Skip to contentSkip to conversation

RDMA/rxe: Bump up default maximum values used via uverbs

PATCH In review Rao Shoaib 6 participants 23 messages Jul 18, 2021 – Sep 14, 2021

Subsystems (2)
lore
Share & tools
lore

Rao Shoaib

author
#0 lore raw

v3 cover RDMA/rxe: Bump up default maximum values used via uverbs

Changes since 1st rev:
Fixed an issue reported by kernel robot build
Fixed index not being calculated properly by [email protected]

Rao Shoaib (1):
RDMA/rxe: Bump up default maximum values used via uverbs

 drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
 1 file changed, 16 insertions(+), 14 deletions(-)
Signature · 1 line
2.27.0
author
#1 lore raw

PATCH v3 1/1 RDMA/rxe: Bump up default maximum values used via uverbs

From: Rao Shoaib <[email protected]>

In our internal testing we have found that the current maximum values are too small. Ideally there should be no limits but currently, maximum values are reported via ibv_query_device, so we have to keep the maximum values but the default has been set to a large value.

Resubmitting after fixing an issue reported by test robot.

---
 drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
 1 file changed, 16 insertions(+), 14 deletions(-)
drivers/infiniband/sw/rxe/rxe_param.h +16 -14
index 742e6ec93686..d703a54df4ec 100644
@@ -9,6 +9,8 @@
 
 #include <uapi/rdma/rdma_user_rxe.h>
 
+#define DEFAULT_MAX_VALUE (1 << 20)
+
 static inline enum ib_mtu rxe_mtu_int_to_enum(int mtu)
 {
 	if (mtu < 256)
@@ -37,7 +39,7 @@ static inline enum ib_mtu eth_mtu_int_to_enum(int mtu)
 enum rxe_device_param {
 	RXE_MAX_MR_SIZE			= -1ull,
 	RXE_PAGE_SIZE_CAP		= 0xfffff000,
-	RXE_MAX_QP_WR			= 0x4000,
+	RXE_MAX_QP_WR			= DEFAULT_MAX_VALUE,
 	RXE_DEVICE_CAP_FLAGS		= IB_DEVICE_BAD_PKEY_CNTR
 					| IB_DEVICE_BAD_QKEY_CNTR
 					| IB_DEVICE_AUTO_PATH_MIG
@@ -58,42 +60,42 @@ enum rxe_device_param {
 	RXE_MAX_INLINE_DATA		= RXE_MAX_WQE_SIZE -
 					  sizeof(struct rxe_send_wqe),
 	RXE_MAX_SGE_RD			= 32,
-	RXE_MAX_CQ			= 16384,
+	RXE_MAX_CQ			= DEFAULT_MAX_VALUE,
 	RXE_MAX_LOG_CQE			= 15,
-	RXE_MAX_PD			= 0x7ffc,
+	RXE_MAX_PD			= DEFAULT_MAX_VALUE,
 	RXE_MAX_QP_RD_ATOM		= 128,
 	RXE_MAX_RES_RD_ATOM		= 0x3f000,
 	RXE_MAX_QP_INIT_RD_ATOM		= 128,
 	RXE_MAX_MCAST_GRP		= 8192,
 	RXE_MAX_MCAST_QP_ATTACH		= 56,
 	RXE_MAX_TOT_MCAST_QP_ATTACH	= 0x70000,
-	RXE_MAX_AH			= 100,
-	RXE_MAX_SRQ_WR			= 0x4000,
+	RXE_MAX_AH			= DEFAULT_MAX_VALUE,
+	RXE_MAX_SRQ_WR			= DEFAULT_MAX_VALUE,
 	RXE_MIN_SRQ_WR			= 1,
 	RXE_MAX_SRQ_SGE			= 27,
 	RXE_MIN_SRQ_SGE			= 1,
 	RXE_MAX_FMR_PAGE_LIST_LEN	= 512,
-	RXE_MAX_PKEYS			= 1,
+	RXE_MAX_PKEYS			= 64,
 	RXE_LOCAL_CA_ACK_DELAY		= 15,
 
-	RXE_MAX_UCONTEXT		= 512,
+	RXE_MAX_UCONTEXT		= DEFAULT_MAX_VALUE,
 
 	RXE_NUM_PORT			= 1,
 
-	RXE_MAX_QP			= 0x10000,
 	RXE_MIN_QP_INDEX		= 16,
-	RXE_MAX_QP_INDEX		= 0x00020000,
+	RXE_MAX_QP_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_QP			= DEFAULT_MAX_VALUE - RXE_MIN_QP_INDEX,
 
