2 * Copyright 2012 Tilera Corporation. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation, version 2.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
11 * NON INFRINGEMENT. See the GNU General Public License for
15 #include <linux/kernel.h>
16 #include <linux/mmzone.h>
17 #include <linux/pci.h>
18 #include <linux/delay.h>
19 #include <linux/string.h>
20 #include <linux/init.h>
21 #include <linux/capability.h>
22 #include <linux/sched.h>
23 #include <linux/errno.h>
24 #include <linux/irq.h>
25 #include <linux/msi.h>
27 #include <linux/uaccess.h>
28 #include <linux/ctype.h>
30 #include <asm/processor.h>
31 #include <asm/sections.h>
32 #include <asm/byteorder.h>
34 #include <gxio/iorpc_globals.h>
35 #include <gxio/kiorpc.h>
36 #include <gxio/trio.h>
37 #include <gxio/iorpc_trio.h>
38 #include <hv/drv_trio_intf.h>
43 * This file containes the routines to search for PCI buses,
44 * enumerate the buses, and configure any attached devices.
47 #define DEBUG_PCI_CFG 0
50 #define TRACE_CFG_WR(size, val, bus, dev, func, offset) \
51 pr_info("CFG WR %d-byte VAL %#x to bus %d dev %d func %d addr %u\n", \
52 size, val, bus, dev, func, offset & 0xFFF);
53 #define TRACE_CFG_RD(size, val, bus, dev, func, offset) \
54 pr_info("CFG RD %d-byte VAL %#x from bus %d dev %d func %d addr %u\n", \
55 size, val, bus, dev, func, offset & 0xFFF);
57 #define TRACE_CFG_WR(...)
58 #define TRACE_CFG_RD(...)
61 static int __devinitdata pci_probe = 1;
63 /* Information on the PCIe RC ports configuration. */
64 static int __devinitdata pcie_rc[TILEGX_NUM_TRIO][TILEGX_TRIO_PCIES];
67 * On some platforms with one or more Gx endpoint ports, we need to
68 * delay the PCIe RC port probe for a few seconds to work around
69 * a HW PCIe link-training bug. The exact delay is specified with
70 * a kernel boot argument in the form of "pcie_rc_delay=T,P,S",
71 * where T is the TRIO instance number, P is the port number and S is
72 * the delay in seconds. If the delay is not provided, the value
73 * will be DEFAULT_RC_DELAY.
75 static int __devinitdata rc_delay[TILEGX_NUM_TRIO][TILEGX_TRIO_PCIES];
77 /* Default number of seconds that the PCIe RC port probe can be delayed. */
78 #define DEFAULT_RC_DELAY 10
80 /* Max number of seconds that the PCIe RC port probe can be delayed. */
81 #define MAX_RC_DELAY 20
83 /* Array of the PCIe ports configuration info obtained from the BIB. */
84 struct pcie_port_property pcie_ports[TILEGX_NUM_TRIO][TILEGX_TRIO_PCIES];
86 /* All drivers share the TRIO contexts defined here. */
87 gxio_trio_context_t trio_contexts[TILEGX_NUM_TRIO];
89 /* Pointer to an array of PCIe RC controllers. */
90 struct pci_controller pci_controllers[TILEGX_NUM_TRIO * TILEGX_TRIO_PCIES];
91 int num_rc_controllers;
92 static int num_ep_controllers;
94 static struct pci_ops tile_cfg_ops;
96 /* Mask of CPUs that should receive PCIe interrupts. */
97 static struct cpumask intr_cpus_map;
100 * We don't need to worry about the alignment of resources.
102 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
103 resource_size_t size, resource_size_t align)
107 EXPORT_SYMBOL(pcibios_align_resource);
111 * Pick a CPU to receive and handle the PCIe interrupts, based on the IRQ #.
112 * For now, we simply send interrupts to non-dataplane CPUs.
113 * We may implement methods to allow user to specify the target CPUs,
114 * e.g. via boot arguments.
116 static int tile_irq_cpu(int irq)
122 count = cpumask_weight(&intr_cpus_map);
123 if (unlikely(count == 0)) {
124 pr_warning("intr_cpus_map empty, interrupts will be"
125 " delievered to dataplane tiles\n");
126 return irq % (smp_height * smp_width);
130 for_each_cpu(cpu, &intr_cpus_map) {
138 * Open a file descriptor to the TRIO shim.
140 static int __devinit tile_pcie_open(int trio_index)
142 gxio_trio_context_t *context = &trio_contexts[trio_index];
146 * This opens a file descriptor to the TRIO shim.
148 ret = gxio_trio_init(context, trio_index);
153 * Allocate an ASID for the kernel.
155 ret = gxio_trio_alloc_asids(context, 1, 0, 0);
157 pr_err("PCI: ASID alloc failure on TRIO %d, give up\n",
159 goto asid_alloc_failure;
164 #ifdef USE_SHARED_PCIE_CONFIG_REGION
166 * Alloc a PIO region for config access, shared by all MACs per TRIO.
167 * This shouldn't fail since the kernel is supposed to the first
168 * client of the TRIO's PIO regions.
