From: John Feeney <jfeeney@redhat.com>
Date: Tue, 21 Apr 2009 19:02:49 -0400
Subject: [misc] I/O AT: new ioat*.c
Message-id: 49EE5099.3030009@redhat.com
O-Subject: [RHEL5.4 PATCH 7/8] Update I/O AT: new ioat*.c
Bugzilla: 436048
RH-Acked-by: Prarit Bhargava <prarit@redhat.com>
RH-Acked-by: John W. Linville <linville@redhat.com>
bz436048 Update I/O AT code to upstream
This patch adds the new ioat*.c files in drivers/dma_v3.
ioat.c | 226 +++++++
ioat_dca.c | 657 ++++++++++++++++++++++
ioat_dma.c | 1718
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
iovlock_v3.c | 274 +++++++++
4 files changed, 2875 insertions(+)
Note: the existing RHEL5 code has a iovlock.c, hence iovlock_v3.c.
The commits to each file and any modifications are listed below.
Note: upstream commits come from drivers/dma but patch
adds files to the new directory, drivers/dma_v3.
ioat*.c files in dma_v3 directory
drivers/dma_v3/iovlock_v3.c
Same as following commits except for:
use base_address_v3 instead of base_address
added "_v3" suffix to
dma_pin_iovec_pages_v3()
dma_unpin_iovec_pages_v3()
dma_memcpy_pg_to_iovec_v3()
de5506e155276d385712c2aa1c2d9a27cd4ed947 Leech 6/18/06
56e0873b7b146564a0c36e225624f5a9b2d63791 Bunk 7/4/06
342ff7b24f42d01b27d884c699855c713d720fcb Nelson 7/31/07
drivers/dma_v3/ioat_dma.c
Same as the below commits, except for:
Uses common.device_v3 vs. common.device
Uses dma_chan->device_v3 vs. dma_chan->device
Uses all new dmaengine functions (dma_async_register_v3,
dma_async_unregister_v3, dma_async_tx_descriptor_init)
Add struct pt_regs argument to function called on interrupt.
for_each_bit() macro not supported so kept find_first_bit()
Replaced INIT_DELAYED_WORK() with INIT_WORK()
Started as ioatdma.c
0bbd5f4e97ff9c057b385a1886b4aed1fb0300f1 Leech 5/24/06
223758c77a67b1eb383a92b35d67de29502a9f55 Nelson 10/16/2007
File partially split and renamed to ioat_dma.c
1fda5f4e96225c3ed0baded942704c0ae399da23 Nelson 10/16/07
43d6e369d43ff175e1e0e80caaedb1e53829247c Nelson 10/16/07
8ab89567da0cea9bae2c1b5dad47b51c424479e4 Nelson 10/16/07
3e037454bcfa4b187e8293d2121bd8c0f5a5c31c Nelson 10/16/07
7df7cf0676060d778486359676734447347e1caf Nelson 10/18/07
dfe2299e7b35a0adfc87f04d3e725ccc508d7626 Nelson 10/18/07
5149fd010f404889b7d8f79159057791fbb817b1 Nelson 10/18/07
7f2b291f56d08e001454d16d3c92e175434898b3 Nelson 10/18/07
952184304fbf030f0133d8b66a91b2847dce729e Nelson 10/18/07
7bb67c14fd3778504fb77da30ce11582336dfced Nelson 11/15/07
711924b1052a280bd2452c3babb9816e4a77c72 Nelson 12/18/07
bb8e8bcce7eceacb52eb0a3ebb64202ad6bcc438 Nelson 12/18/07
0036731c88fdb5bf4f04a796a30b5e445fc57f54 Williams 2/6/08
d4c56f97ff21df405d0cebe11f49e3c3c79662b5 Williams 2/6/08
6497dcffe07b7c3d863f9899280c4f6eae999161 Williams 3/1/08
ce4d65a5db77e1568c82d5151a746f627c4f6ed5 Williams 4/17/08
636bdeaa1243327501edfd2a597ed7443eb4239a Williams 4/17/08
848c536a37b8db4e461f14ca15fe29850151c822 Skinnemoen 7/8/08
e1d181efb14a93cf263d6c588a5395518edf3294 Williams 7/8/08
09177e85d6a0bffac8b55afd28ed8b82bd873f0b Sosnowski 7/22/08
16a37acaaf4aaa631ba3f83710ed6cdb1a597520 Sosnowski 7/23/08
7f1b358a236ee9c19657a619ac6f2dcabcaa0924 Sosnowski 7/23/08
3eb62a1a8c1655cf88e44158938a28ef774fdc08 Williams 12/4/08
per Intel's request.
ceaccb6a0b1816a05cccb77a0f5343a1a900d3cb Williams 1/6/09
per Intel's request.
drivers/dma_v3/ioat.c
.name of ioat_pci_driver struct has "_v3" suffix
Split from ioatdma.c
47b16539e1072afd2c964b97be4a95b5cf2ff801 Viro 10/10/06 (ioatdma.c)
8ab89567da0cea9bae2c1b5dad47b51c424479e4 Nelson 10/16/07
2ed6dc34f9ed39bb8e4c81ea1056f0ba56315841 Nelson 10/16/07
7df7cf0676060d778486359676734447347e1caf Nelson 10/18/07
dfe2299e7b35a0adfc87f04d3e725ccc508d7626 Nelson 10/18/07
5149fd010f404889b7d8f79159057791fbb817b1 Nelson 10/18/07
7bb67c14fd3778504fb77da30ce11582336dfced Nelson 11/15/07
7f1b358a236ee9c19657a619ac6f2dcabcaa0924 Sosnowski 8/23/08
drivers/dma_v3/ioat_dca.c
Same as upstream
2ed6dc34f9ed39bb8e4c81ea1056f0ba56315841 Nelson 10/16/07
5149fd010f404889b7d8f79159057791fbb817b1 Nelson 10/18/07
7bb67c14fd3778504fb77da30ce11582336dfced Nelson 11/15/07
53a0c98e117272125183138aefc6b13b4a5f38a1 Viro 3/30/08
7f1b358a236ee9c19657a619ac6f2dcabcaa0924 Sosnowski 8/23/08
diff --git a/drivers/dma_v3/ioat.c b/drivers/dma_v3/ioat.c
new file mode 100644
index 0000000..8947880
--- /dev/null
+++ b/drivers/dma_v3/ioat.c
@@ -0,0 +1,226 @@
+/*
+ * Intel I/OAT DMA Linux driver
+ * Copyright(c) 2007 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ */
+
+/*
+ * This driver supports an Intel I/OAT DMA engine, which does asynchronous
+ * copy operations.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/interrupt.h>
+#include <linux/dca.h>
+#include "ioatdma.h"
+#include "ioatdma_registers.h"
+#include "ioatdma_hw.h"
+
+MODULE_VERSION(IOAT_DMA_VERSION);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Intel Corporation");
+
+static struct pci_device_id ioat_pci_tbl[] = {
+ /* I/OAT v1 platforms */
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_CNB) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_SCNB) },
+ { PCI_DEVICE(PCI_VENDOR_ID_UNISYS, PCI_DEVICE_ID_UNISYS_DMA_DIRECTOR) },
+
+ /* I/OAT v2 platforms */
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB) },
+
+ /* I/OAT v3 platforms */
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG0) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG1) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG2) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG3) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG4) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG5) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG6) },
+ { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG7) },
+ { 0, }
+};
+
+struct ioat_device {
+ struct pci_dev *pdev;
+ void __iomem *iobase;
+ struct ioatdma_device *dma;
+ struct dca_provider *dca;
+};
+
+static int __devinit ioat_probe(struct pci_dev *pdev,
+ const struct pci_device_id *id);
+static void __devexit ioat_remove(struct pci_dev *pdev);
+
+static int ioat_dca_enabled = 1;
+module_param(ioat_dca_enabled, int, 0644);
+MODULE_PARM_DESC(ioat_dca_enabled, "control support of dca service (default: 1)");
+
+static int ioat_setup_functionality(struct pci_dev *pdev, void __iomem *iobase)
+{
+ struct ioat_device *device = pci_get_drvdata(pdev);
+ u8 version;
+ int err = 0;
+
+ version = readb(iobase + IOAT_VER_OFFSET);
+ switch (version) {
+ case IOAT_VER_1_2:
+ device->dma = ioat_dma_probe(pdev, iobase);
+ if (device->dma && ioat_dca_enabled)
+ device->dca = ioat_dca_init(pdev, iobase);
+ break;
+ case IOAT_VER_2_0:
+ device->dma = ioat_dma_probe(pdev, iobase);
+ if (device->dma && ioat_dca_enabled)
+ device->dca = ioat2_dca_init(pdev, iobase);
+ break;
+ case IOAT_VER_3_0:
+ device->dma = ioat_dma_probe(pdev, iobase);
+ if (device->dma && ioat_dca_enabled)
+ device->dca = ioat3_dca_init(pdev, iobase);
+ break;
+ default:
+ err = -ENODEV;
+ break;
+ }
+ if (!device->dma)
+ err = -ENODEV;
+ return err;
+}
+
+static void ioat_shutdown_functionality(struct pci_dev *pdev)
+{
+ struct ioat_device *device = pci_get_drvdata(pdev);
+
+ dev_err(&pdev->dev, "Removing dma and dca services\n");
+ if (device->dca) {
+ unregister_dca_provider(device->dca);
+ free_dca_provider(device->dca);
+ device->dca = NULL;
+ }
+
+ if (device->dma) {
+ ioat_dma_remove(device->dma);
+ device->dma = NULL;
+ }
+}
+
+static struct pci_driver ioat_pci_driver = {
+ .name = "ioatdma_v3",
+ .id_table = ioat_pci_tbl,
+ .probe = ioat_probe,
+ .shutdown = ioat_shutdown_functionality,
+ .remove = __devexit_p(ioat_remove),
+};
+
+static int __devinit ioat_probe(struct pci_dev *pdev,
+ const struct pci_device_id *id)
+{
+ void __iomem *iobase;
+ struct ioat_device *device;
+ unsigned long mmio_start, mmio_len;
+ int err;
+
+ err = pci_enable_device(pdev);
+ if (err)
+ goto err_enable_device;
+
+ err = pci_request_regions(pdev, ioat_pci_driver.name);
+ if (err)
+ goto err_request_regions;
+
+ err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
+ if (err)
+ err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+ if (err)
+ goto err_set_dma_mask;
+
+ err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
+ if (err)
+ err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
+ if (err)
+ goto err_set_dma_mask;
+
+ mmio_start = pci_resource_start(pdev, 0);
+ mmio_len = pci_resource_len(pdev, 0);
+ iobase = ioremap(mmio_start, mmio_len);
+ if (!iobase) {
+ err = -ENOMEM;
+ goto err_ioremap;
+ }
+
+ device = kzalloc(sizeof(*device), GFP_KERNEL);
+ if (!device) {
+ err = -ENOMEM;
+ goto err_kzalloc;
+ }
+ device->pdev = pdev;
+ pci_set_drvdata(pdev, device);
+ device->iobase = iobase;
+
+ pci_set_master(pdev);
+
+ err = ioat_setup_functionality(pdev, iobase);
+ if (err)
+ goto err_version;
+
+ return 0;
+
+err_version:
+ kfree(device);
+err_kzalloc:
+ iounmap(iobase);
+err_ioremap:
+err_set_dma_mask:
+ pci_release_regions(pdev);
+ pci_disable_device(pdev);
+err_request_regions:
+err_enable_device:
+ return err;
+}
+
+/*
+ * It is unsafe to remove this module: if removed while a requested
+ * dma is outstanding, esp. from tcp, it is possible to hang while
+ * waiting for something that will never finish. However, if you're
+ * feeling lucky, this usually works just fine.
