#include "sdma.h"
#include "eprom.h"
#include "efivar.h"
+#include "platform.h"
+#include "aspm.h"
#define NUM_IB_PORTS 1
SEC_SPC_FREEZE,
SEND_PIO_ERR_STATUS_PIO_STATE_MACHINE_ERR_SMASK),
/*23*/ FLAG_ENTRY("PioWriteQwValidParity",
- SEC_WRITE_DROPPED|SEC_SPC_FREEZE,
+ SEC_WRITE_DROPPED | SEC_SPC_FREEZE,
SEND_PIO_ERR_STATUS_PIO_WRITE_QW_VALID_PARITY_ERR_SMASK),
/*24*/ FLAG_ENTRY("PioBlockQwCountParity",
- SEC_WRITE_DROPPED|SEC_SPC_FREEZE,
+ SEC_WRITE_DROPPED | SEC_SPC_FREEZE,
SEND_PIO_ERR_STATUS_PIO_BLOCK_QW_COUNT_PARITY_ERR_SMASK),
/*25*/ FLAG_ENTRY("PioVlfVlLenParity",
SEC_SPC_FREEZE,
| SEND_DMA_ERR_STATUS_SDMA_CSR_PARITY_ERR_SMASK \
| SEND_DMA_ERR_STATUS_SDMA_PCIE_REQ_TRACKING_UNC_ERR_SMASK)
+/* SendEgressErrInfo bits that correspond to a PortXmitDiscard counter */
+#define PORT_DISCARD_EGRESS_ERRS \
+ (SEND_EGRESS_ERR_INFO_TOO_LONG_IB_PACKET_ERR_SMASK \
+ | SEND_EGRESS_ERR_INFO_VL_MAPPING_ERR_SMASK \
+ | SEND_EGRESS_ERR_INFO_VL_ERR_SMASK)
+
/*
* TXE Egress Error flags
*/
FLAG_ENTRY0("Link going down", 0x0100),
};
-
static u32 encoded_size(u32 size);
static u32 chip_to_opa_lstate(struct hfi1_devdata *dd, u32 chip_lstate);
static int set_physical_link_state(struct hfi1_devdata *dd, u64 state);
#define OVR_LBL(ctx) C_RCV_HDR_OVF_ ## ctx
#define OVR_ELM(ctx) \
CNTR_ELEM("RcvHdrOvr" #ctx, \
- (RCV_HDR_OVFL_CNT + ctx*0x100), \
+ (RCV_HDR_OVFL_CNT + ctx * 0x100), \
0, CNTR_NORMAL, port_access_u64_csr)
/* 32bit TXE */
{
u64 ret;
-
if (mode == CNTR_MODE_R) {
ret = read_csr(dd, csr);
} else if (mode == CNTR_MODE_W) {
void *context, int vl, int mode, u64 data)
{
struct hfi1_devdata *dd = context;
+ u64 csr = entry->csr;
- if (vl != CNTR_INVALID_VL)
- return 0;
- return read_write_csr(dd, entry->csr, mode, data);
+ if (entry->flags & CNTR_SDMA) {
+ if (vl == CNTR_INVALID_VL)
+ return 0;
+ csr += 0x100 * vl;
+ } else {
+ if (vl != CNTR_INVALID_VL)
+ return 0;
+ }
+ return read_write_csr(dd, csr, mode, data);
+}
+
+static u64 access_sde_err_cnt(const struct cntr_entry *entry,
+ void *context, int idx, int mode, u64 data)
+{
+ struct hfi1_devdata *dd = (struct hfi1_devdata *)context;
+
+ if (dd->per_sdma && idx < dd->num_sdma)
+ return dd->per_sdma[idx].err_cnt;
+ return 0;
+}
+
+static u64 access_sde_int_cnt(const struct cntr_entry *entry,
+ void *context, int idx, int mode, u64 data)
+{
+ struct hfi1_devdata *dd = (struct hfi1_devdata *)context;
+
+ if (dd->per_sdma && idx < dd->num_sdma)
+ return dd->per_sdma[idx].sdma_int_cnt;
+ return 0;
+}
+
+static u64 access_sde_idle_int_cnt(const struct cntr_entry *entry,
+ void *context, int idx, int mode, u64 data)
+{
+ struct hfi1_devdata *dd = (struct hfi1_devdata *)context;
+
+ if (dd->per_sdma && idx < dd->num_sdma)
+ return dd->per_sdma[idx].idle_int_cnt;
+ return 0;
+}
+
+static u64 access_sde_progress_int_cnt(const struct cntr_entry *entry,
+ void *context, int idx, int mode,
+ u64 data)
+{
+ struct hfi1_devdata *dd = (struct hfi1_devdata *)context;
+
+ if (dd->per_sdma && idx < dd->num_sdma)
+ return dd->per_sdma[idx].progress_int_cnt;
+ return 0;
}
static u64 dev_access_u64_csr(const struct cntr_entry *entry, void *context,
static u64 access_sw_xmit_discards(const struct cntr_entry *entry,
void *context, int vl, int mode, u64 data)
{
- struct hfi1_pportdata *ppd = context;
+ struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)context;
+ u64 zero = 0;
+ u64 *counter;
- if (vl != CNTR_INVALID_VL)
- return 0;
+ if (vl == CNTR_INVALID_VL)
+ counter = &ppd->port_xmit_discards;
+ else if (vl >= 0 && vl < C_VL_COUNT)
+ counter = &ppd->port_xmit_discards_vl[vl];
+ else
+ counter = &zero;
- return read_write_sw(ppd->dd, &ppd->port_xmit_discards, mode, data);
+ return read_write_sw(ppd->dd, counter, mode, data);
}
static u64 access_xmit_constraint_errs(const struct cntr_entry *entry,
return dd->verbs_dev.n_piowait;
}
+static u64 access_sw_pio_drain(const struct cntr_entry *entry,
+ void *context, int vl, int mode, u64 data)
+{
+ struct hfi1_devdata *dd = (struct hfi1_devdata *)context;
+
+ return dd->verbs_dev.n_piodrain;
+}
+
static u64 access_sw_vtx_wait(const struct cntr_entry *entry,
void *context, int vl, int mode, u64 data)
{
{
struct hfi1_devdata *dd = (struct hfi1_devdata *)context;
- return dd->verbs_dev.n_send_schedule;
+ return read_write_cpu(dd, &dd->z_send_schedule, dd->send_schedule, vl,
+ mode, data);
}
/* Software counters for the error status bits within MISC_ERR_STATUS */
void *context, int vl, int mode, u64 data) \
{ \
struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)context; \
- return read_write_cpu(ppd->dd, &ppd->ibport_data.z_ ##cntr, \
- ppd->ibport_data.cntr, vl, \
+ return read_write_cpu(ppd->dd, &ppd->ibport_data.rvp.z_ ##cntr, \
+ ppd->ibport_data.rvp.cntr, vl, \
mode, data); \
}
if (vl != CNTR_INVALID_VL) \
return 0; \
\
- return read_write_sw(ppd->dd, &ppd->ibport_data.n_ ##cntr, \
+ return read_write_sw(ppd->dd, &ppd->ibport_data.rvp.n_ ##cntr, \
mode, data); \
}
access_sw_vtx_wait),
[C_SW_PIO_WAIT] = CNTR_ELEM("PioWait", 0, 0, CNTR_NORMAL,
access_sw_pio_wait),
+[C_SW_PIO_DRAIN] = CNTR_ELEM("PioDrain", 0, 0, CNTR_NORMAL,
+ access_sw_pio_drain),
[C_SW_KMEM_WAIT] = CNTR_ELEM("KmemWait", 0, 0, CNTR_NORMAL,
access_sw_kmem_wait),
[C_SW_SEND_SCHED] = CNTR_ELEM("SendSched", 0, 0, CNTR_NORMAL,
access_sw_send_schedule),
+[C_SDMA_DESC_FETCHED_CNT] = CNTR_ELEM("SDEDscFdCn",
+ SEND_DMA_DESC_FETCHED_CNT, 0,
+ CNTR_NORMAL | CNTR_32BIT | CNTR_SDMA,
+ dev_access_u32_csr),
+[C_SDMA_INT_CNT] = CNTR_ELEM("SDMAInt", 0, 0,
+ CNTR_NORMAL | CNTR_32BIT | CNTR_SDMA,
+ access_sde_int_cnt),
+[C_SDMA_ERR_CNT] = CNTR_ELEM("SDMAErrCt", 0, 0,
+ CNTR_NORMAL | CNTR_32BIT | CNTR_SDMA,
+ access_sde_err_cnt),
+[C_SDMA_IDLE_INT_CNT] = CNTR_ELEM("SDMAIdInt", 0, 0,
+ CNTR_NORMAL | CNTR_32BIT | CNTR_SDMA,
+ access_sde_idle_int_cnt),
+[C_SDMA_PROGRESS_INT_CNT] = CNTR_ELEM("SDMAPrIntCn", 0, 0,
+ CNTR_NORMAL | CNTR_32BIT | CNTR_SDMA,
+ access_sde_progress_int_cnt),
/* MISC_ERR_STATUS */
[C_MISC_PLL_LOCK_FAIL_ERR] = CNTR_ELEM("MISC_PLL_LOCK_FAIL_ERR", 0, 0,
CNTR_NORMAL,
if (source < ARRAY_SIZE(various_names))
strncpy(buf, various_names[source], bsize);
else
- snprintf(buf, bsize, "Reserved%u", source+IS_VARIOUS_START);
+ snprintf(buf, bsize, "Reserved%u", source + IS_VARIOUS_START);
return buf;
}
OPA_LINKDOWN_REASON_EXCESSIVE_BUFFER_OVERRUN);
queue_work(ppd->hfi1_wq, &ppd->link_bounce_work);
}
- dd->rcv_ovfl_cnt = (u32) cur_ovfl_cnt;
+ dd->rcv_ovfl_cnt = (u32)cur_ovfl_cnt;
mod_timer(&dd->rcverr_timer, jiffies + HZ * RCVERR_CHECK_TIME);
}
}
}
-static void count_port_inactive(struct hfi1_devdata *dd)
+static inline void __count_port_discards(struct hfi1_pportdata *ppd)
{
- struct hfi1_pportdata *ppd = dd->pport;
+ incr_cntr64(&ppd->port_xmit_discards);
+}
- if (ppd->port_xmit_discards < ~(u64)0)
- ppd->port_xmit_discards++;
+static void count_port_inactive(struct hfi1_devdata *dd)
+{
+ __count_port_discards(dd->pport);
}
/*
* egress error if more than one packet fails the same integrity check
* since we cleared the corresponding bit in SEND_EGRESS_ERR_INFO.
