summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/qib/qib_wc_x86_64.c
blob: edd0ddbd44815d8e48ed49d3dec46e37545aa031 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
/*
 * Copyright (c) 2012 Intel Corporation. All rights reserved.
 * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - 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.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

/*
 * This file is conditionally built on x86_64 only.  Otherwise weak symbol
 * versions of the functions exported from here are used.
 */

#include <linux/pci.h>
#include <asm/mtrr.h>
#include <asm/processor.h>

#include "qib.h"

/**
 * qib_enable_wc - enable write combining for MMIO writes to the device
 * @dd: qlogic_ib device
 *
 * This routine is x86_64-specific; it twiddles the CPU's MTRRs to enable
 * write combining.
 */
int qib_enable_wc(struct qib_devdata *dd)
{
	int ret = 0;
	u64 pioaddr, piolen;
	unsigned bits;
	const unsigned long addr = pci_resource_start(dd->pcidev, 0);
	const size_t len = pci_resource_len(dd->pcidev, 0);

	/*
	 * Set the PIO buffers to be WCCOMB, so we get HT bursts to the
	 * chip.  Linux (possibly the hardware) requires it to be on a power
	 * of 2 address matching the length (which has to be a power of 2).
	 * For rev1, that means the base address, for rev2, it will be just
	 * the PIO buffers themselves.
	 * For chips with two sets of buffers, the calculations are
	 * somewhat more complicated; we need to sum, and the piobufbase
	 * register has both offsets, 2K in low 32 bits, 4K in high 32 bits.
	 * The buffers are still packed, so a single range covers both.
	 */
	if (dd->piobcnt2k && dd->piobcnt4k) {
		/* 2 sizes for chip */
		unsigned long pio2kbase, pio4kbase;

		pio2kbase = dd->piobufbase & 0xffffffffUL;
		pio4kbase = (dd->piobufbase >> 32) & 0xffffffffUL;
		if (pio2kbase < pio4kbase) {
			/* all current chips */
			pioaddr = addr + pio2kbase;
			piolen = pio4kbase - pio2kbase +
				dd->piobcnt4k * dd->align4k;
		} else {
			pioaddr = addr + pio4kbase;
			piolen = pio2kbase - pio4kbase +
				dd->piobcnt2k * dd->palign;
		}
	} else {  /* single buffer size (2K, currently) */
		pioaddr = addr + dd->piobufbase;
		piolen = dd->piobcnt2k * dd->palign +
			dd->piobcnt4k * dd->align4k;
	}

	for (bits = 0; !(piolen & (1ULL << bits)); bits++)
		; /* do nothing */

	if (piolen != (1ULL << bits)) {
		piolen >>= bits;
		while (piolen >>= 1)
			bits++;
		piolen = 1ULL << (bits + 1);
	}
	if (pioaddr & (piolen - 1)) {
		u64 atmp = pioaddr & ~(piolen - 1);

		if (atmp < addr || (atmp + piolen) > (addr + len)) {
			qib_dev_err(dd,
				"No way to align address/size (%llx/%llx), no WC mtrr\n",
				(unsigned long long) atmp,
				(unsigned long long) piolen << 1);
			ret = -ENODEV;
		} else {
			pioaddr = atmp;
			piolen <<= 1;
		}
	}

	if (!ret) {
		dd->wc_cookie = arch_phys_wc_add(pioaddr, piolen);
		if (dd->wc_cookie < 0)
			/* use error from routine */
			ret = dd->wc_cookie;
	}

	return ret;
}

/**
 * qib_disable_wc - disable write combining for MMIO writes to the device
 * @dd: qlogic_ib device
 */
void qib_disable_wc(struct qib_devdata *dd)
{
	arch_phys_wc_del(dd->wc_cookie);
}

/**
 * qib_unordered_wc - indicate whether write combining is ordered
 *
 * Because our performance depends on our ability to do write combining mmio
 * writes in the most efficient way, we need to know if we are on an Intel
 * or AMD x86_64 processor.  AMD x86_64 processors flush WC buffers out in
 * the order completed, and so no special flushing is required to get
 * correct ordering.  Intel processors, however, will flush write buffers
 * out in "random" orders, and so explicit ordering is needed at times.
 */
int qib_unordered_wc(void)
{
	return boot_cpu_data.x86_vendor != X86_VENDOR_AMD;
}