+ */
+static void __devexit ioat_remove(struct pci_dev *pdev)
+{
+ struct ioat_device *device = pci_get_drvdata(pdev);
+
+ ioat_shutdown_functionality(pdev);
+
+ kfree(device);
+}
+
+static int __init ioat_init_module(void)
+{
+ return pci_register_driver(&ioat_pci_driver);
+}
+module_init(ioat_init_module);
+
+static void __exit ioat_exit_module(void)
+{
+ pci_unregister_driver(&ioat_pci_driver);
+}
+module_exit(ioat_exit_module);
diff --git a/drivers/dma_v3/ioat_dca.c b/drivers/dma_v3/ioat_dca.c
new file mode 100644
index 0000000..85bc467
--- /dev/null
+++ b/drivers/dma_v3/ioat_dca.c
@@ -0,0 +1,657 @@
+/*
+ * Intel I/OAT DMA Linux driver
+ * Copyright(c) 2007 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/smp.h>
+#include <linux/interrupt.h>
+#include <linux/dca.h>
+
+/* either a kernel change is needed, or we need something like this in kernel */
+#ifndef CONFIG_SMP
+#include <asm/smp.h>
+#undef cpu_physical_id
+#define cpu_physical_id(cpu) (cpuid_ebx(1) >> 24)
+#endif
+
+#include "ioatdma.h"
+#include "ioatdma_registers.h"
+
+/*
+ * Bit 7 of a tag map entry is the "valid" bit, if it is set then bits 0:6
+ * contain the bit number of the APIC ID to map into the DCA tag. If the valid
+ * bit is not set, then the value must be 0 or 1 and defines the bit in the tag.
+ */
+#define DCA_TAG_MAP_VALID 0x80
+
+#define DCA3_TAG_MAP_BIT_TO_INV 0x80
+#define DCA3_TAG_MAP_BIT_TO_SEL 0x40
+#define DCA3_TAG_MAP_LITERAL_VAL 0x1
+
+#define DCA_TAG_MAP_MASK 0xDF
+
+/*
+ * "Legacy" DCA systems do not implement the DCA register set in the
+ * I/OAT device. Software needs direct support for their tag mappings.
+ */
+
+#define APICID_BIT(x) (DCA_TAG_MAP_VALID | (x))
+#define IOAT_TAG_MAP_LEN 8
+
+static u8 ioat_tag_map_BNB[IOAT_TAG_MAP_LEN] = {
+ 1, APICID_BIT(1), APICID_BIT(2), APICID_BIT(2), };
+static u8 ioat_tag_map_SCNB[IOAT_TAG_MAP_LEN] = {
+ 1, APICID_BIT(1), APICID_BIT(2), APICID_BIT(2), };
+static u8 ioat_tag_map_CNB[IOAT_TAG_MAP_LEN] = {
+ 1, APICID_BIT(1), APICID_BIT(3), APICID_BIT(4), APICID_BIT(2), };
+static u8 ioat_tag_map_UNISYS[IOAT_TAG_MAP_LEN] = { 0 };
+
+/* pack PCI B/D/F into a u16 */
+static inline u16 dcaid_from_pcidev(struct pci_dev *pci)
+{
+ return (pci->bus->number << 8) | pci->devfn;
+}
+
+static int dca_enabled_in_bios(struct pci_dev *pdev)
+{
+ /* CPUID level 9 returns DCA configuration */
+ /* Bit 0 indicates DCA enabled by the BIOS */
+ unsigned long cpuid_level_9;
+ int res;
+
+ cpuid_level_9 = cpuid_eax(9);
+ res = test_bit(0, &cpuid_level_9);
+ if (!res)
+ dev_err(&pdev->dev, "DCA is disabled in BIOS\n");
+
+ return res;
+}
+
+static int system_has_dca_enabled(struct pci_dev *pdev)
+{
+ if (boot_cpu_has(X86_FEATURE_DCA))
+ return dca_enabled_in_bios(pdev);
+
+ dev_err(&pdev->dev, "boot cpu doesn't have X86_FEATURE_DCA\n");
+ return 0;
+}
+
+struct ioat_dca_slot {
+ struct pci_dev *pdev; /* requester device */
+ u16 rid; /* requester id, as used by IOAT */
+};
+
+#define IOAT_DCA_MAX_REQ 6
+#define IOAT3_DCA_MAX_REQ 2
+
+struct ioat_dca_priv {
+ void __iomem *iobase;
+ void __iomem *dca_base;
+ int max_requesters;
+ int requester_count;
+ u8 tag_map[IOAT_TAG_MAP_LEN];
+ struct ioat_dca_slot req_slots[0];
+};
+
+/* 5000 series chipset DCA Port Requester ID Table Entry Format
+ * [15:8] PCI-Express Bus Number
+ * [7:3] PCI-Express Device Number
+ * [2:0] PCI-Express Function Number
+ *
+ * 5000 series chipset DCA control register format
+ * [7:1] Reserved (0)
+ * [0] Ignore Function Number
+ */
+
+static int ioat_dca_add_requester(struct dca_provider *dca, struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+ u16 id;
+
+ /* This implementation only supports PCI-Express */
+ if (dev->bus != &pci_bus_type)
+ return -ENODEV;
+ pdev = to_pci_dev(dev);
+ id = dcaid_from_pcidev(pdev);
+
+ if (ioatdca->requester_count == ioatdca->max_requesters)
+ return -ENODEV;
+
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == NULL) {
+ /* found an empty slot */
+ ioatdca->requester_count++;
+ ioatdca->req_slots[i].pdev = pdev;
+ ioatdca->req_slots[i].rid = id;
+ writew(id, ioatdca->dca_base + (i * 4));
+ /* make sure the ignore function bit is off */
+ writeb(0, ioatdca->dca_base + (i * 4) + 2);
+ return i;
+ }
+ }
+ /* Error, ioatdma->requester_count is out of whack */
+ return -EFAULT;
+}
+
+static int ioat_dca_remove_requester(struct dca_provider *dca,
+ struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+
+ /* This implementation only supports PCI-Express */
+ if (dev->bus != &pci_bus_type)
+ return -ENODEV;
+ pdev = to_pci_dev(dev);
+
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == pdev) {
+ writew(0, ioatdca->dca_base + (i * 4));
+ ioatdca->req_slots[i].pdev = NULL;
+ ioatdca->req_slots[i].rid = 0;
+ ioatdca->requester_count--;
+ return i;
+ }
+ }
+ return -ENODEV;
+}
+
+static u8 ioat_dca_get_tag(struct dca_provider *dca,
+ struct device *dev,
+ int cpu)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ int i, apic_id, bit, value;
+ u8 entry, tag;
+
+ tag = 0;
+ apic_id = cpu_physical_id(cpu);
+
+ for (i = 0; i < IOAT_TAG_MAP_LEN; i++) {
+ entry = ioatdca->tag_map[i];
+ if (entry & DCA_TAG_MAP_VALID) {
+ bit = entry & ~DCA_TAG_MAP_VALID;
+ value = (apic_id & (1 << bit)) ? 1 : 0;
+ } else {
+ value = entry ? 1 : 0;
+ }
+ tag |= (value << i);
+ }
+ return tag;
+}
+
+static int ioat_dca_dev_managed(struct dca_provider *dca,
+ struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+
+ pdev = to_pci_dev(dev);
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == pdev)
+ return 1;
+ }
+ return 0;
+}
+
+static struct dca_ops ioat_dca_ops = {
+ .add_requester = ioat_dca_add_requester,
+ .remove_requester = ioat_dca_remove_requester,
+ .get_tag = ioat_dca_get_tag,
+ .dev_managed = ioat_dca_dev_managed,
+};
+
+
+struct dca_provider *ioat_dca_init(struct pci_dev *pdev, void __iomem *iobase)
+{
+ struct dca_provider *dca;
+ struct ioat_dca_priv *ioatdca;
+ u8 *tag_map = NULL;
+ int i;
+ int err;
+ u8 version;
+ u8 max_requesters;
+
+ if (!system_has_dca_enabled(pdev))
+ return NULL;
+
+ /* I/OAT v1 systems must have a known tag_map to support DCA */
+ switch (pdev->vendor) {
+ case PCI_VENDOR_ID_INTEL:
+ switch (pdev->device) {
+ case PCI_DEVICE_ID_INTEL_IOAT:
+ tag_map = ioat_tag_map_BNB;
+ break;
+ case PCI_DEVICE_ID_INTEL_IOAT_CNB:
+ tag_map = ioat_tag_map_CNB;
+ break;
+ case PCI_DEVICE_ID_INTEL_IOAT_SCNB:
+ tag_map = ioat_tag_map_SCNB;
+ break;
+ }
+ break;
+ case PCI_VENDOR_ID_UNISYS:
+ switch (pdev->device) {
+ case PCI_DEVICE_ID_UNISYS_DMA_DIRECTOR:
+ tag_map = ioat_tag_map_UNISYS;
+ break;
+ }
+ break;
+ }
+ if (tag_map == NULL)
+ return NULL;
+
+ version = readb(iobase + IOAT_VER_OFFSET);
+ if (version == IOAT_VER_3_0)
+ max_requesters = IOAT3_DCA_MAX_REQ;
+ else
+ max_requesters = IOAT_DCA_MAX_REQ;
+
+ dca = alloc_dca_provider(&ioat_dca_ops,
+ sizeof(*ioatdca) +
+ (sizeof(struct ioat_dca_slot) * max_requesters));
+ if (!dca)
+ return NULL;
+
+ ioatdca = dca_priv(dca);
+ ioatdca->max_requesters = max_requesters;
+ ioatdca->dca_base = iobase + 0x54;
+
+ /* copy over the APIC ID to DCA tag mapping */
+ for (i = 0; i < IOAT_TAG_MAP_LEN; i++)
+ ioatdca->tag_map[i] = tag_map[i];
+
+ err = register_dca_provider(dca, &pdev->dev);
+ if (err) {
+ free_dca_provider(dca);
+ return NULL;
+ }
+
+ return dca;
+}
+
+
+static int ioat2_dca_add_requester(struct dca_provider *dca, struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+ u16 id;
+ u16 global_req_table;
+
+ /* This implementation only supports PCI-Express */
+ if (dev->bus != &pci_bus_type)
+ return -ENODEV;
+ pdev = to_pci_dev(dev);
+ id = dcaid_from_pcidev(pdev);
+
+ if (ioatdca->requester_count == ioatdca->max_requesters)
+ return -ENODEV;
+
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == NULL) {
+ /* found an empty slot */
+ ioatdca->requester_count++;
+ ioatdca->req_slots[i].pdev = pdev;
+ ioatdca->req_slots[i].rid = id;
+ global_req_table =
+ readw(ioatdca->dca_base + IOAT_DCA_GREQID_OFFSET);
+ writel(id | IOAT_DCA_GREQID_VALID,
+ ioatdca->iobase + global_req_table + (i * 4));
+ return i;
+ }
+ }
+ /* Error, ioatdma->requester_count is out of whack */
+ return -EFAULT;
+}
+
+static int ioat2_dca_remove_requester(struct dca_provider *dca,
+ struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+ u16 global_req_table;
+
+ /* This implementation only supports PCI-Express */
+ if (dev->bus != &pci_bus_type)
+ return -ENODEV;
+ pdev = to_pci_dev(dev);
+
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == pdev) {
+ global_req_table =
+ readw(ioatdca->dca_base + IOAT_DCA_GREQID_OFFSET);
+ writel(0, ioatdca->iobase + global_req_table + (i * 4));
+ ioatdca->req_slots[i].pdev = NULL;
+ ioatdca->req_slots[i].rid = 0;
+ ioatdca->requester_count--;
+ return i;
+ }
+ }
+ return -ENODEV;
+}
+
+static u8 ioat2_dca_get_tag(struct dca_provider *dca,
+ struct device *dev,
+ int cpu)
+{
+ u8 tag;
+
+ tag = ioat_dca_get_tag(dca, dev, cpu);
+ tag = (~tag) & 0x1F;
+ return tag;
+}
+
+static struct dca_ops ioat2_dca_ops = {
+ .add_requester = ioat2_dca_add_requester,
+ .remove_requester = ioat2_dca_remove_requester,
+ .get_tag = ioat2_dca_get_tag,
