freebsd-dev/sys/ia64/include/bus.h

821 lines
25 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2009 Marcel Moolenaar
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/* $NetBSD: bus.h,v 1.12 1997/10/01 08:25:15 fvdl Exp $ */
/*-
* Copyright (c) 1996, 1997 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
* NASA Ames Research Center.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
/*-
* Copyright (c) 1996 Charles M. Hannum. All rights reserved.
* Copyright (c) 1996 Christopher G. Demetriou. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Christopher G. Demetriou
* for the NetBSD Project.
* 4. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/* $FreeBSD$ */
#ifndef _MACHINE_BUS_H_
#define _MACHINE_BUS_H_
#include <machine/_bus.h>
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
#include <machine/cpufunc.h>
/*
* I/O port reads with ia32 semantics.
*/
#define inb bus_space_read_io_1
#define inw bus_space_read_io_2
#define inl bus_space_read_io_4
#define outb bus_space_write_io_1
#define outw bus_space_write_io_2
#define outl bus_space_write_io_4
/*
* Values for the ia64 bus space tag, not to be used directly by MI code.
*/
#define IA64_BUS_SPACE_IO 0 /* space is i/o space */
#define IA64_BUS_SPACE_MEM 1 /* space is mem space */
#define BUS_SPACE_BARRIER_READ 0x01 /* force read barrier */
#define BUS_SPACE_BARRIER_WRITE 0x02 /* force write barrier */
#define BUS_SPACE_MAXSIZE_24BIT 0xFFFFFF
#define BUS_SPACE_MAXSIZE_32BIT 0xFFFFFFFF
#define BUS_SPACE_MAXSIZE 0xFFFFFFFFFFFFFFFF
#define BUS_SPACE_MAXADDR_24BIT 0xFFFFFF
#define BUS_SPACE_MAXADDR_32BIT 0xFFFFFFFF
#define BUS_SPACE_MAXADDR 0xFFFFFFFFFFFFFFFF
#define BUS_SPACE_UNRESTRICTED (~0)
/*
* Map and unmap a region of device bus space into CPU virtual address space.
*/
int
bus_space_map(bus_space_tag_t, bus_addr_t, bus_size_t, int,
bus_space_handle_t *);
void
bus_space_unmap(bus_space_tag_t, bus_space_handle_t, bus_size_t size);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
/*
* Get a new handle for a subregion of an already-mapped area of bus space.
*/
static __inline int
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_subregion(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, bus_size_t size __unused, bus_space_handle_t *nbshp)
{
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
*nbshp = bsh + ofs;
return (0);
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
/*
* Allocate a region of memory that is accessible to devices in bus space.
*/
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
int
bus_space_alloc(bus_space_tag_t bst, bus_addr_t rstart, bus_addr_t rend,
bus_size_t size, bus_size_t align, bus_size_t boundary, int flags,
bus_addr_t *addrp, bus_space_handle_t *bshp);
/*
* Free a region of bus space accessible memory.
*/
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
void
bus_space_free(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t size);
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Bus read/write barrier method.
*/
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_barrier(bus_space_tag_t bst __unused, bus_space_handle_t bsh __unused,
bus_size_t ofs __unused, bus_size_t size __unused, int flags __unused)
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
{
ia64_mf_a();
ia64_mf();
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
/*
* Read 1 unit of data from bus space described by the tag, handle and ofs
* tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is returned.