-	RXE_MAX_SRQ			= 0x00001000,
 	RXE_MIN_SRQ_INDEX		= 0x00020001,
-	RXE_MAX_SRQ_INDEX		= 0x00040000,
+	RXE_MAX_SRQ_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_SRQ			= DEFAULT_MAX_VALUE - RXE_MIN_SRQ_INDEX,
 
-	RXE_MAX_MR			= 0x00001000,
-	RXE_MAX_MW			= 0x00001000,
 	RXE_MIN_MR_INDEX		= 0x00000001,
-	RXE_MAX_MR_INDEX		= 0x00010000,
+	RXE_MAX_MR_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_MR			= DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,
 	RXE_MIN_MW_INDEX		= 0x00010001,
 	RXE_MAX_MW_INDEX		= 0x00020000,
+	RXE_MAX_MW			= 0x00001000,
 
 	RXE_MAX_PKT_PER_ACK		= 64,
 
Signature · 1 line
2.27.0
author
#2 lore raw
In reply to Rao Shoaib

Hi Jason et al,

Can I please get an up or down comment on my patch?

Shoaib

Quoted 10 lines · Rao Shoaib

Changes since 1st rev:
Fixed an issue reported by kernel robot build
Fixed index not being calculated properly by [email protected]

Rao Shoaib (1):
RDMA/rxe: Bump up default maximum values used via uverbs

  drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
  1 file changed, 16 insertions(+), 14 deletions(-)
#3 lore raw
Quoted 3 lines · Shoaib Rao

Hi Jason et al,

Can I please get an up or down comment on my patch?

Bob and Zhu should check it

Jason

maintainer
#4 lore raw
Quoted 7 lines · Jason Gunthorpe
Quoted 3 lines · Shoaib Rao

Hi Jason et al,

Can I please get an up or down comment on my patch?

Bob and Zhu should check it

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

Zhu Yanjun

Quoted 2 lines

Jason

author
#5 lore raw
Quoted 9 lines · Zhu Yanjun
Quoted 5 lines · Jason Gunthorpe
Quoted 3 lines · Shoaib Rao

Hi Jason et al,

Can I please get an up or down comment on my patch?

Bob and Zhu should check it

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

Can you paste the stack?

Quoted 5 lines

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

That is strange because all the large values do is initialize the pool with large values. Nothing else. So unless large values are used there should be no issues. Is it possible that the issue is with 5.14-rc3. Do things work between 5.12-stable systems. Anyways, please post the stack trace and also information on the setup and rping commands used.

Shoaib

Quoted 3 lines

Zhu Yanjun

Quoted 1 line

Jason

maintainer
#6 lore raw
Quoted 5 lines · Shoaib Rao
Quoted 2 lines · Zhu Yanjun
Quoted 1 line · Jason Gunthorpe

On Tue, Jul 27, 2021 at 09:15:45AM -0700, Shoaib Rao wrote:

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines
Quoted 3 lines

Hi Jason et al,

Can I please get an up or down comment on my patch?

Bob and Zhu should check it

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

Can you paste the stack?