+ .dev_managed = ioat_dca_dev_managed,
+};
+
+static int ioat2_dca_count_dca_slots(void __iomem *iobase, u16 dca_offset)
+{
+ int slots = 0;
+ u32 req;
+ u16 global_req_table;
+
+ global_req_table = readw(iobase + dca_offset + IOAT_DCA_GREQID_OFFSET);
+ if (global_req_table == 0)
+ return 0;
+ do {
+ req = readl(iobase + global_req_table + (slots * sizeof(u32)));
+ slots++;
+ } while ((req & IOAT_DCA_GREQID_LASTID) == 0);
+
+ return slots;
+}
+
+struct dca_provider *ioat2_dca_init(struct pci_dev *pdev, void __iomem *iobase)
+{
+ struct dca_provider *dca;
+ struct ioat_dca_priv *ioatdca;
+ int slots;
+ int i;
+ int err;
+ u32 tag_map;
+ u16 dca_offset;
+ u16 csi_fsb_control;
+ u16 pcie_control;
+ u8 bit;
+
+ if (!system_has_dca_enabled(pdev))
+ return NULL;
+
+ dca_offset = readw(iobase + IOAT_DCAOFFSET_OFFSET);
+ if (dca_offset == 0)
+ return NULL;
+
+ slots = ioat2_dca_count_dca_slots(iobase, dca_offset);
+ if (slots == 0)
+ return NULL;
+
+ dca = alloc_dca_provider(&ioat2_dca_ops,
+ sizeof(*ioatdca)
+ + (sizeof(struct ioat_dca_slot) * slots));
+ if (!dca)
+ return NULL;
+
+ ioatdca = dca_priv(dca);
+ ioatdca->iobase = iobase;
+ ioatdca->dca_base = iobase + dca_offset;
+ ioatdca->max_requesters = slots;
+
+ /* some bios might not know to turn these on */
+ csi_fsb_control = readw(ioatdca->dca_base + IOAT_FSB_CAP_ENABLE_OFFSET);
+ if ((csi_fsb_control & IOAT_FSB_CAP_ENABLE_PREFETCH) == 0) {
+ csi_fsb_control |= IOAT_FSB_CAP_ENABLE_PREFETCH;
+ writew(csi_fsb_control,
+ ioatdca->dca_base + IOAT_FSB_CAP_ENABLE_OFFSET);
+ }
+ pcie_control = readw(ioatdca->dca_base + IOAT_PCI_CAP_ENABLE_OFFSET);
+ if ((pcie_control & IOAT_PCI_CAP_ENABLE_MEMWR) == 0) {
+ pcie_control |= IOAT_PCI_CAP_ENABLE_MEMWR;
+ writew(pcie_control,
+ ioatdca->dca_base + IOAT_PCI_CAP_ENABLE_OFFSET);
+ }
+
+
+ /* TODO version, compatibility and configuration checks */
+
+ /* copy out the APIC to DCA tag map */
+ tag_map = readl(ioatdca->dca_base + IOAT_APICID_TAG_MAP_OFFSET);
+ for (i = 0; i < 5; i++) {
+ bit = (tag_map >> (4 * i)) & 0x0f;
+ if (bit < 8)
+ ioatdca->tag_map[i] = bit | DCA_TAG_MAP_VALID;
+ else
+ ioatdca->tag_map[i] = 0;
+ }
+
+ err = register_dca_provider(dca, &pdev->dev);
+ if (err) {
+ free_dca_provider(dca);
+ return NULL;
+ }
+
+ return dca;
+}
+
+static int ioat3_dca_add_requester(struct dca_provider *dca, struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+ u16 id;
+ u16 global_req_table;
+
+ /* This implementation only supports PCI-Express */
+ if (dev->bus != &pci_bus_type)
+ return -ENODEV;
+ pdev = to_pci_dev(dev);
+ id = dcaid_from_pcidev(pdev);
+
+ if (ioatdca->requester_count == ioatdca->max_requesters)
+ return -ENODEV;
+
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == NULL) {
+ /* found an empty slot */
+ ioatdca->requester_count++;
+ ioatdca->req_slots[i].pdev = pdev;
+ ioatdca->req_slots[i].rid = id;
+ global_req_table =
+ readw(ioatdca->dca_base + IOAT3_DCA_GREQID_OFFSET);
+ writel(id | IOAT_DCA_GREQID_VALID,
+ ioatdca->iobase + global_req_table + (i * 4));
+ return i;
+ }
+ }
+ /* Error, ioatdma->requester_count is out of whack */
+ return -EFAULT;
+}
+
+static int ioat3_dca_remove_requester(struct dca_provider *dca,
+ struct device *dev)
+{
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ struct pci_dev *pdev;
+ int i;
+ u16 global_req_table;
+
+ /* This implementation only supports PCI-Express */
+ if (dev->bus != &pci_bus_type)
+ return -ENODEV;
+ pdev = to_pci_dev(dev);
+
+ for (i = 0; i < ioatdca->max_requesters; i++) {
+ if (ioatdca->req_slots[i].pdev == pdev) {
+ global_req_table =
+ readw(ioatdca->dca_base + IOAT3_DCA_GREQID_OFFSET);
+ writel(0, ioatdca->iobase + global_req_table + (i * 4));
+ ioatdca->req_slots[i].pdev = NULL;
+ ioatdca->req_slots[i].rid = 0;
+ ioatdca->requester_count--;
+ return i;
+ }
+ }
+ return -ENODEV;
+}
+
+static u8 ioat3_dca_get_tag(struct dca_provider *dca,
+ struct device *dev,
+ int cpu)
+{
+ u8 tag;
+
+ struct ioat_dca_priv *ioatdca = dca_priv(dca);
+ int i, apic_id, bit, value;
+ u8 entry;
+
+ tag = 0;
+ apic_id = cpu_physical_id(cpu);
+
+ for (i = 0; i < IOAT_TAG_MAP_LEN; i++) {
+ entry = ioatdca->tag_map[i];
+ if (entry & DCA3_TAG_MAP_BIT_TO_SEL) {
+ bit = entry &
+ ~(DCA3_TAG_MAP_BIT_TO_SEL | DCA3_TAG_MAP_BIT_TO_INV);
+ value = (apic_id & (1 << bit)) ? 1 : 0;
+ } else if (entry & DCA3_TAG_MAP_BIT_TO_INV) {
+ bit = entry & ~DCA3_TAG_MAP_BIT_TO_INV;
+ value = (apic_id & (1 << bit)) ? 0 : 1;
+ } else {
+ value = (entry & DCA3_TAG_MAP_LITERAL_VAL) ? 1 : 0;
+ }
+ tag |= (value << i);
+ }
+
+ return tag;
+}
+
+static struct dca_ops ioat3_dca_ops = {
+ .add_requester = ioat3_dca_add_requester,
+ .remove_requester = ioat3_dca_remove_requester,
+ .get_tag = ioat3_dca_get_tag,
+ .dev_managed = ioat_dca_dev_managed,
+};
+
+static int ioat3_dca_count_dca_slots(void *iobase, u16 dca_offset)
+{
+ int slots = 0;
+ u32 req;
+ u16 global_req_table;
+
+ global_req_table = readw(iobase + dca_offset + IOAT3_DCA_GREQID_OFFSET);
+ if (global_req_table == 0)
+ return 0;
+
+ do {
+ req = readl(iobase + global_req_table + (slots * sizeof(u32)));
+ slots++;
+ } while ((req & IOAT_DCA_GREQID_LASTID) == 0);
+
+ return slots;
+}
+
+struct dca_provider *ioat3_dca_init(struct pci_dev *pdev, void __iomem *iobase)
+{
+ struct dca_provider *dca;
+ struct ioat_dca_priv *ioatdca;
+ int slots;
+ int i;
+ int err;
+ u16 dca_offset;
+ u16 csi_fsb_control;
+ u16 pcie_control;
+ u8 bit;
+
+ union {
+ u64 full;
+ struct {
+ u32 low;
+ u32 high;
+ };
+ } tag_map;
+
+ if (!system_has_dca_enabled(pdev))
+ return NULL;
+
+ dca_offset = readw(iobase + IOAT_DCAOFFSET_OFFSET);
+ if (dca_offset == 0)
+ return NULL;
+
+ slots = ioat3_dca_count_dca_slots(iobase, dca_offset);
+ if (slots == 0)
+ return NULL;
+
+ dca = alloc_dca_provider(&ioat3_dca_ops,
+ sizeof(*ioatdca)
+ + (sizeof(struct ioat_dca_slot) * slots));
+ if (!dca)
+ return NULL;
+
+ ioatdca = dca_priv(dca);
+ ioatdca->iobase = iobase;
+ ioatdca->dca_base = iobase + dca_offset;
+ ioatdca->max_requesters = slots;
+
+ /* some bios might not know to turn these on */
+ csi_fsb_control = readw(ioatdca->dca_base + IOAT3_CSI_CONTROL_OFFSET);
+ if ((csi_fsb_control & IOAT3_CSI_CONTROL_PREFETCH) == 0) {
+ csi_fsb_control |= IOAT3_CSI_CONTROL_PREFETCH;
+ writew(csi_fsb_control,
+ ioatdca->dca_base + IOAT3_CSI_CONTROL_OFFSET);
+ }
+ pcie_control = readw(ioatdca->dca_base + IOAT3_PCI_CONTROL_OFFSET);
+ if ((pcie_control & IOAT3_PCI_CONTROL_MEMWR) == 0) {
+ pcie_control |= IOAT3_PCI_CONTROL_MEMWR;
+ writew(pcie_control,
+ ioatdca->dca_base + IOAT3_PCI_CONTROL_OFFSET);
+ }
+
+
+ /* TODO version, compatibility and configuration checks */
+
+ /* copy out the APIC to DCA tag map */
+ tag_map.low =
+ readl(ioatdca->dca_base + IOAT3_APICID_TAG_MAP_OFFSET_LOW);
+ tag_map.high =
+ readl(ioatdca->dca_base + IOAT3_APICID_TAG_MAP_OFFSET_HIGH);
+ for (i = 0; i < 8; i++) {
+ bit = tag_map.full >> (8 * i);
+ ioatdca->tag_map[i] = bit & DCA_TAG_MAP_MASK;
+ }
+
+ err = register_dca_provider(dca, &pdev->dev);
+ if (err) {
+ free_dca_provider(dca);
+ return NULL;
+ }
+
+ return dca;
+}
diff --git a/drivers/dma_v3/ioat_dma.c b/drivers/dma_v3/ioat_dma.c
new file mode 100644
index 0000000..f62c95f
--- /dev/null
+++ b/drivers/dma_v3/ioat_dma.c
@@ -0,0 +1,1718 @@
+/*
+ * Intel I/OAT DMA Linux driver
+ * Copyright(c) 2004 - 2007 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ */
+
+/*
+ * This driver supports an Intel I/OAT DMA engine, which does asynchronous
+ * copy operations.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/interrupt.h>
+#include <linux/dmaengine.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/workqueue.h>
+#include "ioatdma.h"
+#include "ioatdma_registers.h"
+#include "ioatdma_hw.h"
+
+#define to_ioat_chan(chan) container_of(chan, struct ioat_dma_chan, common)
+#define to_ioatdma_device(dev) container_of(dev, struct ioatdma_device, common)
+#define to_ioat_desc(lh) container_of(lh, struct ioat_desc_sw, node)
+#define tx_to_ioat_desc(tx) container_of(tx, struct ioat_desc_sw, async_tx)
+
+#define chan_num(ch) ((int)((ch)->reg_base - (ch)->device->reg_base) / 0x80)
+static int ioat_pending_level = 4;
+module_param(ioat_pending_level, int, 0644);
+MODULE_PARM_DESC(ioat_pending_level,
+ "high-water mark for pushing ioat descriptors (default: 4)");
+
+#define RESET_DELAY msecs_to_jiffies(100)
+#define WATCHDOG_DELAY round_jiffies(msecs_to_jiffies(2000))
+static void ioat_dma_chan_reset_part2(struct work_struct *work);
+static void ioat_dma_chan_watchdog(struct work_struct *work);
+
+/*
+ * workaround for IOAT ver.3.0 null descriptor issue
+ * (channel returns error when size is 0)
+ */
+#define NULL_DESC_BUFFER_SIZE 1
+
+/* internal functions */
+static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat_chan);
+static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan);
+
+static struct ioat_desc_sw *
+ioat1_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan);
+static struct ioat_desc_sw *
+ioat2_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan);
+
+static inline struct ioat_dma_chan *ioat_lookup_chan_by_index(
+ struct ioatdma_device *device,
+ int index)