*/
-static void handle_send_egress_err_info(struct hfi1_devdata *dd)
+static void handle_send_egress_err_info(struct hfi1_devdata *dd,
+ int vl)
{
struct hfi1_pportdata *ppd = dd->pport;
u64 src = read_csr(dd, SEND_EGRESS_ERR_SOURCE); /* read first */
info, egress_err_info_string(buf, sizeof(buf), info), src);
/* Eventually add other counters for each bit */
+ if (info & PORT_DISCARD_EGRESS_ERRS) {
+ int weight, i;
- if (info & SEND_EGRESS_ERR_INFO_TOO_LONG_IB_PACKET_ERR_SMASK) {
- if (ppd->port_xmit_discards < ~(u64)0)
- ppd->port_xmit_discards++;
+ /*
+ * Count all, in case multiple bits are set. Reminder:
+ * since there is only one info register for many sources,
+ * these may be attributed to the wrong VL if they occur
+ * too close together.
+ */
+ weight = hweight64(info);
+ for (i = 0; i < weight; i++) {
+ __count_port_discards(ppd);
+ if (vl >= 0 && vl < TXE_NUM_DATA_VL)
+ incr_cntr64(&ppd->port_xmit_discards_vl[vl]);
+ else if (vl == 15)
+ incr_cntr64(&ppd->port_xmit_discards_vl
+ [C_VL_15]);
+ }
}
}
* Input value is a bit position within the SEND_EGRESS_ERR_STATUS
* register. Does it represent a 'disallowed packet' error?
*/
-static inline int disallowed_pkt_err(u64 posn)
+static inline int disallowed_pkt_err(int posn)
{
return (posn >= SEES(TX_SDMA0_DISALLOWED_PACKET) &&
posn <= SEES(TX_SDMA15_DISALLOWED_PACKET));
}
+/*
+ * Input value is a bit position of one of the SDMA engine disallowed
+ * packet errors. Return which engine. Use of this must be guarded by
+ * disallowed_pkt_err().
+ */
+static inline int disallowed_pkt_engine(int posn)
+{
+ return posn - SEES(TX_SDMA0_DISALLOWED_PACKET);
+}
+
+/*
+ * Translate an SDMA engine to a VL. Return -1 if the tranlation cannot
+ * be done.
+ */
+static int engine_to_vl(struct hfi1_devdata *dd, int engine)
+{
+ struct sdma_vl_map *m;
+ int vl;
+
+ /* range check */
+ if (engine < 0 || engine >= TXE_NUM_SDMA_ENGINES)
+ return -1;
+
+ rcu_read_lock();
+ m = rcu_dereference(dd->sdma_map);
+ vl = m->engine_to_vl[engine];
+ rcu_read_unlock();
+
+ return vl;
+}
+
+/*
+ * Translate the send context (sofware index) into a VL. Return -1 if the
+ * translation cannot be done.
+ */
+static int sc_to_vl(struct hfi1_devdata *dd, int sw_index)
+{
+ struct send_context_info *sci;
+ struct send_context *sc;
+ int i;
+
+ sci = &dd->send_contexts[sw_index];
+
+ /* there is no information for user (PSM) and ack contexts */
+ if (sci->type != SC_KERNEL)
+ return -1;
+
+ sc = sci->sc;
+ if (!sc)
+ return -1;
+ if (dd->vld[15].sc == sc)
+ return 15;
+ for (i = 0; i < num_vls; i++)
+ if (dd->vld[i].sc == sc)
+ return i;
+
+ return -1;
+}
+
static void handle_egress_err(struct hfi1_devdata *dd, u32 unused, u64 reg)
{
u64 reg_copy = reg, handled = 0;
if (reg & ALL_TXE_EGRESS_FREEZE_ERR)
start_freeze_handling(dd->pport, 0);
- if (is_ax(dd) && (reg &
- SEND_EGRESS_ERR_STATUS_TX_CREDIT_RETURN_VL_ERR_SMASK)
- && (dd->icode != ICODE_FUNCTIONAL_SIMULATOR))
+ else if (is_ax(dd) &&
+ (reg & SEND_EGRESS_ERR_STATUS_TX_CREDIT_RETURN_VL_ERR_SMASK) &&
+ (dd->icode != ICODE_FUNCTIONAL_SIMULATOR))
start_freeze_handling(dd->pport, 0);
while (reg_copy) {
int posn = fls64(reg_copy);
- /*
- * fls64() returns a 1-based offset, but we generally
- * want 0-based offsets.
- */
+ /* fls64() returns a 1-based offset, we want it zero based */
int shift = posn - 1;
+ u64 mask = 1ULL << shift;
if (port_inactive_err(shift)) {
count_port_inactive(dd);
- handled |= (1ULL << shift);
+ handled |= mask;
} else if (disallowed_pkt_err(shift)) {
- handle_send_egress_err_info(dd);
- handled |= (1ULL << shift);
+ int vl = engine_to_vl(dd, disallowed_pkt_engine(shift));
+
+ handle_send_egress_err_info(dd, vl);
+ handled |= mask;
}
- clear_bit(shift, (unsigned long *)®_copy);
+ reg_copy &= ~mask;
}
reg &= ~handled;
send_context_err_status_string(flags, sizeof(flags), status));
if (status & SEND_CTXT_ERR_STATUS_PIO_DISALLOWED_PACKET_ERR_SMASK)
- handle_send_egress_err_info(dd);
+ handle_send_egress_err_info(dd, sc_to_vl(dd, sw_index));
/*
* Automatically restart halted kernel contexts out of interrupt
dd_dev_err(sde->dd, "CONFIG SDMA(%u) source: %u status 0x%llx\n",
sde->this_idx, source, (unsigned long long)status);
#endif
+ sde->err_cnt++;
sdma_engine_error(sde, status);
/*
static void handle_qsfp_int(struct hfi1_devdata *dd, u32 src_ctx, u64 reg)
{
- /* source is always zero */
+ /* src_ctx is always zero */
struct hfi1_pportdata *ppd = dd->pport;
unsigned long flags;
u64 qsfp_int_mgmt = (u64)(QSFP_HFI0_INT_N | QSFP_HFI0_MODPRST_N);
* an interrupt when a cable is inserted
*/
ppd->qsfp_info.cache_valid = 0;
- ppd->qsfp_info.qsfp_interrupt_functional = 0;
+ ppd->qsfp_info.reset_needed = 0;
+ ppd->qsfp_info.limiting_active = 0;
spin_unlock_irqrestore(&ppd->qsfp_info.qsfp_lock,
flags);
- write_csr(dd,
- dd->hfi1_id ?
- ASIC_QSFP2_INVERT :
- ASIC_QSFP1_INVERT,
- qsfp_int_mgmt);
+ /* Invert the ModPresent pin now to detect plug-in */
+ write_csr(dd, dd->hfi1_id ? ASIC_QSFP2_INVERT :
+ ASIC_QSFP1_INVERT, qsfp_int_mgmt);
+
+ if ((ppd->offline_disabled_reason >
+ HFI1_ODR_MASK(
+ OPA_LINKDOWN_REASON_LOCAL_MEDIA_NOT_INSTALLED)) ||
+ (ppd->offline_disabled_reason ==
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE)))
+ ppd->offline_disabled_reason =
+ HFI1_ODR_MASK(
+ OPA_LINKDOWN_REASON_LOCAL_MEDIA_NOT_INSTALLED);
+
if (ppd->host_link_state == HLS_DN_POLL) {
/*
* The link is still in POLL. This means
spin_unlock_irqrestore(&ppd->qsfp_info.qsfp_lock,
flags);
+ /*
+ * Stop inversion of ModPresent pin to detect
+ * removal of the cable
+ */
qsfp_int_mgmt &= ~(u64)QSFP_HFI0_MODPRST_N;
- write_csr(dd,
- dd->hfi1_id ?
- ASIC_QSFP2_INVERT :
- ASIC_QSFP1_INVERT,
- qsfp_int_mgmt);
+ write_csr(dd, dd->hfi1_id ? ASIC_QSFP2_INVERT :
+ ASIC_QSFP1_INVERT, qsfp_int_mgmt);
+
+ ppd->offline_disabled_reason =
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_TRANSIENT);
}
}
__func__);
spin_lock_irqsave(&ppd->qsfp_info.qsfp_lock, flags);
ppd->qsfp_info.check_interrupt_flags = 1;
- ppd->qsfp_info.qsfp_interrupt_functional = 1;
spin_unlock_irqrestore(&ppd->qsfp_info.qsfp_lock, flags);
}
}
/*
- * Handle requests from the 8051.
+ * Handle host requests from the 8051.
+ *
+ * This is a work-queue function outside of the interrupt.