170 ret = gxio_trio_alloc_pio_regions(context, 1, 0, 0);
172 pr_err("PCI: CFG PIO alloc failure on TRIO %d, give up\n",
174 goto pio_alloc_failure;
177 context->pio_cfg_index = ret;
180 * For PIO CFG, the bus_address_hi parameter is 0. The mac parameter
181 * is also 0 because it is specified in PIO_REGION_SETUP_CFG_ADDR.
183 ret = gxio_trio_init_pio_region_aux(context, context->pio_cfg_index,
184 0, 0, HV_TRIO_PIO_FLAG_CONFIG_SPACE);
186 pr_err("PCI: CFG PIO init failure on TRIO %d, give up\n",
188 goto pio_alloc_failure;
195 #ifdef USE_SHARED_PCIE_CONFIG_REGION
198 hv_dev_close(context->fd);
204 tilegx_legacy_irq_ack(struct irq_data *d)
206 __insn_mtspr(SPR_IPI_EVENT_RESET_K, 1UL << d->irq);
210 tilegx_legacy_irq_mask(struct irq_data *d)
212 __insn_mtspr(SPR_IPI_MASK_SET_K, 1UL << d->irq);
216 tilegx_legacy_irq_unmask(struct irq_data *d)
218 __insn_mtspr(SPR_IPI_MASK_RESET_K, 1UL << d->irq);
221 static struct irq_chip tilegx_legacy_irq_chip = {
222 .name = "tilegx_legacy_irq",
223 .irq_ack = tilegx_legacy_irq_ack,
224 .irq_mask = tilegx_legacy_irq_mask,
225 .irq_unmask = tilegx_legacy_irq_unmask,
227 /* TBD: support set_affinity. */
231 * This is a wrapper function of the kernel level-trigger interrupt
232 * handler handle_level_irq() for PCI legacy interrupts. The TRIO
233 * is configured such that only INTx Assert interrupts are proxied
234 * to Linux which just calls handle_level_irq() after clearing the
235 * MAC INTx Assert status bit associated with this interrupt.
238 trio_handle_level_irq(unsigned int irq, struct irq_desc *desc)
240 struct pci_controller *controller = irq_desc_get_handler_data(desc);
241 gxio_trio_context_t *trio_context = controller->trio;
242 uint64_t intx = (uint64_t)irq_desc_get_chip_data(desc);
243 int mac = controller->mac;
244 unsigned int reg_offset;
247 handle_level_irq(irq, desc);
250 * Clear the INTx Level status, otherwise future interrupts are
253 reg_offset = (TRIO_PCIE_INTFC_MAC_INT_STS <<
254 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
255 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_INTERFACE <<
256 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
257 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
259 level_mask = TRIO_PCIE_INTFC_MAC_INT_STS__INT_LEVEL_MASK << intx;
261 __gxio_mmio_write(trio_context->mmio_base_mac + reg_offset, level_mask);
265 * Create kernel irqs and set up the handlers for the legacy interrupts.
266 * Also some minimum initialization for the MSI support.
268 static int __devinit tile_init_irqs(struct pci_controller *controller)
275 cpumask_copy(&intr_cpus_map, cpu_online_mask);
278 for (i = 0; i < 4; i++) {
279 gxio_trio_context_t *context = controller->trio;
282 /* Ask the kernel to allocate an IRQ. */
285 pr_err("PCI: no free irq vectors, failed for %d\n", i);
289 controller->irq_intx_table[i] = irq;
291 /* Distribute the 4 IRQs to different tiles. */
292 cpu = tile_irq_cpu(irq);
294 /* Configure the TRIO intr binding for this IRQ. */
295 result = gxio_trio_config_legacy_intr(context, cpu_x(cpu),
296 cpu_y(cpu), KERNEL_PL,
297 irq, controller->mac, i);
299 pr_err("PCI: MAC intx config failed for %d\n", i);
305 * Register the IRQ handler with the kernel.
307 irq_set_chip_and_handler(irq, &tilegx_legacy_irq_chip,
308 trio_handle_level_irq);
309 irq_set_chip_data(irq, (void *)(uint64_t)i);
310 irq_set_handler_data(irq, controller);
316 for (j = 0; j < i; j++)
317 destroy_irq(controller->irq_intx_table[j]);
323 * Find valid controllers and fill in pci_controller structs for each
326 * Returns the number of controllers discovered.
328 int __init tile_pci_init(void)
330 int num_trio_shims = 0;
335 pr_info("PCI: disabled by boot argument\n");
339 pr_info("PCI: Searching for controllers...\n");
342 * We loop over all the TRIO shims.
344 for (i = 0; i < TILEGX_NUM_TRIO; i++) {
347 ret = tile_pcie_open(i);
354 if (num_trio_shims == 0 || sim_is_simulator())
358 * Now determine which PCIe ports are configured to operate in RC mode.
359 * We look at the Board Information Block first and then see if there
360 * are any overriding configuration by the HW strapping pin.
