* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (62 commits)
mlx4_core: Deprecate log_num_vlan module param
IB/mlx4: Don't set VLAN in IBoE WQEs' control segment
IB/mlx4: Enable 4K mtu for IBoE
RDMA/cxgb4: Mark QP in error before disabling the queue in firmware
RDMA/cxgb4: Serialize calls to CQ's comp_handler
RDMA/cxgb3: Serialize calls to CQ's comp_handler
IB/qib: Fix issue with link states and QSFP cables
IB/mlx4: Configure extended active speeds
mlx4_core: Add extended port capabilities support
IB/qib: Hold links until tuning data is available
IB/qib: Clean up checkpatch issue
IB/qib: Remove s_lock around header validation
IB/qib: Precompute timeout jiffies to optimize latency
IB/qib: Use RCU for qpn lookup
IB/qib: Eliminate divide/mod in converting idx to egr buf pointer
IB/qib: Decode path MTU optimization
IB/qib: Optimize RC/UC code by IB operation
IPoIB: Use the right function to do DMA unmap pages
RDMA/cxgb4: Use correct QID in insert_recv_cqe()
RDMA/cxgb4: Make sure flush CQ entries are collected on connection close
...
Some kernel components pin user space memory (infiniband and perf) (by
increasing the page count) and account that memory as "mlocked".
The difference between mlocking and pinning is:
A. mlocked pages are marked with PG_mlocked and are exempt from
swapping. Page migration may move them around though.
They are kept on a special LRU list.
B. Pinned pages cannot be moved because something needs to
directly access physical memory. They may not be on any
LRU list.
I recently saw an mlockalled process where mm->locked_vm became
bigger than the virtual size of the process (!) because some
memory was accounted for twice:
Once when the page was mlocked and once when the Infiniband
layer increased the refcount because it needt to pin the RDMA
memory.
This patch introduces a separate counter for pinned pages and
accounts them seperately.
Signed-off-by: Christoph Lameter <cl@linux.com>
Cc: Mike Marciniszyn <infinipath@qlogic.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There's no need to set the vlan-related fields in an IBoE send WQE
control segment:
- the vlan to be used by a UD QP is set in the datagram segment.
- for GSI (CM) QP, all the headers down to 8021q and MAC are built by
the software anyway.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The IBoE port MTU is derived from the corresponding Ethernet netdevice
MTU, which can support jumbo frames of 9K, and hence surely supports
the max IB mtu of 4K.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
QPs need to be moved to error before telling the firwmare to shutdown
the queue. Otherwise, the application can submit WRs that will never
get fetched by the hardware and never flushed by the driver.
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Acked-by: Steve Wise <swsie@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Commit 01e7da6ba5 ("RDMA/cxgb4: Make sure flush CQ entries are
collected on connection close") introduced a potential problem where a
CQ's comp_handler can get called simultaneously from different places
in the iw_cxgb4 driver. This does not comply with
Documentation/infiniband/core_locking.txt, which states that at a
given point of time, there should be only one callback per CQ should
be active.
This problem was reported by Parav Pandit <Parav.Pandit@Emulex.Com>.
Based on discussion between Parav Pandit and Steve Wise, this patch
fixes the above problem by serializing the calls to a CQ's
comp_handler using a spin_lock.
Reported-by: Parav Pandit <Parav.Pandit@Emulex.Com>
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
iw_cxgb3 has a potential problem where a CQ's comp_handler can get
called simultaneously from different places in iw_cxgb3 driver. This
does not comply with Documentation/infiniband/core_locking.txt, which
states that at a given point of time, there should be only one
callback per CQ should be active.
Such problem was reported by Parav Pandit <Parav.Pandit@Emulex.Com>
for iw_cxgb4 driver. Based on discussion between Parav Pandit and
Steve Wise, this patch fixes the above problem by serializing the
calls to a CQ's comp_handler using a spin_lock.
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Fix an issue where the link would come up after replugging a cable
even if it has been DISABLED manually.
Signed-off-by: Mitko Haralanov <mitko@qlogic.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Set the extended active speeds based on the hardware configuration.
Signed-off-by: Marcel Apfelbaum <marcela@dev.mellanox.co.il>
Reviewed-by: Hal Rosenstock <hal@mellanox.com>
[ Move FDR-10 handling into ib_link_query_port(). - Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
Hold the link state machine until the tuning data is read from the
QSFP EEPROM so correct tuning settings are applied before the state
machine attempts to bring the link up. Link is also held on cable
unplug in case a different cable is used.
Signed-off-by: Mitko Haralanov <mitko@qlogic.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This was probably present from initial submission.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Review of qib_ruc_check_hdr() shows that the s_lock is not required in
the normal case. The r_lock is held in all cases, and protects the qp
fields that are read.