*/
uint8_t bus_space_read_io_1(u_long);
uint16_t bus_space_read_io_2(u_long);
uint32_t bus_space_read_io_4(u_long);
uint64_t bus_space_read_io_8(u_long);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline uint8_t
bus_space_read_1(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs)
{
uint8_t val;
val = (__predict_false(bst == IA64_BUS_SPACE_IO))
? bus_space_read_io_1(bsh + ofs)
: ia64_ld1((void *)(bsh + ofs));
return (val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline uint16_t
bus_space_read_2(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs)
{
uint16_t val;
val = (__predict_false(bst == IA64_BUS_SPACE_IO))
? bus_space_read_io_2(bsh + ofs)
: ia64_ld2((void *)(bsh + ofs));
return (val);
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline uint32_t
bus_space_read_4(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs)
{
uint32_t val;
val = (__predict_false(bst == IA64_BUS_SPACE_IO))
? bus_space_read_io_4(bsh + ofs)
: ia64_ld4((void *)(bsh + ofs));
return (val);
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline uint64_t
bus_space_read_8(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs)
{
uint64_t val;
val = (__predict_false(bst == IA64_BUS_SPACE_IO))
? bus_space_read_io_8(bsh + ofs)
: ia64_ld8((void *)(bsh + ofs));
return (val);
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Write 1 unit of data to bus space described by the tag, handle and ofs
* tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is passed by value.
*/
void bus_space_write_io_1(u_long, uint8_t);
void bus_space_write_io_2(u_long, uint16_t);
void bus_space_write_io_4(u_long, uint32_t);
void bus_space_write_io_8(u_long, uint64_t);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_write_1(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs,
uint8_t val)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_io_1(bsh + ofs, val);
else
ia64_st1((void *)(bsh + ofs), val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_2(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs,
uint16_t val)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_io_2(bsh + ofs, val);
else
ia64_st2((void *)(bsh + ofs), val);
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_4(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs,
uint32_t val)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_io_4(bsh + ofs, val);
else
ia64_st4((void *)(bsh + ofs), val);
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_8(bus_space_tag_t bst, bus_space_handle_t bsh, bus_size_t ofs,
uint64_t val)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_io_8(bsh + ofs, val);
else
ia64_st8((void *)(bsh + ofs), val);
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Read count units of data from bus space described by the tag, handle and
* ofs tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is returned in the buffer passed by reference.
*/
void bus_space_read_multi_io_1(u_long, uint8_t *, size_t);
void bus_space_read_multi_io_2(u_long, uint16_t *, size_t);
void bus_space_read_multi_io_4(u_long, uint32_t *, size_t);
void bus_space_read_multi_io_8(u_long, uint64_t *, size_t);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_read_multi_1(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint8_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_multi_io_1(bsh + ofs, bufp, count);
else {
while (count-- > 0)
*bufp++ = ia64_ld1((void *)(bsh + ofs));
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_read_multi_2(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint16_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_multi_io_2(bsh + ofs, bufp, count);
else {
while (count-- > 0)
*bufp++ = ia64_ld2((void *)(bsh + ofs));
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_read_multi_4(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint32_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_multi_io_4(bsh + ofs, bufp, count);
else {
while (count-- > 0)
*bufp++ = ia64_ld4((void *)(bsh + ofs));
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_read_multi_8(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint64_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_multi_io_8(bsh + ofs, bufp, count);
else {
while (count-- > 0)
*bufp++ = ia64_ld8((void *)(bsh + ofs));
}
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Write count units of data to bus space described by the tag, handle and
* ofs tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is read from the buffer passed by reference.
*/
void bus_space_write_multi_io_1(u_long, const uint8_t *, size_t);
void bus_space_write_multi_io_2(u_long, const uint16_t *, size_t);
void bus_space_write_multi_io_4(u_long, const uint32_t *, size_t);
void bus_space_write_multi_io_8(u_long, const uint64_t *, size_t);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_write_multi_1(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint8_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_multi_io_1(bsh + ofs, bufp, count);
else {
while (count-- > 0)
ia64_st1((void *)(bsh + ofs), *bufp++);
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_multi_2(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint16_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_multi_io_2(bsh + ofs, bufp, count);
else {
while (count-- > 0)
ia64_st2((void *)(bsh + ofs), *bufp++);
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_multi_4(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint32_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_multi_io_4(bsh + ofs, bufp, count);
else {
while (count-- > 0)
ia64_st4((void *)(bsh + ofs), *bufp++);
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_multi_8(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint64_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_multi_io_8(bsh + ofs, bufp, count);
else {
while (count-- > 0)
ia64_st8((void *)(bsh + ofs), *bufp++);
}
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Read count units of data from bus space described by the tag, handle and
* ofs tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is written to the buffer passed by reference and read from successive
* bus space addresses. Access is unordered.
