/*
+ * Copyright(c) 2015, 2016 Intel Corporation.
*
* This file is provided under a dual BSD/GPLv2 license. When using or
* redistributing this file, you may do so under either license.
*
* GPL LICENSE SUMMARY
*
- * Copyright(c) 2015 Intel Corporation.
- *
* This program is free software; you can redistribute it and/or modify
* it under the terms of version 2 of the GNU General Public License as
* published by the Free Software Foundation.
*
* BSD LICENSE
*
- * Copyright(c) 2015 Intel Corporation.
- *
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
#include <linux/module.h>
#include <linux/printk.h>
#include <linux/hrtimer.h>
+#include <rdma/rdma_vt.h>
#include "hfi.h"
#include "device.h"
#include "sdma.h"
#include "debugfs.h"
#include "verbs.h"
+#include "aspm.h"
#undef pr_fmt
#define pr_fmt(fmt) DRIVER_NAME ": " fmt
#define HFI1_MIN_USER_CTXT_BUFCNT 7
#define HFI1_MIN_HDRQ_EGRBUF_CNT 2
+#define HFI1_MAX_HDRQ_EGRBUF_CNT 16352
#define HFI1_MIN_EAGER_BUFFER_SIZE (4 * 1024) /* 4KB */
#define HFI1_MAX_EAGER_BUFFER_SIZE (256 * 1024) /* 256KB */
MODULE_PARM_DESC(
num_user_contexts, "Set max number of user contexts to use");
-u8 krcvqs[RXE_NUM_DATA_VL];
+uint krcvqs[RXE_NUM_DATA_VL];
int krcvqsset;
-module_param_array(krcvqs, byte, &krcvqsset, S_IRUGO);
+module_param_array(krcvqs, uint, &krcvqsset, S_IRUGO);
MODULE_PARM_DESC(krcvqs, "Array of the number of non-control kernel receive queues by VL");
/* computed based on above array */
{
unsigned i;
int ret;
- int local_node_id = pcibus_to_node(dd->pcidev->bus);
/* Control context has to be always 0 */
BUILD_BUG_ON(HFI1_CTRL_CTXT != 0);
- if (local_node_id < 0)
- local_node_id = numa_node_id();
- dd->assigned_node_id = local_node_id;
-
- dd->rcd = kcalloc(dd->num_rcv_contexts, sizeof(*dd->rcd), GFP_KERNEL);
+ dd->rcd = kzalloc_node(dd->num_rcv_contexts * sizeof(*dd->rcd),
+ GFP_KERNEL, dd->node);
if (!dd->rcd)
goto nomem;
struct hfi1_ctxtdata *rcd;
ppd = dd->pport + (i % dd->num_pports);
- rcd = hfi1_create_ctxtdata(ppd, i);
+ rcd = hfi1_create_ctxtdata(ppd, i, dd->node);
if (!rcd) {
dd_dev_err(dd,
- "Unable to allocate kernel receive context, failing\n");
+ "Unable to allocate kernel receive context, failing\n");
goto nomem;
}
/*
rcd->sc = sc_alloc(dd, SC_ACK, rcd->rcvhdrqentsize, dd->node);
if (!rcd->sc) {
dd_dev_err(dd,
- "Unable to allocate kernel send context, failing\n");
+ "Unable to allocate kernel send context, failing\n");
dd->rcd[rcd->ctxt] = NULL;
hfi1_free_ctxtdata(dd, rcd);
goto nomem;
}
}
+ /*
+ * Initialize aspm, to be done after gen3 transition and setting up
+ * contexts and before enabling interrupts
+ */
+ aspm_init(dd);
+
return 0;
nomem:
ret = -ENOMEM;
/*
* Common code for user and kernel context setup.
*/
-struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, u32 ctxt)
+struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, u32 ctxt,
+ int numa)
{
struct hfi1_devdata *dd = ppd->dd;
struct hfi1_ctxtdata *rcd;
rcd->cnt = 1;
rcd->ctxt = ctxt;
dd->rcd[ctxt] = rcd;
- rcd->numa_id = numa_node_id();
+ rcd->numa_id = numa;
rcd->rcv_array_groups = dd->rcv_entries.ngroups;
- spin_lock_init(&rcd->exp_lock);
+ mutex_init(&rcd->exp_lock);
/*
* Calculate the context's RcvArray entry starting point.