[  381.068203] rdma_rxe: qp#17 moved to error state
[  421.464485] BUG: unable to handle page fault for address: ffff9e5de298d180
[  421.464515] #PF: supervisor write access in kernel mode
[  421.464532] #PF: error_code(0x0002) - not-present page
[  421.464549] PGD 100c00067 P4D 100c00067 PUD 100dc1067 PMD 125e78067 PTE 0
[  421.464572] Oops: 0002 [#1] SMP PTI
[  421.464585] CPU: 25 PID: 0 Comm: swapper/25 Kdump: loaded Tainted:
G S      W  OE     5.13.1-rxe+ #17
[  421.464613] Hardware name: Intel Corporation S2600WFT/S2600WFT,
BIOS SE5C620.86B.02.01.0008.031920191559 03/19/2019
[  421.464642] RIP: 0010:rxe_cq_post+0x98/0x210 [rdma_rxe]
[  421.464667] Code: 8b b3 48 01 00 00 4d 8b 48 08 41 8b 48 28 49 8d
b9 80 01 00 00 85 f6 0f 84 78 01 00 00 41 8b 50 34 d3 e2 48 01 fa 48
8b 4d 00 <48> 89 0a 48 8b 4d 08 48 89 4a 08 48 8b 4d 10 48 89 4a 10 48
8b 4d
[  421.464718] RSP: 0018:ffff9e5dc6ce0918 EFLAGS: 00010082
[  421.464735] RAX: 0000000000000246 RBX: ffff8b200cabd800 RCX: 0000000000000000
[  421.464756] RDX: ffff9e5de298d180 RSI: 0000000000000001 RDI: ffff9e5dc698b180
[  421.464777] RBP: ffff9e5dc6ce09c0 R08: ffff8b2014d85a80 R09: ffff9e5dc698b000
[  421.464797] R10: ffffffff8bc90940 R11: 0000000000000001 R12: 0000000000000000
[  421.464817] R13: ffff8b200cabd940 R14: ffff8b206e014008 R15: 000000000000001a
[  421.464838] FS:  0000000000000000(0000) GS:ffff8b1fd1040000(0000)
knlGS:0000000000000000
[  421.464861] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  421.464879] CR2: ffff9e5de298d180 CR3: 0000000c4df4e006 CR4: 00000000007706e0
[  421.464899] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  421.464920] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  421.464941] PKRU: 55555554
[  421.464950] Call Trace:
[  421.464961]  <IRQ>
[  421.464971]  rxe_responder+0x621/0x2480 [rdma_rxe]
[  421.464993]  ? __fib_validate_source+0x2e9/0x450
[  421.465013]  rxe_do_task+0x89/0x100 [rdma_rxe]
[  421.465033]  rxe_rcv+0x2eb/0x900 [rdma_rxe]
[  421.465050]  ? __udp4_lib_lookup+0x2c8/0x440
[  421.465065]  rxe_udp_encap_recv+0x68/0xc0 [rdma_rxe]
[  421.465085]  ? rxe_enable_task+0x10/0x10 [rdma_rxe]
[  421.465104]  udp_queue_rcv_one_skb+0x1df/0x4e0
[  421.465120]  udp_unicast_rcv_skb.isra.67+0x74/0x90
[  421.465135]  __udp4_lib_rcv+0x555/0xb90
[  421.465150]  ? nf_ct_deliver_cached_events+0xc1/0x120 [nf_conntrack]
[  421.465181]  ip_protocol_deliver_rcu+0xe8/0x1b0
[  421.465199]  ip_local_deliver_finish+0x44/0x50
[  421.465215]  ip_local_deliver+0xf1/0x100
[  421.465229]  ? coalesce_fill_reply+0x2c1/0x480
[  421.465249]  ? ip_protocol_deliver_rcu+0x1b0/0x1b0
[  421.465265]  ip_sublist_rcv_finish+0x75/0x80
[  421.465281]  ip_sublist_rcv+0x196/0x220
[  421.465296]  ? ip_local_deliver+0x100/0x100
[  421.465312]  ip_list_rcv+0x137/0x160
[  421.465325]  __netif_receive_skb_list_core+0x29b/0x2c0
[  421.465344]  netif_receive_skb_list_internal+0x1c3/0x2f0
[  421.465361]  gro_normal_list.part.158+0x19/0x40
[  421.465376]  napi_complete_done+0x67/0x160
[  421.465391]  i40e_napi_poll+0x53b/0x840 [i40e]
[  421.465426]  __napi_poll+0x2b/0x120
[  421.466123]  net_rx_action+0x236/0x300
[  421.466783]  __do_softirq+0xc9/0x285
[  421.467440]  irq_exit_rcu+0xba/0xd0
[  421.468091]  common_interrupt+0x7f/0xa0
[  421.468737]  </IRQ>
[  421.469366]  asm_common_interrupt+0x1e/0x40
[  421.469990] RIP: 0010:cpuidle_enter_state+0xd6/0x350
[  421.470608] Code: 49 89 c4 0f 1f 44 00 00 31 ff e8 45 49 99 ff 45
84 ff 74 12 9c 58 f6 c4 02 0f 85 32 02 00 00 31 ff e8 ae c8 9f ff fb
45 85 f6 <0f> 88 e0 00 00 00 49 63 d6 4c 2b 24 24 48 8d 04 52 48 8d 04
82 49
[  421.471935] RSP: 0018:ffff9e5dc679fe80 EFLAGS: 00000202
[  421.472599] RAX: ffff8b1fd106bc40 RBX: 0000000000000002 RCX: 000000000000001f
[  421.473266] RDX: 00000062213d764d RSI: 000000003351fed6 RDI: 0000000000000000
[  421.473920] RBP: ffffbe51c1040000 R08: 0000000000000002 R09: 000000000002b480
[  421.474558] R10: 0000a82bea904be8 R11: ffff8b1fd106a984 R12: 00000062213d764d
[  421.475172] R13: ffffffff8c6c6d80 R14: 0000000000000002 R15: 0000000000000000
[  421.475763]  cpuidle_enter+0x29/0x40
[  421.476348]  do_idle+0x257/0x2a0
[  421.476926]  cpu_startup_entry+0x19/0x20
[  421.477497]  start_secondary+0x116/0x150
[  421.478067]  secondary_startup_64_no_verify+0xc2/0xcb
[  421.478640] Modules linked in: rdma_rxe(OE) ip6_udp_tunnel