*/
-static void handle_8051_request(struct hfi1_devdata *dd)
+void handle_8051_request(struct work_struct *work)
{
+ struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
+ dc_host_req_work);
+ struct hfi1_devdata *dd = ppd->dd;
u64 reg;
- u16 data;
- u8 type;
+ u16 data = 0;
+ u8 type, i, lanes, *cache = ppd->qsfp_info.cache;
+ u8 cdr_ctrl_byte = cache[QSFP_CDR_CTRL_BYTE_OFFS];
reg = read_csr(dd, DC_DC8051_CFG_EXT_DEV_1);
if ((reg & DC_DC8051_CFG_EXT_DEV_1_REQ_NEW_SMASK) == 0)
case HREQ_READ_CONFIG:
case HREQ_SET_TX_EQ_ABS:
case HREQ_SET_TX_EQ_REL:
- case HREQ_ENABLE:
dd_dev_info(dd, "8051 request: request 0x%x not supported\n",
type);
hreq_response(dd, HREQ_NOT_SUPPORTED, 0);
break;
+ case HREQ_ENABLE:
+ lanes = data & 0xF;
+ for (i = 0; lanes; lanes >>= 1, i++) {
+ if (!(lanes & 1))
+ continue;
+ if (data & 0x200) {
+ /* enable TX CDR */
+ if (cache[QSFP_MOD_PWR_OFFS] & 0x8 &&
+ cache[QSFP_CDR_INFO_OFFS] & 0x80)
+ cdr_ctrl_byte |= (1 << (i + 4));
+ } else {
+ /* disable TX CDR */
+ if (cache[QSFP_MOD_PWR_OFFS] & 0x8 &&
+ cache[QSFP_CDR_INFO_OFFS] & 0x80)
+ cdr_ctrl_byte &= ~(1 << (i + 4));
+ }
+
+ if (data & 0x800) {
+ /* enable RX CDR */
+ if (cache[QSFP_MOD_PWR_OFFS] & 0x4 &&
+ cache[QSFP_CDR_INFO_OFFS] & 0x40)
+ cdr_ctrl_byte |= (1 << i);
+ } else {
+ /* disable RX CDR */
+ if (cache[QSFP_MOD_PWR_OFFS] & 0x4 &&
+ cache[QSFP_CDR_INFO_OFFS] & 0x40)
+ cdr_ctrl_byte &= ~(1 << i);
+ }
+ }
+ qsfp_write(ppd, ppd->dd->hfi1_id, QSFP_CDR_CTRL_BYTE_OFFS,
+ &cdr_ctrl_byte, 1);
+ hreq_response(dd, HREQ_SUCCESS, data);
+ refresh_qsfp_cache(ppd, &ppd->qsfp_info);
+ break;
+
case HREQ_CONFIG_DONE:
hreq_response(dd, HREQ_SUCCESS, 0);
break;
/* remove all previous VL credit limits */
for (i = 0; i < TXE_NUM_DATA_VL; i++)
- write_csr(dd, SEND_CM_CREDIT_VL + (8*i), 0);
+ write_csr(dd, SEND_CM_CREDIT_VL + (8 * i), 0);
write_csr(dd, SEND_CM_CREDIT_VL15, 0);
write_global_credit(dd, 0, 0, 0);
/* reset the CM block */
reg
| (1ull << DCC_CFG_RESET_RESET_LCB_SHIFT)
| (1ull << DCC_CFG_RESET_RESET_RX_FPE_SHIFT));
- (void) read_csr(dd, DCC_CFG_RESET); /* make sure the write completed */
+ (void)read_csr(dd, DCC_CFG_RESET); /* make sure the write completed */
if (!abort) {
udelay(1); /* must hold for the longer of 16cclks or 20ns */
write_csr(dd, DCC_CFG_RESET, reg);
*/
static void rxe_kernel_unfreeze(struct hfi1_devdata *dd)
{
+ u32 rcvmask;
int i;
/* enable all kernel contexts */
- for (i = 0; i < dd->n_krcv_queues; i++)
- hfi1_rcvctrl(dd, HFI1_RCVCTRL_CTXT_ENB, i);
+ for (i = 0; i < dd->n_krcv_queues; i++) {
+ rcvmask = HFI1_RCVCTRL_CTXT_ENB;
+ /* HFI1_RCVCTRL_TAILUPD_[ENB|DIS] needs to be set explicitly */
+ rcvmask |= HFI1_CAP_KGET_MASK(dd->rcd[i]->flags, DMA_RTAIL) ?
+ HFI1_RCVCTRL_TAILUPD_ENB : HFI1_RCVCTRL_TAILUPD_DIS;
+ hfi1_rcvctrl(dd, rcvmask, i);
+ }
/* enable port */
add_rcvctrl(dd, RCV_CTRL_RCV_PORT_ENABLE_SMASK);
set_link_down_reason(ppd, OPA_LINKDOWN_REASON_SPEED_POLICY, 0,
OPA_LINKDOWN_REASON_SPEED_POLICY);
set_link_state(ppd, HLS_DN_OFFLINE);
+ tune_serdes(ppd);
start_link(ppd);
}
}
struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
link_down_work);
- /* go offline first, then deal with reasons */
+ if ((ppd->host_link_state &
+ (HLS_DN_POLL | HLS_VERIFY_CAP | HLS_GOING_UP)) &&
+ ppd->port_type == PORT_TYPE_FIXED)
+ ppd->offline_disabled_reason =
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NOT_INSTALLED);
+
+ /* Go offline first, then deal with reading/writing through 8051 */
set_link_state(ppd, HLS_DN_OFFLINE);
lcl_reason = 0;
/* If there is no cable attached, turn the DC off. Otherwise,
* start the link bring up. */
- if (!qsfp_mod_present(ppd))
+ if (!qsfp_mod_present(ppd)) {
dc_shutdown(ppd->dd);
- else
+ } else {
+ tune_serdes(ppd);
start_link(ppd);
+ }
}
void handle_link_bounce(struct work_struct *work)
*/
if (ppd->host_link_state & HLS_UP) {
set_link_state(ppd, HLS_DN_OFFLINE);
+ tune_serdes(ppd);
start_link(ppd);
} else {
dd_dev_info(ppd->dd, "%s: link not up (%s), nothing to do\n",
set_link_down_reason(ppd, OPA_LINKDOWN_REASON_WIDTH_POLICY, 0,
OPA_LINKDOWN_REASON_WIDTH_POLICY);
set_link_state(ppd, HLS_DN_OFFLINE);
+ tune_serdes(ppd);
start_link(ppd);
}
}
host_msg &= ~(u64)LINKUP_ACHIEVED;
}
if (host_msg & EXT_DEVICE_CFG_REQ) {
- handle_8051_request(dd);
+ queue_work(ppd->hfi1_wq, &ppd->dc_host_req_work);
host_msg &= ~(u64)EXT_DEVICE_CFG_REQ;
}
if (host_msg & VERIFY_CAP_FRAME) {
/* if the link is already going down or disabled, do not
* queue another */
if ((ppd->host_link_state
- & (HLS_GOING_OFFLINE|HLS_LINK_COOLDOWN))
+ & (HLS_GOING_OFFLINE | HLS_LINK_COOLDOWN))
|| ppd->link_enabled == 0) {
dd_dev_info(dd, "%s: not queuing link down\n",
__func__);
/* phase 2: call the appropriate handler */
for_each_set_bit(bit, (unsigned long *)®s[0],
- CCE_NUM_INT_CSRS*64) {
+ CCE_NUM_INT_CSRS * 64) {
is_interrupt(dd, bit);
}
/* This read_csr is really bad in the hot path */
status = read_csr(dd,
- CCE_INT_STATUS + (8*(IS_SDMA_START/64)))
+ CCE_INT_STATUS + (8 * (IS_SDMA_START / 64)))
& sde->imask;
if (likely(status)) {
/* clear the interrupt(s) */
write_csr(dd,
- CCE_INT_CLEAR + (8*(IS_SDMA_START/64)),
+ CCE_INT_CLEAR + (8 * (IS_SDMA_START / 64)),
status);
/* handle the interrupt(s) */
}
/*
- * Clear the receive interrupt, forcing the write and making sure
- * we have data from the chip, pushing everything in front of it
- * back to the host.
+ * Clear the receive interrupt. Use a read of the interrupt clear CSR
+ * to insure that the write completed. This does NOT guarantee that
+ * queued DMA writes to memory from the chip are pushed.
*/
static inline void clear_recv_intr(struct hfi1_ctxtdata *rcd)
{
}
/* force the receive interrupt */
-static inline void force_recv_intr(struct hfi1_ctxtdata *rcd)
+void force_recv_intr(struct hfi1_ctxtdata *rcd)
{
write_csr(rcd->dd, CCE_INT_FORCE + (8 * rcd->ireg), rcd->imask);
}
-/* return non-zero if a packet is present */
+/*
+ * Return non-zero if a packet is present.
+ *
+ * This routine is called when rechecking for packets after the RcvAvail
+ * interrupt has been cleared down. First, do a quick check of memory for
+ * a packet present. If not found, use an expensive CSR read of the context
+ * tail to determine the actual tail. The CSR read is necessary because there
+ * is no method to push pending DMAs to memory other than an interrupt and we
+ * are trying to determine if we need to force an interrupt.
+ */
static inline int check_packet_present(struct hfi1_ctxtdata *rcd)
{
+ u32 tail;
+ int present;
+
if (!HFI1_CAP_IS_KSET(DMA_RTAIL))
- return (rcd->seq_cnt ==
+ present = (rcd->seq_cnt ==
rhf_rcv_seq(rhf_to_cpu(get_rhf_addr(rcd))));
+ else /* is RDMA rtail */
+ present = (rcd->head != get_rcvhdrtail(rcd));
- /* else is RDMA rtail */
- return (rcd->head != get_rcvhdrtail(rcd));
+ if (present)
+ return 1;
+
+ /* fall back to a CSR read, correct indpendent of DMA_RTAIL */
+ tail = (u32)read_uctxt_csr(rcd->dd, rcd->ctxt, RCV_HDR_TAIL);
+ return rcd->head != tail;
}
/*
trace_hfi1_receive_interrupt(dd, rcd->ctxt);
this_cpu_inc(*dd->int_counter);
+ aspm_ctx_disable(rcd);
/* receive interrupt remains blocked while processing packets */
disposition = rcd->do_interrupt(rcd, 0);
& DC_DC8051_STS_CUR_STATE_PORT_MASK;
}
-static u32 read_logical_state(struct hfi1_devdata *dd)
+u32 read_logical_state(struct hfi1_devdata *dd)
{
u64 reg;
return do_8051_command(dd, HCMD_CHANGE_PHY_STATE, state, NULL);
}
-static int load_8051_config(struct hfi1_devdata *dd, u8 field_id,
- u8 lane_id, u32 config_data)
+int load_8051_config(struct hfi1_devdata *dd, u8 field_id,
+ u8 lane_id, u32 config_data)
{
u64 data;
int ret;
* set the result, even on error.