The s_lock will be needed to around the call to qib_migrate_qp() to
insure that the send engine sees a consistent set of fields.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
A new field is added to qib_qp called timeout_jiffies. It is
initialized upon create and modify.
The field is now used instead of a computation based on qp->timeout.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The heavy weight spinlock in qib_lookup_qpn() is replaced with RCU.
The hash list itself is now accessed via jhash functions instead of mod.
The changes should benefit multiple receive contexts in different
processors by not contending for the lock just to read the hash
structures.
The patch also adds a lookaside_qp (pointer) and a lookaside_qpn in
the context. The interrupt handler will test the current packet's qpn
against lookaside_qpn if the lookaside_qp pointer is non-NULL. The
pointer is NULL'ed when the interrupt handler exits.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The context init now saves a shift from rcvegrbufs_perchunk
rcvegrbufs_perchunk_shift using ilog2. A BUG_ON() protects the
power of 2 assumption.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Store both the encoded and decoded MTU in the QP structure as a minor
optimization for UC/RC receive routines.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The memset for zeroing work completions had been unconditional.
This patch removes the memset and moves the zeroing into the work
completion with a more explicit field by field set. With this patch,
non-ONLY/non-LAST packets will avoid the overhead since they will not
generate a completion.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
To ease skb->truesize sanitization, its better to be able to localize
all references to skb frags size.
Define accessors : skb_frag_size() to fetch frag size, and
skb_frag_size_{set|add|sub}() to manipulate it.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When creating flushed receive CQEs, set the QPID field in the t4_cqe
to the SQ QID and not the RQ QID. Otherwise the poll code will not
find the correct QP context.
Signed-off by: Jonathan Lallinger <jonathan@ogc.us>
Signed-off by: Steve Wise <swise@ogc.us>
Signed-off-by: Roland Dreier <roland@purestorage.com>
At the time when a peer closes the connection, iw_cxgb4 will not send
a cq event if ibqp.uobject exists. In that case, its possible for a
user application to get blocked in ibv_get_cq_event().
To resolve this, call the cq's comp_handler to unblock any read from
ibv_get_cq_event(). This will trigger userspace to poll the cq and
collect flush status completions for any pending work requests.
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Allow processes that share the same XRC domain to open an existing
shareable QP. This permits those processes to receive events on the
shared QP and transfer ownership, so that any process may modify the
QP. The latter allows the creating process to exit, while a remaining
process can still transition it for path migration purposes.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Support the creation of XRC INI and TGT QPs. To handle the case where
a CQ or PD is not provided, we allocate them internally with the xrcd.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Allow the user to create XRC SRQs. This patch is based on a patch
from Jack Morgenstrein <jackm@dev.mellanox.co.il>.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Support creating and destroying XRC domains. Any sharing of the XRCD
is managed above the low-level driver.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Currently, there is only a single ("basic") type of SRQ, but with XRC
support we will add a second. Prepare for this by defining an SRQ type
and setting all current users to IB_SRQT_BASIC.
Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Fix build errors:
drivers/infiniband/hw/ipath/ipath_init_chip.c:54:1: error: 'S_IRUGO' undeclared here (not in a function)
drivers/infiniband/hw/ipath/ipath_init_chip.c:54:1: error: bit-field '<anonymous>' width not an integer constant
drivers/infiniband/hw/ipath/ipath_init_chip.c:67:1: error: 'S_IWUSR' undeclared here (not in a function)
drivers/infiniband/hw/ipath/ipath_init_chip.c:67:1: error: bit-field '<anonymous>' width not an integer constant
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Support for Packed and Unaligned (PAU) FPDUs is needed for
interoperability between NES and non-NES nodes. When the NES hardware
detects a PAU frame, it will pass it to the driver to process the
frame. NES driver creates a new frame for each FPDU and forwards it
to the hardware to be sent to its associated qp.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Faisal Latif <Faisal.Latif@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Print the IP address of the remote host when a critical asynchronous event is
received.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Faisal Latif <Faisal.Latif@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Fixes a crash that occurs during close when error async event is received.
Terminate message is not sent to the remote node if already processing close.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Faisal Latif <Faisal.Latif@intel.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
When I converted some drivers from pci_map_page to skb_frag_dma_map I
neglected to convert PCI_DMA_xDEVICE into DMA_x_DEVICE and
pci_dma_mapping_error into dma_mapping_error.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for Enhanced RDMA Connection Establishment
(draft-ietf-storm-mpa-peer-connect-06), aka MPAv2. Details of draft
can be obtained from:
<http://www.ietf.org/id/draft-ietf-storm-mpa-peer-connect-06.txt>
For backwards compatibility, the MPAv2 enabled driver reverts to MPAv1
if the remote node doesn't support MPAv2.
