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/compat/ndis/subr_hal.c

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  * Copyright (c) 2003
    3  *      Bill Paul <wpaul@windriver.com>.  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  * 3. All advertising materials mentioning features or use of this software
   14  *    must display the following acknowledgement:
   15  *      This product includes software developed by Bill Paul.
   16  * 4. Neither the name of the author nor the names of any co-contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
   24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
   25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
   26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
   27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
   28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
   29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
   30  * THE POSSIBILITY OF SUCH DAMAGE.
   31  */
   32 
   33 #include <sys/cdefs.h>
   34 __FBSDID("$FreeBSD$");
   35 
   36 #include <sys/param.h>
   37 #include <sys/types.h>
   38 #include <sys/errno.h>
   39 
   40 #include <sys/callout.h>
   41 #include <sys/kernel.h>
   42 #include <sys/lock.h>
   43 #include <sys/mutex.h>
   44 #include <sys/proc.h>
   45 #include <sys/sched.h>
   46 #include <sys/module.h>
   47 
   48 #include <sys/systm.h>
   49 #include <machine/bus.h>
   50 
   51 #include <sys/bus.h>
   52 #include <sys/rman.h>
   53 
   54 #include <compat/ndis/pe_var.h>
   55 #include <compat/ndis/resource_var.h>
   56 #include <compat/ndis/cfg_var.h>
   57 #include <compat/ndis/ntoskrnl_var.h>
   58 #include <compat/ndis/hal_var.h>
   59 
   60 static void KeStallExecutionProcessor(uint32_t);
   61 static void WRITE_PORT_BUFFER_ULONG(uint32_t *,
   62         uint32_t *, uint32_t);
   63 static void WRITE_PORT_BUFFER_USHORT(uint16_t *,
   64         uint16_t *, uint32_t);
   65 static void WRITE_PORT_BUFFER_UCHAR(uint8_t *,
   66         uint8_t *, uint32_t);
   67 static void WRITE_PORT_ULONG(uint32_t *, uint32_t);
   68 static void WRITE_PORT_USHORT(uint16_t *, uint16_t);
   69 static void WRITE_PORT_UCHAR(uint8_t *, uint8_t);
   70 static uint32_t READ_PORT_ULONG(uint32_t *);
   71 static uint16_t READ_PORT_USHORT(uint16_t *);
   72 static uint8_t READ_PORT_UCHAR(uint8_t *);
   73 static void READ_PORT_BUFFER_ULONG(uint32_t *,
   74         uint32_t *, uint32_t);
   75 static void READ_PORT_BUFFER_USHORT(uint16_t *,
   76         uint16_t *, uint32_t);
   77 static void READ_PORT_BUFFER_UCHAR(uint8_t *,
   78         uint8_t *, uint32_t);
   79 static uint64_t KeQueryPerformanceCounter(uint64_t *);
   80 static void _KeLowerIrql(uint8_t);
   81 static uint8_t KeRaiseIrqlToDpcLevel(void);
   82 static void dummy (void);
   83 
   84 #define NDIS_MAXCPUS 64
   85 static struct mtx disp_lock[NDIS_MAXCPUS];
   86 
   87 int
   88 hal_libinit()
   89 {
   90         image_patch_table       *patch;
   91         int                     i;
   92 
   93         for (i = 0; i < NDIS_MAXCPUS; i++)
   94                 mtx_init(&disp_lock[i], "HAL preemption lock",
   95                     "HAL lock", MTX_RECURSE|MTX_DEF);
   96 
   97         patch = hal_functbl;
   98         while (patch->ipt_func != NULL) {
   99                 windrv_wrap((funcptr)patch->ipt_func,
  100                     (funcptr *)&patch->ipt_wrap,
  101                     patch->ipt_argcnt, patch->ipt_ftype);
  102                 patch++;
  103         }
  104 
  105         
  106         return(0);
  107 }
  108 
  109 int
  110 hal_libfini()
  111 {
  112         image_patch_table       *patch;
  113         int                     i;
  114 
  115         for (i = 0; i < NDIS_MAXCPUS; i++)
  116                 mtx_destroy(&disp_lock[i]);
  117 
  118         patch = hal_functbl;