/* Validate and initialize Rcv Hdr Q variables */
if (rcvhdrcnt % HDRQ_INCREMENT) {
dd_dev_err(dd,
- "ctxt%u: header queue count %d must be divisible by %d\n",
+ "ctxt%u: header queue count %d must be divisible by %lu\n",
rcd->ctxt, rcvhdrcnt, HDRQ_INCREMENT);
goto bail;
}
cc_state = get_cc_state(ppd);
- if (cc_state == NULL)
+ if (!cc_state)
/*
* This should _never_ happen - rcu_read_lock() is held,
* and set_link_ipg() should not be called if cc_state
cc_state = get_cc_state(ppd);
- if (cc_state == NULL) {
+ if (!cc_state) {
rcu_read_unlock();
return HRTIMER_NORESTART;
}
INIT_WORK(&ppd->link_vc_work, handle_verify_cap);
INIT_WORK(&ppd->link_up_work, handle_link_up);
INIT_WORK(&ppd->link_down_work, handle_link_down);
+ INIT_WORK(&ppd->dc_host_req_work, handle_8051_request);
INIT_WORK(&ppd->freeze_work, handle_freeze);
INIT_WORK(&ppd->link_downgrade_work, handle_link_downgrade);
INIT_WORK(&ppd->sma_message_work, handle_sma_message);
INIT_WORK(&ppd->link_bounce_work, handle_link_bounce);
+ INIT_WORK(&ppd->linkstate_active_work, receive_interrupt_work);
+ INIT_WORK(&ppd->qsfp_info.qsfp_work, qsfp_event);
+
mutex_init(&ppd->hls_lock);
spin_lock_init(&ppd->sdma_alllock);
spin_lock_init(&ppd->qsfp_info.qsfp_lock);
+ ppd->qsfp_info.ppd = ppd;
ppd->sm_trap_qp = 0x0;
ppd->sa_qp = 0x1;
* Enable kernel ctxts' receive and receive interrupt.
* Other ctxts done as user opens and initializes them.
*/
- rcvmask = HFI1_RCVCTRL_CTXT_ENB | HFI1_RCVCTRL_INTRAVAIL_ENB;
for (i = 0; i < dd->first_user_ctxt; ++i) {
+ rcvmask = HFI1_RCVCTRL_CTXT_ENB | HFI1_RCVCTRL_INTRAVAIL_ENB;
rcvmask |= HFI1_CAP_KGET_MASK(dd->rcd[i]->flags, DMA_RTAIL) ?
HFI1_RCVCTRL_TAILUPD_ENB : HFI1_RCVCTRL_TAILUPD_DIS;
if (!HFI1_CAP_KGET_MASK(dd->rcd[i]->flags, MULTI_PKT_EGR))
lastfail = hfi1_setup_eagerbufs(rcd);
if (lastfail)
dd_dev_err(dd,
- "failed to allocate kernel ctxt's rcvhdrq and/or egr bufs\n");
+ "failed to allocate kernel ctxt's rcvhdrq and/or egr bufs\n");
}
if (lastfail)
ret = lastfail;
/* enable chip even if we have an error, so we can debug cause */
enable_chip(dd);
- ret = hfi1_cq_init(dd);
done:
/*
* Set status even if port serdes is not initialized
for (pidx = 0; pidx < dd->num_pports; ++pidx) {
ppd = dd->pport + pidx;
- /* initialize the qsfp if it exists
- * Requires interrupts to be enabled so we are notified
- * when the QSFP completes reset, and has
- * to be done before bringing up the SERDES
+ /*
+ * start the serdes - must be after interrupts are
+ * enabled so we are notified when the link goes up
*/
- init_qsfp(ppd);
-
- /* start the serdes - must be after interrupts are
- enabled so we are notified when the link goes up */
lastfail = bringup_serdes(ppd);
if (lastfail)
dd_dev_info(dd,
- "Failed to bring up port %u\n",
- ppd->port);
+ "Failed to bring up port %u\n",
+ ppd->port);
/*
* Set status even if port serdes is not initialized
/* disable the send device */
pio_send_control(dd, PSC_GLOBAL_DISABLE);
+ shutdown_led_override(ppd);
+
/*
* Clear SerdesEnable.
* We can't count on interrupts since we are stopping.
kfree(rcd->egrbufs.buffers);
sc_free(rcd->sc);
- vfree(rcd->physshadow);
- vfree(rcd->tid_pg_list);
vfree(rcd->user_event_mask);
vfree(rcd->subctxt_uregbase);
vfree(rcd->subctxt_rcvegrbuf);
vfree(rcd->subctxt_rcvhdr_base);
- kfree(rcd->tidusemap);
kfree(rcd->opstats);
kfree(rcd);
}
+/*
+ * Release our hold on the shared asic data. If we are the last one,
+ * free the structure. Must be holding hfi1_devs_lock.