+{
+ return device->idx[index];
+}
+
+/**
+ * ioat_dma_do_interrupt - handler used for single vector interrupt mode
+ * @irq: interrupt id
+ * @data: interrupt data
+ */
+static irqreturn_t ioat_dma_do_interrupt(int irq, void *data,
+ struct pt_regs *fp)
+{
+ struct ioatdma_device *instance = data;
+ struct ioat_dma_chan *ioat_chan;
+ unsigned long attnstatus;
+ int bit;
+ u8 intrctrl;
+
+ intrctrl = readb(instance->reg_base + IOAT_INTRCTRL_OFFSET);
+
+ if (!(intrctrl & IOAT_INTRCTRL_MASTER_INT_EN))
+ return IRQ_NONE;
+
+ if (!(intrctrl & IOAT_INTRCTRL_INT_STATUS)) {
+ writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
+ return IRQ_NONE;
+ }
+
+ attnstatus = readl(instance->reg_base + IOAT_ATTNSTATUS_OFFSET);
+ for((bit = find_first_bit(&attnstatus, BITS_PER_LONG));
+ (bit) < (BITS_PER_LONG);
+ (bit) = find_next_bit((&attnstatus), (BITS_PER_LONG), (bit)+1)) {
+ ioat_chan = ioat_lookup_chan_by_index(instance, bit);
+ tasklet_schedule(&ioat_chan->cleanup_task);
+ }
+
+ writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
+ return IRQ_HANDLED;
+}
+
+/**
+ * ioat_dma_do_interrupt_msix - handler used for vector-per-channel interrupt mode
+ * @irq: interrupt id
+ * @data: interrupt data
+ */
+static irqreturn_t ioat_dma_do_interrupt_msix(int irq, void *data,
+ struct pt_regs *fp)
+{
+ struct ioat_dma_chan *ioat_chan = data;
+
+ tasklet_schedule(&ioat_chan->cleanup_task);
+
+ return IRQ_HANDLED;
+}
+
+static void ioat_dma_cleanup_tasklet(unsigned long data);
+
+/**
+ * ioat_dma_enumerate_channels - find and initialize the device's channels
+ * @device: the device to be enumerated
+ */
+static int ioat_dma_enumerate_channels(struct ioatdma_device *device)
+{
+ u8 xfercap_scale;
+ u32 xfercap;
+ int i;
+ struct ioat_dma_chan *ioat_chan;
+
+ /*
+ * IOAT ver.3 workarounds
+ */
+ if (device->version == IOAT_VER_3_0) {
+ u32 chan_err_mask;
+ u16 dev_id;
+ u32 dmauncerrsts;
+
+ /*
+ * Write CHANERRMSK_INT with 3E07h to mask out the errors
+ * that can cause stability issues for IOAT ver.3
+ */
+ chan_err_mask = 0x3E07;
+ pci_write_config_dword(device->pdev,
+ IOAT_PCI_CHANERRMASK_INT_OFFSET,
+ chan_err_mask);
+
+ /*
+ * Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit
+ * (workaround for spurious config parity error after restart)
+ */
+ pci_read_config_word(device->pdev,
+ IOAT_PCI_DEVICE_ID_OFFSET,
+ &dev_id);
+ if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0) {
+ dmauncerrsts = 0x10;
+ pci_write_config_dword(device->pdev,
+ IOAT_PCI_DMAUNCERRSTS_OFFSET,
+ dmauncerrsts);
+ }
+ }
+
+ device->common.chancnt = readb(device->reg_base + IOAT_CHANCNT_OFFSET);
+ xfercap_scale = readb(device->reg_base + IOAT_XFERCAP_OFFSET);
+ xfercap = (xfercap_scale == 0 ? -1 : (1UL << xfercap_scale));
+
+ for (i = 0; i < device->common.chancnt; i++) {
+ ioat_chan = kzalloc(sizeof(*ioat_chan), GFP_KERNEL);
+ if (!ioat_chan) {
+ device->common.chancnt = i;
+ break;
+ }
+
+ ioat_chan->device = device;
+ ioat_chan->reg_base = device->reg_base + (0x80 * (i + 1));
+ ioat_chan->xfercap = xfercap;
+ ioat_chan->desccount = 0;
+ INIT_WORK(&ioat_chan->work, (void*)ioat_dma_chan_reset_part2,
+ &ioat_chan->work);
+ if (ioat_chan->device->version != IOAT_VER_1_2) {
+ writel(IOAT_DCACTRL_CMPL_WRITE_ENABLE
+ | IOAT_DMA_DCA_ANY_CPU,
+ ioat_chan->reg_base + IOAT_DCACTRL_OFFSET);
+ }
+ spin_lock_init(&ioat_chan->cleanup_lock);
+ spin_lock_init(&ioat_chan->desc_lock);
+ INIT_LIST_HEAD(&ioat_chan->free_desc);
+ INIT_LIST_HEAD(&ioat_chan->used_desc);
+ /* This should be made common somewhere in dmaengine.c */
+ ioat_chan->common.device_v3 = &device->common;
+ list_add_tail(&ioat_chan->common.device_node,
+ &device->common.channels);
+ device->idx[i] = ioat_chan;
+ tasklet_init(&ioat_chan->cleanup_task,
+ ioat_dma_cleanup_tasklet,
+ (unsigned long) ioat_chan);
+ tasklet_disable(&ioat_chan->cleanup_task);
+ }
+ return device->common.chancnt;
+}
+
+/**
+ * ioat_dma_memcpy_issue_pending - push potentially unrecognized appended
+ * descriptors to hw
+ * @chan: DMA channel handle
+ */
+static inline void __ioat1_dma_memcpy_issue_pending(
+ struct ioat_dma_chan *ioat_chan)
+{
+ ioat_chan->pending = 0;
+ writeb(IOAT_CHANCMD_APPEND, ioat_chan->reg_base + IOAT1_CHANCMD_OFFSET);
+}
+
+static void ioat1_dma_memcpy_issue_pending(struct dma_chan *chan)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+
+ if (ioat_chan->pending > 0) {
+ spin_lock_bh(&ioat_chan->desc_lock);
+ __ioat1_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ }
+}
+
+static inline void __ioat2_dma_memcpy_issue_pending(
+ struct ioat_dma_chan *ioat_chan)
+{
+ ioat_chan->pending = 0;
+ writew(ioat_chan->dmacount,
+ ioat_chan->reg_base + IOAT_CHAN_DMACOUNT_OFFSET);
+}
+
+static void ioat2_dma_memcpy_issue_pending(struct dma_chan *chan)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+
+ if (ioat_chan->pending > 0) {
+ spin_lock_bh(&ioat_chan->desc_lock);
+ __ioat2_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ }
+}
+
+
+/**
+ * ioat_dma_chan_reset_part2 - reinit the channel after a reset
+ */
+static void ioat_dma_chan_reset_part2(struct work_struct *work)
+{
+ struct ioat_dma_chan *ioat_chan =
+ container_of(work, struct ioat_dma_chan, work);
+ struct ioat_desc_sw *desc;
+
+ spin_lock_bh(&ioat_chan->cleanup_lock);
+ spin_lock_bh(&ioat_chan->desc_lock);
+
+ ioat_chan->completion_virt->low = 0;
+ ioat_chan->completion_virt->high = 0;
+ ioat_chan->pending = 0;
+
+ /*
+ * count the descriptors waiting, and be sure to do it
+ * right for both the CB1 line and the CB2 ring
+ */
+ ioat_chan->dmacount = 0;
+ if (ioat_chan->used_desc.prev) {
+ desc = to_ioat_desc(ioat_chan->used_desc.prev);
+ do {
+ ioat_chan->dmacount++;
+ desc = to_ioat_desc(desc->node.next);
+ } while (&desc->node != ioat_chan->used_desc.next);
+ }
+
+ /*
+ * write the new starting descriptor address
+ * this puts channel engine into ARMED state
+ */
+ desc = to_ioat_desc(ioat_chan->used_desc.prev);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_HIGH);
+
+ writeb(IOAT_CHANCMD_START, ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ break;
+ case IOAT_VER_2_0:
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH);
+
+ /* tell the engine to go with what's left to be done */
+ writew(ioat_chan->dmacount,
+ ioat_chan->reg_base + IOAT_CHAN_DMACOUNT_OFFSET);
+
+ break;
+ }
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d reset - %d descs waiting, %d total desc\n",
+ chan_num(ioat_chan), ioat_chan->dmacount, ioat_chan->desccount);
+
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+}
+
+/**
+ * ioat_dma_reset_channel - restart a channel
+ * @ioat_chan: IOAT DMA channel handle
+ */
+static void ioat_dma_reset_channel(struct ioat_dma_chan *ioat_chan)
+{
+ u32 chansts, chanerr;
+
+ if (!ioat_chan->used_desc.prev)
+ return;
+
+ chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
+ chansts = (ioat_chan->completion_virt->low
+ & IOAT_CHANSTS_DMA_TRANSFER_STATUS);
+ if (chanerr) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d, CHANSTS = 0x%08x CHANERR = 0x%04x, clearing\n",
+ chan_num(ioat_chan), chansts, chanerr);
+ writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
+ }
+
+ /*
+ * whack it upside the head with a reset
+ * and wait for things to settle out.
+ * force the pending count to a really big negative
+ * to make sure no one forces an issue_pending
+ * while we're waiting.
+ */
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+ ioat_chan->pending = INT_MIN;
+ writeb(IOAT_CHANCMD_RESET,
+ ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ /* schedule the 2nd half instead of sleeping a long time */
+ schedule_delayed_work(&ioat_chan->work, RESET_DELAY);
+}
+
+/**
+ * ioat_dma_chan_watchdog - watch for stuck channels
+ */
+static void ioat_dma_chan_watchdog(struct work_struct *work)
+{
+ struct ioatdma_device *device =
+ container_of(work, struct ioatdma_device, work);
+ struct ioat_dma_chan *ioat_chan;
+ int i;
+
+ union {
+ u64 full;
+ struct {
+ u32 low;
+ u32 high;
+ };
+ } completion_hw;
+ unsigned long compl_desc_addr_hw;
+
+ for (i = 0; i < device->common.chancnt; i++) {
+ ioat_chan = ioat_lookup_chan_by_index(device, i);
+
+ if (ioat_chan->device->version == IOAT_VER_1_2
+ /* have we started processing anything yet */
+ && ioat_chan->last_completion
+ /* have we completed any since last watchdog cycle? */
+ && (ioat_chan->last_completion ==
+ ioat_chan->watchdog_completion)
+ /* has TCP stuck on one cookie since last watchdog? */
+ && (ioat_chan->watchdog_tcp_cookie ==
+ ioat_chan->watchdog_last_tcp_cookie)
+ && (ioat_chan->watchdog_tcp_cookie !=
+ ioat_chan->completed_cookie)
+ /* is there something in the chain to be processed? */
+ /* CB1 chain always has at least the last one processed */
+ && (ioat_chan->used_desc.prev != ioat_chan->used_desc.next)
+ && ioat_chan->pending == 0) {
+
+ /*
+ * check CHANSTS register for completed
+ * descriptor address.
+ * if it is different than completion writeback,
+ * it is not zero
+ * and it has changed since the last watchdog
+ * we can assume that channel
+ * is still working correctly
+ * and the problem is in completion writeback.
