diff options
Diffstat (limited to 'gdb/osf-share/HP800')
-rw-r--r-- | gdb/osf-share/HP800/.Sanitize | 44 | ||||
-rw-r--r-- | gdb/osf-share/HP800/cma_thread_io.h | 432 |
2 files changed, 0 insertions, 476 deletions
diff --git a/gdb/osf-share/HP800/.Sanitize b/gdb/osf-share/HP800/.Sanitize deleted file mode 100644 index 17e1389..0000000 --- a/gdb/osf-share/HP800/.Sanitize +++ /dev/null @@ -1,44 +0,0 @@ -# .Sanitize for devo/gdb/osf-share/HP800. - -# Each directory to survive its way into a release will need a file -# like this one called "./.Sanitize". All keyword lines must exist, -# and must exist in the order specified by this file. Each directory -# in the tree will be processed, top down, in the following order. - -# Hash started lines like this one are comments and will be deleted -# before anything else is done. Blank lines will also be squashed -# out. - -# The lines between the "Do-first:" line and the "Things-to-keep:" -# line are executed as a /bin/sh shell script before anything else is -# done in this directory. - -Do-first: - -# All files listed between the "Things-to-keep:" line and the -# "Files-to-sed:" line will be kept. All other files will be removed. -# Directories listed in this section will have their own Sanitize -# called. Directories not listed will be removed in their entirety -# with rm -rf. - -Things-to-keep: - -cma_thread_io.h - -# Things which are explicitly *not* kept, for now. - -Things-to-lose: - -Do-last: - -# Don't try to clean directories here, as the 'mv' command will fail. -# Also, grep fails on NFS mounted directories. - -for i in * ; do - if test ! -d $i && (grep sanitize $i > /dev/null) ; then - echo '***' Some mentions of Sanitize are still left in $i! 1>&2 - fi -done - -# -# End of file. diff --git a/gdb/osf-share/HP800/cma_thread_io.h b/gdb/osf-share/HP800/cma_thread_io.h deleted file mode 100644 index e9bef3b..0000000 --- a/gdb/osf-share/HP800/cma_thread_io.h +++ /dev/null @@ -1,432 +0,0 @@ -/* - * (c) Copyright 1990-1996 OPEN SOFTWARE FOUNDATION, INC. - * (c) Copyright 1990-1996 HEWLETT-PACKARD COMPANY - * (c) Copyright 1990-1996 DIGITAL EQUIPMENT CORPORATION - * (c) Copyright 1991, 1992 Siemens-Nixdorf Information Systems - * To anyone who acknowledges that this file is provided "AS IS" without - * any express or implied warranty: permission to use, copy, modify, and - * distribute this file for any purpose is hereby granted without fee, - * provided that the above copyright notices and this notice appears in - * all source code copies, and that none of the names listed above be used - * in advertising or publicity pertaining to distribution of the software - * without specific, written prior permission. None of these organizations - * makes any representations about the suitability of this software for - * any purpose. - */ -/* - * - * Header file for thread synchrounous I/O - */ - -#ifndef CMA_THREAD_IO -#define CMA_THREAD_IO - -/* - * INCLUDE FILES - */ - -#include <cma_config.h> -#include <sys/file.h> -#include <cma.h> -#include <sys/types.h> -#include <sys/time.h> -#include <cma_init.h> -#include <cma_errors.h> - -/* - * CONSTANTS - */ - - - -/* - * Maximum number of files (ie, max_fd+1) - */ -#define cma__c_mx_file FD_SETSIZE - -/* - * Number of bits per file descriptor bit mask (ie number of bytes * bits/byte) - */ -#define cma__c_nbpm NFDBITS - -/* - * TYPE DEFINITIONS - */ - -typedef enum CMA__T_IO_TYPE { - cma__c_io_read = 0, - cma__c_io_write = 1, - cma__c_io_except = 2 - } cma__t_io_type; -#define cma__c_max_io_type 2 - -/* - * From our local <sys/types.h>: - * - * typedef long fd_mask; - * - * typedef struct fd_set { - * fd_mask fds_bits[howmany(FD_SETSIZE, NFDBITS)]; - * } fd_set; - * - */ -typedef fd_mask cma__t_mask; -typedef fd_set cma__t_file_mask; - - - -/* - * GLOBAL DATA - */ - -/* - * Maximum number of files (ie, max_fd+1) as determined by getdtablesize(). - */ -extern int cma__g_mx_file; - -/* - * Number of submasks (ie "int" sized chunks) per file descriptor mask as - * determined by getdtablesize(). - */ -extern int cma__g_nspm; - -/* - * MACROS - */ - -/* - * Define a constant for the errno value which indicates that the requested - * operation was not performed because it would block the process. - */ -# define cma__is_blocking(s) \ - ((s == EAGAIN) || (s == EWOULDBLOCK) || (s == EINPROGRESS) || \ - (s == EALREADY) || (s == EDEADLK)) - -/* -* It is necessary to issue an I/O function, before calling cma__io_wait() -* in the following cases: -* -* * This file descriptor has been set non-blocking by CMA -* * This file descriptor has been set non-blocking by the user. -*/ - -#define cma__issue_io_call(fd) \ - ( (cma__g_file[fd]->non_blocking) || \ - (cma__g_file[fd]->user_fl.user_non_blocking) ) - - -#define cma__set_user_nonblocking(flags) \ - -/* - * Determine if the file is open - */ -/* - * If the file gets closed while waiting for the mutex cma__g_file[rfd] - * gets set to null. This results in a crash if NDEBUG is set to 0 - * since cma__int_lock tries to dereference it to set the mutex ownership - * after it gets the mutex. The following will still set the ownership - * in cma__int_lock so we'll set it back to noone if cma__g_file is null - * when we come back just in case it matters. It shouldn't since its no - * longer in use but..... - * Callers of this should recheck cma__g_file after the reservation to - * make sure continueing makes sense. - */ -#define cma__fd_reserve(rfd) \ - { \ - cma__t_int_mutex *__mutex__; \ - __mutex__ = cma__g_file[rfd]->mutex; \ - cma__int_lock (__mutex__); \ - if(cma__g_file[rfd] == (cma__t_file_obj *)cma_c_null_ptr) \ - cma__int_unlock(__mutex__); \ - } - - -/* - * Unreserve a file descriptor - */ -#define cma__fd_unreserve(ufd) cma__int_unlock (cma__g_file[ufd]->mutex) - -/* - * AND together two select file descriptor masks - */ -#define cma__fdm_and(target,a,b) \ - { \ - int __i__ = cma__g_nspm; \ - while (__i__--) \ - (target)->fds_bits[__i__] = \ - (a)->fds_bits[__i__] & (b)->fds_bits[__i__]; \ - } - -/* - * Clear a bit in a select file descriptor mask - * - * FD_CLR(n, p) := ((p)->fds_bits[(n)/NFDBITS] &= ~(1 << ((n) % NFDBITS))) - */ -#define cma__fdm_clr_bit(n,p) FD_CLR (n, p) - -/* - * Copy the contents of one file descriptor mask into another. If the - * destination operand is null, do nothing; if the source operand is null, - * simply zero the destination. - */ -#define cma__fdm_copy(src,dst,nfds) { \ - if (dst) \ - if (src) { \ - cma__t_mask *__s__ = (cma__t_mask *)(src); \ - cma__t_mask *__d__ = (cma__t_mask *)(dst); \ - int __i__; \ - for (__i__ = 0; __i__ < (nfds); __i__ += cma__c_nbpm) \ - *__d__++ = *__s__++; \ - } \ - else \ - cma__fdm_zero (dst); \ - } - -/* - * To increment count for each bit set in fd - mask - */ -#define cma__fdm_count_bits(map,count) \ - { \ - int __i__ = cma__g_nspm; \ - while (__i__--) { \ - cma__t_mask __tm__; \ - __tm__ = (map)->fds_bits[__i__]; \ - while(__tm__) { \ - (count)++; \ - __tm__ &= ~(__tm__ & (-__tm__)); /* Assumes 2's comp */ \ - } \ - } \ - } - -/* - * Test if a bit is set in a select file descriptor mask - * - * FD_ISSET(n,p) := ((p)->fds_bits[(n)/NFDBITS] & (1 << ((n) % NFDBITS))) - */ -#define cma__fdm_is_set(n,p) FD_ISSET (n, p) - -/* - * OR together two select file descriptor masks - */ -#define cma__fdm_or(target,a,b) \ - { \ - int __i__ = cma__g_nspm; \ - while (__i__--) \ - (target)->fds_bits[__i__] = \ - (a)->fds_bits[__i__] | (b)->fds_bits[__i__]; \ - } - -/* - * Set a bit in a select file descriptor mask - * - * FD_SET(n,p) := ((p)->fds_bits[(n)/NFDBITS] |= (1 << ((n) % NFDBITS))) - */ -#define cma__fdm_set_bit(n,p) FD_SET (n, p) - -/* - * Clear a select file descriptor mask. - */ -#define cma__fdm_zero(n) \ - cma__memset ((char *) n, 0, cma__g_nspm * sizeof(cma__t_mask)) - - - - - -/* - * CMA "thread-synchronous" I/O read/write operations - */ - - /* - * Since all CMA "thread-synchronous" I/O (read or write) operations on - * U*ix follow the exact same structure, the wrapper routines have been - * condensed into a macro. - * - * The steps performed are as follows: - * 1. Check that the file descriptor is a legitimate value. - * 2. Check that the entry in the CMA file "database" which corresponds to - * the file descriptor indicates that the "file" was "opened" by CMA. - * 3. Reserve the file, to serialized access to files. This not only - * simplifies things, but also defends against non-reentrancy. - * 4. If the "file" is "set" for non-blocking I/O, check if we - * have actually set the file non-blocking yet, and if not do so. - * Then, issue the I/O operantion. - * Success or failure is returned immediately, after unreserving the - * file. If the error indicates that the operation would have caused - * the process to block, continue to the next step. - * 5. The I/O prolog adds this "file" to the global bit mask, which - * represents all "files" which have threads waiting to perform I/O on - * them, and causes the thread to block on the condition variable for - * this "file". Periodically, a select is done on this global bit - * mask, and the condition variables corresponding to "files" which - * are ready for I/O are signaled, releasing those waiting threads to - * perform their I/O. - * 6. When the thread returns from the I/O prolog, it can (hopefully) - * perform its operation without blocking the process. - * 7. The I/O epilog clears the bit in the global mask and/or signals the - * the next thread waiting for this "file", as appropriate. - * 8. If the I/O failed, continue to loop. - * 9. Finally, the "file" is unreserved, as we're done with it, and the - * result of the operation is returned. - * - * - * Note: currently, we believe that timeslicing which is based on the - * virtual-time timer does not cause system calls to return EINTR. - * Threfore, any EINTR returns are relayed directly to the caller. - * On platforms which do not support a virtual-time timer, the code - * should probably catch EINTR returns and restart the system call. - */ - -/* - * This macro is used for both read-type and write-type functions. - * - * Note: the second call to "func" may require being bracketed in a - * cma__interrupt_disable/cma__interrupt_enable pair, but we'll - * wait and see if this is necessary. - */ -#define cma__ts_func(func,fd,arglist,type,post_process) { \ - cma_t_integer __res__; \ - cma_t_boolean __done__ = cma_c_false; \ - if ((fd < 0) || (fd >= cma__g_mx_file)) return (cma__set_errno (EBADF), -1); \ - if (!cma__is_open(fd)) return (cma__set_errno (EBADF), -1); \ - cma__fd_reserve (fd); \ - if (!cma__is_open(fd)) return (cma__set_errno (EBADF), -1); \ - if (cma__issue_io_call(fd)) {\ - if ((!cma__g_file[fd]->set_non_blocking) && \ - (cma__g_file[fd]->non_blocking == cma_c_true)) \ - cma__set_nonblocking(fd); \ - cma__interrupt_disable (0); \ - TRY { \ - __res__ = func arglist; \ - } \ - CATCH_ALL { \ - cma__interrupt_enable (0); \ - cma__fd_unreserve (fd); \ - RERAISE; \ - } \ - ENDTRY \ - cma__interrupt_enable (0); \ - if ((__res__ != -1) \ - || (!cma__is_blocking (errno)) \ - || (cma__g_file[fd]->user_fl.user_non_blocking)) \ - __done__ = cma_c_true; \ - } \ - if (__done__) { \ - cma__fd_unreserve (fd); \ - } \ - else { \ - TRY { \ - cma__io_prolog (type, fd); \ - while (!