+ */
+static void release_asic_data(struct hfi1_devdata *dd)
+{
+ int other;
+
+ if (!dd->asic_data)
+ return;
+ dd->asic_data->dds[dd->hfi1_id] = NULL;
+ other = dd->hfi1_id ? 0 : 1;
+ if (!dd->asic_data->dds[other]) {
+ /* we are the last holder, free it */
+ kfree(dd->asic_data);
+ }
+ dd->asic_data = NULL;
+}
+
void hfi1_free_devdata(struct hfi1_devdata *dd)
{
unsigned long flags;
spin_lock_irqsave(&hfi1_devs_lock, flags);
idr_remove(&hfi1_unit_table, dd->unit);
list_del(&dd->list);
+ release_asic_data(dd);
spin_unlock_irqrestore(&hfi1_devs_lock, flags);
- hfi1_dbg_ibdev_exit(&dd->verbs_dev);
+ free_platform_config(dd);
rcu_barrier(); /* wait for rcu callbacks to complete */
free_percpu(dd->int_counter);
free_percpu(dd->rcv_limit);
- ib_dealloc_device(&dd->verbs_dev.ibdev);
+ hfi1_dev_affinity_free(dd);
+ free_percpu(dd->send_schedule);
+ ib_dealloc_device(&dd->verbs_dev.rdi.ibdev);
}
/*
{
unsigned long flags;
struct hfi1_devdata *dd;
- int ret;
+ int ret, nports;
- dd = (struct hfi1_devdata *)ib_alloc_device(sizeof(*dd) + extra);
+ /* extra is * number of ports */
+ nports = extra / sizeof(struct hfi1_pportdata);
+
+ dd = (struct hfi1_devdata *)rvt_alloc_device(sizeof(*dd) + extra,
+ nports);
if (!dd)
return ERR_PTR(-ENOMEM);
- /* extra is * number of ports */
- dd->num_pports = extra / sizeof(struct hfi1_pportdata);
+ dd->num_pports = nports;
dd->pport = (struct hfi1_pportdata *)(dd + 1);
INIT_LIST_HEAD(&dd->list);
- dd->node = dev_to_node(&pdev->dev);
- if (dd->node < 0)
- dd->node = 0;
idr_preload(GFP_KERNEL);
spin_lock_irqsave(&hfi1_devs_lock, flags);
spin_lock_init(&dd->sc_init_lock);
spin_lock_init(&dd->dc8051_lock);
spin_lock_init(&dd->dc8051_memlock);
- mutex_init(&dd->qsfp_i2c_mutex);
seqlock_init(&dd->sc2vl_lock);
spin_lock_init(&dd->sde_map_lock);
+ spin_lock_init(&dd->pio_map_lock);
init_waitqueue_head(&dd->event_queue);
dd->int_counter = alloc_percpu(u64);
goto bail;
}
+ dd->send_schedule = alloc_percpu(u64);
+ if (!dd->send_schedule) {
+ ret = -ENOMEM;
+ hfi1_early_err(&pdev->dev,
+ "Could not allocate per-cpu int_counter\n");
+ goto bail;
+ }
+
if (!hfi1_cpulist_count) {
u32 count = num_online_cpus();
&pdev->dev,
"Could not alloc cpulist info, cpu affinity might be wrong\n");
}
- hfi1_dbg_ibdev_init(&dd->verbs_dev);
return dd;
bail:
if (!list_empty(&dd->list))
list_del_init(&dd->list);
- ib_dealloc_device(&dd->verbs_dev.ibdev);
+ ib_dealloc_device(&dd->verbs_dev.rdi.ibdev);
return ERR_PTR(ret);
}
user_credit_return_threshold = 100;
compute_krcvqs();
- /* sanitize receive interrupt count, time must wait until after
- the hardware type is known */
+ /*
+ * sanitize receive interrupt count, time must wait until after
+ * the hardware type is known
+ */
if (rcv_intr_count > RCV_HDR_HEAD_COUNTER_MASK)
rcv_intr_count = RCV_HDR_HEAD_COUNTER_MASK;
/* reject invalid combinations */
idr_init(&hfi1_unit_table);
hfi1_dbg_init();
+ ret = hfi1_wss_init();
+ if (ret < 0)
+ goto bail_wss;
ret = pci_register_driver(&hfi1_pci_driver);
if (ret < 0) {
pr_err("Unable to register driver: error %d\n", -ret);
goto bail; /* all OK */
bail_dev:
+ hfi1_wss_exit();
+bail_wss:
hfi1_dbg_exit();
idr_destroy(&hfi1_unit_table);
dev_cleanup();
static void __exit hfi1_mod_cleanup(void)
{
pci_unregister_driver(&hfi1_pci_driver);
+ hfi1_wss_exit();
hfi1_dbg_exit();
hfi1_cpulist_count = 0;
kfree(hfi1_cpulist);
}
}
kfree(tmp);
+ free_pio_map(dd);
/* must follow rcv context free - need to remove rcv's hooks */
for (ctxt = 0; ctxt < dd->num_send_contexts; ctxt++)
sc_free(dd->send_contexts[ctxt].sc);
dd->num_send_contexts = 0;
kfree(dd->send_contexts);
dd->send_contexts = NULL;
+ kfree(dd->hw_to_sw);
+ dd->hw_to_sw = NULL;
kfree(dd->boardname);
vfree(dd->events);
vfree(dd->status);
- hfi1_cq_exit(dd);
}
/*
ret = -EINVAL;
goto bail;
}
+ if (rcvhdrcnt > HFI1_MAX_HDRQ_EGRBUF_CNT) {
+ hfi1_early_err(&pdev->dev,
+ "Receive header queue count cannot be greater than %u\n",
+ HFI1_MAX_HDRQ_EGRBUF_CNT);
+ ret = -EINVAL;
+ goto bail;
+ }
/* use the encoding function as a sanitization check */
if (!encode_rcv_header_entry_size(hfi1_hdrq_entsize)) {
hfi1_early_err(&pdev->dev, "Invalid HdrQ Entry size %u\n",
* we still create devices, so diags, etc. can be used
* to determine cause of problem.