udp_tunnel xt_CHECKSUM xt_MASQUERADE xt_conntrack ipt_REJECT nf_reject_ipv4 nft_compat nft_counter nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables nfnetlink tun bridge stp llc nls_utf8 isofs cdrom loop rfkill ib_isert iscsi_target_mod ib_srpt ext4 target_core_mod ib_srp scsi_transport_srp mbcache jbd2 rpcrdma sunrpc intel_rapl_msr intel_rapl_common rdma_ucm isst_if_common ib_iser ib_umad rdma_cm ib_ipoib iw_cm skx_edac libiscsi ib_cm nfit libnvdimm scsi_transport_iscsi x86_pkg_temp_thermal intel_powerclamp mlx5_ib coretemp crct10dif_pclmul crc32_pclmul iTCO_wdt iTCO_vendor_support ib_uverbs ghash_clmulni_intel rapl ipmi_ssif intel_cstate ib_core mei_me acpi_ipmi i2c_i801 joydev intel_uncore pcspkr mei i2c_smbus lpc_ich ioatdma ipmi_si intel_pch_thermal dca ipmi_devintf ipmi_msghandler acpi_pad acpi_power_meter ip_tables xfs libcrc32c sd_mod t10_pi sg mlx5_core ast i2c_algo_bit drm_vram_helper

[  421.478702]  drm_kms_helper syscopyarea sysfillrect sysimgblt
fb_sys_fops drm_ttm_helper ttm mlxfw ahci libahci pci_hyperv_intf ice
drm i40e tls crc32c_intel libata psample wmi dm_mirror dm_region_hash
dm_log dm_mod fuse [last unloaded: ip6_udp_tunnel]
[  421.483665] CR2: ffff9e5de298d180
Quoted 17 lines
Quoted 5 lines

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

That is strange because all the large values do is initialize the pool with large values. Nothing else. So unless large values are used there should be no issues. Is it possible that the issue is with 5.14-rc3. Do things work between 5.12-stable systems. Anyways, please post the stack trace and also information on the setup and rping commands used.

Shoaib

Quoted 3 lines

Zhu Yanjun

Quoted 1 line

Jason

author
#7 lore raw

I switched the values to the old values and compiled rdma_rxe module. I could not get rping to work. First I get CRC errors and then one node panics. Both nodes are running 5.14.0-rc1. So the issue you are seeing is not caused by my changes, rxe is already broken in 5.14.0-rc1.