+ * update completion writeback
+ * with actual CHANSTS value
+ * else
+ * try resetting the channel
+ */
+
+ completion_hw.low = readl(ioat_chan->reg_base +
+ IOAT_CHANSTS_OFFSET_LOW(ioat_chan->device->version));
+ completion_hw.high = readl(ioat_chan->reg_base +
+ IOAT_CHANSTS_OFFSET_HIGH(ioat_chan->device->version));
+#if (BITS_PER_LONG == 64)
+ compl_desc_addr_hw =
+ completion_hw.full
+ & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
+#else
+ compl_desc_addr_hw =
+ completion_hw.low & IOAT_LOW_COMPLETION_MASK;
+#endif
+
+ if ((compl_desc_addr_hw != 0)
+ && (compl_desc_addr_hw != ioat_chan->watchdog_completion)
+ && (compl_desc_addr_hw != ioat_chan->last_compl_desc_addr_hw)) {
+ ioat_chan->last_compl_desc_addr_hw = compl_desc_addr_hw;
+ ioat_chan->completion_virt->low = completion_hw.low;
+ ioat_chan->completion_virt->high = completion_hw.high;
+ } else {
+ ioat_dma_reset_channel(ioat_chan);
+ ioat_chan->watchdog_completion = 0;
+ ioat_chan->last_compl_desc_addr_hw = 0;
+ }
+
+ /*
+ * for version 2.0 if there are descriptors yet to be processed
+ * and the last completed hasn't changed since the last watchdog
+ * if they haven't hit the pending level
+ * issue the pending to push them through
+ * else
+ * try resetting the channel
+ */
+ } else if (ioat_chan->device->version == IOAT_VER_2_0
+ && ioat_chan->used_desc.prev
+ && ioat_chan->last_completion
+ && ioat_chan->last_completion == ioat_chan->watchdog_completion) {
+
+ if (ioat_chan->pending < ioat_pending_level)
+ ioat2_dma_memcpy_issue_pending(&ioat_chan->common);
+ else {
+ ioat_dma_reset_channel(ioat_chan);
+ ioat_chan->watchdog_completion = 0;
+ }
+ } else {
+ ioat_chan->last_compl_desc_addr_hw = 0;
+ ioat_chan->watchdog_completion
+ = ioat_chan->last_completion;
+ }
+
+ ioat_chan->watchdog_last_tcp_cookie =
+ ioat_chan->watchdog_tcp_cookie;
+ }
+
+ schedule_delayed_work(&device->work, WATCHDOG_DELAY);
+}
+
+static dma_cookie_t ioat1_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(tx->chan);
+ struct ioat_desc_sw *first = tx_to_ioat_desc(tx);
+ struct ioat_desc_sw *prev, *new;
+ struct ioat_dma_descriptor *hw;
+ dma_cookie_t cookie;
+ LIST_HEAD(new_chain);
+ u32 copy;
+ size_t len;
+ dma_addr_t src, dst;
+ unsigned long orig_flags;
+ unsigned int desc_count = 0;
+
+ /* src and dest and len are stored in the initial descriptor */
+ len = first->len;
+ src = first->src;
+ dst = first->dst;
+ orig_flags = first->async_tx.flags;
+ new = first;
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+ prev = to_ioat_desc(ioat_chan->used_desc.prev);
+ prefetch(prev->hw);
+ do {
+ copy = min_t(size_t, len, ioat_chan->xfercap);
+
+ async_tx_ack(&new->async_tx);
+
+ hw = new->hw;
+ hw->size = copy;
+ hw->ctl = 0;
+ hw->src_addr = src;
+ hw->dst_addr = dst;
+ hw->next = 0;
+
+ /* chain together the physical address list for the HW */
+ wmb();
+ prev->hw->next = (u64) new->async_tx.phys;
+
+ len -= copy;
+ dst += copy;
+ src += copy;
+
+ list_add_tail(&new->node, &new_chain);
+ desc_count++;
+ prev = new;
+ } while (len && (new = ioat1_dma_get_next_descriptor(ioat_chan)));
+
+ if (!new) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "tx submit failed\n");
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ return -ENOMEM;
+ }
+
+ hw->ctl = IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
+ if (new->async_tx.callback) {
+ hw->ctl |= IOAT_DMA_DESCRIPTOR_CTL_INT_GN;
+ if (first != new) {
+ /* move callback into to last desc */
+ new->async_tx.callback = first->async_tx.callback;
+ new->async_tx.callback_param
+ = first->async_tx.callback_param;
+ first->async_tx.callback = NULL;
+ first->async_tx.callback_param = NULL;
+ }
+ }
+
+ new->tx_cnt = desc_count;
+ new->async_tx.flags = orig_flags; /* client is in control of this ack */
+
+ /* store the original values for use in later cleanup */
+ if (new != first) {
+ new->src = first->src;
+ new->dst = first->dst;
+ new->len = first->len;
+ }
+
+ /* cookie incr and addition to used_list must be atomic */
+ cookie = ioat_chan->common.cookie;
+ cookie++;
+ if (cookie < 0)
+ cookie = 1;
+ ioat_chan->common.cookie = new->async_tx.cookie = cookie;
+
+ /* write address into NextDescriptor field of last desc in chain */
+ to_ioat_desc(ioat_chan->used_desc.prev)->hw->next =
+ first->async_tx.phys;
+ __list_splice(&new_chain, ioat_chan->used_desc.prev);
+
+ ioat_chan->dmacount += desc_count;
+ ioat_chan->pending += desc_count;
+ if (ioat_chan->pending >= ioat_pending_level)
+ __ioat1_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ return cookie;
+}
+
+static dma_cookie_t ioat2_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(tx->chan);
+ struct ioat_desc_sw *first = tx_to_ioat_desc(tx);
+ struct ioat_desc_sw *new;
+ struct ioat_dma_descriptor *hw;
+ dma_cookie_t cookie;
+ u32 copy;
+ size_t len;
+ dma_addr_t src, dst;
+ unsigned long orig_flags;
+ unsigned int desc_count = 0;
+
+ /* src and dest and len are stored in the initial descriptor */
+ len = first->len;
+ src = first->src;
+ dst = first->dst;
+ orig_flags = first->async_tx.flags;
+ new = first;
+
+ /*
+ * ioat_chan->desc_lock is still in force in version 2 path
+ * it gets unlocked at end of this function
+ */
+ do {
+ copy = min_t(size_t, len, ioat_chan->xfercap);
+
+ async_tx_ack(&new->async_tx);
+
+ hw = new->hw;
+ hw->size = copy;
+ hw->ctl = 0;
+ hw->src_addr = src;
+ hw->dst_addr = dst;
+
+ len -= copy;
+ dst += copy;
+ src += copy;
+ desc_count++;
+ } while (len && (new = ioat2_dma_get_next_descriptor(ioat_chan)));
+
+ if (!new) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "tx submit failed\n");
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ return -ENOMEM;
+ }
+
+ hw->ctl |= IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
+ if (new->async_tx.callback) {
+ hw->ctl |= IOAT_DMA_DESCRIPTOR_CTL_INT_GN;
+ if (first != new) {
+ /* move callback into to last desc */
+ new->async_tx.callback = first->async_tx.callback;
+ new->async_tx.callback_param
+ = first->async_tx.callback_param;
+ first->async_tx.callback = NULL;
+ first->async_tx.callback_param = NULL;
+ }
+ }
+
+ new->tx_cnt = desc_count;
+ new->async_tx.flags = orig_flags; /* client is in control of this ack */
+
+ /* store the original values for use in later cleanup */
+ if (new != first) {
+ new->src = first->src;
+ new->dst = first->dst;
+ new->len = first->len;
+ }
+
+ /* cookie incr and addition to used_list must be atomic */
+ cookie = ioat_chan->common.cookie;
+ cookie++;
+ if (cookie < 0)
+ cookie = 1;
+ ioat_chan->common.cookie = new->async_tx.cookie = cookie;
+
+ ioat_chan->dmacount += desc_count;
+ ioat_chan->pending += desc_count;
+ if (ioat_chan->pending >= ioat_pending_level)
+ __ioat2_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ return cookie;
+}
+
+/**
+ * ioat_dma_alloc_descriptor - allocate and return a sw and hw descriptor pair
+ * @ioat_chan: the channel supplying the memory pool for the descriptors
+ * @flags: allocation flags
+ */
+static struct ioat_desc_sw *ioat_dma_alloc_descriptor(
+ struct ioat_dma_chan *ioat_chan,
+ gfp_t flags)
+{
+ struct ioat_dma_descriptor *desc;
+ struct ioat_desc_sw *desc_sw;
+ struct ioatdma_device *ioatdma_device;
+ dma_addr_t phys;
+
+ ioatdma_device = to_ioatdma_device(ioat_chan->common.device_v3);
+ desc = pci_pool_alloc(ioatdma_device->dma_pool, flags, &phys);
+ if (unlikely(!desc))
+ return NULL;
+
+ desc_sw = kzalloc(sizeof(*desc_sw), flags);
+ if (unlikely(!desc_sw)) {
+ pci_pool_free(ioatdma_device->dma_pool, desc, phys);
+ return NULL;
+ }
+
+ memset(desc, 0, sizeof(*desc));
+ dma_async_tx_descriptor_init_v3(&desc_sw->async_tx, &ioat_chan->common);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ desc_sw->async_tx.tx_submit = ioat1_tx_submit;
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ desc_sw->async_tx.tx_submit = ioat2_tx_submit;
+ break;
+ }
+ INIT_LIST_HEAD(&desc_sw->async_tx.tx_list);
+
+ desc_sw->hw = desc;
+ desc_sw->async_tx.phys = phys;
+
+ return desc_sw;
+}
+
+static int ioat_initial_desc_count = 256;
+module_param(ioat_initial_desc_count, int, 0644);
+MODULE_PARM_DESC(ioat_initial_desc_count,
+ "initial descriptors per channel (default: 256)");
+
+/**
+ * ioat2_dma_massage_chan_desc - link the descriptors into a circle
+ * @ioat_chan: the channel to be massaged
+ */
+static void ioat2_dma_massage_chan_desc(struct ioat_dma_chan *ioat_chan)
+{
+ struct ioat_desc_sw *desc, *_desc;
+
+ /* setup used_desc */
+ ioat_chan->used_desc.next = ioat_chan->free_desc.next;
+ ioat_chan->used_desc.prev = NULL;
+
+ /* pull free_desc out of the circle so that every node is a hw
+ * descriptor, but leave it pointing to the list
+ */
+ ioat_chan->free_desc.prev->next = ioat_chan->free_desc.next;
+ ioat_chan->free_desc.next->prev = ioat_chan->free_desc.prev;
+
+ /* circle link the hw descriptors */
+ desc = to_ioat_desc(ioat_chan->free_desc.next);
+ desc->hw->next = to_ioat_desc(desc->node.next)->async_tx.phys;
+ list_for_each_entry_safe(desc, _desc, ioat_chan->free_desc.next, node) {
+ desc->hw->next = to_ioat_desc(desc->node.next)->async_tx.phys;
+ }
+}
+
+/**
+ * ioat_dma_alloc_chan_resources - returns the number of allocated descriptors
+ * @chan: the channel to be filled out
+ */
+static int ioat_dma_alloc_chan_resources(struct dma_chan *chan,
+ struct dma_client *client)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ struct ioat_desc_sw *desc;
+ u16 chanctrl;
+ u32 chanerr;
+ int i;
+ LIST_HEAD(tmp_list);
+
+ /* have we already been set up? */
+ if (!list_empty(&ioat_chan->free_desc))
+ return ioat_chan->desccount;
+
+ /* Setup register to interrupt and write completion status on error */
+ chanctrl = IOAT_CHANCTRL_ERR_INT_EN |
+ IOAT_CHANCTRL_ANY_ERR_ABORT_EN |
+ IOAT_CHANCTRL_ERR_COMPLETION_EN;
+ writew(chanctrl, ioat_chan->reg_base + IOAT_CHANCTRL_OFFSET);
+
+ chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
+ if (chanerr) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "CHANERR = %x, clearing\n", chanerr);
+ writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
+ }
+
+ /* Allocate descriptors */
+ for (i = 0; i < ioat_initial_desc_count; i++) {
+ desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL);
+ if (!desc) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "Only %d initial descriptors\n", i);
+ break;
+ }
+ list_add_tail(&desc->node, &tmp_list);
+ }
+ spin_lock_bh(&ioat_chan->desc_lock);
+ ioat_chan->desccount = i;
+ list_splice(&tmp_list, &ioat_chan->free_desc);
+ if (ioat_chan->device->version != IOAT_VER_1_2)
+ ioat2_dma_massage_chan_desc(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ /* allocate a completion writeback area */
+ /* doing 2 32bit writes to mmio since 1 64b write doesn't work */
+ ioat_chan->completion_virt =
+ pci_pool_alloc(ioat_chan->device->completion_pool,
+ GFP_KERNEL,
+ &ioat_chan->completion_addr);
+ memset(ioat_chan->completion_virt, 0,
+ sizeof(*ioat_chan->completion_virt));
+ writel(((u64) ioat_chan->completion_addr) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_LOW);
+ writel(((u64) ioat_chan->completion_addr) >> 32,
+ ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH);
+
+ tasklet_enable(&ioat_chan->cleanup_task);
+ ioat_dma_start_null_desc(ioat_chan); /* give chain to dma device */
+ return ioat_chan->desccount;
+}
+
+/**
+ * ioat_dma_free_chan_resources - release all the descriptors
+ * @chan: the channel to be cleaned
+ */
+static void ioat_dma_free_chan_resources(struct dma_chan *chan)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ struct ioatdma_device *ioatdma_device =
+ to_ioatdma_device(chan->device_v3);
+ struct ioat_desc_sw *desc, *_desc;
+ int in_use_descs = 0;
+
+ tasklet_disable(&ioat_chan->cleanup_task);
+ ioat_dma_memcpy_cleanup(ioat_chan);
+
+ /* Delay 100ms after reset to allow internal DMA logic to quiesce
+ * before removing DMA descriptor resources.
