Clean ups, refactror, additions
Subsystems (2)
cover Clean ups, refactror, additions
These patches add some recactoring and code clean ups to make things more organized. There is a performance optimization and new counter/debugging stats added as well. The new "API" that is added is a driver internal API not an actual "API" that is exposed to the outside.
---
Grzegorz Andrejczuk (3):
IB/hfi1: Move common receive IRQ code to function IB/hfi1: Decouple IRQ name from type IB/hfi1: Return void in packet receiving functions
Mike Marciniszyn (6):
IB/hfi1: Move chip specific functions to chip.c IB/hfi1: Add fast and slow handlers for receive context IB/hfi1: IB/hfi1: Add an API to handle special case drop IB/hfi1: Create API for auto activate
IB/hfi1: Add software counter for ctxt0 seq drop IB/hfi1: Add RcvShortLengthErrCnt to hfi1stats
drivers/infiniband/hw/hfi1/chip.c | 171 ++++++++++++++++++++++----- drivers/infiniband/hw/hfi1/chip.h | 8 + drivers/infiniband/hw/hfi1/chip_registers.h | 1 drivers/infiniband/hw/hfi1/driver.c | 151 +++++++++--------------- drivers/infiniband/hw/hfi1/hfi.h | 66 ++++++++++ drivers/infiniband/hw/hfi1/init.c | 81 ++----------- drivers/infiniband/hw/hfi1/msix.c | 106 +++++++++-------- drivers/infiniband/hw/hfi1/msix.h | 1 drivers/infiniband/hw/hfi1/trace_rx.h | 6 - 9 files changed, 337 insertions(+), 254 deletions(-)
--
-Denny
PATCH 1/9 IB/hfi1: Move chip specific functions to chip.c
From: Mike Marciniszyn <[email protected]>
Move routines and defines associated with hdrq size validation to a chip specific routine since the limits are specific to the device.
Fix incorrect value for min size 2 -> 32
CSR writes should also be in chip.c.
Create a chip routine to write the hdrq specific CSRs and call as appropriate.
--- drivers/infiniband/hw/hfi1/chip.c | 78 +++++++++++++++++++++++++++++++++++++ drivers/infiniband/hw/hfi1/chip.h | 6 +++ drivers/infiniband/hw/hfi1/init.c | 72 +--------------------------------- 3 files changed, 87 insertions(+), 69 deletions(-)
drivers/infiniband/hw/hfi1/chip.c +78 -0
@@ -11858,6 +11858,84 @@ static u32 encoded_size(u32 size)return 0x1 ;/* if invalid, go with the minimum size */ } +/** +* encode_rcv_header_entry_size - return chip specific encoding for size +* @size: size in dwords +* +* Convert a receive header entry size that to the encoding used in the CSR. +* +* Return a zero if the given size is invalid, otherwise the encoding. +*/ +u8 encode_rcv_header_entry_size (u8 size) +{ +/* there are only 3 valid receive header entry sizes */ +if (size ==2 ) +return 1 ; +if (size ==16 ) +return 2 ; +if (size ==32 ) +return 4 ; +return 0 ;/* invalid */ +} + +/** +* hfi1_validate_rcvhdrcnt - validate hdrcnt +* @dd: the device data +* @thecnt: the header count +*/ +int hfi1_validate_rcvhdrcnt (struct hfi1_devdata *dd,uint thecnt) +{ +if (thecnt <=HFI1_MIN_HDRQ_EGRBUF_CNT ) { +dd_dev_err (dd,"Receive header queue count too small\n" ); +return -EINVAL ; + } + +if (thecnt >HFI1_MAX_HDRQ_EGRBUF_CNT ) { +dd_dev_err (dd, +"Receive header queue count cannot be greater than %u\n" , +HFI1_MAX_HDRQ_EGRBUF_CNT ); +return -EINVAL ; + } + +if (thecnt %HDRQ_INCREMENT ) { +dd_dev_err (dd,"Receive header queue count %d must be divisible by %lu\n" , + thecnt,HDRQ_INCREMENT ); +return -EINVAL ; + } + +return 0 ; +} + +/** +* set_hdrq_regs - set header queue registers for context +* @dd: the device data +* @ctxt: the context +* @entsize: the dword entry size +* @hdrcnt: the number of header entries +*/ +void set_hdrq_regs (struct hfi1_devdata *dd,u8 ctxt,u8 entsize,u16 hdrcnt) +{ +u64 reg; + + reg = (((u64 )hdrcnt >>HDRQ_SIZE_SHIFT ) &RCV_HDR_CNT_CNT_MASK ) << +RCV_HDR_CNT_CNT_SHIFT ; +write_kctxt_csr (dd, ctxt,RCV_HDR_CNT , reg); + reg = ((u64 )encode_rcv_header_entry_size (entsize) & +RCV_HDR_ENT_SIZE_ENT_SIZE_MASK ) << +RCV_HDR_ENT_SIZE_ENT_SIZE_SHIFT ; +write_kctxt_csr (dd, ctxt,RCV_HDR_ENT_SIZE , reg); + reg = ((u64 )DEFAULT_RCVHDRSIZE &RCV_HDR_SIZE_HDR_SIZE_MASK ) << +RCV_HDR_SIZE_HDR_SIZE_SHIFT ; +write_kctxt_csr (dd, ctxt,RCV_HDR_SIZE , reg); + +/* +* Program dummy tail address for every receive context +* before enabling any receive context +*/ +write_kctxt_csr (dd, ctxt,RCV_HDR_TAIL_ADDR , + dd->rcvhdrtail_dummy_dma); +} +void hfi1_rcvctrl (struct hfi1_devdata *dd,unsigned int op,struct hfi1_ctxtdata *rcd) {
drivers/infiniband/hw/hfi1/chip.h +6 -0
@@ -358,6 +358,8 @@#define MAX_EAGER_BUFFER (256 * 1024)#define MAX_EAGER_BUFFER_TOTAL (64 * (1 << 20))/* max per ctxt 64MB */ #define MAX_EXPECTED_BUFFER (2048 * 1024) +#define HFI1_MIN_HDRQ_EGRBUF_CNT 32 +#define HFI1_MAX_HDRQ_EGRBUF_CNT 16352/* * Receive expected base and count and eager base and count increment -
