Hi Doug and Jason,
Mike and Ira have identified some fixes that they would like to see land in the RC, these are serious problems and are marked as stable.
Changes since v1: Update Lukasz' email address and fix subject tag.
---
Ira Weiny (1): IB/hfi1: Fix SL array bounds check
Michael J. Ruhl (3): IB/hfi1: Invalid user input can result in crash IB/hfi1: Fix context recovery when PBC has an UnsupportedVL IB/hfi1: Fix destroy_qp hang after a link down
drivers/infiniband/hw/hfi1/chip.c | 6 +++- drivers/infiniband/hw/hfi1/pio.c | 51 ++++++++++++++++++++++++++------ drivers/infiniband/hw/hfi1/pio.h | 2 + drivers/infiniband/hw/hfi1/user_sdma.c | 2 + drivers/infiniband/hw/hfi1/verbs.c | 8 ++++- 5 files changed, 56 insertions(+), 13 deletions(-)
-- -Denny
From: Ira Weiny ira.weiny@intel.com
The SL specified by a user needs to be a valid SL.
Add a range check to the user specified SL value which protects from running off the end of the SL to SC table.
CC: stable@vger.kernel.org Fixes: 7724105686e7 ("IB/hfi1: add driver files") Signed-off-by: Ira Weiny ira.weiny@intel.com Signed-off-by: Dennis Dalessandro dennis.dalessandro@intel.com --- drivers/infiniband/hw/hfi1/verbs.c | 8 +++++++- 1 files changed, 7 insertions(+), 1 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c index 13374c7..a7c586a 100644 --- a/drivers/infiniband/hw/hfi1/verbs.c +++ b/drivers/infiniband/hw/hfi1/verbs.c @@ -1582,6 +1582,7 @@ static int hfi1_check_ah(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr) struct hfi1_pportdata *ppd; struct hfi1_devdata *dd; u8 sc5; + u8 sl;
if (hfi1_check_mcast(rdma_ah_get_dlid(ah_attr)) && !(rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH)) @@ -1590,8 +1591,13 @@ static int hfi1_check_ah(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr) /* test the mapping for validity */ ibp = to_iport(ibdev, rdma_ah_get_port_num(ah_attr)); ppd = ppd_from_ibp(ibp); - sc5 = ibp->sl_to_sc[rdma_ah_get_sl(ah_attr)]; dd = dd_from_ppd(ppd); + + sl = rdma_ah_get_sl(ah_attr); + if (sl >= ARRAY_SIZE(ibp->sl_to_sc)) + return -EINVAL; + + sc5 = ibp->sl_to_sc[sl]; if (sc_to_vlt(dd, sc5) > num_vls && sc_to_vlt(dd, sc5) != 0xf) return -EINVAL; return 0;
From: Michael J. Ruhl michael.j.ruhl@intel.com
If the number of packets in a user sdma request does not match the actual iovectors being sent, sdma_cleanup can be called on an uninitialized request structure, resulting in a crash similar to this:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008 IP: [<ffffffffc0ae8bb7>] __sdma_txclean+0x57/0x1e0 [hfi1] PGD 8000001044f61067 PUD 1052706067 PMD 0 Oops: 0000 [#1] SMP CPU: 30 PID: 69912 Comm: upsm Kdump: loaded Tainted: G OE ------------ 3.10.0-862.el7.x86_64 #1 Hardware name: Intel Corporation S2600KPR/S2600KPR, BIOS SE5C610.86B.01.01.0019.101220160604 10/12/2016 task: ffff8b331c890000 ti: ffff8b2ed1f98000 task.ti: ffff8b2ed1f98000 RIP: 0010:[<ffffffffc0ae8bb7>] [<ffffffffc0ae8bb7>] __sdma_txclean+0x57/0x1e0 [hfi1] RSP: 