+ */
+ writeb(IOAT_CHANCMD_RESET,
+ ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ mdelay(100);
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ list_for_each_entry_safe(desc, _desc,
+ &ioat_chan->used_desc, node) {
+ in_use_descs++;
+ list_del(&desc->node);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ }
+ list_for_each_entry_safe(desc, _desc,
+ &ioat_chan->free_desc, node) {
+ list_del(&desc->node);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ }
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ list_for_each_entry_safe(desc, _desc,
+ ioat_chan->free_desc.next, node) {
+ list_del(&desc->node);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ }
+ desc = to_ioat_desc(ioat_chan->free_desc.next);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ INIT_LIST_HEAD(&ioat_chan->free_desc);
+ INIT_LIST_HEAD(&ioat_chan->used_desc);
+ break;
+ }
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ pci_pool_free(ioatdma_device->completion_pool,
+ ioat_chan->completion_virt,
+ ioat_chan->completion_addr);
+
+ /* one is ok since we left it on there on purpose */
+ if (in_use_descs > 1)
+ dev_err(&ioat_chan->device->pdev->dev,
+ "Freeing %d in use descriptors!\n",
+ in_use_descs - 1);
+
+ ioat_chan->last_completion = ioat_chan->completion_addr = 0;
+ ioat_chan->pending = 0;
+ ioat_chan->dmacount = 0;
+ ioat_chan->watchdog_completion = 0;
+ ioat_chan->last_compl_desc_addr_hw = 0;
+ ioat_chan->watchdog_tcp_cookie =
+ ioat_chan->watchdog_last_tcp_cookie = 0;
+}
+
+/**
+ * ioat_dma_get_next_descriptor - return the next available descriptor
+ * @ioat_chan: IOAT DMA channel handle
+ *
+ * Gets the next descriptor from the chain, and must be called with the
+ * channel's desc_lock held. Allocates more descriptors if the channel
+ * has run out.
+ */
+static struct ioat_desc_sw *
+ioat1_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan)
+{
+ struct ioat_desc_sw *new;
+
+ if (!list_empty(&ioat_chan->free_desc)) {
+ new = to_ioat_desc(ioat_chan->free_desc.next);
+ list_del(&new->node);
+ } else {
+ /* try to get another desc */
+ new = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC);
+ if (!new) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "alloc failed\n");
+ return NULL;
+ }
+ }
+
+ prefetch(new->hw);
+ return new;
+}
+
+static struct ioat_desc_sw *
+ioat2_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan)
+{
+ struct ioat_desc_sw *new;
+
+ /*
+ * used.prev points to where to start processing
+ * used.next points to next free descriptor
+ * if used.prev == NULL, there are none waiting to be processed
+ * if used.next == used.prev.prev, there is only one free descriptor,
+ * and we need to use it to as a noop descriptor before
+ * linking in a new set of descriptors, since the device
+ * has probably already read the pointer to it
+ */
+ if (ioat_chan->used_desc.prev &&
+ ioat_chan->used_desc.next == ioat_chan->used_desc.prev->prev) {
+
+ struct ioat_desc_sw *desc;
+ struct ioat_desc_sw *noop_desc;
+ int i;
+
+ /* set up the noop descriptor */
+ noop_desc = to_ioat_desc(ioat_chan->used_desc.next);
+ /* set size to non-zero value (channel returns error when size is 0) */
+ noop_desc->hw->size = NULL_DESC_BUFFER_SIZE;
+ noop_desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL;
+ noop_desc->hw->src_addr = 0;
+ noop_desc->hw->dst_addr = 0;
+
+ ioat_chan->used_desc.next = ioat_chan->used_desc.next->next;
+ ioat_chan->pending++;
+ ioat_chan->dmacount++;
+
+ /* try to get a few more descriptors */
+ for (i = 16; i; i--) {
+ desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC);
+ if (!desc) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "alloc failed\n");
+ break;
+ }
+ list_add_tail(&desc->node, ioat_chan->used_desc.next);
+
+ desc->hw->next
+ = to_ioat_desc(desc->node.next)->async_tx.phys;
+ to_ioat_desc(desc->node.prev)->hw->next
+ = desc->async_tx.phys;
+ ioat_chan->desccount++;
+ }
+
+ ioat_chan->used_desc.next = noop_desc->node.next;
+ }
+ new = to_ioat_desc(ioat_chan->used_desc.next);
+ prefetch(new);
+ ioat_chan->used_desc.next = new->node.next;
+
+ if (ioat_chan->used_desc.prev == NULL)
+ ioat_chan->used_desc.prev = &new->node;
+
+ prefetch(new->hw);
+ return new;
+}
+
+static struct ioat_desc_sw *ioat_dma_get_next_descriptor(
+ struct ioat_dma_chan *ioat_chan)
+{
+ if (!ioat_chan)
+ return NULL;
+
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ return ioat1_dma_get_next_descriptor(ioat_chan);
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ return ioat2_dma_get_next_descriptor(ioat_chan);
+ break;
+ }
+ return NULL;
+}
+
+static struct dma_async_tx_descriptor *ioat1_dma_prep_memcpy(
+ struct dma_chan *chan,
+ dma_addr_t dma_dest,
+ dma_addr_t dma_src,
+ size_t len,
+ unsigned long flags)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ struct ioat_desc_sw *new;
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+ new = ioat_dma_get_next_descriptor(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ if (new) {
+ new->len = len;
+ new->dst = dma_dest;
+ new->src = dma_src;
+ new->async_tx.flags = flags;
+ return &new->async_tx;
+ } else {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d - get_next_desc failed: %d descs waiting, %d total desc\n",
+ chan_num(ioat_chan), ioat_chan->dmacount, ioat_chan->desccount);
+ return NULL;
+ }
+}
+
+static struct dma_async_tx_descriptor *ioat2_dma_prep_memcpy(
+ struct dma_chan *chan,
+ dma_addr_t dma_dest,
+ dma_addr_t dma_src,
+ size_t len,
+ unsigned long flags)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ struct ioat_desc_sw *new;
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+ new = ioat2_dma_get_next_descriptor(ioat_chan);
+
+ /*
+ * leave ioat_chan->desc_lock set in ioat 2 path
+ * it will get unlocked at end of tx_submit
+ */
+
+ if (new) {
+ new->len = len;
+ new->dst = dma_dest;
+ new->src = dma_src;
+ new->async_tx.flags = flags;
+ return &new->async_tx;
+ } else {
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d - get_next_desc failed: %d descs waiting, %d total desc\n",
+ chan_num(ioat_chan), ioat_chan->dmacount, ioat_chan->desccount);
+ return NULL;
+ }
+}
+
+static void ioat_dma_cleanup_tasklet(unsigned long data)
+{
+ struct ioat_dma_chan *chan = (void *)data;
+ ioat_dma_memcpy_cleanup(chan);
+ writew(IOAT_CHANCTRL_INT_DISABLE,
+ chan->reg_base + IOAT_CHANCTRL_OFFSET);
+}
+
+static void
+ioat_dma_unmap(struct ioat_dma_chan *ioat_chan, struct ioat_desc_sw *desc)
+{
+ /*
+ * yes we are unmapping both _page and _single
+ * alloc'd regions with unmap_page. Is this
+ * *really* that bad?
+ */
+ if (!(desc->async_tx.flags & DMA_COMPL_SKIP_DEST_UNMAP))
+ pci_unmap_page(ioat_chan->device->pdev,
+ pci_unmap_addr(desc, dst),
+ pci_unmap_len(desc, len),
+ PCI_DMA_FROMDEVICE);
+
+ if (!(desc->async_tx.flags & DMA_COMPL_SKIP_SRC_UNMAP))
+ pci_unmap_page(ioat_chan->device->pdev,
+ pci_unmap_addr(desc, src),
+ pci_unmap_len(desc, len),
+ PCI_DMA_TODEVICE);
+}
+
+/**
+ * ioat_dma_memcpy_cleanup - cleanup up finished descriptors
+ * @chan: ioat channel to be cleaned up
+ */
+static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan)
+{
+ unsigned long phys_complete;
+ struct ioat_desc_sw *desc, *_desc;
+ dma_cookie_t cookie = 0;
+ unsigned long desc_phys;
+ struct ioat_desc_sw *latest_desc;
+
+ prefetch(ioat_chan->completion_virt);
+
+ if (!spin_trylock_bh(&ioat_chan->cleanup_lock))
+ return;
+
+ /* The completion writeback can happen at any time,
+ so reads by the driver need to be atomic operations
+ The descriptor physical addresses are limited to 32-bits
+ when the CPU can only do a 32-bit mov */
+
+#if (BITS_PER_LONG == 64)
+ phys_complete =
+ ioat_chan->completion_virt->full
+ & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
+#else
+ phys_complete =
+ ioat_chan->completion_virt->low & IOAT_LOW_COMPLETION_MASK;
+#endif
+
+ if ((ioat_chan->completion_virt->full
+ & IOAT_CHANSTS_DMA_TRANSFER_STATUS) ==
+ IOAT_CHANSTS_DMA_TRANSFER_STATUS_HALTED) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "Channel halted, chanerr = %x\n",
+ readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET));
+
+ /* TODO do something to salvage the situation */
+ }
+
+ if (phys_complete == ioat_chan->last_completion) {
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+ /*
+ * perhaps we're stuck so hard that the watchdog can't go off?
+ * try to catch it after 2 seconds
+ */
+ if (ioat_chan->device->version != IOAT_VER_3_0) {
+ if (time_after(jiffies,
+ ioat_chan->last_completion_time + HZ*WATCHDOG_DELAY)) {
+ ioat_dma_chan_watchdog(&(ioat_chan->device->work));
+ ioat_chan->last_completion_time = jiffies;
+ }
+ }
+ return;
+ }
+ ioat_chan->last_completion_time = jiffies;
+
+ cookie = 0;
+ if (!spin_trylock_bh(&ioat_chan->desc_lock)) {
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+ return;
+ }
+
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ list_for_each_entry_safe(desc, _desc,
+ &ioat_chan->used_desc, node) {
+
+ /*
+ * Incoming DMA requests may use multiple descriptors,
+ * due to exceeding xfercap, perhaps. If so, only the
+ * last one will have a cookie, and require unmapping.