Jason,

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

Regards,

Shoaib

[ 2105.071603] rdma_rxe: bad ICRC from 10.129.135.22
[ 2106.979538] rdma_rxe: bad ICRC from 10.129.135.22
[ 2109.155417] rdma_rxe: bad ICRC from 10.129.135.22
[ 2111.331292] rdma_rxe: bad ICRC from 10.129.135.22
[ 2113.507169] rdma_rxe: bad ICRC from 10.129.135.22
[ 2115.683046] rdma_rxe: bad ICRC from 10.129.135.22
[ 2117.858927] rdma_rxe: bad ICRC from 10.129.135.22
[ 2120.034798] rdma_rxe: bad ICRC from 10.129.135.22
[ 2122.210691] BUG: unable to handle page fault for address: ffffbd8562275180
[ 2122.292744] #PF: supervisor write access in kernel mode
[ 2122.355063] #PF: error_code(0x0002) - not-present page
[ 2122.416342] PGD 100000067 P4D 100000067 PUD 1001c7067 PMD 142a84067 PTE 0
[ 2122.497361] Oops: 0002 [#1] SMP PTI
[ 2122.538913] CPU: 8 PID: 0 Comm: swapper/8 Not tainted 5.14.0-rc1_rxe_values+ #4
[ 2122.626155] Hardware name: Oracle Corporation SUN FIRE X4170 M2 SERVER        /ASSY,MOTHERBOARD,X4170, BIOS 08140115 07/04/2018
[ 2122.763248] RIP: 0010:rxe_cq_post+0x9e/0x220 [rdma_rxe]
[ 2122.825578] Code: 44 8b 8b 48 01 00 00 4c 8b 47 08 8b 4f 28 49 8d b0 80 01 00 00 45 85 c9 0f 84 7d 01 00 00 8b 57 34 d3 e2 48 01 f2 49 8b 0c 24 <48> 89 0a 49 8b 4c 24 08 48 89 4a 08 49 8b 4c 24 10 48 89 4a 10 49
[ 2123.049907] RSP: 0018:ffffbd85464f0800 EFLAGS: 00010082
[ 2123.112225] RAX: 0000000000000246 RBX: ffff9dad4fd3f800 RCX: 0000000000000000
[ 2123.197389] RDX: ffffbd8562275180 RSI: ffffbd8546273180 RDI: ffff9da2c3cee840
[ 2123.282555] RBP: ffffbd85464f0840 R08: ffffbd8546273000 R09: 0000000000000001
[ 2123.367722] R10: 0000000000000001 R11: 0000000000000001 R12: ffffbd85464f08b8
[ 2123.452886] R13: 0000000000000000 R14: ffff9dad4fd3f940 R15: ffff9da3002ac008
[ 2123.538053] FS:  0000000000000000(0000) GS:ffff9db94fa80000(0000) knlGS:0000000000000000
[ 2123.634642] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2123.703190] CR2: ffffbd8562275180 CR3: 00000013f340a005 CR4: 00000000000206e0
[ 2123.788357] Call Trace:
[ 2123.817443]  <IRQ>
[ 2123.841335]  rxe_responder+0x5d9/0x2490 [rdma_rxe]
[ 2123.898467]  ? native_apic_mem_write+0x10/0x10
[ 2123.951445]  ? native_apic_wait_icr_idle+0x22/0x30
[ 2124.008575]  ? arch_irq_work_raise+0x3a/0x40
[ 2124.059476]  ? __irq_work_queue_local+0x48/0x60
[ 2124.113486]  ? fib_table_lookup+0x21e/0x640
[ 2124.163348]  ? wake_up_klogd.part.31+0x34/0x40
[ 2124.216319]  rxe_do_task+0x94/0x110 [rdma_rxe]
[ 2124.269297]  rxe_run_task+0x2a/0x40 [rdma_rxe]
[ 2124.322275]  rxe_resp_queue_pkt+0x44/0x50 [rdma_rxe]
[ 2124.381485]  rxe_rcv+0x2eb/0x900 [rdma_rxe]
[ 2124.431347]  rxe_udp_encap_recv+0x6d/0xd0 [rdma_rxe]
[ 2124.490555]  ? rxe_enable_task+0x10/0x10 [rdma_rxe]
[ 2124.548725]  udp_queue_rcv_one_skb+0x1f2/0x500
[ 2124.601697]  udp_queue_rcv_skb+0x50/0x210
[ 2124.649475]  udp_unicast_rcv_skb.isra.67+0x78/0x90
[ 2124.706600]  __udp4_lib_rcv+0x57c/0xbe0
[ 2124.752303]  udp_rcv+0x1a/0x20
Quoted 128 lines · Zhu Yanjun
Quoted 4 lines · Shoaib Rao
Quoted 2 lines · Zhu Yanjun
Quoted 1 line · Jason Gunthorpe

On Tue, Jul 27, 2021 at 09:15:45AM -0700, Shoaib Rao wrote:

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines
Quoted 3 lines

Hi Jason et al,

Can I please get an up or down comment on my patch?

Bob and Zhu should check it

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

Can you paste the stack?