0018:ffff8b2ed1f9bab0 EFLAGS: 00010286 RAX: 0000000000008b2b RBX: ffff8b2adf6e0000 RCX: 0000000000000000 RDX: 00000000000000a0 RSI: ffff8b2e9eedc540 RDI: ffff8b2adf6e0000 RBP: ffff8b2ed1f9bad8 R08: 0000000000000000 R09: ffffffffc0b04a06 R10: ffff8b331c890190 R11: ffffe6ed00bf1840 R12: ffff8b3315480000 R13: ffff8b33154800f0 R14: 00000000fffffff2 R15: ffff8b2e9eedc540 FS: 00007f035ac47740(0000) GS:ffff8b331e100000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000000000008 CR3: 0000000c03fe6000 CR4: 00000000001607e0 Call Trace: [<ffffffffc0b0570d>] user_sdma_send_pkts+0xdcd/0x1990 [hfi1] [<ffffffff9fe75fb0>] ? gup_pud_range+0x140/0x290 [<ffffffffc0ad3105>] ? hfi1_mmu_rb_insert+0x155/0x1b0 [hfi1] [<ffffffffc0b0777b>] hfi1_user_sdma_process_request+0xc5b/0x11b0 [hfi1] [<ffffffffc0ac193a>] hfi1_aio_write+0xba/0x110 [hfi1] [<ffffffffa001a2bb>] do_sync_readv_writev+0x7b/0xd0 [<ffffffffa001bede>] do_readv_writev+0xce/0x260 [<ffffffffa022b089>] ? tty_ldisc_deref+0x19/0x20 [<ffffffffa02268c0>] ? n_tty_ioctl+0xe0/0xe0 [<ffffffffa001c105>] vfs_writev+0x35/0x60 [<ffffffffa001c2bf>] SyS_writev+0x7f/0x110 [<ffffffffa051f7d5>] system_call_fastpath+0x1c/0x21 Code: 06 49 c7 47 18 00 00 00 00 0f 87 89 01 00 00 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 2e 0f 1f 84 00 00 00 00 00 48 8b 4e 10 48 89 fb <48> 8b 51 08 49 89 d4 83 e2 0c 41 81 e4 00 e0 00 00 48 c1 ea 02 RIP [<ffffffffc0ae8bb7>] __sdma_txclean+0x57/0x1e0 [hfi1] RSP <ffff8b2ed1f9bab0> CR2: 0000000000000008
There are two exit points from user_sdma_send_pkts(). One (free_tx) merely frees the slab entry and one (free_txreq) cleans the sdma_txreq prior to freeing the slab entry. The free_txreq variation can only be called after one of the sdma_init*() variations has been called.
In the panic case, the slab entry had been allocated but not inited.
Fix the issue by exiting through free_tx thus avoiding sdma_clean().
Cc: stable@vger.kernel.org # 4.9.x+ Fixes: 7724105686e7 ("IB/hfi1: add driver files") Reviewed-by: Mike Marciniszyn mike.marciniszyn@intel.com Reviewed-by: Lukasz Odzioba lukasz.odzioba@intel.com Signed-off-by: Michael J. Ruhl michael.j.ruhl@intel.com Signed-off-by: Dennis Dalessandro dennis.dalessandro@intel.com --- drivers/infiniband/hw/hfi1/user_sdma.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c index a3a7b33..5c88706 100644 --- a/drivers/infiniband/hw/hfi1/user_sdma.c +++ b/drivers/infiniband/hw/hfi1/user_sdma.c @@ -828,7 +828,7 @@ static int user_sdma_send_pkts(struct user_sdma_request *req, unsigned maxpkts) if (READ_ONCE(iovec->offset) == iovec->iov.iov_len) { if (++req->iov_idx == req->data_iovs) { ret = -EFAULT; - goto free_txreq; + goto free_tx; } iovec = &req->iovs[req->iov_idx]; WARN_ON(iovec->offset);
From: Michael J. Ruhl michael.j.ruhl@intel.com
If a packet stream uses an UnsupportedVL (virtual lane), the send engine will not send the packet, and it will not indicate that an error has occurred. This will cause the packet stream to block.