362 for (i = 0; i < TILEGX_NUM_TRIO; i++) {
363 gxio_trio_context_t *context = &trio_contexts[i];
369 ret = hv_dev_pread(context->fd, 0,
370 (HV_VirtAddr)&pcie_ports[i][0],
371 sizeof(struct pcie_port_property) * TILEGX_TRIO_PCIES,
372 GXIO_TRIO_OP_GET_PORT_PROPERTY);
374 pr_err("PCI: PCIE_GET_PORT_PROPERTY failure, error %d,"
375 " on TRIO %d\n", ret, i);
379 for (j = 0; j < TILEGX_TRIO_PCIES; j++) {
380 if (pcie_ports[i][j].allow_rc) {
382 num_rc_controllers++;
384 else if (pcie_ports[i][j].allow_ep) {
385 num_ep_controllers++;
391 * Return if no PCIe ports are configured to operate in RC mode.
393 if (num_rc_controllers == 0)
397 * Set the TRIO pointer and MAC index for each PCIe RC port.
399 for (i = 0; i < TILEGX_NUM_TRIO; i++) {
400 for (j = 0; j < TILEGX_TRIO_PCIES; j++) {
402 pci_controllers[ctl_index].trio =
404 pci_controllers[ctl_index].mac = j;
405 pci_controllers[ctl_index].trio_index = i;
407 if (ctl_index == num_rc_controllers)
415 * Configure each PCIe RC port.
417 for (i = 0; i < num_rc_controllers; i++) {
419 * Configure the PCIe MAC to run in RC mode.
422 struct pci_controller *controller = &pci_controllers[i];
424 controller->index = i;
425 controller->ops = &tile_cfg_ops;
428 * The PCI memory resource is located above the PA space.
429 * For every host bridge, the BAR window or the MMIO aperture
430 * is in range [3GB, 4GB - 1] of a 4GB space beyond the
434 controller->mem_offset = TILE_PCI_MEM_START +
435 (i * TILE_PCI_BAR_WINDOW_TOP);
436 controller->mem_space.start = controller->mem_offset +
437 TILE_PCI_BAR_WINDOW_TOP - TILE_PCI_BAR_WINDOW_SIZE;
438 controller->mem_space.end = controller->mem_offset +
439 TILE_PCI_BAR_WINDOW_TOP - 1;
440 controller->mem_space.flags = IORESOURCE_MEM;
441 snprintf(controller->mem_space_name,
442 sizeof(controller->mem_space_name),
443 "PCI mem domain %d", i);
444 controller->mem_space.name = controller->mem_space_name;
447 return num_rc_controllers;
451 * (pin - 1) converts from the PCI standard's [1:4] convention to
452 * a normal [0:3] range.
454 static int tile_map_irq(const struct pci_dev *dev, u8 device, u8 pin)
456 struct pci_controller *controller =
457 (struct pci_controller *)dev->sysdata;
458 return controller->irq_intx_table[pin - 1];
462 static void __devinit fixup_read_and_payload_sizes(struct pci_controller *
465 gxio_trio_context_t *trio_context = controller->trio;
466 struct pci_bus *root_bus = controller->root_bus;
467 TRIO_PCIE_RC_DEVICE_CONTROL_t dev_control;
468 TRIO_PCIE_RC_DEVICE_CAP_t rc_dev_cap;
469 unsigned int reg_offset;
470 struct pci_bus *child;
474 mac = controller->mac;
477 * Set our max read request size to be 4KB.
480 (TRIO_PCIE_RC_DEVICE_CONTROL <<
481 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
482 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_STANDARD <<
483 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
484 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
486 dev_control.word = __gxio_mmio_read32(trio_context->mmio_base_mac +
488 dev_control.max_read_req_sz = 5;
489 __gxio_mmio_write32(trio_context->mmio_base_mac + reg_offset,
493 * Set the max payload size supported by this Gx PCIe MAC.
494 * Though Gx PCIe supports Max Payload Size of up to 1024 bytes,
495 * experiments have shown that setting MPS to 256 yields the
499 (TRIO_PCIE_RC_DEVICE_CAP <<
500 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
501 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_STANDARD <<
502 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
503 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
505 rc_dev_cap.word = __gxio_mmio_read32(trio_context->mmio_base_mac +
507 rc_dev_cap.mps_sup = 1;
508 __gxio_mmio_write32(trio_context->mmio_base_mac + reg_offset,
511 /* Configure PCI Express MPS setting. */
512 list_for_each_entry(child, &root_bus->children, node) {
513 struct pci_dev *self = child->self;
517 pcie_bus_configure_settings(child, self->pcie_mpss);
521 * Set the mac_config register in trio based on the MPS/MRS of the link.