@@ -699,6 +701,10 @@ static inline u32 chip_rcv_array_count(struct hfi1_devdata *dd)return read_csr (dd,RCV_ARRAY_CNT ); } +u8 encode_rcv_header_entry_size (u8 size); +int hfi1_validate_rcvhdrcnt (struct hfi1_devdata *dd,uint thecnt); +void set_hdrq_regs (struct hfi1_devdata *dd,u8 ctxt,u8 entsize,u16 hdrcnt); +u64 create_pbc (struct hfi1_pportdata *ppd,u64 flags,int srate_mbs,u32 vl,u32 dw_len);
drivers/infiniband/hw/hfi1/init.c +3 -69
115 lines not on this page · raw message
PATCH 2/9 IB/hfi1: Add fast and slow handlers for receive context
In reply to Dennis DalessandroFrom: Mike Marciniszyn <[email protected]>
This patch eliminate special cases by adding a fast_handler member to the receive context and changes to the fast handler as specified in the new variable. Initialize the variable as soon as the setting for dma tail is known when the context is created.
Setting fast path is called every time when any context has entered slow path. Add function to check if contexts is using fast path and do not set fast path when it is already done to improve RCD fastpath setting.
--- drivers/infiniband/hw/hfi1/driver.c | 66 ++++++--------------------------- drivers/infiniband/hw/hfi1/hfi.h | 41 ++++++++++++++++++++- drivers/infiniband/hw/hfi1/init.c | 5 +++ drivers/infiniband/hw/hfi1/trace_rx.h | 6 +-- 4 files changed, 58 insertions(+), 60 deletions(-)
drivers/infiniband/hw/hfi1/driver.c +12 -54
@@ -883,9 +883,8 @@ int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread)return last; } -static inline void set_nodma_rtail (struct hfi1_devdata *dd,u16 ctxt) +static void set_all_fastpath (struct hfi1_devdata *dd,struct hfi1_ctxtdata *rcd) { -struct hfi1_ctxtdata *rcd;u16 i;/*
@@ -893,50 +892,17 @@ static inline void set_nodma_rtail(struct hfi1_devdata *dd, u16 ctxt)* interrupt handler only for that context. Otherwise, switch * interrupt handler for all statically allocated kernel contexts. */ -if (ctxt >= dd->first_dyn_alloc_ctxt) { - rcd =hfi1_rcd_get_by_index_safe (dd, ctxt); -if (rcd) { - rcd->do_interrupt = - &handle_receive_interrupt_nodma_rtail; -hfi1_rcd_put (rcd); - } -return ; - } - -for (i =HFI1_CTRL_CTXT +1 ; i < dd->first_dyn_alloc_ctxt; i++) { - rcd =hfi1_rcd_get_by_index (dd, i); -if (rcd) - rcd->do_interrupt = - &handle_receive_interrupt_nodma_rtail; +if (rcd->ctxt >= dd->first_dyn_alloc_ctxt && !rcd->is_vnic) { +hfi1_rcd_get (rcd); +hfi1_set_fast (rcd);hfi1_rcd_put (rcd); - } -} - -static inline void set_dma_rtail (struct hfi1_devdata *dd,u16 ctxt) -{ -struct hfi1_ctxtdata *rcd; -u16 i; - -/* -* For dynamically allocated kernel contexts (like vnic) switch -* interrupt handler only for that context. Otherwise, switch -* interrupt handler for all statically allocated kernel contexts. -*/ -if (ctxt >= dd->first_dyn_alloc_ctxt) { - rcd =hfi1_rcd_get_by_index_safe (dd, ctxt); -if (rcd) { - rcd->do_interrupt = - &handle_receive_interrupt_dma_rtail; -hfi1_rcd_put (rcd); - }return ; } -for (i =HFI1_CTRL_CTXT +1 ; i < dd->first_dyn_alloc_ctxt; i++) { +for (i =HFI1_CTRL_CTXT +1 ; i < dd->num_rcv_contexts; i++) { rcd =hfi1_rcd_get_by_index (dd, i); -if (rcd) - rcd->do_interrupt = - &handle_receive_interrupt_dma_rtail; +if (rcd && (i < dd->first_dyn_alloc_ctxt || rcd->is_vnic)) +hfi1_set_fast (rcd);hfi1_rcd_put (rcd); } }
@@ -952,7 +918,7 @@ void set_all_slowpath(struct hfi1_devdata *dd)if (!rcd)continue ;if (i < dd->first_dyn_alloc_ctxt || rcd->is_vnic) - rcd->do_interrupt = &handle_receive_interrupt; + rcd->do_interrupt = rcd->slow_handler;hfi1_rcd_put (rcd); }
@@ -1065,11 +1031,6 @@ int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)if (!get_dma_rtail_setting (rcd)) {if (hfi1_seq_incr (rcd,rhf_rcv_seq (packet.rhf))) last =RCV_PKT_DONE ; -if (needset) { -dd_dev_info (dd,"Switching to NO_DMA_RTAIL\n" ); -set_nodma_rtail (dd, rcd->ctxt); - needset =0 ; - } }else {if (packet.rhqoff == hdrqtail) last =RCV_PKT_DONE ;
@@ -1085,15 +1046,12 @@ int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)if (!last && lseq) skip_pkt =1 ; } - -if (needset) { -dd_dev_info (dd, -"Switching to DMA_RTAIL\n" ); -set_dma_rtail (dd, rcd->ctxt); - needset =0 ; - } } +if (needset) { + needset = false; +set_all_fastpath (dd, rcd); + }process_rcv_update (last, &packet); }
drivers/infiniband/hw/hfi1/hfi.h +40 -1
59 lines not on this page · raw message
drivers/infiniband/hw/hfi1/init.c +5 -0
@@ -150,6 +150,11 @@ static int hfi1_create_kctxt(struct hfi1_devdata *dd,/* Control context must use DMA_RTAIL */ if (rcd->ctxt ==HFI1_CTRL_CTXT ) rcd->flags |=HFI1_CAP_DMA_RTAIL ; + rcd->fast_handler =get_dma_rtail_setting (rcd) ? + handle_receive_interrupt_dma_rtail : + handle_receive_interrupt_nodma_rtail; + rcd->slow_handler = handle_receive_interrupt; +hfi1_set_seq_cnt (rcd,1 ); rcd->sc =sc_alloc (dd,SC_ACK , rcd->rcvhdrqentsize, dd->node);