  119         while (patch->ipt_func != NULL) {
  120                 windrv_unwrap(patch->ipt_wrap);
  121                 patch++;
  122         }
  123 
  124         return(0);
  125 }
  126 
  127 static void
  128 KeStallExecutionProcessor(usecs)
  129         uint32_t                usecs;
  130 {
  131         DELAY(usecs);
  132         return;
  133 }
  134 
  135 static void
  136 WRITE_PORT_ULONG(port, val)
  137         uint32_t                *port;
  138         uint32_t                val;
  139 {
  140         bus_space_write_4(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
  141         return;
  142 }
  143 
  144 static void
  145 WRITE_PORT_USHORT(uint16_t *port, uint16_t val)
  146 {
  147         bus_space_write_2(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
  148         return;
  149 }
  150 
  151 static void
  152 WRITE_PORT_UCHAR(uint8_t *port, uint8_t val)
  153 {
  154         bus_space_write_1(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
  155         return;
  156 }
  157 
  158 static void
  159 WRITE_PORT_BUFFER_ULONG(port, val, cnt)
  160         uint32_t                *port;
  161         uint32_t                *val;
  162         uint32_t                cnt;
  163 {
  164         bus_space_write_multi_4(NDIS_BUS_SPACE_IO, 0x0,
  165             (bus_size_t)port, val, cnt);
  166         return;
  167 }
  168 
  169 static void
  170 WRITE_PORT_BUFFER_USHORT(port, val, cnt)
  171         uint16_t                *port;
  172         uint16_t                *val;
  173         uint32_t                cnt;
  174 {
  175         bus_space_write_multi_2(NDIS_BUS_SPACE_IO, 0x0,
  176             (bus_size_t)port, val, cnt);
  177         return;
  178 }
  179 
  180 static void
  181 WRITE_PORT_BUFFER_UCHAR(port, val, cnt)
  182         uint8_t                 *port;
  183         uint8_t                 *val;
  184         uint32_t                cnt;
  185 {
  186         bus_space_write_multi_1(NDIS_BUS_SPACE_IO, 0x0,
  187             (bus_size_t)port, val, cnt);
  188         return;
  189 }
  190 
  191 static uint16_t
  192 READ_PORT_USHORT(port)
  193         uint16_t                *port;
  194 {
  195         return(bus_space_read_2(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
  196 }
  197 
  198 static uint32_t
  199 READ_PORT_ULONG(port)
  200         uint32_t                *port;
  201 {
  202         return(bus_space_read_4(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
  203 }
  204 
  205 static uint8_t
  206 READ_PORT_UCHAR(port)
  207         uint8_t                 *port;
  208 {
  209         return(bus_space_read_1(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
  210 }
  211 
  212 static void
  213 READ_PORT_BUFFER_ULONG(port, val, cnt)
  214         uint32_t                *port;
  215         uint32_t                *val;
  216         uint32_t                cnt;
  217 {
  218         bus_space_read_multi_4(NDIS_BUS_SPACE_IO, 0x0,
  219             (bus_size_t)port, val, cnt);
  220         return;
  221 }
  222 
  223 static void
  224 READ_PORT_BUFFER_USHORT(port, val, cnt)
  225         uint16_t                *port;
  226         uint16_t                *val;
  227         uint32_t                cnt;
  228 {
  229         bus_space_read_multi_2(NDIS_BUS_SPACE_IO, 0x0,
  230             (bus_size_t)port, val, cnt);
  231         return;
  232 }
  233 
  234 static void
  235 READ_PORT_BUFFER_UCHAR(port, val, cnt)
  236         uint8_t                 *port;
  237         uint8_t                 *val;
  238         uint32_t                cnt;
  239 {
  240         bus_space_read_multi_1(NDIS_BUS_SPACE_IO, 0x0,
  241             (bus_size_t)port, val, cnt);
  242         return;
  243 }
  244 
  245 /*
  246  * The spinlock implementation in Windows differs from that of FreeBSD.