* Return 0 on success, -errno on failure
*/
-static int read_8051_config(struct hfi1_devdata *dd, u8 field_id, u8 lane_id,
- u32 *result)
+int read_8051_config(struct hfi1_devdata *dd, u8 field_id, u8 lane_id,
+ u32 *result)
{
u64 big_data;
u32 addr;
u16 from;
u16 to;
} opa_link_xlate[] = {
- { OPA_LINK_WIDTH_1X, 1 << (1-1) },
- { OPA_LINK_WIDTH_2X, 1 << (2-1) },
- { OPA_LINK_WIDTH_3X, 1 << (3-1) },
- { OPA_LINK_WIDTH_4X, 1 << (4-1) },
+ { OPA_LINK_WIDTH_1X, 1 << (1 - 1) },
+ { OPA_LINK_WIDTH_2X, 1 << (2 - 1) },
+ { OPA_LINK_WIDTH_3X, 1 << (3 - 1) },
+ { OPA_LINK_WIDTH_4X, 1 << (4 - 1) },
};
for (i = 0; i < ARRAY_SIZE(opa_link_xlate); i++) {
return -EAGAIN;
}
-static void reset_qsfp(struct hfi1_pportdata *ppd)
+static void wait_for_qsfp_init(struct hfi1_pportdata *ppd)
+{
+ struct hfi1_devdata *dd = ppd->dd;
+ u64 mask;
+ unsigned long timeout;
+
+ /*
+ * Check for QSFP interrupt for t_init (SFF 8679)
+ */
+ timeout = jiffies + msecs_to_jiffies(2000);
+ while (1) {
+ mask = read_csr(dd, dd->hfi1_id ?
+ ASIC_QSFP2_IN : ASIC_QSFP1_IN);
+ if (!(mask & QSFP_HFI0_INT_N)) {
+ write_csr(dd, dd->hfi1_id ? ASIC_QSFP2_CLEAR :
+ ASIC_QSFP1_CLEAR, QSFP_HFI0_INT_N);
+ break;
+ }
+ if (time_after(jiffies, timeout)) {
+ dd_dev_info(dd, "%s: No IntN detected, reset complete\n",
+ __func__);
+ break;
+ }
+ udelay(2);
+ }
+}
+
+static void set_qsfp_int_n(struct hfi1_pportdata *ppd, u8 enable)
+{
+ struct hfi1_devdata *dd = ppd->dd;
+ u64 mask;
+
+ mask = read_csr(dd, dd->hfi1_id ? ASIC_QSFP2_MASK : ASIC_QSFP1_MASK);
+ if (enable)
+ mask |= (u64)QSFP_HFI0_INT_N;
+ else
+ mask &= ~(u64)QSFP_HFI0_INT_N;
+ write_csr(dd, dd->hfi1_id ? ASIC_QSFP2_MASK : ASIC_QSFP1_MASK, mask);
+}
+
+void reset_qsfp(struct hfi1_pportdata *ppd)
{
struct hfi1_devdata *dd = ppd->dd;
u64 mask, qsfp_mask;
+ /* Disable INT_N from triggering QSFP interrupts */
+ set_qsfp_int_n(ppd, 0);
+
+ /* Reset the QSFP */
mask = (u64)QSFP_HFI0_RESET_N;
- qsfp_mask = read_csr(dd,
- dd->hfi1_id ? ASIC_QSFP2_OE : ASIC_QSFP1_OE);
+ qsfp_mask = read_csr(dd, dd->hfi1_id ? ASIC_QSFP2_OE : ASIC_QSFP1_OE);
qsfp_mask |= mask;
write_csr(dd,
- dd->hfi1_id ? ASIC_QSFP2_OE : ASIC_QSFP1_OE,
- qsfp_mask);
+ dd->hfi1_id ? ASIC_QSFP2_OE : ASIC_QSFP1_OE, qsfp_mask);
- qsfp_mask = read_csr(dd,
- dd->hfi1_id ? ASIC_QSFP2_OUT : ASIC_QSFP1_OUT);
+ qsfp_mask = read_csr(dd, dd->hfi1_id ?
+ ASIC_QSFP2_OUT : ASIC_QSFP1_OUT);
qsfp_mask &= ~mask;
write_csr(dd,
- dd->hfi1_id ? ASIC_QSFP2_OUT : ASIC_QSFP1_OUT,
- qsfp_mask);
+ dd->hfi1_id ? ASIC_QSFP2_OUT : ASIC_QSFP1_OUT, qsfp_mask);
udelay(10);
qsfp_mask |= mask;
write_csr(dd,
- dd->hfi1_id ? ASIC_QSFP2_OUT : ASIC_QSFP1_OUT,
- qsfp_mask);
+ dd->hfi1_id ? ASIC_QSFP2_OUT : ASIC_QSFP1_OUT, qsfp_mask);
+
+ wait_for_qsfp_init(ppd);
+
+ /*
+ * Allow INT_N to trigger the QSFP interrupt to watch
+ * for alarms and warnings
+ */
+ set_qsfp_int_n(ppd, 1);
}
static int handle_qsfp_error_conditions(struct hfi1_pportdata *ppd,
return 0;
}
-static int do_pre_lni_host_behaviors(struct hfi1_pportdata *ppd)
-{
- refresh_qsfp_cache(ppd, &ppd->qsfp_info);
-
- return 0;
-}
-
-static int do_qsfp_intr_fallback(struct hfi1_pportdata *ppd)
-{
- struct hfi1_devdata *dd = ppd->dd;
- u8 qsfp_interrupt_status = 0;
-
- if (qsfp_read(ppd, dd->hfi1_id, 2, &qsfp_interrupt_status, 1)
- != 1) {
- dd_dev_info(dd,
- "%s: Failed to read status of QSFP module\n",
- __func__);
- return -EIO;
- }
-
- /* We don't care about alarms & warnings with a non-functional INT_N */
- if (!(qsfp_interrupt_status & QSFP_DATA_NOT_READY))
- do_pre_lni_host_behaviors(ppd);
-
- return 0;
-}
-
/* This routine will only be scheduled if the QSFP module is present */
-static void qsfp_event(struct work_struct *work)
+void qsfp_event(struct work_struct *work)
{
struct qsfp_data *qd;
struct hfi1_pportdata *ppd;
dc_start(dd);
if (qd->cache_refresh_required) {
- msleep(3000);
- reset_qsfp(ppd);
- /* Check for QSFP interrupt after t_init (SFF 8679)
- * + extra
+ set_qsfp_int_n(ppd, 0);
+
+ wait_for_qsfp_init(ppd);
+
+ /*
+ * Allow INT_N to trigger the QSFP interrupt to watch
+ * for alarms and warnings
*/
- msleep(3000);
- if (!qd->qsfp_interrupt_functional) {
- if (do_qsfp_intr_fallback(ppd) < 0)
- dd_dev_info(dd, "%s: QSFP fallback failed\n",
- __func__);
- ppd->driver_link_ready = 1;
- start_link(ppd);
- }
+ set_qsfp_int_n(ppd, 1);
+
+ tune_serdes(ppd);
+
+ start_link(ppd);
}
if (qd->check_interrupt_flags) {
__func__);
} else {
unsigned long flags;
- u8 data_status;
+ handle_qsfp_error_conditions(
+ ppd, qsfp_interrupt_status);
spin_lock_irqsave(&ppd->qsfp_info.qsfp_lock, flags);
ppd->qsfp_info.check_interrupt_flags = 0;
spin_unlock_irqrestore(&ppd->qsfp_info.qsfp_lock,
flags);
-
- if (qsfp_read(ppd, dd->hfi1_id, 2, &data_status, 1)
- != 1) {
- dd_dev_info(dd,
- "%s: Failed to read status of QSFP module\n",
- __func__);
- }
- if (!(data_status & QSFP_DATA_NOT_READY)) {
- do_pre_lni_host_behaviors(ppd);
- start_link(ppd);
- } else
- handle_qsfp_error_conditions(ppd,
- qsfp_interrupt_status);
}
}
}
-void init_qsfp(struct hfi1_pportdata *ppd)
+static void init_qsfp_int(struct hfi1_devdata *dd)
{
- struct hfi1_devdata *dd = ppd->dd;
- u64 qsfp_mask;
+ struct hfi1_pportdata *ppd = dd->pport;
+ u64 qsfp_mask, cce_int_mask;
+ const int qsfp1_int_smask = QSFP1_INT % 64;
+ const int qsfp2_int_smask = QSFP2_INT % 64;
- if (loopback == LOOPBACK_SERDES || loopback == LOOPBACK_LCB ||
- ppd->dd->icode == ICODE_FUNCTIONAL_SIMULATOR) {
- ppd->driver_link_ready = 1;
- return;
+ /*
+ * disable QSFP1 interrupts for HFI1, QSFP2 interrupts for HFI0
+ * Qsfp1Int and Qsfp2Int are adjacent bits in the same CSR,
+ * therefore just one of QSFP1_INT/QSFP2_INT can be used to find
+ * the index of the appropriate CSR in the CCEIntMask CSR array
+ */
+ cce_int_mask = read_csr(dd, CCE_INT_MASK +
+ (8 * (QSFP1_INT / 64)));
+ if (dd->hfi1_id) {
+ cce_int_mask &= ~((u64)1 << qsfp1_int_smask);
+ write_csr(dd, CCE_INT_MASK + (8 * (QSFP1_INT / 64)),
+ cce_int_mask);
+ } else {
+ cce_int_mask &= ~((u64)1 << qsfp2_int_smask);
+ write_csr(dd, CCE_INT_MASK + (8 * (QSFP2_INT / 64)),
+ cce_int_mask);
}
- ppd->qsfp_info.ppd = ppd;
- INIT_WORK(&ppd->qsfp_info.qsfp_work, qsfp_event);
-
qsfp_mask = (u64)(QSFP_HFI0_INT_N | QSFP_HFI0_MODPRST_N);
/* Clear current status to avoid spurious interrupts */
- write_csr(dd,
- dd->hfi1_id ?