524 (TRIO_PCIE_RC_DEVICE_CONTROL <<
525 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
526 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_STANDARD <<
527 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
528 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
530 dev_control.word = __gxio_mmio_read32(trio_context->mmio_base_mac +
533 err = gxio_trio_set_mps_mrs(trio_context,
534 dev_control.max_payload_size,
535 dev_control.max_read_req_sz,
538 pr_err("PCI: PCIE_CONFIGURE_MAC_MPS_MRS failure, "
539 "MAC %d on TRIO %d\n",
540 mac, controller->trio_index);
544 static int __devinit setup_pcie_rc_delay(char *str)
546 unsigned long delay = 0;
547 unsigned long trio_index;
550 if (str == NULL || !isdigit(*str))
552 trio_index = simple_strtoul(str, (char **)&str, 10);
553 if (trio_index >= TILEGX_NUM_TRIO)
562 mac = simple_strtoul(str, (char **)&str, 10);
563 if (mac >= TILEGX_TRIO_PCIES)
573 delay = simple_strtoul(str, (char **)&str, 10);
574 if (delay > MAX_RC_DELAY)
578 rc_delay[trio_index][mac] = delay ? : DEFAULT_RC_DELAY;
579 pr_info("Delaying PCIe RC link training for %u sec"
580 " on MAC %lu on TRIO %lu\n", rc_delay[trio_index][mac],
584 early_param("pcie_rc_delay", setup_pcie_rc_delay);
587 * PCI initialization entry point, called by subsys_initcall.
589 int __init pcibios_init(void)
591 resource_size_t offset;
592 LIST_HEAD(resources);
598 if (num_rc_controllers == 0 && num_ep_controllers == 0)
602 * We loop over all the TRIO shims and set up the MMIO mappings.
604 for (i = 0; i < TILEGX_NUM_TRIO; i++) {
605 gxio_trio_context_t *context = &trio_contexts[i];
611 * Map in the MMIO space for the MAC.
614 context->mmio_base_mac =
615 iorpc_ioremap(context->fd, offset,
616 HV_TRIO_CONFIG_IOREMAP_SIZE);
617 if (context->mmio_base_mac == NULL) {
618 pr_err("PCI: MAC map failure on TRIO %d\n", i);
620 hv_dev_close(context->fd);
627 * Delay a bit in case devices aren't ready. Some devices are
628 * known to require at least 20ms here, but we use a more
629 * conservative value.
633 /* Scan all of the recorded PCI controllers. */
634 for (next_busno = 0, i = 0; i < num_rc_controllers; i++) {
635 struct pci_controller *controller = &pci_controllers[i];
636 gxio_trio_context_t *trio_context = controller->trio;
637 TRIO_PCIE_INTFC_PORT_CONFIG_t port_config;
638 TRIO_PCIE_INTFC_PORT_STATUS_t port_status;
639 TRIO_PCIE_INTFC_TX_FIFO_CTL_t tx_fifo_ctl;
641 unsigned int reg_offset;
642 unsigned int class_code_revision;
647 if (trio_context->fd < 0)
650 trio_index = controller->trio_index;
651 mac = controller->mac;
654 * Check the port strap state which will override the BIB
659 (TRIO_PCIE_INTFC_PORT_CONFIG <<
660 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
661 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_INTERFACE <<
662 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
663 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
666 __gxio_mmio_read(trio_context->mmio_base_mac +
669 if ((port_config.strap_state !=
670 TRIO_PCIE_INTFC_PORT_CONFIG__STRAP_STATE_VAL_AUTO_CONFIG_RC) &&
671 (port_config.strap_state !=
672 TRIO_PCIE_INTFC_PORT_CONFIG__STRAP_STATE_VAL_AUTO_CONFIG_RC_G1)) {
674 * If this is really intended to be an EP port,
675 * record it so that the endpoint driver will know about it.
677 if (port_config.strap_state ==
678 TRIO_PCIE_INTFC_PORT_CONFIG__STRAP_STATE_VAL_AUTO_CONFIG_ENDPOINT ||
679 port_config.strap_state ==
680 TRIO_PCIE_INTFC_PORT_CONFIG__STRAP_STATE_VAL_AUTO_CONFIG_ENDPOINT_G1)
681 pcie_ports[trio_index][mac].allow_ep = 1;
687 * Delay the RC link training if needed.
689 if (rc_delay[trio_index][mac])
690 msleep(rc_delay[trio_index][mac] * 1000);
692 ret = gxio_trio_force_rc_link_up(trio_context, mac);
694 pr_err("PCI: PCIE_FORCE_LINK_UP failure, "
695 "MAC %d on TRIO %d\n", mac, trio_index);
697 pr_info("PCI: Found PCI controller #%d on TRIO %d MAC %d\n", i,
698 trio_index, controller->mac);
701 * Wait a bit here because some EP devices take longer
707 * Check for PCIe link-up status.
711 (TRIO_PCIE_INTFC_PORT_STATUS <<
712 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
713 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_INTERFACE <<
714 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
715 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
718 __gxio_mmio_read(trio_context->mmio_base_mac +
720 if (!port_status.dl_up) {
721 pr_err("PCI: link is down, MAC %d on TRIO %d\n",
727 * Ensure that the link can come out of L1 power down state.
728 * Strictly speaking, this is needed only in the case of
729 * heavy RC-initiated DMAs.
732 (TRIO_PCIE_INTFC_TX_FIFO_CTL <<
733 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
734 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_INTERFACE <<
735 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
736 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
738 __gxio_mmio_read(trio_context->mmio_base_mac +
740 tx_fifo_ctl.min_p_credits = 0;
741 __gxio_mmio_write(trio_context->mmio_base_mac + reg_offset,
745 * Change the device ID so that Linux bus crawl doesn't confuse
746 * the internal bridge with any Tilera endpoints.