  247  * The basic operation of spinlocks involves two steps: 1) spin in a
  248  * tight loop while trying to acquire a lock, 2) after obtaining the
  249  * lock, disable preemption. (Note that on uniprocessor systems, you're
  250  * allowed to skip the first step and just lock out pre-emption, since
  251  * it's not possible for you to be in contention with another running
  252  * thread.) Later, you release the lock then re-enable preemption.
  253  * The difference between Windows and FreeBSD lies in how preemption
  254  * is disabled. In FreeBSD, it's done using critical_enter(), which on
  255  * the x86 arch translates to a cli instruction. This masks off all
  256  * interrupts, and effectively stops the scheduler from ever running
  257  * so _nothing_ can execute except the current thread. In Windows,
  258  * preemption is disabled by raising the processor IRQL to DISPATCH_LEVEL.
  259  * This stops other threads from running, but does _not_ block device
  260  * interrupts. This means ISRs can still run, and they can make other
  261  * threads runable, but those other threads won't be able to execute
  262  * until the current thread lowers the IRQL to something less than
  263  * DISPATCH_LEVEL.
  264  *
  265  * There's another commonly used IRQL in Windows, which is APC_LEVEL.
  266  * An APC is an Asynchronous Procedure Call, which differs from a DPC
  267  * (Defered Procedure Call) in that a DPC is queued up to run in
  268  * another thread, while an APC runs in the thread that scheduled
  269  * it (similar to a signal handler in a UNIX process). We don't
  270  * actually support the notion of APCs in FreeBSD, so for now, the
  271  * only IRQLs we're interested in are DISPATCH_LEVEL and PASSIVE_LEVEL.
  272  *
  273  * To simulate DISPATCH_LEVEL, we raise the current thread's priority
  274  * to PI_REALTIME, which is the highest we can give it. This should,
  275  * if I understand things correctly, prevent anything except for an
  276  * interrupt thread from preempting us. PASSIVE_LEVEL is basically
  277  * everything else.
  278  *
  279  * Be aware that, at least on the x86 arch, the Windows spinlock
  280  * functions are divided up in peculiar ways. The actual spinlock
  281  * functions are KfAcquireSpinLock() and KfReleaseSpinLock(), and
  282  * they live in HAL.dll. Meanwhile, KeInitializeSpinLock(),
  283  * KefAcquireSpinLockAtDpcLevel() and KefReleaseSpinLockFromDpcLevel()
  284  * live in ntoskrnl.exe. Most Windows source code will call
  285  * KeAcquireSpinLock() and KeReleaseSpinLock(), but these are just
  286  * macros that call KfAcquireSpinLock() and KfReleaseSpinLock().
  287  * KefAcquireSpinLockAtDpcLevel() and KefReleaseSpinLockFromDpcLevel()
  288  * perform the lock aquisition/release functions without doing the
  289  * IRQL manipulation, and are used when one is already running at
  290  * DISPATCH_LEVEL. Make sense? Good.
  291  *
  292  * According to the Microsoft documentation, any thread that calls
  293  * KeAcquireSpinLock() must be running at IRQL <= DISPATCH_LEVEL. If
  294  * we detect someone trying to acquire a spinlock from DEVICE_LEVEL
  295  * or HIGH_LEVEL, we panic.
  296  *
  297  * Alternate sleep-lock-based spinlock implementation
  298  * --------------------------------------------------
  299  *
  300  * The earlier spinlock implementation was arguably a bit of a hack
  301  * and presented several problems. It was basically designed to provide
  302  * the functionality of spinlocks without incurring the wrath of
  303  * WITNESS. We could get away with using both our spinlock implementation
  304  * and FreeBSD sleep locks at the same time, but if WITNESS knew what
  305  * we were really up to, it would have spanked us rather severely.
  306  *
  307  * There's another method we can use based entirely on sleep locks.