- ASIC_QSFP2_CLEAR :
- ASIC_QSFP1_CLEAR,
- qsfp_mask);
+ write_csr(dd, dd->hfi1_id ? ASIC_QSFP2_CLEAR : ASIC_QSFP1_CLEAR,
+ qsfp_mask);
+ write_csr(dd, dd->hfi1_id ? ASIC_QSFP2_MASK : ASIC_QSFP1_MASK,
+ qsfp_mask);
+
+ set_qsfp_int_n(ppd, 0);
/* Handle active low nature of INT_N and MODPRST_N pins */
if (qsfp_mod_present(ppd))
write_csr(dd,
dd->hfi1_id ? ASIC_QSFP2_INVERT : ASIC_QSFP1_INVERT,
qsfp_mask);
-
- /* Allow only INT_N and MODPRST_N to trigger QSFP interrupts */
- qsfp_mask |= (u64)QSFP_HFI0_MODPRST_N;
- write_csr(dd,
- dd->hfi1_id ? ASIC_QSFP2_MASK : ASIC_QSFP1_MASK,
- qsfp_mask);
-
- if (qsfp_mod_present(ppd)) {
- msleep(3000);
- reset_qsfp(ppd);
-
- /* Check for QSFP interrupt after t_init (SFF 8679)
- * + extra
- */
- msleep(3000);
- if (!ppd->qsfp_info.qsfp_interrupt_functional) {
- if (do_qsfp_intr_fallback(ppd) < 0)
- dd_dev_info(dd,
- "%s: QSFP fallback failed\n",
- __func__);
- ppd->driver_link_ready = 1;
- }
- }
}
/*
*/
static void init_lcb(struct hfi1_devdata *dd)
{
+ /* simulator does not correctly handle LCB cclk loopback, skip */
+ if (dd->icode == ICODE_FUNCTIONAL_SIMULATOR)
+ return;
+
/* the DC has been reset earlier in the driver load */
/* set LCB for cclk loopback on the port */
ppd->guid = guid;
}
- /* the link defaults to enabled */
- ppd->link_enabled = 1;
/* Set linkinit_reason on power up per OPA spec */
ppd->linkinit_reason = OPA_LINKINIT_REASON_LINKUP;
return ret;
}
+ /* tune the SERDES to a ballpark setting for
+ * optimal signal and bit error rate
+ * Needs to be done before starting the link
+ */
+ tune_serdes(ppd);
+
return start_link(ppd);
}
ppd->driver_link_ready = 0;
ppd->link_enabled = 0;
+ ppd->offline_disabled_reason =
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
set_link_down_reason(ppd, OPA_LINKDOWN_REASON_SMA_DISABLED, 0,
OPA_LINKDOWN_REASON_SMA_DISABLED);
set_link_state(ppd, HLS_DN_OFFLINE);
ppd = (struct hfi1_pportdata *)(dd + 1);
for (i = 0; i < dd->num_pports; i++, ppd++) {
- ppd->ibport_data.rc_acks = NULL;
- ppd->ibport_data.rc_qacks = NULL;
- ppd->ibport_data.rc_acks = alloc_percpu(u64);
- ppd->ibport_data.rc_qacks = alloc_percpu(u64);
- ppd->ibport_data.rc_delayed_comp = alloc_percpu(u64);
- if ((ppd->ibport_data.rc_acks == NULL) ||
- (ppd->ibport_data.rc_delayed_comp == NULL) ||
- (ppd->ibport_data.rc_qacks == NULL))
+ ppd->ibport_data.rvp.rc_acks = NULL;
+ ppd->ibport_data.rvp.rc_qacks = NULL;
+ ppd->ibport_data.rvp.rc_acks = alloc_percpu(u64);
+ ppd->ibport_data.rvp.rc_qacks = alloc_percpu(u64);
+ ppd->ibport_data.rvp.rc_delayed_comp = alloc_percpu(u64);
+ if (!ppd->ibport_data.rvp.rc_acks ||
+ !ppd->ibport_data.rvp.rc_delayed_comp ||
+ !ppd->ibport_data.rvp.rc_qacks)
return -ENOMEM;
}
c1 &= ~(DCC_CFG_PORT_CONFIG1_TARGET_DLID_SMASK
| DCC_CFG_PORT_CONFIG1_DLID_MASK_SMASK);
c1 |= ((ppd->lid & DCC_CFG_PORT_CONFIG1_TARGET_DLID_MASK)
- << DCC_CFG_PORT_CONFIG1_TARGET_DLID_SHIFT)|
+ << DCC_CFG_PORT_CONFIG1_TARGET_DLID_SHIFT) |
((mask & DCC_CFG_PORT_CONFIG1_DLID_MASK_MASK)
<< DCC_CFG_PORT_CONFIG1_DLID_MASK_SHIFT);
write_csr(ppd->dd, DCC_CFG_PORT_CONFIG1, c1);
ret);
return -EINVAL;
}
- if (ppd->offline_disabled_reason == OPA_LINKDOWN_REASON_NONE)
+ if (ppd->offline_disabled_reason ==
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE))
ppd->offline_disabled_reason =
- OPA_LINKDOWN_REASON_TRANSIENT;
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_TRANSIENT);
}
if (do_wait) {
write_csr(dd, DC_LCB_ERR_EN, ~0ull); /* watch LCB errors */
ppd->host_link_state = HLS_LINK_COOLDOWN; /* LCB access allowed */
+ if (ppd->port_type == PORT_TYPE_QSFP &&
+ ppd->qsfp_info.limiting_active &&
+ qsfp_mod_present(ppd)) {
+ set_qsfp_tx(ppd, 0);
+ }
+
/*
* The LNI has a mandatory wait time after the physical state
* moves to Offline.Quiet. The wait time may be different
sdma_all_running(dd);
/* Signal the IB layer that the port has went active */
- event.device = &dd->verbs_dev.ibdev;
+ event.device = &dd->verbs_dev.rdi.ibdev;
event.element.port_num = ppd->port;
event.event = IB_EVENT_PORT_ACTIVE;
}
ppd->link_enabled = 1;
}
+ set_all_slowpath(ppd->dd);
ret = set_local_link_attributes(ppd);
if (ret)
break;
ret = -EINVAL;
}
}
- ppd->offline_disabled_reason = OPA_LINKDOWN_REASON_NONE;
+ ppd->offline_disabled_reason =
+ HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE);
/*
* If an error occurred above, go back to offline. The
* caller may reschedule another attempt.
* The VL Arbitrator high limit is sent in units of 4k
* bytes, while HFI stores it in units of 64 bytes.
*/
- val *= 4096/64;
+ val *= 4096 / 64;
reg = ((u64)val & SEND_HIGH_PRIORITY_LIMIT_LIMIT_MASK)
<< SEND_HIGH_PRIORITY_LIMIT_LIMIT_SHIFT;
write_csr(ppd->dd, SEND_HIGH_PRIORITY_LIMIT, reg);
ppd->vls_operational = val;
if (!ppd->port)
ret = -EINVAL;
- else
- ret = sdma_map_init(
- ppd->dd,
- ppd->port - 1,
- val,
- NULL);
}
break;
/*
/* OPA and HFI have a 1-1 mapping */
for (i = 0; i < TXE_NUM_DATA_VL; i++)
- read_one_cm_vl(dd, SEND_CM_CREDIT_VL + (8*i), &bc->vl[i]);
+ read_one_cm_vl(dd, SEND_CM_CREDIT_VL + (8 * i), &bc->vl[i]);
/* NOTE: assumes that VL* and VL15 CSRs are bit-wise identical */
read_one_cm_vl(dd, SEND_CM_CREDIT_VL15, &bc->vl[15]);
* raise = if the new limit is higher than the current value (may be changed
* earlier in the algorithm), set the new limit to the new value
*/
-static int set_buffer_control(struct hfi1_devdata *dd,
- struct buffer_control *new_bc)
+int set_buffer_control(struct hfi1_pportdata *ppd,
+ struct buffer_control *new_bc)
{
+ struct hfi1_devdata *dd = ppd->dd;
u64 changing_mask, ld_mask, stat_mask;
int change_count;
int i, use_all_mask;
int this_shared_changing;
+ int vl_count = 0, ret;
/*
* A0: add the variable any_shared_limit_changing below and in the
* algorithm above. If removing A0 support, it can be removed.
#define valid_vl(idx) ((idx) < TXE_NUM_DATA_VL || (idx) == 15)
#define NUM_USABLE_VLS 16 /* look at VL15 and less */
-
/* find the new total credits, do sanity check on unused VLs */
for (i = 0; i < OPA_MAX_VLS; i++) {
if (valid_vl(i)) {
/* bracket the credit change with a total adjustment */
if (new_total < cur_total)
set_global_limit(dd, new_total);
+
+ /*
+ * Determine the actual number of operational VLS using the number of
+ * dedicated and shared credits for each VL.
+ */
+ if (change_count > 0) {
+ for (i = 0; i < TXE_NUM_DATA_VL; i++)
+ if (be16_to_cpu(new_bc->vl[i].dedicated) > 0 ||
+ be16_to_cpu(new_bc->vl[i].shared) > 0)
+ vl_count++;
+ ppd->actual_vls_operational = vl_count;
+ ret = sdma_map_init(dd, ppd->port - 1, vl_count ?
+ ppd->actual_vls_operational :
+ ppd->vls_operational,
+ NULL);
+ if (ret == 0)
+ ret = pio_map_init(dd, ppd->port - 1, vl_count ?