750 (TRIO_PCIE_RC_DEVICE_ID_VEN_ID <<
751 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
752 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_STANDARD <<
753 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
754 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
756 __gxio_mmio_write32(trio_context->mmio_base_mac + reg_offset,
757 (TILERA_GX36_RC_DEV_ID <<
758 TRIO_PCIE_RC_DEVICE_ID_VEN_ID__DEV_ID_SHIFT) |
762 * Set the internal P2P bridge class code.
766 (TRIO_PCIE_RC_REVISION_ID <<
767 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
768 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_STANDARD <<
769 TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
770 (mac << TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
772 class_code_revision =
773 __gxio_mmio_read32(trio_context->mmio_base_mac +
775 class_code_revision = (class_code_revision & 0xff ) |
776 (PCI_CLASS_BRIDGE_PCI << 16);
778 __gxio_mmio_write32(trio_context->mmio_base_mac +
779 reg_offset, class_code_revision);
781 #ifdef USE_SHARED_PCIE_CONFIG_REGION
784 * Map in the MMIO space for the PIO region.
786 offset = HV_TRIO_PIO_OFFSET(trio_context->pio_cfg_index) |
787 (((unsigned long long)mac) <<
788 TRIO_TILE_PIO_REGION_SETUP_CFG_ADDR__MAC_SHIFT);
793 * Alloc a PIO region for PCI config access per MAC.
795 ret = gxio_trio_alloc_pio_regions(trio_context, 1, 0, 0);
797 pr_err("PCI: PCI CFG PIO alloc failure for mac %d "
798 "on TRIO %d, give up\n", mac, trio_index);
803 trio_context->pio_cfg_index[mac] = ret;
806 * For PIO CFG, the bus_address_hi parameter is 0.
808 ret = gxio_trio_init_pio_region_aux(trio_context,
809 trio_context->pio_cfg_index[mac],
810 mac, 0, HV_TRIO_PIO_FLAG_CONFIG_SPACE);
812 pr_err("PCI: PCI CFG PIO init failure for mac %d "
813 "on TRIO %d, give up\n", mac, trio_index);
818 offset = HV_TRIO_PIO_OFFSET(trio_context->pio_cfg_index[mac]) |
819 (((unsigned long long)mac) <<
820 TRIO_TILE_PIO_REGION_SETUP_CFG_ADDR__MAC_SHIFT);
824 trio_context->mmio_base_pio_cfg[mac] =
825 iorpc_ioremap(trio_context->fd, offset,
826 (1 << TRIO_TILE_PIO_REGION_SETUP_CFG_ADDR__MAC_SHIFT));
827 if (trio_context->mmio_base_pio_cfg[mac] == NULL) {
828 pr_err("PCI: PIO map failure for mac %d on TRIO %d\n",
835 * Initialize the PCIe interrupts.
837 if (tile_init_irqs(controller)) {
838 pr_err("PCI: IRQs init failure for mac %d on TRIO %d\n",
845 * The PCI memory resource is located above the PA space.
846 * The memory range for the PCI root bus should not overlap
847 * with the physical RAM
849 pci_add_resource_offset(&resources, &controller->mem_space,
850 controller->mem_offset);
852 controller->first_busno = next_busno;
853 bus = pci_scan_root_bus(NULL, next_busno, controller->ops,
854 controller, &resources);
855 controller->root_bus = bus;
856 next_busno = bus->busn_res.end + 1;
860 /* Do machine dependent PCI interrupt routing */
861 pci_fixup_irqs(pci_common_swizzle, tile_map_irq);
864 * This comes from the generic Linux PCI driver.
866 * It allocates all of the resources (I/O memory, etc)
867 * associated with the devices read in above.
870 pci_assign_unassigned_resources();
872 /* Record the I/O resources in the PCI controller structure. */
873 for (i = 0; i < num_rc_controllers; i++) {
874 struct pci_controller *controller = &pci_controllers[i];
875 gxio_trio_context_t *trio_context = controller->trio;
876 struct pci_bus *root_bus = pci_controllers[i].root_bus;
877 struct pci_bus *next_bus;
878 uint32_t bus_address_hi;
884 * Skip controllers that are not properly initialized or
887 if (root_bus == NULL)
890 /* Configure the max_payload_size values for this domain. */
891 fixup_read_and_payload_sizes(controller);
893 list_for_each_entry(dev, &root_bus->devices, bus_list) {
894 /* Find the PCI host controller, ie. the 1st bridge. */
895 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI &&
896 (PCI_SLOT(dev->devfn) == 0)) {
897 next_bus = dev->subordinate;
898 pci_controllers[i].mem_resources[0] =
899 *next_bus->resource[0];
900 pci_controllers[i].mem_resources[1] =
901 *next_bus->resource[1];
902 pci_controllers[i].mem_resources[2] =
903 *next_bus->resource[2];
909 if (pci_controllers[i].mem_resources[1].flags & IORESOURCE_MEM)
911 pci_controllers[i].mem_resources[1].start >> 32;
912 else if (pci_controllers[i].mem_resources[2].flags & IORESOURCE_PREFETCH)
914 pci_controllers[i].mem_resources[2].start >> 32;
916 /* This is unlikely. */
917 pr_err("PCI: no memory resources on TRIO %d mac %d\n",
918 controller->trio_index, controller->mac);
923 * Alloc a PIO region for PCI memory access for each RC port.