HFI has 8 virtual lanes available for packet streams. Each lane can be enabled or disabled using the UnsupportedVL mask. If a lane is disabled, adding a packet to the send context must be disallowed.
The current mask for determining unsupported VLs defaults to 0 (allow all). This is incorrect. Only the VLs that are defined should be allowed.
Determine which VLs are disabled (mtu == 0), and set the appropriate unsupported bit in the mask. The correct mask will allow the send engine to error on the invalid VL, and error recovery will work correctly.
Cc: stable@vger.kernel.org # 4.9.x+ Fixes: 7724105686e7 ("IB/hfi1: add driver files") Reviewed-by: Mike Marciniszyn mike.marciniszyn@intel.com Reviewed-by: Lukasz Odzioba lukasz.odzioba@intel.com Signed-off-by: Michael J. Ruhl michael.j.ruhl@intel.com Signed-off-by: Dennis Dalessandro dennis.dalessandro@intel.com --- drivers/infiniband/hw/hfi1/pio.c | 9 +++++++-- 1 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c index c2c1cba..cd962c9 100644 --- a/drivers/infiniband/hw/hfi1/pio.c +++ b/drivers/infiniband/hw/hfi1/pio.c @@ -86,6 +86,7 @@ void pio_send_control(struct hfi1_devdata *dd, int op) unsigned long flags; int write = 1; /* write sendctrl back */ int flush = 0; /* re-read sendctrl to make sure it is flushed */ + int i;
spin_lock_irqsave(&dd->sendctrl_lock, flags);
@@ -95,9 +96,13 @@ void pio_send_control(struct hfi1_devdata *dd, int op) reg |= SEND_CTRL_SEND_ENABLE_SMASK; /* Fall through */ case PSC_DATA_VL_ENABLE: + mask = 0; + for (i = 0; i < ARRAY_SIZE(dd->vld); i++) + if (!dd->vld[i].mtu) + mask |= BIT_ULL(i); /* Disallow sending on VLs not enabled */ - mask = (((~0ull) << num_vls) & SEND_CTRL_UNSUPPORTED_VL_MASK) << - SEND_CTRL_UNSUPPORTED_VL_SHIFT; + mask = (mask & SEND_CTRL_UNSUPPORTED_VL_MASK) << + SEND_CTRL_UNSUPPORTED_VL_SHIFT; reg = (reg & ~SEND_CTRL_UNSUPPORTED_VL_SMASK) | mask; break; case PSC_GLOBAL_DISABLE:
From: Michael J. Ruhl michael.j.ruhl@intel.com
rvt_destroy_qp() cannot complete until all in process packets have been released from the underlying hardware. If a link down event occurs, an application can hang with a kernel stack similar to:
cat /proc/<app PID>/stack quiesce_qp+0x178/0x250 [hfi1] rvt_reset_qp+0x23d/0x400 [rdmavt] rvt_destroy_qp+0x69/0x210 [rdmavt] ib_destroy_qp+0xba/0x1c0 [ib_core] nvme_rdma_destroy_queue_ib+0x46/0x80 [nvme_rdma] nvme_rdma_free_queue+0x3c/0xd0 [nvme_rdma] nvme_rdma_destroy_io_queues+0x88/0xd0 [nvme_rdma] nvme_rdma_error_recovery_work+0x52/0xf0 [nvme_rdma] process_one_work+0x17a/0x440 worker_thread+0x126/0x3c0 kthread+0xcf/0xe0 ret_from_fork+0x58/0x90 0xffffffffffffffff
quiesce_qp() waits until all outstanding packets have been freed. This wait should be momentary. During a link down event, the cleanup handling does not ensure that all packets caught by the link down are flushed properly.
This is caused by the fact that the freeze path and the link down event is handled the same. This is not correct. The freeze path waits until the HFI is unfrozen and then restarts PIO. A link down is not a freeze event. The link down path cannot restart the PIO until link is restored. If the PIO path is restarted before the link comes up, the application (QP) using the PIO path will hang (until link is restored).