  308  * First, it's important to realize that everything we're locking
  309  * resides inside Project Evil itself: any critical data being locked
  310  * by drivers belongs to the drivers, and should not be referenced
  311  * by any other OS code outside of the NDISulator. The priority-based
  312  * locking scheme has system-wide effects, just like real spinlocks
  313  * (blocking preemption affects the whole CPU), but since we keep all
  314  * our critical data private, we can use a simpler mechanism that
  315  * affects only code/threads directly related to Project Evil.
  316  *
  317  * The idea is to create a sleep lock mutex for each CPU in the system.
  318  * When a CPU running in the NDISulator wants to acquire a spinlock, it
  319  * does the following:
  320  * - Pin ourselves to the current CPU
  321  * - Acquire the mutex for the current CPU
  322  * - Spin on the spinlock variable using atomic test and set, just like
  323  *   a real spinlock.
  324  * - Once we have the lock, we execute our critical code
  325  *
  326  * To give up the lock, we do:
  327  * - Clear the spinlock variable with an atomic op
  328  * - Release the per-CPU mutex
  329  * - Unpin ourselves from the current CPU.
  330  *
  331  * On a uniprocessor system, this means all threads that access protected
  332  * data are serialized through the per-CPU mutex. After one thread
  333  * acquires the 'spinlock,' any other thread that uses a spinlock on the
  334  * current CPU will block on the per-CPU mutex, which has the same general
  335  * effect of blocking pre-emption, but _only_ for those threads that are
  336  * running NDISulator code.
  337  *
  338  * On a multiprocessor system, threads on different CPUs all block on
  339  * their respective per-CPU mutex, and the atomic test/set operation
  340  * on the spinlock variable provides inter-CPU synchronization, though
  341  * only for threads running NDISulator code.
  342  *
  343  * This method solves an important problem. In Windows, you're allowed
  344  * to do an ExAllocatePoolWithTag() with a spinlock held, provided you
  345  * allocate from NonPagedPool. This implies an atomic heap allocation
  346  * that will not cause the current thread to sleep. (You can't sleep
  347  * while holding real spinlock: clowns will eat you.) But in FreeBSD,
  348  * malloc(9) _always_ triggers the acquisition of a sleep lock, even
  349  * when you use M_NOWAIT. This is not a problem for FreeBSD native
  350  * code: you're allowed to sleep in things like interrupt threads. But
  351  * it is a problem with the old priority-based spinlock implementation:
  352  * even though we get away with it most of the time, we really can't
  353  * do a malloc(9) after doing a KeAcquireSpinLock() or KeRaiseIrql().
  354  * With the new implementation, it's not a problem: you're allowed to
  355  * acquire more than one sleep lock (as long as you avoid lock order
  356  * reversals).
  357  *
  358  * The one drawback to this approach is that now we have a lot of
  359  * contention on one per-CPU mutex within the NDISulator code. Whether
  360  * or not this is preferable to the expected Windows spinlock behavior
  361  * of blocking pre-emption is debatable.
