The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/i386/isa/vector.s

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  *      from: vector.s, 386BSD 0.1 unknown origin
    3  * $FreeBSD: releng/5.1/sys/i386/isa/vector.s 88322 2001-12-20 23:48:31Z jhb $
    4  */
    5 
    6 /*
    7  * modified for PC98 by Kakefuda
    8  */
    9 
   10 #include "opt_auto_eoi.h"
   11 
   12 #include <i386/isa/icu.h>
   13 #ifdef PC98
   14 #include <pc98/pc98/pc98.h>
   15 #else
   16 #include <i386/isa/isa.h>
   17 #endif
   18 #include <i386/isa/intr_machdep.h>
   19 
   20 #define FAST_INTR_HANDLER_USES_ES 1
   21 #ifdef FAST_INTR_HANDLER_USES_ES
   22 #define ACTUALLY_PUSHED         1
   23 #define MAYBE_MOVW_AX_ES        movw    %ax,%es
   24 #define MAYBE_POPL_ES           popl    %es
   25 #define MAYBE_PUSHL_ES          pushl   %es
   26 #else
   27 /*
   28  * We can usually skip loading %es for fastintr handlers.  %es should
   29  * only be used for string instructions, and fastintr handlers shouldn't
   30  * do anything slow enough to justify using a string instruction.
   31  */
   32 #define ACTUALLY_PUSHED         0
   33 #define MAYBE_MOVW_AX_ES
   34 #define MAYBE_POPL_ES
   35 #define MAYBE_PUSHL_ES
   36 #endif
   37 
   38         .data
   39         ALIGN_DATA
   40 
   41 /*
   42  * Interrupt counters and names for export to vmstat(8) and friends.
   43  *
   44  * XXX this doesn't really belong here; everything except the labels
   45  * for the endpointers is almost machine-independent.
   46  */
   47 
   48         .globl  intrcnt, eintrcnt
   49 intrcnt:
   50         .space  INTRCNT_COUNT * 4
   51 eintrcnt:
   52 
   53         .globl  intrnames, eintrnames
   54 intrnames:
   55         .space  INTRCNT_COUNT * 16
   56 eintrnames:
   57         .text
   58 
   59 /*
   60  * Macros for interrupt interrupt entry, call to handler, and exit.
   61  *
   62  * XXX - the interrupt frame is set up to look like a trap frame.  This is
   63  * usually a waste of time.  The only interrupt handlers that want a frame
   64  * are the clock handler (it wants a clock frame), the npx handler (it's
   65  * easier to do right all in assembler).  The interrupt return routine
   66  * needs a trap frame for rare AST's (it could easily convert the frame).
   67  * The direct costs of setting up a trap frame are two pushl's (error
   68  * code and trap number), an addl to get rid of these, and pushing and
   69  * popping the call-saved regs %esi, %edi and %ebp twice,  The indirect
   70  * costs are making the driver interface nonuniform so unpending of
   71  * interrupts is more complicated and slower (call_driver(unit) would
   72  * be easier than ensuring an interrupt frame for all handlers.  Finally,
   73  * there are some struct copies in the npx handler and maybe in the clock
   74  * handler that could be avoided by working more with pointers to frames
   75  * instead of frames.
   76  *
   77  * XXX - should we do a cld on every system entry to avoid the requirement
   78  * for scattered cld's?
   79  *
   80  * Coding notes for *.s:
   81  *
   82  * If possible, avoid operations that involve an operand size override.
   83  * Word-sized operations might be smaller, but the operand size override
   84  * makes them slower on on 486's and no faster on 386's unless perhaps
   85  * the instruction pipeline is depleted.  E.g.,
   86  *
   87  *      Use movl to seg regs instead of the equivalent but more descriptive
   88  *      movw - gas generates an irelevant (slower) operand size override.
   89  *
   90  *      Use movl to ordinary regs in preference to movw and especially
   91  *      in preference to movz[bw]l.  Use unsigned (long) variables with the
   92  *      top bits clear instead of unsigned short variables to provide more
   93  *      opportunities for movl.
   94  *
   95  * If possible, use byte-sized operations.  They are smaller and no slower.
   96  *
   97  * Use (%reg) instead of 0(%reg) - gas generates larger code for the latter.
   98  *
   99  * If the interrupt frame is made more flexible,  INTR can push %eax first
  100  * and decide the ipending case with less overhead, e.g., by avoiding
  101  * loading segregs.
  102  */
  103 
  104 #ifdef APIC_IO
  105 #include "i386/isa/apic_vector.s"
  106 #else
  107 #include "i386/isa/icu_vector.s"
  108 #endif  /* APIC_IO */

Cache object: 7a24ecb6d416eb95d2b738ec06ab634d


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.