drivers/infiniband/hw/hfi1/trace_rx.h +1 -5
@@ -106,11 +106,7 @@ ), TP_fast_assign(DD_DEV_ASSIGN(dd); __entry->ctxt = rcd->ctxt; -if (rcd->do_interrupt == - &handle_receive_interrupt) - __entry->slow_path =1 ; -else - __entry->slow_path =0 ; + __entry->slow_path =hfi1_is_slowpath (rcd); __entry->dma_rtail =get_dma_rtail_setting (rcd); ), TP_printk("[%s] ctxt %d SlowPath: %d DmaRtail: %d",
PATCH 3/9 IB/hfi1: Move common receive IRQ code to function
In reply to Dennis DalessandroFrom: Grzegorz Andrejczuk <[email protected]>
Tracing interrupts, incrementing interrupt counter and ASPM are part that will be reused by HFI1 receive IRQ handlers.
Create common function to have shared code in one place.
--- drivers/infiniband/hw/hfi1/chip.c | 82 +++++++++++++++++++++++-------------- 1 file changed, 52 insertions(+), 30 deletions(-)
drivers/infiniband/hw/hfi1/chip.c +52 -30
@@ -8403,6 +8403,55 @@ static inline int check_packet_present(struct hfi1_ctxtdata *rcd)return hfi1_rcd_head (rcd) != tail; } +/** +* Common code for receive contexts interrupt handlers. +* Update traces, increment kernel IRQ counter and +* setup ASPM when needed. +*/ +static void receive_interrupt_common (struct hfi1_ctxtdata *rcd) +{ +struct hfi1_devdata *dd = rcd->dd; + +trace_hfi1_receive_interrupt (dd, rcd); +this_cpu_inc (*dd->int_counter); +aspm_ctx_disable (rcd); +} + +/** +* __hfi1_rcd_eoi_intr() - Make HW issue receive interrupt +* when there are packets present in the queue. When calling +* with interrupts enabled please use hfi1_rcd_eoi_intr. +* +* @rcd: valid receive context +*/ +static void __hfi1_rcd_eoi_intr (struct hfi1_ctxtdata *rcd) +{ +clear_recv_intr (rcd); +if (check_packet_present (rcd)) +force_recv_intr (rcd); +} + +/** +* hfi1_rcd_eoi_intr() - End of Interrupt processing action +* +* @rcd: Ptr to hfi1_ctxtdata of receive context +* +* Hold IRQs so we can safely clear the interrupt and +* recheck for a packet that may have arrived after the previous +* check and the interrupt clear. If a packet arrived, force another +* interrupt. This routine can be called at the end of receive packet +* processing in interrupt service routines, interrupt service thread +* and softirqs +*/ +static void hfi1_rcd_eoi_intr (struct hfi1_ctxtdata *rcd) +{ +unsigned long flags; + +local_irq_save (flags); +__hfi1_rcd_eoi_intr (rcd); +local_irq_restore (flags); +} +/* * Receive packet IRQ handler. This routine expects to be on its own IRQ. * This routine will try to handle packets immediately (latency), but if
@@ -8414,13 +8463,9 @@ static inline int check_packet_present(struct hfi1_ctxtdata *rcd)irqreturn_t receive_context_interrupt (int irq,void *data) {struct hfi1_ctxtdata *rcd = data; -struct hfi1_devdata *dd = rcd->dd;int disposition; -int present; -trace_hfi1_receive_interrupt (dd, rcd); -this_cpu_inc (*dd->int_counter); -aspm_ctx_disable (rcd); +receive_interrupt_common (rcd);/* receive interrupt remains blocked while processing packets */ disposition = rcd->do_interrupt (rcd,0 );
@@ -8433,17 +8478,7 @@ irqreturn_t receive_context_interrupt(int irq, void *data)if (disposition ==RCV_PKT_LIMIT )return IRQ_WAKE_THREAD ; -/* -* The packet processor detected no more packets. Clear the receive -* interrupt and recheck for a packet packet that may have arrived -* after the previous check and interrupt clear. If a packet arrived, -* force another interrupt. -*/ -clear_recv_intr (rcd); - present =check_packet_present (rcd); -if (present) -force_recv_intr (rcd); - +__hfi1_rcd_eoi_intr (rcd);return IRQ_HANDLED ; }
@@ -8454,24 +8489,11 @@ irqreturn_t receive_context_interrupt(int irq, void *data)irqreturn_t receive_context_thread (int irq,void *data) {struct hfi1_ctxtdata *rcd = data; -int present;/* receive interrupt is still blocked from the IRQ handler */ (void )rcd->do_interrupt (rcd,1 ); -/* -* The packet processor will only return if it detected no more -* packets. Hold IRQs here so we can safely clear the interrupt and -* recheck for a packet that may have arrived after the previous -* check and the interrupt clear. If a packet arrived, force another -* interrupt. -*/ -local_irq_disable (); -clear_recv_intr (rcd); - present =check_packet_present (rcd); -if (present) -force_recv_intr (rcd); -local_irq_enable (); +hfi1_rcd_eoi_intr (rcd);return IRQ_HANDLED ; }
PATCH 4/9 IB/hfi1: IB/hfi1: Add an API to handle special case drop
In reply to Dennis DalessandroFrom: Mike Marciniszyn <[email protected]>
This patch pushes special case drop logic into an API to be shared by all interrupt handlers.