__done__) { \ - cma__io_wait (type, fd); \ - __res__ = func arglist; \ - if ((__res__ != -1) \ - || (!cma__is_blocking (errno)) \ - || (cma__g_file[fd]->user_fl.user_non_blocking)) \ - __done__ = cma_c_true; \ - } \ - } \ - FINALLY { \ - cma__io_epilog (type, fd); \ - cma__fd_unreserve (fd); \ - } \ - ENDTRY \ - } \ - if (__res__ != -1) post_process; \ - return __res__; \ - } - - /* - * Since most CMA "thread-synchronous" I/O ("open"-type) operations on - * U*ix follow the exact same structure, the wrapper routines have been - * condensed into a macro. - * - * The steps performed are as follows: - * 1. Issue the open function. - * 2. If the value returned indicates an error, return it to the caller. - * 3. If the file descriptor returned is larger than what we think is the - * maximum value (ie if it is too big for our database) then bugcheck. - * 4. "Open" the "file" in the CMA file database. - * 5. Return the file descriptor value to the caller. - * - * FIX-ME: for the time being, if the I/O operation returns EINTR, we - * simply return it to the caller; eventually, we should catch this - * and "do the right thing" (if we can figure out what that is). - */ - -/* - * This macro is used for all "open"-type functions which return a single file - * desciptor by immediate value. - */ -#define cma__ts_open(func,arglist,post_process) { \ - int __fd__; \ - TRY { \ - cma__int_init (); \ - cma__int_lock (cma__g_io_data_mutex); \ - __fd__ = func arglist; \ - cma__int_unlock (cma__g_io_data_mutex); \ - if (__fd__ >= 0 && __fd__ < cma__g_mx_file) \ - post_process; \ - } \ - CATCH_ALL \ - { \ - cma__set_errno (EBADF); \ - __fd__ = -1; \ - } \ - ENDTRY \ - if (__fd__ >= cma__g_mx_file) \ - cma__bugcheck ("cma__ts_open: fd is too large"); \ - return __fd__; \ - } -/* - * This macro is used for all "open"-type functions which return a pair of file - * desciptors by reference parameter. - */ -#define cma__ts_open2(func,fdpair,arglist,post_process) { \ - int __res__; \ - TRY { \ - cma__int_init (); \ - cma__int_lock (cma__g_io_data_mutex); \ - __res__ = func arglist; \ - cma__int_unlock (cma__g_io_data_mutex); \ - if (__res__ >= 0 && fdpair[0] < cma__g_mx_file \ - && fdpair[1] < cma__g_mx_file) \ - post_process; \ - } \ - CATCH_ALL \ - { \ - cma__set_errno (EBADF); \ - __res__ = -1; \ - } \ - ENDTRY \ - if ((fdpair[0] >= cma__g_mx_file) || (fdpair[1] >= cma__g_mx_file)) \ - cma__bugcheck ("cma__ts_open2: one of fd's is too large"); \ - return __res__; \ - } - -/* - * INTERNAL INTERFACES - */ -extern void cma__close_general (int); - -extern void cma__init_thread_io (void); - -extern cma_t_boolean cma__io_available (cma__t_io_type,int,struct timeval *); - -extern void cma__io_epilog (cma__t_io_type,int); - -extern void cma__io_prolog (cma__t_io_type,int); - -extern void cma__io_wait (cma__t_io_type,int); - -extern void cma__open_general (int); - -extern void cma__reinit_thread_io (int); - -extern void cma__set_nonblocking (int); - -extern void cma__set_user_nonblock_flags (int,int); - -extern cma_t_boolean cma__is_open (int); - -#endif |