[  381.068203] rdma_rxe: qp#17 moved to error state
[  421.464485] BUG: unable to handle page fault for address: ffff9e5de298d180
[  421.464515] #PF: supervisor write access in kernel mode
[  421.464532] #PF: error_code(0x0002) - not-present page
[  421.464549] PGD 100c00067 P4D 100c00067 PUD 100dc1067 PMD 125e78067 PTE 0
[  421.464572] Oops: 0002 [#1] SMP PTI
[  421.464585] CPU: 25 PID: 0 Comm: swapper/25 Kdump: loaded Tainted:
G S      W  OE     5.13.1-rxe+ #17
[  421.464613] Hardware name: Intel Corporation S2600WFT/S2600WFT,
BIOS SE5C620.86B.02.01.0008.031920191559 03/19/2019
[  421.464642] RIP: 0010:rxe_cq_post+0x98/0x210 [rdma_rxe]
[  421.464667] Code: 8b b3 48 01 00 00 4d 8b 48 08 41 8b 48 28 49 8d
b9 80 01 00 00 85 f6 0f 84 78 01 00 00 41 8b 50 34 d3 e2 48 01 fa 48
8b 4d 00 <48> 89 0a 48 8b 4d 08 48 89 4a 08 48 8b 4d 10 48 89 4a 10 48
8b 4d
[  421.464718] RSP: 0018:ffff9e5dc6ce0918 EFLAGS: 00010082
[  421.464735] RAX: 0000000000000246 RBX: ffff8b200cabd800 RCX: 0000000000000000
[  421.464756] RDX: ffff9e5de298d180 RSI: 0000000000000001 RDI: ffff9e5dc698b180
[  421.464777] RBP: ffff9e5dc6ce09c0 R08: ffff8b2014d85a80 R09: ffff9e5dc698b000
[  421.464797] R10: ffffffff8bc90940 R11: 0000000000000001 R12: 0000000000000000
[  421.464817] R13: ffff8b200cabd940 R14: ffff8b206e014008 R15: 000000000000001a
[  421.464838] FS:  0000000000000000(0000) GS:ffff8b1fd1040000(0000)
knlGS:0000000000000000
[  421.464861] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  421.464879] CR2: ffff9e5de298d180 CR3: 0000000c4df4e006 CR4: 00000000007706e0
[  421.464899] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  421.464920] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  421.464941] PKRU: 55555554
[  421.464950] Call Trace:
[  421.464961]  <IRQ>
[  421.464971]  rxe_responder+0x621/0x2480 [rdma_rxe]
[  421.464993]  ? __fib_validate_source+0x2e9/0x450
[  421.465013]  rxe_do_task+0x89/0x100 [rdma_rxe]
[  421.465033]  rxe_rcv+0x2eb/0x900 [rdma_rxe]
[  421.465050]  ? __udp4_lib_lookup+0x2c8/0x440
[  421.465065]  rxe_udp_encap_recv+0x68/0xc0 [rdma_rxe]
[  421.465085]  ? rxe_enable_task+0x10/0x10 [rdma_rxe]
[  421.465104]  udp_queue_rcv_one_skb+0x1df/0x4e0
[  421.465120]  udp_unicast_rcv_skb.isra.67+0x74/0x90
[  421.465135]  __udp4_lib_rcv+0x555/0xb90
[  421.465150]  ? nf_ct_deliver_cached_events+0xc1/0x120 [nf_conntrack]
[  421.465181]  ip_protocol_deliver_rcu+0xe8/0x1b0
[  421.465199]  ip_local_deliver_finish+0x44/0x50
[  421.465215]  ip_local_deliver+0xf1/0x100
[  421.465229]  ? coalesce_fill_reply+0x2c1/0x480
[  421.465249]  ? ip_protocol_deliver_rcu+0x1b0/0x1b0
[  421.465265]  ip_sublist_rcv_finish+0x75/0x80
[  421.465281]  ip_sublist_rcv+0x196/0x220
[  421.465296]  ? ip_local_deliver+0x100/0x100
[  421.465312]  ip_list_rcv+0x137/0x160
[  421.465325]  __netif_receive_skb_list_core+0x29b/0x2c0
[  421.465344]  netif_receive_skb_list_internal+0x1c3/0x2f0
[  421.465361]  gro_normal_list.part.158+0x19/0x40
[  421.465376]  napi_complete_done+0x67/0x160
[  421.465391]  i40e_napi_poll+0x53b/0x840 [i40e]
[  421.465426]  __napi_poll+0x2b/0x120
[  421.466123]  net_rx_action+0x236/0x300
[  421.466783]  __do_softirq+0xc9/0x285
[  421.467440]  irq_exit_rcu+0xba/0xd0
[  421.468091]  common_interrupt+0x7f/0xa0
[  421.468737]  </IRQ>
[  421.469366]  asm_common_interrupt+0x1e/0x40
[  421.469990] RIP: 0010:cpuidle_enter_state+0xd6/0x350
[  421.470608] Code: 49 89 c4 0f 1f 44 00 00 31 ff e8 45 49 99 ff 45
84 ff 74 12 9c 58 f6 c4 02 0f 85 32 02 00 00 31 ff e8 ae c8 9f ff fb
45 85 f6 <0f> 88 e0 00 00 00 49 63 d6 4c 2b 24 24 48 8d 04 52 48 8d 04
82 49
[  421.471935] RSP: 0018:ffff9e5dc679fe80 EFLAGS: 00000202
[  421.472599] RAX: ffff8b1fd106bc40 RBX: 0000000000000002 RCX: 000000000000001f
[  421.473266] RDX: 00000062213d764d RSI: 000000003351fed6 RDI: 0000000000000000
[  421.473920] RBP: ffffbe51c1040000 R08: 0000000000000002 R09: 000000000002b480
[  421.474558] R10: 0000a82bea904be8 R11: ffff8b1fd106a984 R12: 00000062213d764d
[  421.475172] R13: ffffffff8c6c6d80 R14: 0000000000000002 R15: 0000000000000000
[  421.475763]  cpuidle_enter+0x29/0x40
[  421.476348]  do_idle+0x257/0x2a0
[  421.476926]  cpu_startup_entry+0x19/0x20
[  421.477497]  start_secondary+0x116/0x150
[  421.478067]  secondary_startup_64_no_verify+0xc2/0xcb
[  421.478640] Modules linked in: rdma_rxe(OE) ip6_udp_tunnel