Additionally, convert do_drop to a bool.
--- drivers/infiniband/hw/hfi1/driver.c | 6 +----- drivers/infiniband/hw/hfi1/hfi.h | 21 ++++++++++++++++++++- drivers/infiniband/hw/hfi1/init.c | 4 ++-- 3 files changed, 23 insertions(+), 8 deletions(-)
drivers/infiniband/hw/hfi1/driver.c +1 -5
@@ -1004,11 +1004,7 @@ int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)prescan_rxq (rcd, &packet);while (last ==RCV_PKT_OK ) { -if (unlikely (dd->do_drop && -atomic_xchg (&dd->drop_packet,DROP_PACKET_OFF ) == -DROP_PACKET_ON )) { - dd->do_drop =0 ; - +if (hfi1_need_drop (dd)) {/* On to the next packet */ packet.rhqoff += packet.rsize; packet.rhf_addr = (__le32 *)rcd->rcvhdrq +
drivers/infiniband/hw/hfi1/hfi.h +20 -1
@@ -1316,7 +1316,7 @@ struct hfi1_devdata {struct err_info_constraint err_info_xmit_constraint;atomic_t drop_packet; -u8 do_drop; +bool do_drop;u8 err_info_uncorrectable;u8 err_info_fmconfig;
@@ -2462,6 +2462,25 @@ static inline bool is_integrated(struct hfi1_devdata *dd)return dd->pcidev->device ==PCI_DEVICE_ID_INTEL1 ; } +/** +* hfi1_need_drop - detect need for drop +* @dd: - the device +* +* In some cases, the first packet needs to be dropped. +* +* Return true is the current packet needs to be dropped and false otherwise. +*/ +static inline bool hfi1_need_drop (struct hfi1_devdata *dd) +{ +if (unlikely (dd->do_drop && +atomic_xchg (&dd->drop_packet,DROP_PACKET_OFF ) == +DROP_PACKET_ON )) { + dd->do_drop = false; +return true; + } +return false; +} +int hfi1_tempsense_rd (struct hfi1_devdata *dd,struct hfi1_temp *temp);#define DD_DEV_ENTRY (dd) __string(dev, dev_name(&(dd)->pcidev->dev))
drivers/infiniband/hw/hfi1/init.c +2 -2
@@ -876,10 +876,10 @@ int hfi1_init(struct hfi1_devdata *dd, int reinit)if (is_ax (dd)) {atomic_set (&dd->drop_packet,DROP_PACKET_ON ); - dd->do_drop =1 ; + dd->do_drop = true; }else {atomic_set (&dd->drop_packet,DROP_PACKET_OFF ); - dd->do_drop =0 ; + dd->do_drop = false; }/* make sure the link is not "up" */
PATCH 5/9 IB/hfi1: Create API for auto activate
In reply to Dennis DalessandroFrom: Mike Marciniszyn <[email protected]>
Add an auto activate routine for use by the interrupt handler.
--- drivers/infiniband/hw/hfi1/driver.c | 37 ++++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 14 deletions(-)
drivers/infiniband/hw/hfi1/driver.c +23 -14
@@ -924,11 +924,8 @@ void set_all_slowpath(struct hfi1_devdata *dd) } } -static inline int set_armed_to_active (struct hfi1_ctxtdata *rcd, -struct hfi1_packet *packet, -struct hfi1_devdata *dd) +static bool __set_armed_to_active (struct hfi1_packet *packet) { -struct work_struct *lsaw = &rcd->ppd->linkstate_active_work;u8 etype =rhf_rcv_type (packet->rhf);u8 sc =SC15_PACKET ;
@@ -943,19 +940,34 @@ static inline int set_armed_to_active(struct hfi1_ctxtdata *rcd, sc =hfi1_16B_get_sc (hdr); }if (sc !=SC15_PACKET ) { -int hwstate =driver_lstate (rcd->ppd); +int hwstate =driver_lstate (packet->rcd->ppd); +struct work_struct *lsaw = + &packet->rcd->ppd->linkstate_active_work;if (hwstate !=IB_PORT_ACTIVE ) { -dd_dev_info (dd, +dd_dev_info (packet->rcd->dd,"Unexpected link state %s\n" ,opa_lstate_name (hwstate)); -return 0 ; +return false; } -queue_work (rcd->ppd->link_wq, lsaw); -return 1 ; +queue_work (packet->rcd->ppd->link_wq, lsaw); +return true; } -return 0 ; +return false; +} + +/** +* armed to active - the fast path for armed to active +* @packet: the packet structure +* +* Return true if packet processing needs to bail. +*/ +static bool set_armed_to_active (struct hfi1_packet *packet) +{ +if (likely (packet->rcd->ppd->host_link_state !=HLS_UP_ARMED )) +return false; +return __set_armed_to_active (packet); }/*
@@ -1016,10 +1028,7 @@ int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread) last =skip_rcv_packet (&packet, thread); skip_pkt =0 ; }else { -/* Auto activate link on non-SC15 packet receive */ -if (unlikely (rcd->ppd->host_link_state == -HLS_UP_ARMED ) && -set_armed_to_active (rcd, &packet, dd)) +if (set_armed_to_active (&packet)) goto bail; last =process_rcv_packet (&packet, thread); }
PATCH 6/9 IB/hfi1: Decouple IRQ name from type
In reply to Dennis DalessandroFrom: Grzegorz Andrejczuk <[email protected]>
IRQ name was connected to IRQ type, this is not sufficient and it would be better to use name as argument to msix_request_irq instead of assigning it to variables when function is called.
