************
+* GLOBAL DESCRIPTOR TABLE *
+**************************/
+
+        .p2align 4
+gdtptr:
+        /* we use the NULL descriptor to store the GDT pointer - a trick quite
+           nifty due to: Robert Collins (rcollins@x86.org) */
+        .word   gdtlen - 1
+        .long   gdtptr
+        .word   0x0000
+
+        /* code segment */
+        .word   0xffff, 0
+        .byte   0, 0x9f, 0xcf, 0
+
+        /* data segment */
+        .word   0xffff, 0
+        .byte   0, 0x93, 0xcf, 0
+
+     	/* gs segment */
+        .word 0xffff, 0
+        .byte 0, 0x92, 0xcf, 0
+
+        .set    gdtlen, . - gdtptr      # length of GDT
diff --git a/c/src/lib/libbsp/i386/pc386/Makefile.am b/c/src/lib/libbsp/i386/pc386/Makefile.am
index 218e6bc065..c71f672063 100644
--- a/c/src/lib/libbsp/i386/pc386/Makefile.am
+++ b/c/src/lib/libbsp/i386/pc386/Makefile.am
@@ -114,8 +114,8 @@ librtemsbsp_a_SOURCES += ../../../../../../bsps/i386/pc386/start/getcpuid.c
 librtemsbsp_a_SOURCES += ../../../../../../bsps/i386/pc386/start/smp-imps.c
 
 project_lib_DATA += appstart.$(OBJEXT)
-appcpustart.$(OBJEXT): ../../../../../../bsps/i386/pc386/start/start16.S
-	$(CPPASCOMPILE) $(AM_CPPFLAGS) -DSMP_SECONDARY_CORE -o $@ -c $<
+appcpustart.$(OBJEXT): ../../../../../../bsps/i386/pc386/start/startAP.S
+	$(CPPASCOMPILE) $(AM_CPPFLAGS) -o $@ -c $<
 
 appstart.$(OBJEXT): appcpustart.$(OBJEXT)
 	$(LD) -N  \