925 ret = gxio_trio_alloc_pio_regions(trio_context, 1, 0, 0);
927 pr_err("PCI: MEM PIO alloc failure on TRIO %d mac %d, "
928 "give up\n", controller->trio_index,
934 controller->pio_mem_index = ret;
937 * For PIO MEM, the bus_address_hi parameter is hard-coded 0
938 * because we always assign 32-bit PCI bus BAR ranges.
940 ret = gxio_trio_init_pio_region_aux(trio_context,
941 controller->pio_mem_index,
946 pr_err("PCI: MEM PIO init failure on TRIO %d mac %d, "
947 "give up\n", controller->trio_index,
954 * Configure a Mem-Map region for each memory controller so
955 * that Linux can map all of its PA space to the PCI bus.
956 * Use the IOMMU to handle hash-for-home memory.
958 for_each_online_node(j) {
959 unsigned long start_pfn = node_start_pfn[j];
960 unsigned long end_pfn = node_end_pfn[j];
961 unsigned long nr_pages = end_pfn - start_pfn;
963 ret = gxio_trio_alloc_memory_maps(trio_context, 1, 0,
966 pr_err("PCI: Mem-Map alloc failure on TRIO %d "
967 "mac %d for MC %d, give up\n",
968 controller->trio_index,
971 goto alloc_mem_map_failed;
974 controller->mem_maps[j] = ret;
977 * Initialize the Mem-Map and the I/O MMU so that all
978 * the physical memory can be accessed by the endpoint
979 * devices. The base bus address is set to the base CPA
980 * of this memory controller plus an offset (see pci.h).
981 * The region's base VA is set to the base CPA. The
982 * I/O MMU table essentially translates the CPA to
983 * the real PA. Implicitly, for node 0, we create
984 * a separate Mem-Map region that serves as the inbound
985 * window for legacy 32-bit devices. This is a direct
986 * map of the low 4GB CPA space.
988 ret = gxio_trio_init_memory_map_mmu_aux(trio_context,
989 controller->mem_maps[j],
990 start_pfn << PAGE_SHIFT,
991 nr_pages << PAGE_SHIFT,
994 (start_pfn << PAGE_SHIFT) +
995 TILE_PCI_MEM_MAP_BASE_OFFSET,
997 GXIO_TRIO_ORDER_MODE_UNORDERED);
999 pr_err("PCI: Mem-Map init failure on TRIO %d "
1000 "mac %d for MC %d, give up\n",
1001 controller->trio_index,
1002 controller->mac, j);
1004 goto alloc_mem_map_failed;
1008 alloc_mem_map_failed:
1016 subsys_initcall(pcibios_init);
1018 /* Note: to be deleted after Linux 3.6 merge. */
1019 void __devinit pcibios_fixup_bus(struct pci_bus *bus)
1024 * This can be called from the generic PCI layer, but doesn't need to
1027 char __devinit *pcibios_setup(char *str)
1029 if (!strcmp(str, "off")) {
1037 * Enable memory address decoding, as appropriate, for the
1038 * device described by the 'dev' struct. The I/O decoding
1039 * is disabled, though the TILE-Gx supports I/O addressing.
1041 * This is called from the generic PCI layer, and can be called
1042 * for bridges or endpoints.
1044 int pcibios_enable_device(struct pci_dev *dev, int mask)
1046 return pci_enable_resources(dev, mask);
1049 /* Called for each device after PCI setup is done. */
1051 pcibios_fixup_final(struct pci_dev *pdev)
1053 set_dma_ops(&pdev->dev, gx_pci_dma_map_ops);
1054 set_dma_offset(&pdev->dev, TILE_PCI_MEM_MAP_BASE_OFFSET);
1055 pdev->dev.archdata.max_direct_dma_addr =
1056 TILE_PCI_MAX_DIRECT_DMA_ADDRESS;
1058 DECLARE_PCI_FIXUP_FINAL(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_final);
1060 /* Map a PCI MMIO bus address into VA space. */
1061 void __iomem *ioremap(resource_size_t phys_addr, unsigned long size)
1063 struct pci_controller *controller = NULL;
1064 resource_size_t bar_start;
1065 resource_size_t bar_end;
1066 resource_size_t offset;
1067 resource_size_t start;
1068 resource_size_t end;
1073 end = phys_addr + size - 1;
1076 * In the following, each PCI controller's mem_resources[1]
1077 * represents its (non-prefetchable) PCI memory resource and
1078 * mem_resources[2] refers to its prefetchable PCI memory resource.
1079 * By searching phys_addr in each controller's mem_resources[], we can
1080 * determine the controller that should accept the PCI memory access.