Index argument was required to generate name and now it can be removed.
To generate name correctly helpers function were added and updated.
--- drivers/infiniband/hw/hfi1/msix.c | 106 ++++++++++++++++++++----------------- drivers/infiniband/hw/hfi1/msix.h | 1 2 files changed, 59 insertions(+), 48 deletions(-)
drivers/infiniband/hw/hfi1/msix.c +58 -48
@@ -115,13 +115,11 @@ int msix_initialize(struct hfi1_devdata *dd)*/ static int msix_request_irq (struct hfi1_devdata *dd,void *arg,irq_handler_t handler,irq_handler_t thread, -u32 idx,enum irq_type type) +enum irq_type type,const char *name) {unsigned long nr;int irq;int ret; -const char *err_info; -char name[MAX_NAME_SIZE ];struct hfi1_msix_entry *me;/* Allocate an MSIx vector */
@@ -135,43 +133,15 @@ static int msix_request_irq(struct hfi1_devdata *dd, void *arg,if (nr == dd->msix_info.max_requested)return -ENOSPC ; -/* Specific verification and determine the name */ -switch (type) { -case IRQ_GENERAL : -/* general interrupt must be MSIx vector 0 */ -if (nr) { -spin_lock (&dd->msix_info.msix_lock); -__clear_bit (nr, dd->msix_info.in_use_msix); -spin_unlock (&dd->msix_info.msix_lock); -dd_dev_err (dd,"Invalid index %lu for GENERAL IRQ\n" , - nr); -return -EINVAL ; - } -snprintf (name,sizeof (name),DRIVER_NAME "_%d" , dd->unit); - err_info ="general" ; -break ; -case IRQ_SDMA : -snprintf (name,sizeof (name),DRIVER_NAME "_%d sdma%d" , - dd->unit, idx); - err_info ="sdma" ; -break ; -case IRQ_RCVCTXT : -snprintf (name,sizeof (name),DRIVER_NAME "_%d kctxt%d" , - dd->unit, idx); - err_info ="receive context" ; -break ; -case IRQ_OTHER : -default : +if (type <IRQ_SDMA && type >=IRQ_OTHER )return -EINVAL ; - } - name[sizeof (name) -1 ] =0 ; irq =pci_irq_vector (dd->pcidev, nr); ret =pci_request_irq (dd->pcidev, nr, handler, thread, arg, name);if (ret) {dd_dev_err (dd, -"%s: request for IRQ %d failed, MSIx %d, err %d\n" , - err_info, irq, idx, ret); +"%s: request for IRQ %d failed, MSIx %lu, err %d\n" , + name, irq, nr, ret);spin_lock (&dd->msix_info.msix_lock);__clear_bit (nr, dd->msix_info.in_use_msix);spin_unlock (&dd->msix_info.msix_lock);
@@ -195,17 +165,13 @@ static int msix_request_irq(struct hfi1_devdata *dd, void *arg,return nr; } -/** -* msix_request_rcd_irq() - Helper function for RCVAVAIL IRQs -* @rcd: valid rcd context -* -*/ -int msix_request_rcd_irq (struct hfi1_ctxtdata *rcd) +static int msix_request_rcd_irq_common (struct hfi1_ctxtdata *rcd, +irq_handler_t handler, +irq_handler_t thread, +const char *name) { -int nr; - - nr =msix_request_irq (rcd->dd, rcd, receive_context_interrupt, - receive_context_thread, rcd->ctxt,IRQ_RCVCTXT ); +int nr =msix_request_irq (rcd->dd, rcd, handler, thread, +IRQ_RCVCTXT , name);if (nr <0 )return nr;
@@ -222,6 +188,22 @@ int msix_request_rcd_irq(struct hfi1_ctxtdata *rcd) }/** +* msix_request_rcd_irq() - Helper function for RCVAVAIL IRQs +* @rcd: valid rcd context +* +*/ +int msix_request_rcd_irq (struct hfi1_ctxtdata *rcd) +{ +char name[MAX_NAME_SIZE ]; + +snprintf (name,sizeof (name),DRIVER_NAME "_%d kctxt%d" , + rcd->dd->unit, rcd->ctxt); + +return msix_request_rcd_irq_common (rcd, receive_context_interrupt, + receive_context_thread, name); +} + +/** * msix_request_smda_ira() - Helper for getting SDMA IRQ resources * @sde: valid sdma engine *
@@ -229,9 +211,12 @@ int msix_request_rcd_irq(struct hfi1_ctxtdata *rcd)int msix_request_sdma_irq (struct sdma_engine *sde) {int nr; +char name[MAX_NAME_SIZE ]; +snprintf (name,sizeof (name),DRIVER_NAME "_%d sdma%d" , + sde->dd->unit, sde->this_idx); nr =msix_request_irq (sde->dd, sde, sdma_interrupt,NULL , - sde->this_idx,IRQ_SDMA ); +IRQ_SDMA , name);if (nr <0 )return nr; sde->msix_intr = nr;
