[PATCH 3/3] staging/rdma/hfi1: check for ARMED->ACTIVE transition in receive interrupt

Jubin John jubin.john at intel.com
Wed Dec 23 22:27:17 UTC 2015


On Fri, Dec 18, 2015 at 09:31:39AM +0300, Dan Carpenter wrote:
> Possible off by one, but mostly the whitespace makes me itch.
> 
> Jim was not on the CC list.
> 
> On Thu, Dec 17, 2015 at 07:24:15PM -0500, Jubin John wrote:
> > From: Jim Snow <jim.m.snow at intel.com>
> > 
> > The link state will transition from ARMED to ACTIVE when a non-SC15
> > packet arrives, but the driver might not notice the change.  With this
> > fix, if the slowpath receive interrupt handler sees a non-SC15 packet
> > while in the ARMED state, we queue work to call linkstate_active_work
> > from process context to promote it to ACTIVE.
> > 
> > Signed-off-by: Jim Snow <jim.m.snow at intel.com>
> > Signed-off-by: Brendan Cunningham <brendan.cunningham at intel.com>
> > Reviewed-by: Dean Luick <dean.luick at intel.com>
> > Signed-off-by: Jubin John <jubin.john at intel.com>
> > ---
> >  drivers/staging/rdma/hfi1/chip.c   |    5 ++-
> >  drivers/staging/rdma/hfi1/chip.h   |    2 +
> >  drivers/staging/rdma/hfi1/driver.c |   66 ++++++++++++++++++++++++++++++++++++
> >  drivers/staging/rdma/hfi1/hfi.h    |   12 ++++++
> >  drivers/staging/rdma/hfi1/init.c   |    1 +
> >  5 files changed, 84 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/staging/rdma/hfi1/chip.c b/drivers/staging/rdma/hfi1/chip.c
> > index 78a5f08..f82b848 100644
> > --- a/drivers/staging/rdma/hfi1/chip.c
> > +++ b/drivers/staging/rdma/hfi1/chip.c
> > @@ -4606,7 +4606,7 @@ 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);
> >  }
> > @@ -4705,7 +4705,7 @@ u32 read_physical_state(struct hfi1_devdata *dd)
> >  				& 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;
> >  
> > @@ -6658,6 +6658,7 @@ int set_link_state(struct hfi1_pportdata *ppd, u32 state)
> >  				ppd->link_enabled = 1;
> >  		}
> >  
> > +		set_all_slowpath(ppd->dd);
> >  		ret = set_local_link_attributes(ppd);
> >  		if (ret)
> >  			break;
> > diff --git a/drivers/staging/rdma/hfi1/chip.h b/drivers/staging/rdma/hfi1/chip.h
> > index d74aed8..620cf08 100644
> > --- a/drivers/staging/rdma/hfi1/chip.h
> > +++ b/drivers/staging/rdma/hfi1/chip.h
> > @@ -691,6 +691,8 @@ u64 read_dev_cntr(struct hfi1_devdata *dd, int index, int vl);
> >  u64 write_dev_cntr(struct hfi1_devdata *dd, int index, int vl, u64 data);
> >  u64 read_port_cntr(struct hfi1_pportdata *ppd, int index, int vl);
> >  u64 write_port_cntr(struct hfi1_pportdata *ppd, int index, int vl, u64 data);
> > +u32 read_logical_state(struct hfi1_devdata *dd);
> > +void force_recv_intr(struct hfi1_ctxtdata *rcd);
> >  
> >  /* Per VL indexes */
> >  enum {
> > diff --git a/drivers/staging/rdma/hfi1/driver.c b/drivers/staging/rdma/hfi1/driver.c
> > index 4c52e78..16b1be1 100644
> > --- a/drivers/staging/rdma/hfi1/driver.c
> > +++ b/drivers/staging/rdma/hfi1/driver.c
> > @@ -862,6 +862,15 @@ static inline void set_all_dma_rtail(struct hfi1_devdata *dd)
> >  			&handle_receive_interrupt_dma_rtail;
> >  }
> >  
> > +void set_all_slowpath(struct hfi1_devdata *dd)
> > +{
> > +	int i;
> > +
> > +	for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
> > +		dd->rcd[i]->do_interrupt =
> > +		  &handle_receive_interrupt;
> 
> This fits within the 80 character limit.

Will fix in v2.
> 
> We start counting from HFI1_CTRL_CTXT + 1 but in receive_interrupt_work()
> we start counting from HFI1_CTRL_CTXT.  What's the story?  It's either a
> bug or needs much better documentation.

In set_all_slowpath(), we exclude HFI1_CTRL_CTXT because HFI1_CTRL_CTXT
(the SC15, error, and multicast context) must always use the slowpath
handler. We skip HFI1_CTRL_CTXT in set_all_nodma_rtail() and
set_all_dma_rtail() as well.

receive_interrupt_work() starts at HFI1_CTRL_CTXT because we want to
trigger interrupts on all of the receive contexts, including
HFI1_CTRL_CTXT in case any packets were received on any of the contexts
between the interrupt and when the work function is called.
> 
> > +}
> > +
> >  /*
> >   * handle_receive_interrupt - receive a packet
> >   * @rcd: the context
> > @@ -929,6 +938,27 @@ 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)) {
> > +				struct work_struct *lsaw =
> > +				  &rcd->ppd->linkstate_active_work;
> > +				struct hfi1_message_header *hdr =
> > +				  hfi1_get_msgheader(packet.rcd->dd,
> > +						     packet.rhf_addr);
> > +				if (hdr2sc(hdr, packet.rhf) != 0xf) {
> > +					int hwstate = read_logical_state(dd);
> > +
> > +					if (hwstate != LSTATE_ACTIVE) {
> > +						dd_dev_info(dd, "Unexpected link state %d\n",
> > +							    hwstate);
> > +					} else {
> > +						queue_work(
> > +						  rcd->ppd->hfi1_wq, lsaw);
> > +						goto bail;
> > +					}
> > +				}
> > +			}
> 
> Can we make this an inline function?