*/
void bus_space_read_region_io_1(u_long, uint8_t *, size_t);
void bus_space_read_region_io_2(u_long, uint16_t *, size_t);
void bus_space_read_region_io_4(u_long, uint32_t *, size_t);
void bus_space_read_region_io_8(u_long, uint64_t *, size_t);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_read_region_1(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint8_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_region_io_1(bsh + ofs, bufp, count);
else {
uint8_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
*bufp++ = ia64_ld1(bsp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_read_region_2(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint16_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_region_io_2(bsh + ofs, bufp, count);
else {
uint16_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
*bufp++ = ia64_ld2(bsp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_read_region_4(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint32_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_region_io_4(bsh + ofs, bufp, count);
else {
uint32_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
*bufp++ = ia64_ld4(bsp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_read_region_8(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint64_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_read_region_io_8(bsh + ofs, bufp, count);
else {
uint64_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
*bufp++ = ia64_ld8(bsp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Write count units of data from bus space described by the tag, handle and
* ofs tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is read from the buffer passed by reference and written to successive
* bus space addresses. Access is unordered.
*/
void bus_space_write_region_io_1(u_long, const uint8_t *, size_t);
void bus_space_write_region_io_2(u_long, const uint16_t *, size_t);
void bus_space_write_region_io_4(u_long, const uint32_t *, size_t);
void bus_space_write_region_io_8(u_long, const uint64_t *, size_t);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_write_region_1(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint8_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_region_io_1(bsh + ofs, bufp, count);
else {
uint8_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st1(bsp++, *bufp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_region_2(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint16_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_region_io_2(bsh + ofs, bufp, count);
else {
uint16_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st2(bsp++, *bufp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_region_4(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint32_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_region_io_4(bsh + ofs, bufp, count);
else {
uint32_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st4(bsp++, *bufp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_write_region_8(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, const uint64_t *bufp, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_write_region_io_8(bsh + ofs, bufp, count);
else {
uint64_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st8(bsp++, *bufp++);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Write count units of data from bus space described by the tag, handle and
* ofs tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is passed by value. Writes are unordered.
*/
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_set_multi_1(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint8_t val, size_t count)
{
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
while (count-- > 0)
bus_space_write_1(bst, bsh, ofs, val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_set_multi_2(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint16_t val, size_t count)
{
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
while (count-- > 0)
bus_space_write_2(bst, bsh, ofs, val);
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_set_multi_4(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint32_t val, size_t count)
{
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
while (count-- > 0)
bus_space_write_4(bst, bsh, ofs, val);
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_set_multi_8(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint64_t val, size_t count)
{
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
while (count-- > 0)
bus_space_write_8(bst, bsh, ofs, val);
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Write count units of data from bus space described by the tag, handle and
* ofs tuple. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes. The
* data is passed by value and written to successive bus space addresses.
* Writes are unordered.
*/
void bus_space_set_region_io_1(u_long, uint8_t, size_t);
void bus_space_set_region_io_2(u_long, uint16_t, size_t);
void bus_space_set_region_io_4(u_long, uint32_t, size_t);
void bus_space_set_region_io_8(u_long, uint64_t, size_t);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
static __inline void
bus_space_set_region_1(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint8_t val, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_set_region_io_1(bsh + ofs, val, count);
else {
uint8_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st1(bsp++, val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_set_region_2(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint16_t val, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_set_region_io_2(bsh + ofs, val, count);
else {
uint16_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st2(bsp++, val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_set_region_4(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint32_t val, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_set_region_io_4(bsh + ofs, val, count);
else {
uint32_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st4(bsp++, val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
static __inline void
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
bus_space_set_region_8(bus_space_tag_t bst, bus_space_handle_t bsh,
bus_size_t ofs, uint64_t val, size_t count)
{
if (__predict_false(bst == IA64_BUS_SPACE_IO))
bus_space_set_region_io_4(bsh + ofs, val, count);
else {
uint64_t *bsp = (void *)(bsh + ofs);
while (count-- > 0)
ia64_st8(bsp++, val);
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
}
/*
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
* Copy count units of data from bus space described by the tag and the first
* handle and ofs pair to bus space described by the tag and the second handle
* and ofs pair. A unit of data can be 1 byte, 2 bytes, 4 bytes or 8 bytes.