+ */
+ if (desc->async_tx.cookie) {
+ cookie = desc->async_tx.cookie;
+ ioat_dma_unmap(ioat_chan, desc);
+ if (desc->async_tx.callback) {
+ desc->async_tx.callback(desc->async_tx.callback_param);
+ desc->async_tx.callback = NULL;
+ }
+ }
+
+ if (desc->async_tx.phys != phys_complete) {
+ /*
+ * a completed entry, but not the last, so clean
+ if the client is done with the descriptor
+ */
+ if (async_tx_test_ack(&desc->async_tx)) {
+ list_del(&desc->node);
+ list_add_tail(&desc->node,
+ &ioat_chan->free_desc);
+ } else
+ desc->async_tx.cookie = 0;
+ } else {
+ /*
+ * last used desc. Do not remove, so we can
+ * append from it, but don't look at it next
+ * time, either
+ */
+ desc->async_tx.cookie = 0;
+
+ /* TODO check status bits? */
+ break;
+ }
+ }
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ /* has some other thread has already cleaned up? */
+ if (ioat_chan->used_desc.prev == NULL)
+ break;
+
+ /* work backwards to find latest finished desc */
+ desc = to_ioat_desc(ioat_chan->used_desc.next);
+ latest_desc = NULL;
+ do {
+ desc = to_ioat_desc(desc->node.prev);
+ desc_phys = (unsigned long)desc->async_tx.phys
+ & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
+ if (desc_phys == phys_complete) {
+ latest_desc = desc;
+ break;
+ }
+ } while (&desc->node != ioat_chan->used_desc.prev);
+
+ if (latest_desc != NULL) {
+
+ /* work forwards to clear finished descriptors */
+ for (desc = to_ioat_desc(ioat_chan->used_desc.prev);
+ &desc->node != latest_desc->node.next &&
+ &desc->node != ioat_chan->used_desc.next;
+ desc = to_ioat_desc(desc->node.next)) {
+ if (desc->async_tx.cookie) {
+ cookie = desc->async_tx.cookie;
+ desc->async_tx.cookie = 0;
+ ioat_dma_unmap(ioat_chan, desc);
+ if (desc->async_tx.callback) {
+ desc->async_tx.callback(desc->async_tx.callback_param);
+ desc->async_tx.callback = NULL;
+ }
+ }
+ }
+
+ /* move used.prev up beyond those that are finished */
+ if (&desc->node == ioat_chan->used_desc.next)
+ ioat_chan->used_desc.prev = NULL;
+ else
+ ioat_chan->used_desc.prev = &desc->node;
+ }
+ break;
+ }
+
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ ioat_chan->last_completion = phys_complete;
+ if (cookie != 0)
+ ioat_chan->completed_cookie = cookie;
+
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+}
+
+/**
+ * ioat_dma_is_complete - poll the status of a IOAT DMA transaction
+ * @chan: IOAT DMA channel handle
+ * @cookie: DMA transaction identifier
+ * @done: if not %NULL, updated with last completed transaction
+ * @used: if not %NULL, updated with last used transaction
+ */
+static enum dma_status ioat_dma_is_complete(struct dma_chan *chan,
+ dma_cookie_t cookie,
+ dma_cookie_t *done,
+ dma_cookie_t *used)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ dma_cookie_t last_used;
+ dma_cookie_t last_complete;
+ enum dma_status ret;
+
+ last_used = chan->cookie;
+ last_complete = ioat_chan->completed_cookie;
+ ioat_chan->watchdog_tcp_cookie = cookie;
+
+ if (done)
+ *done = last_complete;
+ if (used)
+ *used = last_used;
+
+ ret = dma_async_is_complete(cookie, last_complete, last_used);
+ if (ret == DMA_SUCCESS)
+ return ret;
+
+ ioat_dma_memcpy_cleanup(ioat_chan);
+
+ last_used = chan->cookie;
+ last_complete = ioat_chan->completed_cookie;
+
+ if (done)
+ *done = last_complete;
+ if (used)
+ *used = last_used;
+
+ return dma_async_is_complete(cookie, last_complete, last_used);
+}
+
+static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat_chan)
+{
+ struct ioat_desc_sw *desc;
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+
+ desc = ioat_dma_get_next_descriptor(ioat_chan);
+
+ if (!desc) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "Unable to start null desc - get next desc failed\n");
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ return;
+ }
+
+ desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL
+ | IOAT_DMA_DESCRIPTOR_CTL_INT_GN
+ | IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
+ /* set size to non-zero value (channel returns error when size is 0) */
+ desc->hw->size = NULL_DESC_BUFFER_SIZE;
+ desc->hw->src_addr = 0;
+ desc->hw->dst_addr = 0;
+ async_tx_ack(&desc->async_tx);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ desc->hw->next = 0;
+ list_add_tail(&desc->node, &ioat_chan->used_desc);
+
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_HIGH);
+
+ writeb(IOAT_CHANCMD_START, ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH);
+
+ ioat_chan->dmacount++;
+ __ioat2_dma_memcpy_issue_pending(ioat_chan);
+ break;
+ }
+ spin_unlock_bh(&ioat_chan->desc_lock);
+}
+
+/*
+ * Perform a IOAT transaction to verify the HW works.
+ */
+#define IOAT_TEST_SIZE 2000
+
+static void ioat_dma_test_callback(void *dma_async_param)
+{
+ struct completion *cmp = dma_async_param;
+ complete(cmp);
+}
+
+/**
+ * ioat_dma_self_test - Perform a IOAT transaction to verify the HW works.
+ * @device: device to be tested
+ */
+static int ioat_dma_self_test(struct ioatdma_device *device)
+{
+ int i;
+ u8 *src;
+ u8 *dest;
+ struct dma_chan *dma_chan;
+ struct dma_async_tx_descriptor *tx;
+ dma_addr_t dma_dest, dma_src;
+ dma_cookie_t cookie;
+ int err = 0;
+ struct completion cmp;
+
+ src = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL);
+ if (!src)
+ return -ENOMEM;
+ dest = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL);
+ if (!dest) {
+ kfree(src);
+ return -ENOMEM;
+ }
+
+ /* Fill in src buffer */
+ for (i = 0; i < IOAT_TEST_SIZE; i++)
+ src[i] = (u8)i;
+
+ /* Start copy, using first DMA channel */
+ dma_chan = container_of(device->common.channels.next,
+ struct dma_chan,
+ device_node);
+ if (device->common.device_alloc_chan_resources(dma_chan, NULL) < 1) {
+ dev_err(&device->pdev->dev,
+ "selftest cannot allocate chan resource\n");
+ err = -ENODEV;
+ goto out;
+ }
+
+ dma_src = dma_map_single(dma_chan->device_v3->dev, src, IOAT_TEST_SIZE,
+ DMA_TO_DEVICE);
+ dma_dest = dma_map_single(dma_chan->device_v3->dev,dest, IOAT_TEST_SIZE,
+ DMA_FROM_DEVICE);
+ tx = device->common.device_prep_dma_memcpy(dma_chan, dma_dest, dma_src,
+ IOAT_TEST_SIZE, 0);
+ if (!tx) {
+ dev_err(&device->pdev->dev,
+ "Self-test prep failed, disabling\n");
+ err = -ENODEV;
+ goto free_resources;
+ }
+
+ async_tx_ack(tx);
+ init_completion(&cmp);
+ tx->callback = ioat_dma_test_callback;
+ tx->callback_param = &cmp;
+ cookie = tx->tx_submit(tx);
+ if (cookie < 0) {
+ dev_err(&device->pdev->dev,
+ "Self-test setup failed, disabling\n");
+ err = -ENODEV;
+ goto free_resources;
+ }
+ device->common.device_issue_pending(dma_chan);
+
+ wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
+
+ if (device->common.device_is_tx_complete(dma_chan, cookie, NULL, NULL)
+ != DMA_SUCCESS) {
+ dev_err(&device->pdev->dev,
+ "Self-test copy timed out, disabling\n");
+ err = -ENODEV;
+ goto free_resources;
+ }
+ if (memcmp(src, dest, IOAT_TEST_SIZE)) {
+ dev_err(&device->pdev->dev,
+ "Self-test copy failed compare, disabling\n");
+ err = -ENODEV;
+ goto free_resources;
+ }
+
+free_resources:
+ device->common.device_free_chan_resources(dma_chan);
+out:
+ kfree(src);
+ kfree(dest);
+ return err;
+}
+
+static char ioat_interrupt_style[32] = "msix";
+module_param_string(ioat_interrupt_style, ioat_interrupt_style,
+ sizeof(ioat_interrupt_style), 0644);
+MODULE_PARM_DESC(ioat_interrupt_style,
+ "set ioat interrupt style: msix (default), "
+ "msix-single-vector, msi, intx)");
+
+/**
+ * ioat_dma_setup_interrupts - setup interrupt handler
+ * @device: ioat device
+ */
+static int ioat_dma_setup_interrupts(struct ioatdma_device *device)
+{
+ struct ioat_dma_chan *ioat_chan;
+ int err, i, j, msixcnt;
+ u8 intrctrl = 0;
+
+ if (!strcmp(ioat_interrupt_style, "msix"))
+ goto msix;
+ if (!strcmp(ioat_interrupt_style, "msix-single-vector"))
+ goto msix_single_vector;
+ if (!strcmp(ioat_interrupt_style, "msi"))
+ goto msi;
+ if (!strcmp(ioat_interrupt_style, "intx"))
+ goto intx;
+ dev_err(&device->pdev->dev, "invalid ioat_interrupt_style %s\n",
+ ioat_interrupt_style);
+ goto err_no_irq;
+
+msix:
+ /* The number of MSI-X vectors should equal the number of channels */
+ msixcnt = device->common.chancnt;
+ for (i = 0; i < msixcnt; i++)
+ device->msix_entries[i].entry = i;
+
+ err = pci_enable_msix(device->pdev, device->msix_entries, msixcnt);
+ if (err < 0)
+ goto msi;
+ if (err > 0)
+ goto msix_single_vector;
+
+ for (i = 0; i < msixcnt; i++) {
+ ioat_chan = ioat_lookup_chan_by_index(device, i);
+ err = request_irq(device->msix_entries[i].vector,
+ ioat_dma_do_interrupt_msix,
+ 0, "ioat-msix", ioat_chan);
+ if (err) {
+ for (j = 0; j < i; j++) {
+ ioat_chan =
+ ioat_lookup_chan_by_index(device, j);
+ free_irq(device->msix_entries[j].vector,
+ ioat_chan);
+ }
+ goto msix_single_vector;
+ }
+ }
+ intrctrl |= IOAT_INTRCTRL_MSIX_VECTOR_CONTROL;
+ device->irq_mode = msix_multi_vector;
+ goto done;
+
+msix_single_vector:
+ device->msix_entries[0].entry = 0;
+ err = pci_enable_msix(device->pdev, device->msix_entries, 1);
+ if (err)
+ goto msi;
+
+ err = request_irq(device->msix_entries[0].vector, ioat_dma_do_interrupt,
+ 0, "ioat-msix", device);
+ if (err) {
+ pci_disable_msix(device->pdev);
+ goto msi;
+ }
+ device->irq_mode = msix_single_vector;
+ goto done;
+
+msi:
+ err = pci_enable_msi(device->pdev);
+ if (err)
+ goto intx;
+
+ err = request_irq(device->pdev->irq, ioat_dma_do_interrupt,
+ 0, "ioat-msi", device);
+ if (err) {
+ pci_disable_msi(device->pdev);
+ goto intx;
+ }
+ /*
+ * CB 1.2 devices need a bit set in configuration space to enable MSI
+ */
+ if (device->version == IOAT_VER_1_2) {
+ u32 dmactrl;
+ pci_read_config_dword(device->pdev,
+ IOAT_PCI_DMACTRL_OFFSET, &dmactrl);
+ dmactrl |= IOAT_PCI_DMACTRL_MSI_EN;
+ pci_write_config_dword(device->pdev,
+ IOAT_PCI_DMACTRL_OFFSET, dmactrl);
+ }
+ device->irq_mode = msi;
+ goto done;
+
+intx:
+ err = request_irq(device->pdev->irq, ioat_dma_do_interrupt,
+ IRQF_SHARED, "ioat-intx", device);
+ if (err)
+ goto err_no_irq;
+ device->irq_mode = intx;
+
+done:
+ intrctrl |= IOAT_INTRCTRL_MASTER_INT_EN;
+ writeb(intrctrl, device->reg_base + IOAT_INTRCTRL_OFFSET);
+ return 0;
+
+err_no_irq:
+ /* Disable all interrupt generation */
+ writeb(0, device->reg_base + IOAT_INTRCTRL_OFFSET);
+ dev_err(&device->pdev->dev, "no usable interrupts\n");
+ device->irq_mode = none;
+ return -1;
+}
+
+/**
+ * ioat_dma_remove_interrupts - remove whatever interrupts were set
+ * @device: ioat device
+ */
+static void ioat_dma_remove_interrupts(struct ioatdma_device *device)
+{
+ struct ioat_dma_chan *ioat_chan;
+ int i;
+
+ /* Disable all interrupt generation */
+ writeb(0, device->reg_base + IOAT_INTRCTRL_OFFSET);
+
+ switch (device->irq_mode) {
+ case msix_multi_vector:
+ for (i = 0; i < device->common.chancnt; i++) {
+ ioat_chan = ioat_lookup_chan_by_index(device, i);