Fix by separating the linkdown path from the freeze path and use the link down path for link down events.
Close a race condition sc_disable() by acquiring both the progress and release locks.
Close a race condition in sc_stop() by moving the setting of the flag bits under the alloc lock.
Cc: stable@vger.kernel.org # 4.9.x+ Fixes: 7724105686e7 ("IB/hfi1: add driver files") Reviewed-by: Mike Marciniszyn mike.marciniszyn@intel.com Signed-off-by: Michael J. Ruhl michael.j.ruhl@intel.com Signed-off-by: Dennis Dalessandro dennis.dalessandro@intel.com --- drivers/infiniband/hw/hfi1/chip.c | 6 ++++- drivers/infiniband/hw/hfi1/pio.c | 42 ++++++++++++++++++++++++++++++------- drivers/infiniband/hw/hfi1/pio.h | 2 ++ 3 files changed, 41 insertions(+), 9 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c index 2c19bf7..e1668bc 100644 --- a/drivers/infiniband/hw/hfi1/chip.c +++ b/drivers/infiniband/hw/hfi1/chip.c @@ -6733,6 +6733,7 @@ void start_freeze_handling(struct hfi1_pportdata *ppd, int flags) struct hfi1_devdata *dd = ppd->dd; struct send_context *sc; int i; + int sc_flags;
if (flags & FREEZE_SELF) write_csr(dd, CCE_CTRL, CCE_CTRL_SPC_FREEZE_SMASK); @@ -6743,11 +6744,13 @@ void start_freeze_handling(struct hfi1_pportdata *ppd, int flags) /* notify all SDMA engines that they are going into a freeze */ sdma_freeze_notify(dd, !!(flags & FREEZE_LINK_DOWN));
+ sc_flags = SCF_FROZEN | SCF_HALTED | (flags & FREEZE_LINK_DOWN ? + SCF_LINK_DOWN : 0); /* do halt pre-handling on all enabled send contexts */ for (i = 0; i < dd->num_send_contexts; i++) { sc = dd->send_contexts[i].sc; if (sc && (sc->flags & SCF_ENABLED)) - sc_stop(sc, SCF_FROZEN | SCF_HALTED); + sc_stop(sc, sc_flags); }
/* Send context are frozen. Notify user space */ @@ -10674,6 +10677,7 @@ int set_link_state(struct hfi1_pportdata *ppd, u32 state) add_rcvctrl(dd, RCV_CTRL_RCV_PORT_ENABLE_SMASK);
handle_linkup_change(dd, 1); + pio_kernel_linkup(dd);
/* * After link up, a new link width will have been set. diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c index cd962c9..7520576 100644 --- a/drivers/infiniband/hw/hfi1/pio.c +++ b/drivers/infiniband/hw/hfi1/pio.c @@ -926,20 +926,18 @@ void sc_free(struct send_context *sc) void sc_disable(struct send_context *sc) { u64 reg; - unsigned long flags; struct pio_buf *pbuf;
if (!sc) return;
/* do all steps, even if already disabled */ - spin_lock_irqsave(&sc->alloc_lock, flags); + spin_lock_irq(&sc->alloc_lock); reg = read_kctxt_csr(sc->dd, sc->hw_context, SC(CTRL)); reg &= ~SC(CTRL_CTXT_ENABLE_SMASK); sc->flags &= ~SCF_ENABLED; sc_wait_for_packet_egress(sc, 1); write_kctxt_csr(sc->dd, sc->hw_context, SC(CTRL), reg); - spin_unlock_irqrestore(&sc->alloc_lock, flags);
/* * Flush any waiters. Once the context is disabled, @@ -949,7 +947,7 @@ void sc_disable(struct send_context *sc) * proceed with the flush. */ udelay(1); - spin_lock_irqsave(&sc->release_lock, flags); + spin_lock(&sc->release_lock); if (sc->sr) { /* this context has a shadow ring */ while (sc->sr_tail != sc->sr_head) { pbuf = &sc->sr[sc->sr_tail].pbuf; @@ -960,7 +958,8 @@ void sc_disable(struct send_context *sc) sc->sr_tail = 0; } } - spin_unlock_irqrestore(&sc->release_lock, flags); + spin_unlock(&sc->release_lock); + spin_unlock_irq(&sc->alloc_lock); }