1083 for (i = 0; i < num_rc_controllers; i++) {
1085 * Skip controllers that are not properly initialized or
1088 if (pci_controllers[i].root_bus == NULL)
1091 for (j = 1; j < 3; j++) {
1093 pci_controllers[i].mem_resources[j].start;
1095 pci_controllers[i].mem_resources[j].end;
1097 if ((start >= bar_start) && (end <= bar_end)) {
1099 controller = &pci_controllers[i];
1106 if (controller == NULL)
1110 trio_fd = controller->trio->fd;
1112 /* Convert the resource start to the bus address offset. */
1113 start = phys_addr - controller->mem_offset;
1115 offset = HV_TRIO_PIO_OFFSET(controller->pio_mem_index) + start;
1118 * We need to keep the PCI bus address's in-page offset in the VA.
1120 return iorpc_ioremap(trio_fd, offset, size) +
1121 (phys_addr & (PAGE_SIZE - 1));
1123 EXPORT_SYMBOL(ioremap);
1125 void pci_iounmap(struct pci_dev *dev, void __iomem *addr)
1129 EXPORT_SYMBOL(pci_iounmap);
1131 /****************************************************************
1133 * Tile PCI config space read/write routines
1135 ****************************************************************/
1138 * These are the normal read and write ops
1139 * These are expanded with macros from pci_bus_read_config_byte() etc.
1141 * devfn is the combined PCI device & function.
1143 * offset is in bytes, from the start of config space for the
1144 * specified bus & device.
1147 static int __devinit tile_cfg_read(struct pci_bus *bus,
1153 struct pci_controller *controller = bus->sysdata;
1154 gxio_trio_context_t *trio_context = controller->trio;
1155 int busnum = bus->number & 0xff;
1156 int device = PCI_SLOT(devfn);
1157 int function = PCI_FUNC(devfn);
1158 int config_type = 1;
1159 TRIO_TILE_PIO_REGION_SETUP_CFG_ADDR_t cfg_addr;
1163 * Map all accesses to the local device on root bus into the
1164 * MMIO space of the MAC. Accesses to the downstream devices
1165 * go to the PIO space.
1167 if (pci_is_root_bus(bus)) {
1170 * This is the internal downstream P2P bridge,
1173 unsigned int reg_offset;
1175 reg_offset = ((offset & 0xFFF) <<
1176 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
1177 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_PROTECTED
1178 << TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
1180 TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
1182 mmio_addr = trio_context->mmio_base_mac + reg_offset;
1188 * We fake an empty device for (device > 0),
1189 * since there is only one device on bus 0.
1191 goto invalid_device;
1196 * Accesses to the directly attached device have to be
1197 * sent as type-0 configs.
1200 if (busnum == (controller->first_busno + 1)) {
1202 * There is only one device off of our built-in P2P bridge.
1205 goto invalid_device;
1211 cfg_addr.reg_addr = (offset & 0xFFF);
1212 cfg_addr.fn = function;
1213 cfg_addr.dev = device;
1214 cfg_addr.bus = busnum;
1215 cfg_addr.type = config_type;
1218 * Note that we don't set the mac field in cfg_addr because the
1219 * mapping is per port.
1222 mmio_addr = trio_context->mmio_base_pio_cfg[controller->mac] +
1229 *val = __gxio_mmio_read32(mmio_addr);
1233 *val = __gxio_mmio_read16(mmio_addr);
1237 *val = __gxio_mmio_read8(mmio_addr);
1241 return PCIBIOS_FUNC_NOT_SUPPORTED;
1244 TRACE_CFG_RD(size, *val, busnum, device, function, offset);
1264 return PCIBIOS_FUNC_NOT_SUPPORTED;
1272 * See tile_cfg_read() for relevent comments.
1273 * Note that "val" is the value to write, not a pointer to that value.
1275 static int __devinit tile_cfg_write(struct pci_bus *bus,
1281 struct pci_controller *controller = bus->sysdata;
1282 gxio_trio_context_t *trio_context = controller->trio;
1283 int busnum = bus->number & 0xff;
1284 int device = PCI_SLOT(devfn);
1285 int function = PCI_FUNC(devfn);
1286 int config_type = 1;
1287 TRIO_TILE_PIO_REGION_SETUP_CFG_ADDR_t cfg_addr;
1289 u32 val_32 = (u32)val;
1290 u16 val_16 = (u16)val;
1294 * Map all accesses to the local device on root bus into the
1295 * MMIO space of the MAC. Accesses to the downstream devices
1296 * go to the PIO space.
1298 if (pci_is_root_bus(bus)) {
1301 * This is the internal downstream P2P bridge,
1304 unsigned int reg_offset;
1306 reg_offset = ((offset & 0xFFF) <<
1307 TRIO_CFG_REGION_ADDR__REG_SHIFT) |
1308 (TRIO_CFG_REGION_ADDR__INTFC_VAL_MAC_PROTECTED
1309 << TRIO_CFG_REGION_ADDR__INTFC_SHIFT ) |
1311 TRIO_CFG_REGION_ADDR__MAC_SEL_SHIFT);
1313 mmio_addr = trio_context->mmio_base_mac + reg_offset;
1319 * We fake an empty device for (device > 0),
1320 * since there is only one device on bus 0.
1322 goto invalid_device;
1327 * Accesses to the directly attached device have to be
1328 * sent as type-0 configs.
1331 if (busnum == (controller->first_busno + 1)) {
1333 * There is only one device off of our built-in P2P bridge.