Yes, will make that change in v2.
> 
> >  			last = process_rcv_packet(&packet, thread);
> >  		}
> >  
> > @@ -984,6 +1014,42 @@ bail:
> >  }
> >  
> >  /*
> > + * We may discover in the interrupt that the hardware link state has
> > + * changed from ARMED to ACTIVE (due to the arrival of a non-SC15 packet),
> > + * and we need to update the driver's notion of the link state.  We cannot
> > + * run set_link_state from interrupt context, so we queue this function on
> > + * a workqueue.
> > + *
> > + * We delay the regular interrupt processing until after the state changes
> > + * so that the link will be in the correct state by the time any application
> > + * we wake up attempts to send a reply to any message it received.
> > + * (Subsequent receive interrupts may possibly force the wakeup before we
> > + * update the link state.)
> > + *
> > + * The rcd is freed in hfi1_free_ctxtdata after hfi1_postinit_cleanup invokes
> > + * dd->f_cleanup(dd) to disable the interrupt handler and flush workqueues,
> > + * so we're safe from use-after-free of the rcd.
> > + */
> > +void receive_interrupt_work(struct work_struct *work)
> > +{
> > +	struct hfi1_pportdata *ppd =
> > +	  container_of(work, struct hfi1_pportdata, linkstate_active_work);
> 
> Normally we would put mostly on the first line.
> 
> 	struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
> 						  linkstate_active_work);

Fixed in v2.
> 
> > +	struct hfi1_devdata *dd = ppd->dd;
> > +	int i;
> > +
> > +	/* Received non-SC15 packet implies neighbor_normal */
> > +	ppd->neighbor_normal = 1;
> > +	set_link_state(ppd, HLS_UP_ACTIVE);
> > +
> > +	/*
> > +	 * Interrupt all kernel contexts that could have had an
> > +	 *  interrupt during auto activation.
> 
> Remove the extra space before "interrupt".

Fixed in v2.
> 
> > +	 */
> > +	for (i = HFI1_CTRL_CTXT; i < dd->first_user_ctxt; i++)
> > +		force_recv_intr(dd->rcd[i]);
> > +}
> > +
> > +/*
> >   * Convert a given MTU size to the on-wire MAD packet enumeration.
> >   * Return -1 if the size is invalid.
> >   */
> > diff --git a/drivers/staging/rdma/hfi1/hfi.h b/drivers/staging/rdma/hfi1/hfi.h
> > index 54ed6b3..bfd9723 100644
> > --- a/drivers/staging/rdma/hfi1/hfi.h
> > +++ b/drivers/staging/rdma/hfi1/hfi.h
> > @@ -712,6 +712,7 @@ struct hfi1_pportdata {
> >  	u8 remote_link_down_reason;
> >  	/* Error events that will cause a port bounce. */
> >  	u32 port_error_action;
> > +	struct work_struct linkstate_active_work;
> >  };
> >  
> >  typedef int (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
> > @@ -1128,6 +1129,7 @@ void hfi1_free_ctxtdata(struct hfi1_devdata *, struct hfi1_ctxtdata *);
> >  int handle_receive_interrupt(struct hfi1_ctxtdata *, int);
> >  int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *, int);
> >  int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *, int);
> > +void set_all_slowpath(struct hfi1_devdata *dd);
> >  
> >  /* receive packet handler dispositions */
> >  #define RCV_PKT_OK      0x0 /* keep going */
> > @@ -1148,6 +1150,16 @@ static inline u32 driver_lstate(struct hfi1_pportdata *ppd)
> >  	return ppd->lstate; /* use the cached value */
> >  }
> >  
> > +void receive_interrupt_work(struct work_struct *work);
> > +
> > +/* extract service channel from header and rhf */
> > +static inline int hdr2sc(struct hfi1_message_header *hdr, u64 rhf)
> > +{
> > +	return
> > +		((be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf) |
> > +		((!!(rhf & RHF_DC_INFO_MASK)) << 4);
> 
> This should be:
> 
> 	return ((be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf) |
> 	       ((!!(rhf & RHF_DC_INFO_MASK)) << 4);

Fixed in v2.
> 
> > +}
> > +
> >  static inline u16 generate_jkey(kuid_t uid)
> >  {
> >  	return from_kuid(current_user_ns(), uid) & 0xffff;
> > diff --git a/drivers/staging/rdma/hfi1/init.c b/drivers/staging/rdma/hfi1/init.c
> > index 8f13d53..ee206ae 100644
> > --- a/drivers/staging/rdma/hfi1/init.c
> > +++ b/drivers/staging/rdma/hfi1/init.c
> > @@ -498,6 +498,7 @@ void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
> >  	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);
> >  	mutex_init(&ppd->hls_lock);
> >  	spin_lock_init(&ppd->sdma_alllock);
> >  	spin_lock_init(&ppd->qsfp_info.qsfp_lock);
> > -- 
> > 1.7.1
> > 
> > _______________________________________________
> > devel mailing list
> > devel at linuxdriverproject.org
> > http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel


More information about the devel mailing list