+ ppd->actual_vls_operational :
+ ppd->vls_operational, NULL);
+ if (ret)
+ return ret;
+ }
return 0;
}
VL_ARB_LOW_PRIO_TABLE_SIZE, t);
break;
case FM_TBL_BUFFER_CONTROL:
- ret = set_buffer_control(ppd->dd, t);
+ ret = set_buffer_control(ppd, t);
break;
case FM_TBL_SC2VLNT:
set_sc2vlnt(ppd->dd, t);
static u32 encoded_size(u32 size)
{
switch (size) {
- case 4*1024: return 0x1;
- case 8*1024: return 0x2;
- case 16*1024: return 0x3;
- case 32*1024: return 0x4;
- case 64*1024: return 0x5;
- case 128*1024: return 0x6;
- case 256*1024: return 0x7;
- case 512*1024: return 0x8;
- case 1*1024*1024: return 0x9;
- case 2*1024*1024: return 0xa;
+ case 4 * 1024: return 0x1;
+ case 8 * 1024: return 0x2;
+ case 16 * 1024: return 0x3;
+ case 32 * 1024: return 0x4;
+ case 64 * 1024: return 0x5;
+ case 128 * 1024: return 0x6;
+ case 256 * 1024: return 0x7;
+ case 512 * 1024: return 0x8;
+ case 1 * 1024 * 1024: return 0x9;
+ case 2 * 1024 * 1024: return 0xa;
}
return 0x1; /* if invalid, go with the minimum size */
}
if (dd->rcvhdrtail_dummy_physaddr) {
write_kctxt_csr(dd, ctxt, RCV_HDR_TAIL_ADDR,
dd->rcvhdrtail_dummy_physaddr);
+ /* Enabling RcvCtxtCtrl.TailUpd is intentional. */
rcvctrl |= RCV_CTXT_CTRL_TAIL_UPD_SMASK;
}
rcvctrl &= ~RCV_CTXT_CTRL_INTR_AVAIL_SMASK;
if (op & HFI1_RCVCTRL_TAILUPD_ENB && rcd->rcvhdrqtailaddr_phys)
rcvctrl |= RCV_CTXT_CTRL_TAIL_UPD_SMASK;
- if (op & HFI1_RCVCTRL_TAILUPD_DIS)
- rcvctrl &= ~RCV_CTXT_CTRL_TAIL_UPD_SMASK;
+ if (op & HFI1_RCVCTRL_TAILUPD_DIS) {
+ /* See comment on RcvCtxtCtrl.TailUpd above */
+ if (!(op & HFI1_RCVCTRL_CTXT_DIS))
+ rcvctrl &= ~RCV_CTXT_CTRL_TAIL_UPD_SMASK;
+ }
if (op & HFI1_RCVCTRL_TIDFLOW_ENB)
rcvctrl |= RCV_CTXT_CTRL_TID_FLOW_ENABLE_SMASK;
if (op & HFI1_RCVCTRL_TIDFLOW_DIS)
dd->cntrs[entry->offset + j] =
val;
}
+ } else if (entry->flags & CNTR_SDMA) {
+ hfi1_cdbg(CNTR,
+ "\t Per SDMA Engine\n");
+ for (j = 0; j < dd->chip_sdma_engines;
+ j++) {
+ val =
+ entry->rw_cntr(entry, dd, j,
+ CNTR_MODE_R, 0);
+ hfi1_cdbg(CNTR,
+ "\t\tRead 0x%llx for %d\n",
+ val, j);
+ dd->cntrs[entry->offset + j] =
+ val;
+ }
} else {
val = entry->rw_cntr(entry, dd,
CNTR_INVALID_VL,
for (i = 0; i < dd->num_pports; i++, ppd++) {
kfree(ppd->cntrs);
kfree(ppd->scntrs);
- free_percpu(ppd->ibport_data.rc_acks);
- free_percpu(ppd->ibport_data.rc_qacks);
- free_percpu(ppd->ibport_data.rc_delayed_comp);
+ free_percpu(ppd->ibport_data.rvp.rc_acks);
+ free_percpu(ppd->ibport_data.rvp.rc_qacks);
+ free_percpu(ppd->ibport_data.rvp.rc_delayed_comp);
ppd->cntrs = NULL;
ppd->scntrs = NULL;
- ppd->ibport_data.rc_acks = NULL;
- ppd->ibport_data.rc_qacks = NULL;
- ppd->ibport_data.rc_delayed_comp = NULL;
+ ppd->ibport_data.rvp.rc_acks = NULL;
+ ppd->ibport_data.rvp.rc_qacks = NULL;
+ ppd->ibport_data.rvp.rc_delayed_comp = NULL;
}
kfree(dd->portcntrnames);
dd->portcntrnames = NULL;
#define C_MAX_NAME 13 /* 12 chars + one for /0 */
static int init_cntrs(struct hfi1_devdata *dd)
{
- int i, rcv_ctxts, index, j;
+ int i, rcv_ctxts, j;
size_t sz;
char *p;
char name[C_MAX_NAME];
struct hfi1_pportdata *ppd;
+ const char *bit_type_32 = ",32";
+ const int bit_type_32_sz = strlen(bit_type_32);
/* set up the stats timer; the add_timer is done at the end */
setup_timer(&dd->synth_stats_timer, update_synth_timer,
/* size names and determine how many we have*/
dd->ndevcntrs = 0;
sz = 0;
- index = 0;
for (i = 0; i < DEV_CNTR_LAST; i++) {
- hfi1_dbg_early("Init cntr %s\n", dev_cntrs[i].name);
if (dev_cntrs[i].flags & CNTR_DISABLED) {
hfi1_dbg_early("\tSkipping %s\n", dev_cntrs[i].name);
continue;
}
if (dev_cntrs[i].flags & CNTR_VL) {
- hfi1_dbg_early("\tProcessing VL cntr\n");
- dev_cntrs[i].offset = index;
+ dev_cntrs[i].offset = dd->ndevcntrs;
for (j = 0; j < C_VL_COUNT; j++) {
- memset(name, '\0', C_MAX_NAME);
snprintf(name, C_MAX_NAME, "%s%d",
dev_cntrs[i].name,
vl_from_idx(j));
sz += strlen(name);
+ /* Add ",32" for 32-bit counters */
+ if (dev_cntrs[i].flags & CNTR_32BIT)
+ sz += bit_type_32_sz;
+ sz++;
+ dd->ndevcntrs++;
+ }
+ } else if (dev_cntrs[i].flags & CNTR_SDMA) {
+ dev_cntrs[i].offset = dd->ndevcntrs;
+ for (j = 0; j < dd->chip_sdma_engines; j++) {
+ snprintf(name, C_MAX_NAME, "%s%d",
+ dev_cntrs[i].name, j);
+ sz += strlen(name);
+ /* Add ",32" for 32-bit counters */
+ if (dev_cntrs[i].flags & CNTR_32BIT)
+ sz += bit_type_32_sz;
sz++;
- hfi1_dbg_early("\t\t%s\n", name);
dd->ndevcntrs++;
- index++;
}
} else {
- /* +1 for newline */
+ /* +1 for newline. */
sz += strlen(dev_cntrs[i].name) + 1;
+ /* Add ",32" for 32-bit counters */
+ if (dev_cntrs[i].flags & CNTR_32BIT)
+ sz += bit_type_32_sz;
+ dev_cntrs[i].offset = dd->ndevcntrs;
dd->ndevcntrs++;
- dev_cntrs[i].offset = index;
- index++;
- hfi1_dbg_early("\tAdding %s\n", dev_cntrs[i].name);
}
}
/* allocate space for the counter values */
- dd->cntrs = kcalloc(index, sizeof(u64), GFP_KERNEL);
+ dd->cntrs = kcalloc(dd->ndevcntrs, sizeof(u64), GFP_KERNEL);
if (!dd->cntrs)
goto bail;
- dd->scntrs = kcalloc(index, sizeof(u64), GFP_KERNEL);
+ dd->scntrs = kcalloc(dd->ndevcntrs, sizeof(u64), GFP_KERNEL);
if (!dd->scntrs)
goto bail;
-
/* allocate space for the counter names */
dd->cntrnameslen = sz;
dd->cntrnames = kmalloc(sz, GFP_KERNEL);
goto bail;
/* fill in the names */
- for (p = dd->cntrnames, i = 0, index = 0; i < DEV_CNTR_LAST; i++) {
+ for (p = dd->cntrnames, i = 0; i < DEV_CNTR_LAST; i++) {
if (dev_cntrs[i].flags & CNTR_DISABLED) {
/* Nothing */
- } else {
- if (dev_cntrs[i].flags & CNTR_VL) {
- for (j = 0; j < C_VL_COUNT; j++) {
- memset(name, '\0', C_MAX_NAME);
- snprintf(name, C_MAX_NAME, "%s%d",
- dev_cntrs[i].name,
- vl_from_idx(j));
- memcpy(p, name, strlen(name));
- p += strlen(name);
- *p++ = '\n';
+ } else if (dev_cntrs[i].flags & CNTR_VL) {
+ for (j = 0; j < C_VL_COUNT; j++) {
+ snprintf(name, C_MAX_NAME, "%s%d",
+ dev_cntrs[i].name,
+ vl_from_idx(j));
+ memcpy(p, name, strlen(name));
+ p += strlen(name);
+
+ /* Counter is 32 bits */
+ if (dev_cntrs[i].flags & CNTR_32BIT) {
+ memcpy(p, bit_type_32, bit_type_32_sz);
+ p += bit_type_32_sz;
}
- } else {
- memcpy(p, dev_cntrs[i].name,
- strlen(dev_cntrs[i].name));
- p += strlen(dev_cntrs[i].name);
+
+ *p++ = '\n';
+ }
+ } else if (dev_cntrs[i].flags & CNTR_SDMA) {
+ for (j = 0; j < dd->chip_sdma_engines; j++) {
+ snprintf(name, C_MAX_NAME, "%s%d",
+ dev_cntrs[i].name, j);
+ memcpy(p, name, strlen(name));
+ p += strlen(name);
+
+ /* Counter is 32 bits */
+ if (dev_cntrs[i].flags & CNTR_32BIT) {
+ memcpy(p, bit_type_32, bit_type_32_sz);
+ p += bit_type_32_sz;
+ }
+
*p++ = '\n';
}
- index++;
+ } else {
+ memcpy(p, dev_cntrs[i].name, strlen(dev_cntrs[i].name));
+ p += strlen(dev_cntrs[i].name);
+
+ /* Counter is 32 bits */
+ if (dev_cntrs[i].flags & CNTR_32BIT) {
+ memcpy(p, bit_type_32, bit_type_32_sz);
+ p += bit_type_32_sz;
+ }
+
+ *p++ = '\n';
}
}
sz = 0;
dd->nportcntrs = 0;
for (i = 0; i < PORT_CNTR_LAST; i++) {
- hfi1_dbg_early("Init pcntr %s\n", port_cntrs[i].name);
if (port_cntrs[i].flags & CNTR_DISABLED) {
hfi1_dbg_early("\tSkipping %s\n", port_cntrs[i].name);
continue;
}
if (port_cntrs[i].flags & CNTR_VL) {
- hfi1_dbg_early("\tProcessing VL cntr\n");
port_cntrs[i].offset = dd->nportcntrs;
for (j = 0; j < C_VL_COUNT; j++) {
- memset(name, '\0', C_MAX_NAME);
snprintf(name, C_MAX_NAME, "%s%d",
port_cntrs[i].name,
vl_from_idx(j));
sz += strlen(name);
+ /* Add ",32" for 32-bit counters */
+ if (port_cntrs[i].flags & CNTR_32BIT)
+ sz += bit_type_32_sz;
sz++;
- hfi1_dbg_early("\t\t%s\n", name);
dd->nportcntrs++;
}
} else {
- /* +1 for newline */
+ /* +1 for newline */
sz += strlen(port_cntrs[i].name) + 1;
+ /* Add ",32" for 32-bit counters */
+ if (port_cntrs[i].flags & CNTR_32BIT)
+ sz += bit_type_32_sz;
port_cntrs[i].offset = dd->nportcntrs;
dd->nportcntrs++;
- hfi1_dbg_early("\tAdding %s\n", port_cntrs[i].name);
}
}
if (port_cntrs[i].flags & CNTR_VL) {
for (j = 0; j < C_VL_COUNT; j++) {
- memset(name, '\0', C_MAX_NAME);
snprintf(name, C_MAX_NAME, "%s%d",
port_cntrs[i].name,
vl_from_idx(j));
memcpy(p, name, strlen(name));
p += strlen(name);
+
+ /* Counter is 32 bits */
+ if (port_cntrs[i].flags & CNTR_32BIT) {
+ memcpy(p, bit_type_32, bit_type_32_sz);
+ p += bit_type_32_sz;
+ }
+
*p++ = '\n';
}
} else {
memcpy(p, port_cntrs[i].name,
strlen(port_cntrs[i].name));
p += strlen(port_cntrs[i].name);
+
+ /* Counter is 32 bits */
+ if (port_cntrs[i].flags & CNTR_32BIT) {
+ memcpy(p, bit_type_32, bit_type_32_sz);
+ p += bit_type_32_sz;
+ }
+
*p++ = '\n';
}
}
return -ENOMEM;
}
-
static u32 chip_to_opa_lstate(struct hfi1_devdata *dd, u32 chip_lstate)
{
switch (chip_lstate) {
u8 hfi1_ibphys_portstate(struct hfi1_pportdata *ppd)
{
- static u32 remembered_state = 0xff;
u32 pstate;
u32 ib_pstate;
pstate = read_physical_state(ppd->dd);
ib_pstate = chip_to_opa_pstate(ppd->dd, pstate);
- if (remembered_state != ib_pstate) {
+ if (ppd->last_pstate != ib_pstate) {
dd_dev_info(ppd->dd,
"%s: physical state changed to %s (0x%x), phy 0x%x\n",
__func__, opa_pstate_name(ib_pstate), ib_pstate,
pstate);
- remembered_state = ib_pstate;
+ ppd->last_pstate = ib_pstate;
}
return ib_pstate;
}
* In HFI, the mask needs to be 1 to allow interrupts.