Signed-off-by: Tatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: Faisal Latif <Faisal.Latif@intel.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This patch adds support for Enhanced RDMA Connection Establishment
(draft-ietf-storm-mpa-peer-connect-06), aka MPAv2. Details of draft
can be obtained from:
<http://www.ietf.org/id/draft-ietf-storm-mpa-peer-connect-06.txt>
The patch updates the following functions for initiator perspective:
- send_mpa_request
- process_mpa_reply
- post_terminate for TERM error codes
- destroy_qp for TERM related change
- adds layer/etype/ecode to c4iw_qp_attrs for sending with TERM
- peer_abort for retrying connection attempt with MPA_v1 message
- added c4iw_reconnect function
The patch updates the following functions for responder perspective:
- process_mpa_request
- send_mpa_reply
- c4iw_accept_cr
- passes ird/ord to upper layers
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
As part of MPAv2 Enhanced RDMA Negotiation, pass max supported ird/ord
values upwards for the time being in iw_cxgb3 and amso1100.
Signed-off-by: Kumar Sanghvi <kumaras@chelsio.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The code that was recently introduced to report the number
of free contexts is flawed for multiple HCAs:
/* Return the number of free user ports (contexts) available. */
return scnprintf(buf, PAGE_SIZE, "%u\n", dd->cfgctxts -
dd->first_user_ctxt - (u32)qib_stats.sps_ctxts);
The qib_stats is global to the module, not per HCA, so the code is broken
for multiple HCAs.
This patch adds a qib_devdata field, freectxts, that reflects the free
contexts for this HCA.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Reviewed-by: Ram Vepa <ram.vepa@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Add calls to ib_umem_release(), as in the other error-handling code in
nes_reg_user_mr().
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Since commit e58aa3d2d0 ("genirq: Run irq handlers with interrupts
disabled"), we run all interrupt handlers with interrupts disabled and
we even check and yell when an interrupt handler returns with
interrupts enabled -- cf commit b738a50a20 ("genirq: Warn when
handler enables interrupts").
So now this flag is a no-op and can be removed.
Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Acked-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The iw_cxgb4 module crashes at init time if the T4 card does not
support RDMA. So clean up the init logic to correctly deal with
non-RDMA cards.
- If any RDMA resources are not available, then fail the initialization
logging an info message.
- Clean up properly on initialization failures.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This oops was reported recently:
d:mon> e
cpu 0xd: Vector: 300 (Data Access) at [c0000000fd4c7120]
pc: d00000000076f194: .t3_l2t_get+0x44/0x524 [cxgb3]
lr: d000000000b02108: .init_act_open+0x150/0x3d4 [cxgb3i]
sp: c0000000fd4c73a0
msr: 8000000000009032
dar: 0
dsisr: 40000000
current = 0xc0000000fd640d40
paca = 0xc00000000054ff80
pid = 5085, comm = iscsid
d:mon> t
[c0000000fd4c7450] d000000000b02108 .init_act_open+0x150/0x3d4 [cxgb3i]
[c0000000fd4c7500] d000000000e45378 .cxgbi_ep_connect+0x784/0x8e8 [libcxgbi]
[c0000000fd4c7650] d000000000db33f0 .iscsi_if_rx+0x71c/0xb18
[scsi_transport_iscsi2]
[c0000000fd4c7740] c000000000370c9c .netlink_data_ready+0x40/0xa4
[c0000000fd4c77c0] c00000000036f010 .netlink_sendskb+0x4c/0x9c
[c0000000fd4c7850] c000000000370c18 .netlink_sendmsg+0x358/0x39c
[c0000000fd4c7950] c00000000033be24 .sock_sendmsg+0x114/0x1b8
[c0000000fd4c7b50] c00000000033d208 .sys_sendmsg+0x218/0x2ac
[c0000000fd4c7d70] c00000000033f55c .sys_socketcall+0x228/0x27c
[c0000000fd4c7e30] c0000000000086a4 syscall_exit+0x0/0x40
--- Exception: c01 (System Call) at 00000080da560cfc
The root cause was an EEH error, which sent us down the offload_close path in
the cxgb3 driver, which in turn sets cdev->l2opt to NULL, without regard for
upper layer driver (like the cxgbi drivers) which might have execution contexts
in the middle of its use. The result is the oops above, when t3_l2t_get attempts
to dereference L2DATA(cdev)->nentries in arp_hash right after the EEH error handler sets it to NULL.
The fix is to prevent the setting of the NULL pointer until after there are no
further users of it. The t3cdev->l2opt pointer is now converted to be an rcu
pointer and the L2DATA macro is now called under the protection of the
rcu_read_lock(). When the EEH error path:
t3_adapter_error->offload_close->cxgb3_offload_deactivate
Is exectured, setting of that l2opt pointer to NULL, is now gated on an rcu
quiescence point, preventing, allowing L2DATA callers to safely check for a NULL
pointer without concern that the underlying data will be freeded before the
pointer is dereferenced.