@@ -241,6 +226,32 @@ int msix_request_sdma_irq(struct sdma_engine *sde) }/** +* msix_request_general_irq(void) - Helper for getting general IRQ +* resources +* @dd: valid device data +*/ +int msix_request_general_irq (struct hfi1_devdata *dd) +{ +int nr; +char name[MAX_NAME_SIZE ]; + +snprintf (name,sizeof (name),DRIVER_NAME "_%d" , dd->unit); + nr =msix_request_irq (dd, dd, general_interrupt,NULL ,IRQ_GENERAL , + name); +if (nr <0 ) +return nr; + +/* general interrupt must be MSIx vector 0 */ +if (nr) { +msix_free_irq (dd, (u8 )nr); +dd_dev_err (dd,"Invalid index %d for GENERAL IRQ\n" , nr); +return -EINVAL ; + } + +return 0 ; +} + +/** * enable_sdma_src() - Helper to enable SDMA IRQ srcs * @dd: valid devdata structure * @i: index of SDMA engine
@@ -265,10 +276,9 @@ static void enable_sdma_srcs(struct hfi1_devdata *dd, int i)int msix_request_irqs (struct hfi1_devdata *dd) {int i; -int ret; +int ret =msix_request_general_irq (dd); - ret =msix_request_irq (dd, dd, general_interrupt,NULL ,0 ,IRQ_GENERAL ); -if (ret <0 ) +if (ret)return ret;for (i =0 ; i < dd->num_sdma; i++) {
drivers/infiniband/hw/hfi1/msix.h +1 -0
@@ -54,6 +54,7 @@int msix_initialize (struct hfi1_devdata *dd);int msix_request_irqs (struct hfi1_devdata *dd);void msix_clean_up_interrupts (struct hfi1_devdata *dd); +int msix_request_general_irq (struct hfi1_devdata *dd);int msix_request_rcd_irq (struct hfi1_ctxtdata *rcd);int msix_request_sdma_irq (struct sdma_engine *sde);void msix_free_irq (struct hfi1_devdata *dd,u8 msix_intr);
PATCH 7/9 IB/hfi1: Return void in packet receiving functions
In reply to Dennis DalessandroFrom: Grzegorz Andrejczuk <[email protected]>
Packet receiving functions returns int value, and yet the return values are not used at all.
This patch converts the functions to return void.
--- drivers/infiniband/hw/hfi1/driver.c | 41 +++++++++++++++-------------------- drivers/infiniband/hw/hfi1/hfi.h | 2 +- 2 files changed, 18 insertions(+), 25 deletions(-)
drivers/infiniband/hw/hfi1/driver.c +17 -24
@@ -1553,23 +1553,22 @@ void handle_eflags(struct hfi1_packet *packet)* The following functions are called by the interrupt handler. They are type * specific handlers for each packet type. */ -static int process_receive_ib (struct hfi1_packet *packet) +static void process_receive_ib (struct hfi1_packet *packet) {if (hfi1_setup_9B_packet (packet)) -return RHF_RCV_CONTINUE ; +return ;if (unlikely (hfi1_dbg_should_fault_rx (packet))) -return RHF_RCV_CONTINUE ; +return ;trace_hfi1_rcvhdr (packet);if (unlikely (rhf_err_flags (packet->rhf))) {handle_eflags (packet); -return RHF_RCV_CONTINUE ; +return ; }hfi1_ib_rcv (packet); -return RHF_RCV_CONTINUE ; }static inline bool hfi1_is_vnic_packet (struct hfi1_packet *packet)
@@ -1585,23 +1584,23 @@ static inline bool hfi1_is_vnic_packet(struct hfi1_packet *packet)return false; } -static int process_receive_bypass (struct hfi1_packet *packet) +static void process_receive_bypass (struct hfi1_packet *packet) {struct hfi1_devdata *dd = packet->rcd->dd;if (hfi1_is_vnic_packet (packet)) {hfi1_vnic_bypass_rcv (packet); -return RHF_RCV_CONTINUE ; +return ; }if (hfi1_setup_bypass_packet (packet)) -return RHF_RCV_CONTINUE ; +return ;trace_hfi1_rcvhdr (packet);if (unlikely (rhf_err_flags (packet->rhf))) {handle_eflags (packet); -return RHF_RCV_CONTINUE ; +return ; }if (hfi1_16B_get_l2 (packet->hdr) ==0x2 ) {
@@ -1624,17 +1623,16 @@ static int process_receive_bypass(struct hfi1_packet *packet) (OPA_EI_STATUS_SMASK |BAD_L2_ERR ); } } -return RHF_RCV_CONTINUE ; } -static int process_receive_error (struct hfi1_packet *packet) +static void process_receive_error (struct hfi1_packet *packet) {/* KHdrHCRCErr -- KDETH packet with a bad HCRC */ if (unlikely (hfi1_dbg_fault_suppress_err (&packet->rcd->dd->verbs_dev) && (rhf_rcv_type_err (packet->rhf) ==RHF_RCV_TYPE_ERROR || packet->rhf &RHF_DC_ERR ))) -return RHF_RCV_CONTINUE ; +return ;hfi1_setup_ib_header (packet);handle_eflags (packet);