*/
if (enable) {
- u64 cce_int_mask;
- const int qsfp1_int_smask = QSFP1_INT % 64;
- const int qsfp2_int_smask = QSFP2_INT % 64;
-
/* enable all interrupts */
for (i = 0; i < CCE_NUM_INT_CSRS; i++)
- write_csr(dd, CCE_INT_MASK + (8*i), ~(u64)0);
+ write_csr(dd, CCE_INT_MASK + (8 * i), ~(u64)0);
- /*
- * disable QSFP1 interrupts for HFI1, QSFP2 interrupts for HFI0
- * Qsfp1Int and Qsfp2Int are adjacent bits in the same CSR,
- * therefore just one of QSFP1_INT/QSFP2_INT can be used to find
- * the index of the appropriate CSR in the CCEIntMask CSR array
- */
- cce_int_mask = read_csr(dd, CCE_INT_MASK +
- (8*(QSFP1_INT/64)));
- if (dd->hfi1_id) {
- cce_int_mask &= ~((u64)1 << qsfp1_int_smask);
- write_csr(dd, CCE_INT_MASK + (8*(QSFP1_INT/64)),
- cce_int_mask);
- } else {
- cce_int_mask &= ~((u64)1 << qsfp2_int_smask);
- write_csr(dd, CCE_INT_MASK + (8*(QSFP2_INT/64)),
- cce_int_mask);
- }
+ init_qsfp_int(dd);
} else {
for (i = 0; i < CCE_NUM_INT_CSRS; i++)
- write_csr(dd, CCE_INT_MASK + (8*i), 0ull);
+ write_csr(dd, CCE_INT_MASK + (8 * i), 0ull);
}
}
int i;
for (i = 0; i < CCE_NUM_INT_CSRS; i++)
- write_csr(dd, CCE_INT_CLEAR + (8*i), ~(u64)0);
+ write_csr(dd, CCE_INT_CLEAR + (8 * i), ~(u64)0);
write_csr(dd, CCE_ERR_CLEAR, ~(u64)0);
write_csr(dd, MISC_ERR_CLEAR, ~(u64)0);
for (i = 0; i < dd->num_msix_entries; i++, me++) {
if (me->arg == NULL) /* => no irq, no affinity */
- break;
- irq_set_affinity_hint(dd->msix_entries[i].msix.vector,
- NULL);
+ continue;
+ hfi1_put_irq_affinity(dd, &dd->msix_entries[i]);
free_irq(me->msix.vector, me->arg);
}
} else {
}
/* clean structures */
- for (i = 0; i < dd->num_msix_entries; i++)
- free_cpumask_var(dd->msix_entries[i].mask);
kfree(dd->msix_entries);
dd->msix_entries = NULL;
dd->num_msix_entries = 0;
/* direct the chip source to the given MSI-X interrupt */
m = isrc / 8;
n = isrc % 8;
- reg = read_csr(dd, CCE_INT_MAP + (8*m));
- reg &= ~((u64)0xff << (8*n));
- reg |= ((u64)msix_intr & 0xff) << (8*n);
- write_csr(dd, CCE_INT_MAP + (8*m), reg);
+ reg = read_csr(dd, CCE_INT_MAP + (8 * m));
+ reg &= ~((u64)0xff << (8 * n));
+ reg |= ((u64)msix_intr & 0xff) << (8 * n);
+ write_csr(dd, CCE_INT_MAP + (8 * m), reg);
}
static void remap_sdma_interrupts(struct hfi1_devdata *dd,
* SDMAProgress
* SDMAIdle
*/
- remap_intr(dd, IS_SDMA_START + 0*TXE_NUM_SDMA_ENGINES + engine,
+ remap_intr(dd, IS_SDMA_START + 0 * TXE_NUM_SDMA_ENGINES + engine,
msix_intr);
- remap_intr(dd, IS_SDMA_START + 1*TXE_NUM_SDMA_ENGINES + engine,
+ remap_intr(dd, IS_SDMA_START + 1 * TXE_NUM_SDMA_ENGINES + engine,
msix_intr);
- remap_intr(dd, IS_SDMA_START + 2*TXE_NUM_SDMA_ENGINES + engine,
+ remap_intr(dd, IS_SDMA_START + 2 * TXE_NUM_SDMA_ENGINES + engine,
msix_intr);
}
static int request_msix_irqs(struct hfi1_devdata *dd)
{
- const struct cpumask *local_mask;
- cpumask_var_t def, rcv;
- bool def_ret, rcv_ret;
int first_general, last_general;
int first_sdma, last_sdma;
int first_rx, last_rx;
- int first_cpu, curr_cpu;
- int rcv_cpu, sdma_cpu;
- int i, ret = 0, possible;
- int ht;
+ int i, ret = 0;
/* calculate the ranges we are going to use */
first_general = 0;
first_rx = last_sdma = first_sdma + dd->num_sdma;
last_rx = first_rx + dd->n_krcv_queues;
- /*
- * Interrupt affinity.
- *
- * non-rcv avail gets a default mask that
- * starts as possible cpus with threads reset
- * and each rcv avail reset.
- *
- * rcv avail gets node relative 1 wrapping back
- * to the node relative 1 as necessary.
- *
- */
- local_mask = cpumask_of_pcibus(dd->pcidev->bus);
- /* if first cpu is invalid, use NUMA 0 */
- if (cpumask_first(local_mask) >= nr_cpu_ids)
- local_mask = topology_core_cpumask(0);
-
- def_ret = zalloc_cpumask_var(&def, GFP_KERNEL);
- rcv_ret = zalloc_cpumask_var(&rcv, GFP_KERNEL);
- if (!def_ret || !rcv_ret)
- goto bail;
- /* use local mask as default */
- cpumask_copy(def, local_mask);
- possible = cpumask_weight(def);
- /* disarm threads from default */
- ht = cpumask_weight(
- topology_sibling_cpumask(cpumask_first(local_mask)));
- for (i = possible/ht; i < possible; i++)
- cpumask_clear_cpu(i, def);
- /* def now has full cores on chosen node*/
- first_cpu = cpumask_first(def);
- if (nr_cpu_ids >= first_cpu)
- first_cpu++;
- curr_cpu = first_cpu;
-
- /* One context is reserved as control context */
- for (i = first_cpu; i < dd->n_krcv_queues + first_cpu - 1; i++) {
- cpumask_clear_cpu(curr_cpu, def);
- cpumask_set_cpu(curr_cpu, rcv);
- curr_cpu = cpumask_next(curr_cpu, def);
- if (curr_cpu >= nr_cpu_ids)
- break;
- }
- /* def mask has non-rcv, rcv has recv mask */
- rcv_cpu = cpumask_first(rcv);
- sdma_cpu = cpumask_first(def);
-
/*
* Sanity check - the code expects all SDMA chip source
* interrupts to be in the same CSR, starting at bit 0. Verify
snprintf(me->name, sizeof(me->name),
DRIVER_NAME "_%d", dd->unit);
err_info = "general";
+ me->type = IRQ_GENERAL;
} else if (first_sdma <= i && i < last_sdma) {
idx = i - first_sdma;
sde = &dd->per_sdma[idx];
DRIVER_NAME "_%d sdma%d", dd->unit, idx);
err_info = "sdma";
remap_sdma_interrupts(dd, idx, i);
+ me->type = IRQ_SDMA;
} else if (first_rx <= i && i < last_rx) {
idx = i - first_rx;
rcd = dd->rcd[idx];
* Set the interrupt register and mask for this
* context's interrupt.
*/
- rcd->ireg = (IS_RCVAVAIL_START+idx) / 64;
+ rcd->ireg = (IS_RCVAVAIL_START + idx) / 64;
rcd->imask = ((u64)1) <<
- ((IS_RCVAVAIL_START+idx) % 64);
+ ((IS_RCVAVAIL_START + idx) % 64);
handler = receive_context_interrupt;
thread = receive_context_thread;
arg = rcd;
DRIVER_NAME "_%d kctxt%d", dd->unit, idx);
err_info = "receive context";
remap_intr(dd, IS_RCVAVAIL_START + idx, i);
+ me->type = IRQ_RCVCTXT;
} else {
/* not in our expected range - complain, then
ignore it */
if (arg == NULL)
continue;
/* make sure the name is terminated */
- me->name[sizeof(me->name)-1] = 0;
+ me->name[sizeof(me->name) - 1] = 0;
ret = request_threaded_irq(me->msix.vector, handler, thread, 0,
me->name, arg);
*/
me->arg = arg;
- if (!zalloc_cpumask_var(
- &dd->msix_entries[i].mask,
- GFP_KERNEL))
- goto bail;
- if (handler == sdma_interrupt) {
- dd_dev_info(dd, "sdma engine %d cpu %d\n",
- sde->this_idx, sdma_cpu);
- sde->cpu = sdma_cpu;
- cpumask_set_cpu(sdma_cpu, dd->msix_entries[i].mask);
- sdma_cpu = cpumask_next(sdma_cpu, def);
- if (sdma_cpu >= nr_cpu_ids)
- sdma_cpu = cpumask_first(def);
- } else if (handler == receive_context_interrupt) {
- dd_dev_info(dd, "rcv ctxt %d cpu %d\n", rcd->ctxt,
- (rcd->ctxt == HFI1_CTRL_CTXT) ?