*/
- if (!initfail && !ret)
+ if (!initfail && !ret) {
dd->flags |= HFI1_INITTED;
+ /* create debufs files after init and ib register */
+ hfi1_dbg_ibdev_init(&dd->verbs_dev);
+ }
j = hfi1_device_create(dd);
if (j)
{
struct hfi1_devdata *dd = pci_get_drvdata(pdev);
+ /* close debugfs files before ib unregister */
+ hfi1_dbg_ibdev_exit(&dd->verbs_dev);
/* unregister from IB core */
hfi1_unregister_ib_device(dd);
if (!rcd->rcvhdrq) {
dd_dev_err(dd,
- "attempt to allocate %d bytes for ctxt %u rcvhdrq failed\n",
- amt, rcd->ctxt);
+ "attempt to allocate %d bytes for ctxt %u rcvhdrq failed\n",
+ amt, rcd->ctxt);
goto bail;
}
- /* Event mask is per device now and is in hfi1_devdata */
- /*if (rcd->ctxt >= dd->first_user_ctxt) {
- rcd->user_event_mask = vmalloc_user(PAGE_SIZE);
- if (!rcd->user_event_mask)
- goto bail_free_hdrq;
- }*/
-
if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
rcd->rcvhdrtail_kvaddr = dma_zalloc_coherent(
&dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail,
bail_free:
dd_dev_err(dd,
- "attempt to allocate 1 page for ctxt %u rcvhdrqtailaddr failed\n",
- rcd->ctxt);
+ "attempt to allocate 1 page for ctxt %u rcvhdrqtailaddr failed\n",
+ rcd->ctxt);
vfree(rcd->user_event_mask);
rcd->user_event_mask = NULL;
dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
if (rcd->egrbufs.rcvtid_size == round_mtu ||
!HFI1_CAP_KGET_MASK(rcd->flags, MULTI_PKT_EGR)) {
dd_dev_err(dd, "ctxt%u: Failed to allocate eager buffers\n",
- rcd->ctxt);
+ rcd->ctxt);
goto bail_rcvegrbuf_phys;
}
rcd->egrbufs.buffers[j].len)) {
j++;
offset = 0;
- } else
+ } else {
offset += new_size;
+ }
}
rcd->egrbufs.rcvtid_size = new_size;
}
rcd->ctxt, rcd->egrbufs.alloced, rcd->egrbufs.rcvtid_size,
rcd->egrbufs.size);
-
/*
* Set the contexts rcv array head update threshold to the closest
* power of 2 (so we can use a mask instead of modulo) below half
for (idx = 0; idx < rcd->egrbufs.alloced; idx++) {
hfi1_put_tid(dd, rcd->eager_base + idx, PT_EAGER,
- rcd->egrbufs.rcvtids[idx].phys, order);
+ rcd->egrbufs.rcvtids[idx].phys, order);
cond_resched();
}
goto bail;
bail_rcvegrbuf_phys:
for (idx = 0; idx < rcd->egrbufs.alloced &&
- rcd->egrbufs.buffers[idx].addr;
+ rcd->egrbufs.buffers[idx].addr;
idx++) {
dma_free_coherent(&dd->pcidev->dev,
rcd->egrbufs.buffers[idx].len,
projects / karo-tx-linux.git / blobdiff