/* return SendEgressCtxtStatus.PacketOccupancy */ @@ -1183,11 +1182,39 @@ void pio_kernel_unfreeze(struct hfi1_devdata *dd) sc = dd->send_contexts[i].sc; if (!sc || !(sc->flags & SCF_FROZEN) || sc->type == SC_USER) continue; + if (sc->flags & SCF_LINK_DOWN) + continue;
sc_enable(sc); /* will clear the sc frozen flag */ } }
+/** + * pio_kernel_linkup() - Re-enable send contexts after linkup event + * @dd: valid devive data + * + * When the link goes down, the freeze path is taken. However, a link down + * event is different from a freeze because if the send context is re-enabled + * whowever is sending data will start sending data again, which will hang + * any QP that is sending data. + * + * The freeze path now looks at the type of event that occurs and takes this + * path for link down event. + */ +void pio_kernel_linkup(struct hfi1_devdata *dd) +{ + struct send_context *sc; + int i; + + for (i = 0; i < dd->num_send_contexts; i++) { + sc = dd->send_contexts[i].sc; + if (!sc || !(sc->flags & SCF_LINK_DOWN) || sc->type == SC_USER) + continue; + + sc_enable(sc); /* will clear the sc link down flag */ + } +} + /* * Wait for the SendPioInitCtxt.PioInitInProgress bit to clear. * Returns: @@ -1387,11 +1414,10 @@ void sc_stop(struct send_context *sc, int flag) { unsigned long flags;
- /* mark the context */ - sc->flags |= flag; - /* stop buffer allocations */ spin_lock_irqsave(&sc->alloc_lock, flags); + /* mark the context */ + sc->flags |= flag; sc->flags &= ~SCF_ENABLED; spin_unlock_irqrestore(&sc->alloc_lock, flags); wake_up(&sc->halt_wait); diff --git a/drivers/infiniband/hw/hfi1/pio.h b/drivers/infiniband/hw/hfi1/pio.h index 058b08f..aaf372c 100644 --- a/drivers/infiniband/hw/hfi1/pio.h +++ b/drivers/infiniband/hw/hfi1/pio.h @@ -139,6 +139,7 @@ struct send_context { #define SCF_IN_FREE 0x02 #define SCF_HALTED 0x04 #define SCF_FROZEN 0x08 +#define SCF_LINK_DOWN 0x10
struct send_context_info { struct send_context *sc; /* allocated working context */ @@ -306,6 +307,7 @@ struct pio_buf *sc_buffer_alloc(struct send_context *sc, u32 dw_len, void pio_reset_all(struct hfi1_devdata *dd); void pio_freeze(struct hfi1_devdata *dd); void pio_kernel_unfreeze(struct hfi1_devdata *dd); +void pio_kernel_linkup(struct hfi1_devdata *dd);
/* global PIO send control operations */ #define PSC_GLOBAL_ENABLE 0
On Thu, Sep 20, 2018 at 12:58:34PM -0700, Dennis Dalessandro wrote:
Hi Doug and Jason,
Mike and Ira have identified some fixes that they would like to see land in the RC, these are serious problems and are marked as stable.
Changes since v1: Update Lukasz' email address and fix subject tag.
Ira Weiny (1): IB/hfi1: Fix SL array bounds check
Michael J. Ruhl (3): IB/hfi1: Invalid user input can result in crash IB/hfi1: Fix context recovery when PBC has an UnsupportedVL IB/hfi1: Fix destroy_qp hang after a link down
Applied to for-rc, thanks
Jason
linux-stable-mirror@lists.linaro.org
-
Dennis Dalessandro -
Jason Gunthorpe