2 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * This file is conditionally built on x86_64 only. Otherwise weak symbol
35 * versions of the functions exported from here are used.
38 #include <linux/pci.h>
40 #include <asm/processor.h>
42 #include "ipath_kernel.h"
45 * ipath_enable_wc - enable write combining for MMIO writes to the device
46 * @dd: infinipath device
48 * This routine is x86_64-specific; it twiddles the CPU's MTRRs to enable
51 int ipath_enable_wc(struct ipath_devdata *dd)
56 const unsigned long addr = pci_resource_start(dd->pcidev, 0);
57 const size_t len = pci_resource_len(dd->pcidev, 0);
60 * Set the PIO buffers to be WCCOMB, so we get HT bursts to the
61 * chip. Linux (possibly the hardware) requires it to be on a power
62 * of 2 address matching the length (which has to be a power of 2).
63 * For rev1, that means the base address, for rev2, it will be just
64 * the PIO buffers themselves.
66 pioaddr = addr + dd->ipath_piobufbase;
67 piolen = (dd->ipath_piobcnt2k +
68 dd->ipath_piobcnt4k) *
69 ALIGN(dd->ipath_piobcnt2k +
70 dd->ipath_piobcnt4k, dd->ipath_palign);
72 for (bits = 0; !(piolen & (1ULL << bits)); bits++)
75 if (piolen != (1ULL << bits)) {
79 piolen = 1ULL << (bits + 1);
81 if (pioaddr & (piolen - 1)) {
83 ipath_dbg("pioaddr %llx not on right boundary for size "
85 (unsigned long long) pioaddr,
86 (unsigned long long) piolen);
87 atmp = pioaddr & ~(piolen - 1);
88 if (atmp < addr || (atmp + piolen) > (addr + len)) {
89 ipath_dev_err(dd, "No way to align address/size "
90 "(%llx/%llx), no WC mtrr\n",
91 (unsigned long long) atmp,
92 (unsigned long long) piolen << 1);
95 ipath_dbg("changing WC base from %llx to %llx, "
96 "len from %llx to %llx\n",
97 (unsigned long long) pioaddr,
98 (unsigned long long) atmp,
99 (unsigned long long) piolen,
100 (unsigned long long) piolen << 1);
108 ipath_cdbg(VERBOSE, "Setting mtrr for chip to WC "
109 "(addr %llx, len=0x%llx)\n",
110 (unsigned long long) pioaddr,
111 (unsigned long long) piolen);
112 cookie = mtrr_add(pioaddr, piolen, MTRR_TYPE_WRCOMB, 0);
115 dev_info(&dd->pcidev->dev,
116 "mtrr_add() WC for PIO bufs "
122 ipath_cdbg(VERBOSE, "Set mtrr for chip to WC, "
123 "cookie is %d\n", cookie);
124 dd->ipath_wc_cookie = cookie;
132 * ipath_disable_wc - disable write combining for MMIO writes to the device
133 * @dd: infinipath device
135 void ipath_disable_wc(struct ipath_devdata *dd)
137 if (dd->ipath_wc_cookie) {
138 ipath_cdbg(VERBOSE, "undoing WCCOMB on pio buffers\n");
139 mtrr_del(dd->ipath_wc_cookie, 0, 0);
140 dd->ipath_wc_cookie = 0;
145 * ipath_unordered_wc - indicate whether write combining is ordered
147 * Because our performance depends on our ability to do write combining mmio
148 * writes in the most efficient way, we need to know if we are on an Intel
149 * or AMD x86_64 processor. AMD x86_64 processors flush WC buffers out in
150 * the order completed, and so no special flushing is required to get
151 * correct ordering. Intel processors, however, will flush write buffers
152 * out in "random" orders, and so explicit ordering is needed at times.
154 int ipath_unordered_wc(void)
156 return boot_cpu_data.x86_vendor != X86_VENDOR_AMD;