+ free_irq(device->msix_entries[i].vector, ioat_chan);
+ }
+ pci_disable_msix(device->pdev);
+ break;
+ case msix_single_vector:
+ free_irq(device->msix_entries[0].vector, device);
+ pci_disable_msix(device->pdev);
+ break;
+ case msi:
+ free_irq(device->pdev->irq, device);
+ pci_disable_msi(device->pdev);
+ break;
+ case intx:
+ free_irq(device->pdev->irq, device);
+ break;
+ case none:
+ dev_warn(&device->pdev->dev,
+ "call to %s without interrupts setup\n", __func__);
+ }
+ device->irq_mode = none;
+}
+
+struct ioatdma_device *ioat_dma_probe(struct pci_dev *pdev,
+ void __iomem *iobase)
+{
+ int err;
+ struct ioatdma_device *device;
+
+ device = kzalloc(sizeof(*device), GFP_KERNEL);
+ if (!device) {
+ err = -ENOMEM;
+ goto err_kzalloc;
+ }
+ device->pdev = pdev;
+ device->reg_base = iobase;
+ device->version = readb(device->reg_base + IOAT_VER_OFFSET);
+
+ /* DMA coherent memory pool for DMA descriptor allocations */
+ device->dma_pool = pci_pool_create("dma_desc_pool", pdev,
+ sizeof(struct ioat_dma_descriptor),
+ 64, 0);
+ if (!device->dma_pool) {
+ err = -ENOMEM;
+ goto err_dma_pool;
+ }
+
+ device->completion_pool = pci_pool_create("completion_pool", pdev,
+ sizeof(u64), SMP_CACHE_BYTES,
+ SMP_CACHE_BYTES);
+ if (!device->completion_pool) {
+ err = -ENOMEM;
+ goto err_completion_pool;
+ }
+
+ INIT_LIST_HEAD(&device->common.channels);
+ ioat_dma_enumerate_channels(device);
+
+ device->common.device_alloc_chan_resources =
+ ioat_dma_alloc_chan_resources;
+ device->common.device_free_chan_resources =
+ ioat_dma_free_chan_resources;
+ device->common.dev = &pdev->dev;
+
+ dma_cap_set(DMA_MEMCPY, device->common.cap_mask);
+ device->common.device_is_tx_complete = ioat_dma_is_complete;
+ switch (device->version) {
+ case IOAT_VER_1_2:
+ device->common.device_prep_dma_memcpy = ioat1_dma_prep_memcpy;
+ device->common.device_issue_pending =
+ ioat1_dma_memcpy_issue_pending;
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ device->common.device_prep_dma_memcpy = ioat2_dma_prep_memcpy;
+ device->common.device_issue_pending =
+ ioat2_dma_memcpy_issue_pending;
+ break;
+ }
+
+ dev_err(&device->pdev->dev,
+ "Intel(R) I/OAT DMA Engine found,"
+ " %d channels, device version 0x%02x, driver version %s\n",
+ device->common.chancnt, device->version, IOAT_DMA_VERSION);
+
+ err = ioat_dma_setup_interrupts(device);
+ if (err)
+ goto err_setup_interrupts;
+
+ err = ioat_dma_self_test(device);
+ if (err)
+ goto err_self_test;
+
+ ioat_set_tcp_copy_break(device);
+
+ dma_async_device_register_v3(&device->common);
+
+ if (device->version != IOAT_VER_3_0) {
+ INIT_WORK(&device->work, (void *)ioat_dma_chan_watchdog,
+ &device->work);
+ schedule_delayed_work(&device->work,
+ WATCHDOG_DELAY);
+ }
+
+ return device;
+
+err_self_test:
+ ioat_dma_remove_interrupts(device);
+err_setup_interrupts:
+ pci_pool_destroy(device->completion_pool);
+err_completion_pool:
+ pci_pool_destroy(device->dma_pool);
+err_dma_pool:
+ kfree(device);
+err_kzalloc:
+ dev_err(&pdev->dev,
+ "Intel(R) I/OAT DMA Engine initialization failed\n");
+ return NULL;
+}
+
+void ioat_dma_remove(struct ioatdma_device *device)
+{
+ struct dma_chan *chan, *_chan;
+ struct ioat_dma_chan *ioat_chan;
+
+ ioat_dma_remove_interrupts(device);
+
+ dma_async_device_unregister_v3(&device->common);
+
+ pci_pool_destroy(device->dma_pool);
+ pci_pool_destroy(device->completion_pool);
+
+ iounmap(device->reg_base);
+ pci_release_regions(device->pdev);
+ pci_disable_device(device->pdev);
+
+ if (device->version != IOAT_VER_3_0) {
+ cancel_delayed_work(&device->work);
+ }
+
+ list_for_each_entry_safe(chan, _chan,
+ &device->common.channels, device_node) {
+ ioat_chan = to_ioat_chan(chan);
+ list_del(&chan->device_node);
+ kfree(ioat_chan);
+ }
+ kfree(device);
+}
+
diff --git a/drivers/dma_v3/iovlock_v3.c b/drivers/dma_v3/iovlock_v3.c
new file mode 100644
index 0000000..da29205
--- /dev/null
+++ b/drivers/dma_v3/iovlock_v3.c
@@ -0,0 +1,274 @@
+/*
+ * Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+ * Portions based on net/core/datagram.c and copyrighted by their authors.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * The full GNU General Public License is included in this distribution in the
+ * file called COPYING.
+ */
+
+/*
+ * This code allows the net stack to make use of a DMA engine for
+ * skb to iovec copies.
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/pagemap.h>
+#include <net/tcp.h> /* for memcpy_toiovec */
+#include <asm/io.h>
+#include <asm/uaccess.h>
+
+static int num_pages_spanned(struct iovec *iov)
+{
+ return
+ ((PAGE_ALIGN((unsigned long)iov->iov_base + iov->iov_len) -
+ ((unsigned long)iov->iov_base & PAGE_MASK)) >> PAGE_SHIFT);
+}
+
+/*
+ * Pin down all the iovec pages needed for len bytes.
+ * Return a struct dma_pinned_list to keep track of pages pinned down.
+ *
+ * We are allocating a single chunk of memory, and then carving it up into
+ * 3 sections, the latter 2 whose size depends on the number of iovecs and the
+ * total number of pages, respectively.
+ */
+struct dma_pinned_list *dma_pin_iovec_pages_v3(struct iovec *iov, size_t len)
+{
+ struct dma_pinned_list *local_list;
+ struct page **pages;
+ int i;
+ int ret;
+ int nr_iovecs = 0;
+ int iovec_len_used = 0;
+ int iovec_pages_used = 0;
+ long err;
+
+ /* don't pin down non-user-based iovecs */
+ if (segment_eq(get_fs(), KERNEL_DS))
+ return NULL;
+
+ /* determine how many iovecs/pages there are, up front */
+ do {
+ iovec_len_used += iov[nr_iovecs].iov_len;
+ iovec_pages_used += num_pages_spanned(&iov[nr_iovecs]);
+ nr_iovecs++;
+ } while (iovec_len_used < len);
+
+ /* single kmalloc for pinned list, page_list[], and the page arrays */
+ local_list = kmalloc(sizeof(*local_list)
+ + (nr_iovecs * sizeof (struct dma_page_list))
+ + (iovec_pages_used * sizeof (struct page*)), GFP_KERNEL);
+ if (!local_list) {
+ err = -ENOMEM;
+ goto out;
+ }
+
+ /* list of pages starts right after the page list array */
+ pages = (struct page **) &local_list->page_list[nr_iovecs];
+
+ for (i = 0; i < nr_iovecs; i++) {
+ struct dma_page_list *page_list = &local_list->page_list[i];
+
+ len -= iov[i].iov_len;
+
+ if (!access_ok(VERIFY_WRITE, iov[i].iov_base, iov[i].iov_len)) {
+ err = -EFAULT;
+ goto unpin;
+ }
+
+ page_list->nr_pages = num_pages_spanned(&iov[i]);
+ page_list->base_address_v3 = iov[i].iov_base;
+
+ page_list->pages = pages;
+ pages += page_list->nr_pages;
+
+ /* pin pages down */
+ down_read(¤t->mm->mmap_sem);
+ ret = get_user_pages(
+ current,
+ current->mm,
+ (unsigned long) iov[i].iov_base,
+ page_list->nr_pages,
+ 1, /* write */
+ 0, /* force */
+ page_list->pages,
+ NULL);
+ up_read(¤t->mm->mmap_sem);
+
+ if (ret != page_list->nr_pages) {
+ err = -ENOMEM;
+ goto unpin;
+ }
+
+ local_list->nr_iovecs = i + 1;
+ }
+
+ return local_list;
+
+unpin:
+ dma_unpin_iovec_pages_v3(local_list);
+out:
+ return ERR_PTR(err);
+}
+
+void dma_unpin_iovec_pages_v3(struct dma_pinned_list *pinned_list)
+{
+ int i, j;
+
+ if (!pinned_list)
+ return;
+
+ for (i = 0; i < pinned_list->nr_iovecs; i++) {
+ struct dma_page_list *page_list = &pinned_list->page_list[i];
+ for (j = 0; j < page_list->nr_pages; j++) {
+ set_page_dirty_lock(page_list->pages[j]);
+ page_cache_release(page_list->pages[j]);
+ }
+ }
+
+ kfree(pinned_list);
+}
+
+
+/*
+ * We have already pinned down the pages we will be using in the iovecs.
+ * Each entry in iov array has corresponding entry in pinned_list->page_list.
+ * Using array indexing to keep iov[] and page_list[] in sync.
+ * Initial elements in iov array's iov->iov_len will be 0 if already copied into
+ * by another call.
+ * iov array length remaining guaranteed to be bigger than len.
+ */
+dma_cookie_t dma_memcpy_to_iovec_v3(struct dma_chan *chan, struct iovec *iov,
+ struct dma_pinned_list *pinned_list, unsigned char *kdata, size_t len)
+{
+ int iov_byte_offset;
+ int copy;
+ dma_cookie_t dma_cookie = 0;
+ int iovec_idx;
+ int page_idx;
+
+ if (!chan)
+ return memcpy_toiovec(iov, kdata, len);
+
+ iovec_idx = 0;
+ while (iovec_idx < pinned_list->nr_iovecs) {
+ struct dma_page_list *page_list;
+
+ /* skip already used-up iovecs */
+ while (!iov[iovec_idx].iov_len)
+ iovec_idx++;
+
+ page_list = &pinned_list->page_list[iovec_idx];
+
+ iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
+ page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
+ - ((unsigned long)page_list->base_address_v3 & PAGE_MASK)) >> PAGE_SHIFT;
+
+ /* break up copies to not cross page boundary */
+ while (iov[iovec_idx].iov_len) {
+ copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
+ copy = min_t(int, copy, iov[iovec_idx].iov_len);
+
+ dma_cookie = dma_async_memcpy_buf_to_pg_v3(chan,
+ page_list->pages[page_idx],
+ iov_byte_offset,
+ kdata,
+ copy);
+
+ len -= copy;
+ iov[iovec_idx].iov_len -= copy;
+ iov[iovec_idx].iov_base += copy;
+
+ if (!len)
+ return dma_cookie;
+
+ kdata += copy;
+ iov_byte_offset = 0;
+ page_idx++;
+ }
+ iovec_idx++;
+ }
+
+ /* really bad if we ever run out of iovecs */
+ BUG();
+ return -EFAULT;
+}
+
+dma_cookie_t dma_memcpy_pg_to_iovec_v3(struct dma_chan *chan, struct iovec *iov,
+ struct dma_pinned_list *pinned_list, struct page *page,
+ unsigned int offset, size_t len)
+{
+ int iov_byte_offset;
+ int copy;
+ dma_cookie_t dma_cookie = 0;
+ int iovec_idx;
+ int page_idx;
+ int err;
+
+ /* this needs as-yet-unimplemented buf-to-buff, so punt. */
+ /* TODO: use dma for this */
+ if (!chan || !pinned_list) {
+ u8 *vaddr = kmap(page);
+ err = memcpy_toiovec(iov, vaddr + offset, len);
+ kunmap(page);
+ return err;
+ }
+
+ iovec_idx = 0;
+ while (iovec_idx < pinned_list->nr_iovecs) {
+ struct dma_page_list *page_list;
+
+ /* skip already used-up iovecs */
+ while (!iov[iovec_idx].iov_len)
+ iovec_idx++;
+
+ page_list = &pinned_list->page_list[iovec_idx];
+
+ iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
+ page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
+ - ((unsigned long)page_list->base_address_v3 & PAGE_MASK)) >> PAGE_SHIFT;
+
+ /* break up copies to not cross page boundary */
+ while (iov[iovec_idx].iov_len) {
+ copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
+ copy = min_t(int, copy, iov[iovec_idx].iov_len);
+
+ dma_cookie = dma_async_memcpy_pg_to_pg_v3(chan,
+ page_list->pages[page_idx],
+ iov_byte_offset,
+ page,
+ offset,
+ copy);
+
+ len -= copy;
+ iov[iovec_idx].iov_len -= copy;
+ iov[iovec_idx].iov_base += copy;
+
+ if (!len)
+ return dma_cookie;
+
+ offset += copy;
+ iov_byte_offset = 0;
+ page_idx++;
+ }
+ iovec_idx++;
+ }
+
+ /* really bad if we ever run out of iovecs */
+ BUG();
+ return -EFAULT;
+}
distrib
>
Scientific%20Linux
>
5x
>
x86_64
>
by-pkgid
>
fc11cd6e1c513a17304da94a5390f3cd
>
files
>
1636
