From: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Date: Sat, 15 Jan 2022 18:02:33 -0500
Subject: IB/hfi1: Fix panic with larger ipoib send_queue_size
Patch-mainline: v5.17-rc3
Git-commit: 8c83d39cc730378bbac64d67a551897b203a606e
References: jsc#SLE-19242
When the ipoib send_queue_size is increased from the default the following
panic happens:
RIP: 0010:hfi1_ipoib_drain_tx_ring+0x45/0xf0 [hfi1]
Code: 31 e4 eb 0f 8b 85 c8 02 00 00 41 83 c4 01 44 39 e0 76 60 8b 8d cc 02 00 00 44 89 e3 be 01 00 00 00 d3 e3 48 03 9d c0 02 00 00 <c7> 83 18 01 00 00 00 00 00 00 48 8b bb 30 01 00 00 e8 25 af a7 e0
RSP: 0018:ffffc9000798f4a0 EFLAGS: 00010286
RAX: 0000000000008000 RBX: ffffc9000aa0f000 RCX: 000000000000000f
RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000000
RBP: ffff88810ff08000 R08: ffff88889476d900 R09: 0000000000000101
R10: 0000000000000000 R11: ffffc90006590ff8 R12: 0000000000000200
R13: ffffc9000798fba8 R14: 0000000000000000 R15: 0000000000000001
FS: 00007fd0f79cc3c0(0000) GS:ffff88885fb00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffc9000aa0f118 CR3: 0000000889c84001 CR4: 00000000001706e0
Call Trace:
<TASK>
hfi1_ipoib_napi_tx_disable+0x45/0x60 [hfi1]
hfi1_ipoib_dev_stop+0x18/0x80 [hfi1]
ipoib_ib_dev_stop+0x1d/0x40 [ib_ipoib]
ipoib_stop+0x48/0xc0 [ib_ipoib]
__dev_close_many+0x9e/0x110
__dev_change_flags+0xd9/0x210
dev_change_flags+0x21/0x60
do_setlink+0x31c/0x10f0
? __nla_validate_parse+0x12d/0x1a0
? __nla_parse+0x21/0x30
? inet6_validate_link_af+0x5e/0xf0
? cpumask_next+0x1f/0x20
? __snmp6_fill_stats64.isra.53+0xbb/0x140
? __nla_validate_parse+0x47/0x1a0
__rtnl_newlink+0x530/0x910
? pskb_expand_head+0x73/0x300
? __kmalloc_node_track_caller+0x109/0x280
? __nla_put+0xc/0x20
? cpumask_next_and+0x20/0x30
? update_sd_lb_stats.constprop.144+0xd3/0x820
? _raw_spin_unlock_irqrestore+0x25/0x37
? __wake_up_common_lock+0x87/0xc0
? kmem_cache_alloc_trace+0x3d/0x3d0
rtnl_newlink+0x43/0x60
The issue happens when the shift that should have been a function of the
txq item size mistakenly used the ring size.
Fix by using the item size.
Cc: stable@vger.kernel.org
Fixes: d47dfc2b00e6 ("IB/hfi1: Remove cache and embed txreq in ring")
Link: https://lore.kernel.org/r/1642287756-182313-2-git-send-email-mike.marciniszyn@cornelisnetworks.com
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
drivers/infiniband/hw/hfi1/ipoib_tx.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
--- a/drivers/infiniband/hw/hfi1/ipoib_tx.c
+++ b/drivers/infiniband/hw/hfi1/ipoib_tx.c
@@ -731,7 +731,7 @@ int hfi1_ipoib_txreq_init(struct hfi1_ip
goto free_txqs;
txq->tx_ring.max_items = tx_ring_size;
- txq->tx_ring.shift = ilog2(tx_ring_size);
+ txq->tx_ring.shift = ilog2(tx_item_size);
txq->tx_ring.avail = hfi1_ipoib_ring_hwat(txq);
netif_tx_napi_add(dev, &txq->napi,