This has been tested by the reporter and shown to fix the reproted oops
[nhorman: fix up unitinialised variable reported by Dan Carpenter]
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Karen Xie <kxie@chelsio.com>
Cc: stable@kernel.org
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Faisal Latif <faisal.latif@intel.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Cc: netdev@vger.kernel.org
Acked-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Tom Tucker <tom@opengridcomputing.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux-rdma@vger.kernel.org
Cc: netdev@vger.kernel.org
Acked-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Moves the drivers for the Chelsio chipsets into
drivers/net/ethernet/chelsio/ and the necessary Kconfig and Makefile
changes.
CC: Divy Le Ray <divy@chelsio.com>
CC: Dimitris Michailidis <dm@chelsio.com>
CC: Casey Leedom <leedom@chelsio.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This allows us to move duplicated code in <asm/atomic.h>
(atomic_inc_not_zero() for now) to <linux/atomic.h>
Signed-off-by: Arun Sharma <asharma@fb.com>
Reviewed-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: David Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (26 commits)
IB/qib: Defer HCA error events to tasklet
mlx4_core: Bump the driver version to 1.0
RDMA/cxgb4: Use printk_ratelimited() instead of printk_ratelimit()
IB/mlx4: Support PMA counters for IBoE
IB/mlx4: Use flow counters on IBoE ports
IB/pma: Add include file for IBA performance counters definitions
mlx4_core: Add network flow counters
mlx4_core: Fix location of counter index in QP context struct
mlx4_core: Read extended capabilities into the flags field
mlx4_core: Extend capability flags to 64 bits
IB/mlx4: Generate GID change events in IBoE code
IB/core: Add GID change event
RDMA/cma: Don't allow IPoIB port space for IBoE
RDMA: Allow for NULL .modify_device() and .modify_port() methods
IB/qib: Update active link width
IB/qib: Fix potential deadlock with link down interrupt
IB/qib: Add sysfs interface to read free contexts
IB/mthca: Remove unnecessary read of PCI_CAP_ID_EXP
IB/qib: Remove double define
IB/qib: Remove unnecessary read of PCI_CAP_ID_EXP
...
With ib_qib options:
options ib_qib krcvqs=1 pcie_caps=0x51 rcvhdrcnt=4096 singleport=1 ibmtu=4
a run of ib_write_bw -a yields the following:
------------------------------------------------------------------
#bytes #iterations BW peak[MB/sec] BW average[MB/sec]
1048576 5000 2910.64 229.80
------------------------------------------------------------------
The top cpu use in a profile is:
CPU: Intel Architectural Perfmon, speed 2400.15 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask
of 0x00 (No unit mask) count 1002300
Counted LLC_MISSES events (Last level cache demand requests from this core that
missed the LLC) with a unit mask of 0x41 (No unit mask) count 10000
samples % samples % app name symbol name
15237 29.2642 964 17.1195 ib_qib.ko qib_7322intr
12320 23.6618 1040 18.4692 ib_qib.ko handle_7322_errors
4106 7.8860 0 0 vmlinux vsnprintf
Analysis of the stats, profile, the code, and the annotated profile indicate:
- All of the overflow interrupts (one per packet overflow) are
serviced on CPU0 with no mitigation on the frequency.
- All of the receive interrupts are being serviced by CPU0. (That is
the way truescale.cmds statically allocates the kctx IRQs to CPU)
- The code is spending all of its time servicing QIB_I_C_ERROR
RcvEgrFullErr interrupts on CPU0, starving the packet receive
processing.
- The decode_err routine is very inefficient, using a printf variant
to format a "%s" and continues to loop when the errs mask has been
cleared.
- Both qib_7322intr and handle_7322_errors read pci registers, which
is very inefficient.
The fix does the following:
- Adds a tasklet to service QIB_I_C_ERROR
- Replaces the very inefficient scnprintf() with a memcpy(). A field
is added to qib_hwerror_msgs to save the sizeof("string") at
compile time so that a strlen is not needed during err_decode().
- The most frequent errors (Overflows) are serviced first to exit the
loop as early as possible.
- The loop now exits as soon as the errs mask is clear rather than
fruitlessly looping through the msp array.
With this fix the performance changes to:
------------------------------------------------------------------
#bytes #iterations BW peak[MB/sec] BW average[MB/sec]
1048576 5000 2990.64 2941.35
------------------------------------------------------------------
During testing of the error handling overflow patch, it was determined
that some CPU's were slower when servicing both overflow and receive
interrupts on CPU0 with different MSI interrupt vectors.
This patch adds an option (krcvq01_no_msi) to not use a dedicated MSI
interrupt for kctx's < 2 and to service them on the default interrupt.
For some CPUs, the cost of the interrupt enter/exit is more costly
than then the additional PCI read in the default handler.
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Linus Torvalds