  362  */
  363 
  364 uint8_t
  365 KfAcquireSpinLock(lock)
  366         kspin_lock              *lock;
  367 {
  368         uint8_t                 oldirql;
  369 
  370         KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
  371         KeAcquireSpinLockAtDpcLevel(lock);
  372 
  373         return(oldirql);
  374 }
  375 
  376 void
  377 KfReleaseSpinLock(kspin_lock *lock, uint8_t newirql)
  378 {
  379         KeReleaseSpinLockFromDpcLevel(lock);
  380         KeLowerIrql(newirql);
  381 
  382         return;
  383 }
  384 
  385 uint8_t
  386 KeGetCurrentIrql()
  387 {
  388         if (mtx_owned(&disp_lock[curthread->td_oncpu]))
  389                 return(DISPATCH_LEVEL);
  390         return(PASSIVE_LEVEL);
  391 }
  392 
  393 static uint64_t
  394 KeQueryPerformanceCounter(freq)
  395         uint64_t                *freq;
  396 {
  397         if (freq != NULL)
  398                 *freq = hz;
  399 
  400         return((uint64_t)ticks);
  401 }
  402 
  403 uint8_t
  404 KfRaiseIrql(uint8_t irql)
  405 {
  406         uint8_t                 oldirql;
  407 
  408         oldirql = KeGetCurrentIrql();
  409 
  410         /* I am so going to hell for this. */
  411         if (oldirql > irql)
  412                 panic("IRQL_NOT_LESS_THAN");
  413 
  414         if (oldirql != DISPATCH_LEVEL) {
  415                 sched_pin();
  416                 mtx_lock(&disp_lock[curthread->td_oncpu]);
  417         }
  418 /*printf("RAISE IRQL: %d %d\n", irql, oldirql);*/
  419 
  420         return(oldirql);
  421 }
  422 
  423 void 
  424 KfLowerIrql(uint8_t oldirql)
  425 {
  426         if (oldirql == DISPATCH_LEVEL)
  427                 return;
  428 
  429         if (KeGetCurrentIrql() != DISPATCH_LEVEL)
  430                 panic("IRQL_NOT_GREATER_THAN");
  431 
  432         mtx_unlock(&disp_lock[curthread->td_oncpu]);
  433         sched_unpin();
  434 
  435         return;
  436 }
  437 
  438 static uint8_t
  439 KeRaiseIrqlToDpcLevel(void)
  440 {
  441         uint8_t                 irql;
  442 
  443         KeRaiseIrql(DISPATCH_LEVEL, &irql);
  444         return(irql);
  445 }
  446 
  447 static void
  448 _KeLowerIrql(uint8_t oldirql)
  449 {
  450         KeLowerIrql(oldirql);
  451         return;
  452 }
  453 
  454 static void dummy()
  455 {
  456         printf ("hal dummy called...\n");
  457         return;
  458 }
  459 
  460 image_patch_table hal_functbl[] = {
  461         IMPORT_SFUNC(KeStallExecutionProcessor, 1),
  462         IMPORT_SFUNC(WRITE_PORT_ULONG, 2),
  463         IMPORT_SFUNC(WRITE_PORT_USHORT, 2),
  464         IMPORT_SFUNC(WRITE_PORT_UCHAR, 2),
  465         IMPORT_SFUNC(WRITE_PORT_BUFFER_ULONG, 3),
  466         IMPORT_SFUNC(WRITE_PORT_BUFFER_USHORT, 3),
  467         IMPORT_SFUNC(WRITE_PORT_BUFFER_UCHAR, 3),
  468         IMPORT_SFUNC(READ_PORT_ULONG, 1),
  469         IMPORT_SFUNC(READ_PORT_USHORT, 1),
  470         IMPORT_SFUNC(READ_PORT_UCHAR, 1),
  471         IMPORT_SFUNC(READ_PORT_BUFFER_ULONG, 3),
  472         IMPORT_SFUNC(READ_PORT_BUFFER_USHORT, 3),
  473         IMPORT_SFUNC(READ_PORT_BUFFER_UCHAR, 3),
  474         IMPORT_FFUNC(KfAcquireSpinLock, 1),
  475         IMPORT_FFUNC(KfReleaseSpinLock, 1),
  476         IMPORT_SFUNC(KeGetCurrentIrql, 0),
  477         IMPORT_SFUNC(KeQueryPerformanceCounter, 1),
  478         IMPORT_FFUNC(KfLowerIrql, 1),
  479         IMPORT_FFUNC(KfRaiseIrql, 1),
  480         IMPORT_SFUNC(KeRaiseIrqlToDpcLevel, 0),
  481 #undef KeLowerIrql
  482         IMPORT_SFUNC_MAP(KeLowerIrql, _KeLowerIrql, 1),
  483 
  484         /*
  485          * This last entry is a catch-all for any function we haven't
  486          * implemented yet. The PE import list patching routine will
  487          * use it for any function that doesn't have an explicit match
  488          * in this table.
  489          */
  490 
  491         { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
  492 
  493         /* End of list. */
  494 
  495         { NULL, NULL, NULL }
  496 };

Cache object: 94b9168132fab3748145713806b9f09b


[ 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.