1336 goto invalid_device;
1342 cfg_addr.reg_addr = (offset & 0xFFF);
1343 cfg_addr.fn = function;
1344 cfg_addr.dev = device;
1345 cfg_addr.bus = busnum;
1346 cfg_addr.type = config_type;
1349 * Note that we don't set the mac field in cfg_addr because the
1350 * mapping is per port.
1353 mmio_addr = trio_context->mmio_base_pio_cfg[controller->mac] +
1360 __gxio_mmio_write32(mmio_addr, val_32);
1361 TRACE_CFG_WR(size, val_32, busnum, device, function, offset);
1365 __gxio_mmio_write16(mmio_addr, val_16);
1366 TRACE_CFG_WR(size, val_16, busnum, device, function, offset);
1370 __gxio_mmio_write8(mmio_addr, val_8);
1371 TRACE_CFG_WR(size, val_8, busnum, device, function, offset);
1375 return PCIBIOS_FUNC_NOT_SUPPORTED;
1384 static struct pci_ops tile_cfg_ops = {
1385 .read = tile_cfg_read,
1386 .write = tile_cfg_write,
1391 * MSI support starts here.
1394 tilegx_msi_startup(struct irq_data *d)
1403 tilegx_msi_ack(struct irq_data *d)
1405 __insn_mtspr(SPR_IPI_EVENT_RESET_K, 1UL << d->irq);
1409 tilegx_msi_mask(struct irq_data *d)
1412 __insn_mtspr(SPR_IPI_MASK_SET_K, 1UL << d->irq);
1416 tilegx_msi_unmask(struct irq_data *d)
1418 __insn_mtspr(SPR_IPI_MASK_RESET_K, 1UL << d->irq);
1422 static struct irq_chip tilegx_msi_chip = {
1423 .name = "tilegx_msi",
1424 .irq_startup = tilegx_msi_startup,
1425 .irq_ack = tilegx_msi_ack,
1426 .irq_mask = tilegx_msi_mask,
1427 .irq_unmask = tilegx_msi_unmask,
1429 /* TBD: support set_affinity. */
1432 int arch_setup_msi_irq(struct pci_dev *pdev, struct msi_desc *desc)
1434 struct pci_controller *controller;
1435 gxio_trio_context_t *trio_context;
1438 uint64_t mem_map_base;
1439 uint64_t mem_map_limit;
1451 * Since we use a 64-bit Mem-Map to accept the MSI write, we fail
1452 * devices that are not capable of generating a 64-bit message address.
1453 * These devices will fall back to using the legacy interrupts.
1454 * Most PCIe endpoint devices do support 64-bit message addressing.
1456 if (desc->msi_attrib.is_64 == 0) {
1457 dev_printk(KERN_INFO, &pdev->dev,
1458 "64-bit MSI message address not supported, "
1459 "falling back to legacy interrupts.\n");
1465 default_irq = desc->msi_attrib.default_irq;
1466 controller = irq_get_handler_data(default_irq);
1468 BUG_ON(!controller);
1470 trio_context = controller->trio;
1473 * Allocate the Mem-Map that will accept the MSI write and
1474 * trigger the TILE-side interrupts.
1476 mem_map = gxio_trio_alloc_memory_maps(trio_context, 1, 0, 0);
1478 dev_printk(KERN_INFO, &pdev->dev,
1479 "%s Mem-Map alloc failure. "
1480 "Failed to initialize MSI interrupts. "
1481 "Falling back to legacy interrupts.\n",
1482 desc->msi_attrib.is_msix ? "MSI-X" : "MSI");
1485 goto msi_mem_map_alloc_failure;
1488 /* We try to distribute different IRQs to different tiles. */
1489 cpu = tile_irq_cpu(irq);
1492 * Now call up to the HV to configure the Mem-Map interrupt and
1493 * set up the IPI binding.
1495 mem_map_base = MEM_MAP_INTR_REGIONS_BASE +
1496 mem_map * MEM_MAP_INTR_REGION_SIZE;
1497 mem_map_limit = mem_map_base + MEM_MAP_INTR_REGION_SIZE - 1;
1499 ret = gxio_trio_config_msi_intr(trio_context, cpu_x(cpu), cpu_y(cpu),
1500 KERNEL_PL, irq, controller->mac,
1501 mem_map, mem_map_base, mem_map_limit,
1502 trio_context->asid);
1504 dev_printk(KERN_INFO, &pdev->dev, "HV MSI config failed.\n");
1506 goto hv_msi_config_failure;
1509 irq_set_msi_desc(irq, desc);
1511 msi_addr = mem_map_base + TRIO_MAP_MEM_REG_INT3 - TRIO_MAP_MEM_REG_INT0;
1513 msg.address_hi = msi_addr >> 32;
1514 msg.address_lo = msi_addr & 0xffffffff;
1518 write_msi_msg(irq, &msg);
1519 irq_set_chip_and_handler(irq, &tilegx_msi_chip, handle_level_irq);
1520 irq_set_handler_data(irq, controller);
1524 hv_msi_config_failure:
1526 msi_mem_map_alloc_failure:
1532 void arch_teardown_msi_irq(unsigned int irq)