udp_tunnel xt_CHECKSUM xt_MASQUERADE xt_conntrack ipt_REJECT nf_reject_ipv4 nft_compat nft_counter nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables nfnetlink tun bridge stp llc nls_utf8 isofs cdrom loop rfkill ib_isert iscsi_target_mod ib_srpt ext4 target_core_mod ib_srp scsi_transport_srp mbcache jbd2 rpcrdma sunrpc intel_rapl_msr intel_rapl_common rdma_ucm isst_if_common ib_iser ib_umad rdma_cm ib_ipoib iw_cm skx_edac libiscsi ib_cm nfit libnvdimm scsi_transport_iscsi x86_pkg_temp_thermal intel_powerclamp mlx5_ib coretemp crct10dif_pclmul crc32_pclmul iTCO_wdt iTCO_vendor_support ib_uverbs ghash_clmulni_intel rapl ipmi_ssif intel_cstate ib_core mei_me acpi_ipmi i2c_i801 joydev intel_uncore pcspkr mei i2c_smbus lpc_ich ioatdma ipmi_si intel_pch_thermal dca ipmi_devintf ipmi_msghandler acpi_pad acpi_power_meter ip_tables xfs libcrc32c sd_mod t10_pi sg mlx5_core ast i2c_algo_bit drm_vram_helper

[  421.478702]  drm_kms_helper syscopyarea sysfillrect sysimgblt
fb_sys_fops drm_ttm_helper ttm mlxfw ahci libahci pci_hyperv_intf ice
drm i40e tls crc32c_intel libata psample wmi dm_mirror dm_region_hash
dm_log dm_mod fuse [last unloaded: ip6_udp_tunnel]
[  421.483665] CR2: ffff9e5de298d180
Quoted 14 lines
Quoted 4 lines

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

That is strange because all the large values do is initialize the pool with large values. Nothing else. So unless large values are used there should be no issues. Is it possible that the issue is with 5.14-rc3. Do things work between 5.12-stable systems. Anyways, please post the stack trace and also information on the setup and rping commands used.

Shoaib

Quoted 2 lines

Zhu Yanjun

Quoted 1 line

Jason

#8 lore raw
Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

author
#9 lore raw
Quoted 9 lines · Jason Gunthorpe
Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

I agree with what you are saying and I plan to help address the issue.

For the record, I just tested my changes on a 5.13.6 kernel and they work.

[root@ca-dev14 ~]# rping -s  -a 10.129.135.23
server DISCONNECT EVENT...
wait for RDMA_READ_ADV state 10
[root@ca-dev14 ~]# uname -a
Linux ca-dev14.us.oracle.com 5.13.0-rc6_rxe_defaults+ #1 SMP Thu Jul 29 12:48:41 PDT 2021 x86_64 x86_64 x86_64 GNU/Linux

[root@ca-dev13 ~]# rping -c -a 10.129.135.23 -C 10 -v
ping data: rdma-ping-0: ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqr
ping data: rdma-ping-1: BCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrs
ping data: rdma-ping-2: CDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrst
ping data: rdma-ping-3: DEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstu
ping data: rdma-ping-4: EFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuv
ping data: rdma-ping-5: FGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvw
ping data: rdma-ping-6: GHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwx
ping data: rdma-ping-7: HIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxy
ping data: rdma-ping-8: IJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz
ping data: rdma-ping-9: JKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyzA
[root@ca-dev13 ~]# uname -a
Linux ca-dev13.us.oracle.com 5.13.0-rc6_rxe_defaults+ #1 SMP Thu Jul 29 12:48:41 PDT 2021 x86_64 x86_64 x86_64 GNU/Linux

Shoaib

Quoted 2 lines

Jason

#10 lore raw
In reply to Jason Gunthorpe

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

author
#11 lore raw

Thanks Bob.