* The data is read from successive bus space addresses and also written to
* successive bus space addresses. Both reads and writes are unordered.
*/
void bus_space_copy_region_io_1(u_long, u_long, size_t);
void bus_space_copy_region_io_2(u_long, u_long, size_t);
void bus_space_copy_region_io_4(u_long, u_long, size_t);
void bus_space_copy_region_io_8(u_long, u_long, size_t);
static __inline void
bus_space_copy_region_1(bus_space_tag_t bst, bus_space_handle_t sbsh,
bus_size_t sofs, bus_space_handle_t dbsh, bus_size_t dofs, size_t count)
{
uint8_t *dst, *src;
if (__predict_false(bst == IA64_BUS_SPACE_IO)) {
bus_space_copy_region_io_1(sbsh + sofs, dbsh + dofs, count);
return;
}
src = (void *)(sbsh + sofs);
dst = (void *)(dbsh + dofs);
if (src < dst) {
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
src += count - 1;
dst += count - 1;
while (count-- > 0)
ia64_st1(dst--, ia64_ld1(src--));
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
} else {
while (count-- > 0)
ia64_st1(dst++, ia64_ld1(src++));
}
}
static __inline void
bus_space_copy_region_2(bus_space_tag_t bst, bus_space_handle_t sbsh,
bus_size_t sofs, bus_space_handle_t dbsh, bus_size_t dofs, size_t count)
{
uint16_t *dst, *src;
if (__predict_false(bst == IA64_BUS_SPACE_IO)) {
bus_space_copy_region_io_2(sbsh + sofs, dbsh + dofs, count);
return;
}
src = (void *)(sbsh + sofs);
dst = (void *)(dbsh + dofs);
if (src < dst) {
src += count - 1;
dst += count - 1;
while (count-- > 0)
ia64_st2(dst--, ia64_ld2(src--));
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
} else {
while (count-- > 0)
ia64_st2(dst++, ia64_ld2(src++));
}
}
static __inline void
bus_space_copy_region_4(bus_space_tag_t bst, bus_space_handle_t sbsh,
bus_size_t sofs, bus_space_handle_t dbsh, bus_size_t dofs, size_t count)
{
uint32_t *dst, *src;
if (__predict_false(bst == IA64_BUS_SPACE_IO)) {
bus_space_copy_region_io_4(sbsh + sofs, dbsh + dofs, count);
return;
}
src = (void *)(sbsh + sofs);
dst = (void *)(dbsh + dofs);
if (src < dst) {
src += count - 1;
dst += count - 1;
while (count-- > 0)
ia64_st4(dst--, ia64_ld4(src--));
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
} else {
while (count-- > 0)
ia64_st4(dst++, ia64_ld4(src++));
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
}
static __inline void
bus_space_copy_region_8(bus_space_tag_t bst, bus_space_handle_t sbsh,
bus_size_t sofs, bus_space_handle_t dbsh, bus_size_t dofs, size_t count)
{
uint64_t *dst, *src;
if (__predict_false(bst == IA64_BUS_SPACE_IO)) {
bus_space_copy_region_io_8(sbsh + sofs, dbsh + dofs, count);
return;
}
src = (void *)(sbsh + sofs);
dst = (void *)(dbsh + dofs);
if (src < dst) {
src += count - 1;
dst += count - 1;
while (count-- > 0)
ia64_st8(dst--, ia64_ld8(src--));
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
} else {
while (count-- > 0)
ia64_st8(dst++, ia64_ld8(src++));
}
}
Revamp the newbus functions: o do not use the in* and out* functions. These functions are used by legacy drivers and thus must have ia32 compatible behaviour. Hence, they need to have fences. Using these functions for newbus would then pessimize performance. o remove the conditional compilation of PIO and/or MEMIO support. It's a PITA without having any significant benefit. We always support them both. Since there are no I/O ports on ia64 (they are simulated by the chipset by translating memory mapped I/O to predefined uncacheable memory regions) the only difference between PIO and MEMIO is in the address calculation. There should be enough ILP that can be exploited here that making these computations compile-time conditional is not worth it. We now also