@@ -1642,32 +1640,29 @@ static int process_receive_error(struct hfi1_packet *packet)if (unlikely (rhf_err_flags (packet->rhf)))dd_dev_err (packet->rcd->dd,"Unhandled error packet received. Dropping.\n" ); - -return RHF_RCV_CONTINUE ; } -static int kdeth_process_expected (struct hfi1_packet *packet) +static void kdeth_process_expected (struct hfi1_packet *packet) {hfi1_setup_9B_packet (packet);if (unlikely (hfi1_dbg_should_fault_rx (packet))) -return RHF_RCV_CONTINUE ; +return ;if (unlikely (rhf_err_flags (packet->rhf))) {struct hfi1_ctxtdata *rcd = packet->rcd;if (hfi1_handle_kdeth_eflags (rcd, rcd->ppd, packet)) -return RHF_RCV_CONTINUE ; +return ; }hfi1_kdeth_expected_rcv (packet); -return RHF_RCV_CONTINUE ; } -static int kdeth_process_eager (struct hfi1_packet *packet) +static void kdeth_process_eager (struct hfi1_packet *packet) {hfi1_setup_9B_packet (packet);if (unlikely (hfi1_dbg_should_fault_rx (packet))) -return RHF_RCV_CONTINUE ; +return ;trace_hfi1_rcvhdr (packet);if (unlikely (rhf_err_flags (packet->rhf))) {
@@ -1675,18 +1670,16 @@ static int kdeth_process_eager(struct hfi1_packet *packet)show_eflags_errs (packet);if (hfi1_handle_kdeth_eflags (rcd, rcd->ppd, packet)) -return RHF_RCV_CONTINUE ; +return ; }hfi1_kdeth_eager_rcv (packet); -return RHF_RCV_CONTINUE ; } -static int process_receive_invalid (struct hfi1_packet *packet) +static void process_receive_invalid (struct hfi1_packet *packet) {dd_dev_err (packet->rcd->dd,"Invalid packet type %d. Dropping\n" ,rhf_rcv_type (packet->rhf)); -return RHF_RCV_CONTINUE ; }#define HFI1_RCVHDR_DUMP_MAX 5
drivers/infiniband/hw/hfi1/hfi.h +1 -1
@@ -199,7 +199,7 @@ struct exp_tid_set {struct hfi1_ctxtdata ;typedef int (*intr_handler )(struct hfi1_ctxtdata *rcd,int data); -typedef int (*rhf_rcv_function_ptr )(struct hfi1_packet *packet); +typedef void (*rhf_rcv_function_ptr )(struct hfi1_packet *packet);struct tid_queue {struct list_head queue_head;
PATCH 8/9 IB/hfi1: Add software counter for ctxt0 seq drop
In reply to Dennis DalessandroFrom: Mike Marciniszyn <[email protected]>
All other code paths increment some form of drop counter.
This was missed in the original implementation.
--- drivers/infiniband/hw/hfi1/chip.c | 10 ++++++++++ drivers/infiniband/hw/hfi1/chip.h | 1 + drivers/infiniband/hw/hfi1/driver.c | 1 + drivers/infiniband/hw/hfi1/hfi.h | 2 ++ 4 files changed, 14 insertions(+)
drivers/infiniband/hw/hfi1/chip.c +10 -0
@@ -1685,6 +1685,14 @@ static u64 access_sw_pio_drain(const struct cntr_entry *entry,return dd->verbs_dev.n_piodrain; } +static u64 access_sw_ctx0_seq_drop (const struct cntr_entry *entry, +void *context,int vl,int mode,u64 data) +{ +struct hfi1_devdata *dd = context; + +return dd->ctx0_seq_drop; +} +static u64 access_sw_vtx_wait (const struct cntr_entry *entry,void *context,int vl,int mode,u64 data) {
@@ -4249,6 +4257,8 @@ static u64 access_dc_rcv_err_cnt(const struct cntr_entry *entry, access_sw_cpu_intr), [C_SW_CPU_RCV_LIM ] =CNTR_ELEM ("RcvLimit" ,0 ,0 ,CNTR_NORMAL , access_sw_cpu_rcv_limit), +[C_SW_CTX0_SEQ_DROP ] =CNTR_ELEM ("SeqDrop0" ,0 ,0 ,CNTR_NORMAL , + access_sw_ctx0_seq_drop), [C_SW_VTX_WAIT ] =CNTR_ELEM ("vTxWait" ,0 ,0 ,CNTR_NORMAL , access_sw_vtx_wait), [C_SW_PIO_WAIT] = CNTR_ELEM("PioWait", 0, 0, CNTR_NORMAL,
drivers/infiniband/hw/hfi1/chip.h +1 -0
@@ -932,6 +932,7 @@ enum {C_DC_PG_STS_TX_MBE_CNT ,C_SW_CPU_INTR ,C_SW_CPU_RCV_LIM , +C_SW_CTX0_SEQ_DROP ,C_SW_VTX_WAIT ,C_SW_PIO_WAIT , C_SW_PIO_DRAIN,
drivers/infiniband/hw/hfi1/driver.c +1 -0
@@ -732,6 +732,7 @@ static noinline int skip_rcv_packet(struct hfi1_packet *packet, int thread) {int ret; + packet->rcd->dd->ctx0_seq_drop++;/* Set up for the next packet */ packet->rhqoff += packet->rsize;if (packet->rhqoff >= packet->maxcnt)
drivers/infiniband/hw/hfi1/hfi.h +2 -0
@@ -1159,6 +1159,8 @@ struct hfi1_devdata {char *boardname;/* human readable board info */ +u64 ctx0_seq_drop; +/* reset value */ u64 z_int_counter;u64 z_rcv_limit;
PATCH 9/9 IB/hfi1: Add RcvShortLengthErrCnt to hfi1stats
In reply to Dennis DalessandroFrom: Mike Marciniszyn <[email protected]>
This counter, RxShrErr, is required for error analysis and debug.