- cpumask_first(def) : rcv_cpu);
- if (rcd->ctxt == HFI1_CTRL_CTXT) {
- /* map to first default */
- cpumask_set_cpu(cpumask_first(def),
- dd->msix_entries[i].mask);
- } else {
- cpumask_set_cpu(rcv_cpu,
- dd->msix_entries[i].mask);
- rcv_cpu = cpumask_next(rcv_cpu, rcv);
- if (rcv_cpu >= nr_cpu_ids)
- rcv_cpu = cpumask_first(rcv);
- }
- } else {
- /* otherwise first def */
- dd_dev_info(dd, "%s cpu %d\n",
- err_info, cpumask_first(def));
- cpumask_set_cpu(
- cpumask_first(def), dd->msix_entries[i].mask);
- }
- irq_set_affinity_hint(
- dd->msix_entries[i].msix.vector,
- dd->msix_entries[i].mask);
+ ret = hfi1_get_irq_affinity(dd, me);
+ if (ret)
+ dd_dev_err(dd,
+ "unable to pin IRQ %d\n", ret);
}
-out:
- free_cpumask_var(def);
- free_cpumask_var(rcv);
return ret;
-bail:
- ret = -ENOMEM;
- goto out;
}
/*
/* all chip interrupts map to MSI-X 0 */
for (i = 0; i < CCE_NUM_INT_MAP_CSRS; i++)
- write_csr(dd, CCE_INT_MAP + (8*i), 0);
+ write_csr(dd, CCE_INT_MAP + (8 * i), 0);
}
static int set_up_interrupts(struct hfi1_devdata *dd)
*/
num_kernel_contexts = n_krcvqs + MIN_KERNEL_KCTXTS - 1;
else
- num_kernel_contexts = num_online_nodes();
+ num_kernel_contexts = num_online_nodes() + 1;
num_kernel_contexts =
max_t(int, MIN_KERNEL_KCTXTS, num_kernel_contexts);
/*
dd->num_rcv_contexts = total_contexts;
dd->n_krcv_queues = num_kernel_contexts;
dd->first_user_ctxt = num_kernel_contexts;
+ dd->num_user_contexts = num_user_contexts;
dd->freectxts = num_user_contexts;
dd_dev_info(dd,
"rcv contexts: chip %d, used %d (kernel %d, user %d)\n",
/* CceIntMap */
for (i = 0; i < CCE_NUM_INT_MAP_CSRS; i++)
- write_csr(dd, CCE_INT_MAP+(8*i), 0);
+ write_csr(dd, CCE_INT_MAP + (8 * i), 0);
/* SendCtxtCreditReturnAddr */
for (i = 0; i < dd->chip_send_contexts; i++)
write_kctxt_csr(dd, i, RCV_HDR_ADDR, 0);
write_kctxt_csr(dd, i, RCV_HDR_TAIL_ADDR, 0);
for (j = 0; j < RXE_NUM_TID_FLOWS; j++)
- write_uctxt_csr(dd, i, RCV_TID_FLOW_TABLE+(8*j), 0);
+ write_uctxt_csr(dd, i, RCV_TID_FLOW_TABLE + (8 * j), 0);
}
/* RcvArray */
for (i = 0; i < dd->chip_rcv_array_count; i++)
- write_csr(dd, RCV_ARRAY + (8*i),
+ write_csr(dd, RCV_ARRAY + (8 * i),
RCV_ARRAY_RT_WRITE_ENABLE_SMASK);
/* RcvQPMapTable */
write_csr(dd, SEND_ERR_CLEAR, ~0ull);
/* SEND_ERR_FORCE read-only */
for (i = 0; i < VL_ARB_LOW_PRIO_TABLE_SIZE; i++)
- write_csr(dd, SEND_LOW_PRIORITY_LIST + (8*i), 0);
+ write_csr(dd, SEND_LOW_PRIORITY_LIST + (8 * i), 0);
for (i = 0; i < VL_ARB_HIGH_PRIO_TABLE_SIZE; i++)
- write_csr(dd, SEND_HIGH_PRIORITY_LIST + (8*i), 0);
- for (i = 0; i < dd->chip_send_contexts/NUM_CONTEXTS_PER_SET; i++)
- write_csr(dd, SEND_CONTEXT_SET_CTRL + (8*i), 0);
+ write_csr(dd, SEND_HIGH_PRIORITY_LIST + (8 * i), 0);
+ for (i = 0; i < dd->chip_send_contexts / NUM_CONTEXTS_PER_SET; i++)
+ write_csr(dd, SEND_CONTEXT_SET_CTRL + (8 * i), 0);
for (i = 0; i < TXE_NUM_32_BIT_COUNTER; i++)
- write_csr(dd, SEND_COUNTER_ARRAY32 + (8*i), 0);
+ write_csr(dd, SEND_COUNTER_ARRAY32 + (8 * i), 0);
for (i = 0; i < TXE_NUM_64_BIT_COUNTER; i++)
- write_csr(dd, SEND_COUNTER_ARRAY64 + (8*i), 0);
+ write_csr(dd, SEND_COUNTER_ARRAY64 + (8 * i), 0);
write_csr(dd, SEND_CM_CTRL, SEND_CM_CTRL_RESETCSR);
write_csr(dd, SEND_CM_GLOBAL_CREDIT,
SEND_CM_GLOBAL_CREDIT_RESETCSR);
write_csr(dd, SEND_CM_REMOTE_AU_TABLE0_TO3, 0);
write_csr(dd, SEND_CM_REMOTE_AU_TABLE4_TO7, 0);
for (i = 0; i < TXE_NUM_DATA_VL; i++)
- write_csr(dd, SEND_CM_CREDIT_VL + (8*i), 0);
+ write_csr(dd, SEND_CM_CREDIT_VL + (8 * i), 0);
write_csr(dd, SEND_CM_CREDIT_VL15, 0);
/* SEND_CM_CREDIT_USED_VL read-only */
/* SEND_CM_CREDIT_USED_VL15 read-only */
write_csr(dd, RCV_CTXT_CTRL, 0);
/* mask all interrupt sources */
for (i = 0; i < CCE_NUM_INT_CSRS; i++)
- write_csr(dd, CCE_INT_MASK + (8*i), 0ull);
+ write_csr(dd, CCE_INT_MASK + (8 * i), 0ull);
/*
* DC Reset: do a full DC reset before the register clear.
* across the clear.
*/
write_csr(dd, CCE_DC_CTRL, CCE_DC_CTRL_DC_RESET_SMASK);
- (void) read_csr(dd, CCE_DC_CTRL);
+ (void)read_csr(dd, CCE_DC_CTRL);
if (use_flr) {
/*
write_csr(dd, CCE_DC_CTRL, 0);
/* Set the LED off */
- if (is_ax(dd))
- setextled(dd, 0);
+ setextled(dd, 0);
+
/*
* Clear the QSFP reset.
* An FLR enforces a 0 on all out pins. The driver does not touch
write_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_JOB_KEY, reg);
/*
* Enable send-side J_KEY integrity check, unless this is A0 h/w
- * (due to A0 erratum).
*/
if (!is_ax(dd)) {
reg = read_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_ENABLE);
*/
void hfi1_start_cleanup(struct hfi1_devdata *dd)
{
+ aspm_exit(dd);
free_cntrs(dd);
free_rcverr(dd);
clean_up_interrupts(dd);
ret = read_hfi1_efi_var(dd, "description", &size,
(void **)&dd->boardname);
if (ret) {
- dd_dev_err(dd, "Board description not found\n");
+ dd_dev_info(dd, "Board description not found\n");
/* use generic description */
dd->boardname = kstrdup(generic, GFP_KERNEL);
if (!dd->boardname)
}
ppd->vls_supported = num_vls;
ppd->vls_operational = ppd->vls_supported;
+ ppd->actual_vls_operational = ppd->vls_supported;
/* Set the default MTU. */
for (vl = 0; vl < num_vls; vl++)
dd->vld[vl].mtu = hfi1_max_mtu;
/* start in offline */
ppd->host_link_state = HLS_DN_OFFLINE;
init_vl_arb_caches(ppd);
+ ppd->last_pstate = 0xff; /* invalid value */
}
dd->link_default = HLS_DN_POLL;
num_vls, dd->chip_sdma_engines);
num_vls = dd->chip_sdma_engines;
ppd->vls_supported = dd->chip_sdma_engines;
+ ppd->vls_operational = ppd->vls_supported;
}
/*
if (ret)
goto bail_cleanup;
+ /* Needs to be called before hfi1_firmware_init */
+ get_platform_config(dd);
+
/* read in firmware */
ret = hfi1_firmware_init(dd);
if (ret)
/* set up KDETH QP prefix in both RX and TX CSRs */
init_kdeth_qp(dd);
+ ret = hfi1_dev_affinity_init(dd);
+ if (ret)
+ goto bail_cleanup;
+
/* send contexts must be set up before receive contexts */
ret = init_send_contexts(dd);
if (ret)
return (u16)delta_cycles;
}
-
/**
* create_pbc - build a pbc for transmission
* @flags: special case flags or-ed in built pbc
dd_dev_emerg(dd,
"Critical temperature reached! Forcing device into freeze mode!\n");
dd->flags |= HFI1_FORCED_FREEZE;
- start_freeze_handling(ppd, FREEZE_SELF|FREEZE_ABORT);
+ start_freeze_handling(ppd, FREEZE_SELF | FREEZE_ABORT);
/*
* Shut DC down as much and as quickly as possible.
*