don't use the read* and write* functions. o Add the missing *_8 variants. They were missing, although not missed. It's for completeness. o Do not add the fences that were present in the low-level support functions here. We're using uncacheable memory, which means that accesses are in program order. Change the barrier implementation to not only do a memory fence, but also an acceptance fence. This should more reliably synchronize drivers with the hardware. The memory fence enforces ordering, but does not imply visibility (ie the access does not necessarily have happened). This is what the acceptance deals with. cpufunc.h cleanup: o Remove the low-level memory mapped I/O support functions. They are not used. Keep the low-level I/O port access functions for legacy drivers and add fences to ensure ia32 compatibility. o Remove the syscons specific functions now that we have moved the proper definitions where they belong. o Replace the ia64_port_address() and ia64_memory_address() functions with macros. There's a bigger change inline functions get inlined when there aren't function callsi and the calculations are simply enough to do it with macros. Replace the one reference to ia64_memory address in mp_machdep.c to use the macro.
2003-04-29 09:50:03 +00:00
/*
* Stream accesses are the same as normal accesses on ia64; there are no
* supported bus systems with an endianess different from the host one.
*/
#define bus_space_read_stream_1 bus_space_read_1
#define bus_space_read_stream_2 bus_space_read_2
#define bus_space_read_stream_4 bus_space_read_4
#define bus_space_read_stream_8 bus_space_read_8
#define bus_space_write_stream_1 bus_space_write_1
#define bus_space_write_stream_2 bus_space_write_2
#define bus_space_write_stream_4 bus_space_write_4
#define bus_space_write_stream_8 bus_space_write_8
#define bus_space_read_multi_stream_1 bus_space_read_multi_1
#define bus_space_read_multi_stream_2 bus_space_read_multi_2
#define bus_space_read_multi_stream_4 bus_space_read_multi_4
#define bus_space_read_multi_stream_8 bus_space_read_multi_8
#define bus_space_write_multi_stream_1 bus_space_write_multi_1
#define bus_space_write_multi_stream_2 bus_space_write_multi_2
#define bus_space_write_multi_stream_4 bus_space_write_multi_4
#define bus_space_write_multi_stream_8 bus_space_write_multi_8
#define bus_space_read_region_stream_1 bus_space_read_region_1
#define bus_space_read_region_stream_2 bus_space_read_region_2
#define bus_space_read_region_stream_4 bus_space_read_region_4
#define bus_space_read_region_stream_8 bus_space_read_region_8
#define bus_space_write_region_stream_1 bus_space_write_region_1
#define bus_space_write_region_stream_2 bus_space_write_region_2
#define bus_space_write_region_stream_4 bus_space_write_region_4
#define bus_space_write_region_stream_8 bus_space_write_region_8
#define bus_space_set_multi_stream_1 bus_space_set_multi_1
#define bus_space_set_multi_stream_2 bus_space_set_multi_2
#define bus_space_set_multi_stream_4 bus_space_set_multi_4
#define bus_space_set_multi_stream_8 bus_space_set_multi_8
#define bus_space_set_region_stream_1 bus_space_set_region_1
#define bus_space_set_region_stream_2 bus_space_set_region_2
#define bus_space_set_region_stream_4 bus_space_set_region_4
#define bus_space_set_region_stream_8 bus_space_set_region_8
#define bus_space_copy_region_stream_1 bus_space_copy_region_1
#define bus_space_copy_region_stream_2 bus_space_copy_region_2
#define bus_space_copy_region_stream_4 bus_space_copy_region_4
#define bus_space_copy_region_stream_8 bus_space_copy_region_8
#include <machine/bus_dma.h>
#endif /* _MACHINE_BUS_H_ */