--- drivers/infiniband/hw/hfi1/chip.c | 1 + drivers/infiniband/hw/hfi1/chip.h | 1 + drivers/infiniband/hw/hfi1/chip_registers.h | 1 + 3 files changed, 3 insertions(+)
drivers/infiniband/hw/hfi1/chip.c +1 -0
@@ -4114,6 +4114,7 @@ static u64 access_dc_rcv_err_cnt(const struct cntr_entry *entry,static struct cntr_entry dev_cntrs[DEV_CNTR_LAST ] = { [C_RCV_OVF ] =RXE32_DEV_CNTR_ELEM (RcvOverflow,RCV_BUF_OVFL_CNT ,CNTR_SYNTH ), [C_RX_LEN_ERR ] =RXE32_DEV_CNTR_ELEM (RxLenErr,RCV_LENGTH_ERR_CNT ,CNTR_SYNTH ), +[C_RX_SHORT_ERR ] =RXE32_DEV_CNTR_ELEM (RxShrErr,RCV_SHORT_ERR_CNT ,CNTR_SYNTH ), [C_RX_ICRC_ERR ] =RXE32_DEV_CNTR_ELEM (RxICrcErr,RCV_ICRC_ERR_CNT ,CNTR_SYNTH ), [C_RX_EBP ] =RXE32_DEV_CNTR_ELEM (RxEbpCnt,RCV_EBP_CNT ,CNTR_SYNTH ), [C_RX_TID_FULL] = RXE32_DEV_CNTR_ELEM(RxTIDFullEr, RCV_TID_FULL_ERR_CNT,
drivers/infiniband/hw/hfi1/chip.h +1 -0
@@ -865,6 +865,7 @@ static inline int idx_from_vl(int vl)enum {C_RCV_OVF =0 ,C_RX_LEN_ERR , +C_RX_SHORT_ERR ,C_RX_ICRC_ERR ,C_RX_EBP ,C_RX_TID_FULL ,
drivers/infiniband/hw/hfi1/chip_registers.h +1 -0
@@ -381,6 +381,7 @@#define DC_LCB_STS_LINK_TRANSFER_ACTIVE (DC_LCB_CSRS + 0x000000000468)#define DC_LCB_STS_ROUND_TRIP_LTP_CNT (DC_LCB_CSRS + 0x0000000004B0)#define RCV_LENGTH_ERR_CNT 0 +#define RCV_SHORT_ERR_CNT 2#define RCV_ICRC_ERR_CNT 6#define RCV_EBP_CNT 9#define RCV_BUF_OVFL_CNT 10
Quoted 32 lines · Dennis Dalessandro
These patches add some recactoring and code clean ups to make things more organized. There is a performance optimization and new counter/debugging stats added as well. The new "API" that is added is a driver internal API not an actual "API" that is exposed to the outside.
---
Grzegorz Andrejczuk (3):
IB/hfi1: Move common receive IRQ code to function IB/hfi1: Decouple IRQ name from type IB/hfi1: Return void in packet receiving functionsMike Marciniszyn (6):
IB/hfi1: Move chip specific functions tochip.cIB/hfi1: Add fast and slow handlers for receive context IB/hfi1: IB/hfi1: Add an API to handle special case drop IB/hfi1: Create API for auto activate
IB/hfi1: Add software counter for ctxt0 seq drop IB/hfi1: Add RcvShortLengthErrCnt to hfi1statsdrivers/infiniband/hw/hfi1/chip.c | 171 ++++++++++++++++++++++----- drivers/infiniband/hw/hfi1/chip.h | 8 + drivers/infiniband/hw/hfi1/chip_registers.h | 1 drivers/infiniband/hw/hfi1/driver.c | 151 +++++++++--------------- drivers/infiniband/hw/hfi1/hfi.h | 66 ++++++++++ drivers/infiniband/hw/hfi1/init.c | 81 ++----------- drivers/infiniband/hw/hfi1/msix.c | 106 +++++++++-------- drivers/infiniband/hw/hfi1/msix.h | 1 drivers/infiniband/hw/hfi1/trace_rx.h | 6 - 9 files changed, 337 insertions(+), 254 deletions(-)
Quoted 3 lines
--
-Denny
Forgot to mention these apply on top of the: wip/jgg-for-next branch.
-Denny
Jason Gunthorpe
AppliedQuoted 11 lines · Dennis Dalessandro
These patches add some recactoring and code clean ups to make things more organized. There is a performance optimization and new counter/debugging stats added as well. The new "API" that is added is a driver internal API not an actual "API" that is exposed to the outside.
Grzegorz Andrejczuk (3):
IB/hfi1: Move common receive IRQ code to function IB/hfi1: Decouple IRQ name from type IB/hfi1: Return void in packet receiving functions
Quoted 7 lines
Mike Marciniszyn (6):
IB/hfi1: Move chip specific functions tochip.cIB/hfi1: Add fast and slow handlers for receive context IB/hfi1: IB/hfi1: Add an API to handle special case drop IB/hfi1: Create API for auto activate
IB/hfi1: Add software counter for ctxt0 seq drop IB/hfi1: Add RcvShortLengthErrCnt to hfi1stats
Applied to for-next, thanks
Jason