Zhu can you please apply those patches and test.

Shoaib

Quoted 19 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

author
#12 lore raw

Hi Zhu,

Any update on your testing after applying Bob's fixes

Shoaib

Quoted 26 lines · Shoaib Rao

Thanks Bob.

Zhu can you please apply those patches and test.

Shoaib

Quoted 19 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

maintainer
#13 lore raw
Quoted 4 lines · Shoaib Rao

Hi Zhu,

Any update on your testing after applying Bob's fixes

Do you read my problem carefully?
I mean that before your commit, the whole rxe can work well. After your commit, the rxe can not work well. Please reproduce this problem in your host and fix it.

Zhu Yanjun

Quoted 35 lines

Shoaib

Quoted 31 lines · Shoaib Rao

Thanks Bob.

Zhu can you please apply those patches and test.

Shoaib

Quoted 24 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 4 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

author
#14 lore raw
Quoted 10 lines · Zhu Yanjun
Quoted 3 lines · Shoaib Rao

Hi Zhu,

Any update on your testing after applying Bob's fixes

Do you read my problem carefully?
I mean that before your commit, the whole rxe can work well. After your commit, the rxe can not work well. Please reproduce this problem in your host and fix it.

Zhu Yanjun

You posted

Quoted 10 lines

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

Zhu Yanjun

I am not sure how you made rxe to work because it did not work for me and neither for Bob. Since then, Bob has posted patches for the issue. I also posted that my changes work on 5.13.6 kernel. emails attached.

Even if rxe in 5.14 is working for you some how, please apply Bob's patches and then mine and test.

Thanks,

Shoaib

Quoted 29 lines
Quoted 29 lines

Shoaib

Quoted 25 lines · Shoaib Rao

Thanks Bob.

Zhu can you please apply those patches and test.

Shoaib

Quoted 18 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 4 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

I switched the values to the old values and compiled rdma_rxe module. I could not get rping to work. First I get CRC errors and then one node panics. Both nodes are running 5.14.0-rc1. So the issue you are seeing is not caused by my changes, rxe is already broken in 5.14.0-rc1.

Jason,

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

Regards,

Shoaib

[ 2105.071603] rdma_rxe: bad ICRC from 10.129.135.22
[ 2106.979538] rdma_rxe: bad ICRC from 10.129.135.22
[ 2109.155417] rdma_rxe: bad ICRC from 10.129.135.22
[ 2111.331292] rdma_rxe: bad ICRC from 10.129.135.22
[ 2113.507169] rdma_rxe: bad ICRC from 10.129.135.22
[ 2115.683046] rdma_rxe: bad ICRC from 10.129.135.22
[ 2117.858927] rdma_rxe: bad ICRC from 10.129.135.22
[ 2120.034798] rdma_rxe: bad ICRC from 10.129.135.22
[ 2122.210691] BUG: unable to handle page fault for address: ffffbd8562275180
[ 2122.292744] #PF: supervisor write access in kernel mode
[ 2122.355063] #PF: error_code(0x0002) - not-present page
[ 2122.416342] PGD 100000067 P4D 100000067 PUD 1001c7067 PMD 142a84067 PTE 0
[ 2122.497361] Oops: 0002 [#1] SMP PTI
[ 2122.538913] CPU: 8 PID: 0 Comm: swapper/8 Not tainted 5.14.0-rc1_rxe_values+ #4
[ 2122.626155] Hardware name: Oracle Corporation SUN FIRE X4170 M2 SERVER        /ASSY,MOTHERBOARD,X4170, BIOS 08140115 07/04/2018
[ 2122.763248] RIP: 0010:rxe_cq_post+0x9e/0x220 [rdma_rxe]
[ 2122.825578] Code: 44 8b 8b 48 01 00 00 4c 8b 47 08 8b 4f 28 49 8d b0 80 01 00 00 45 85 c9 0f 84 7d 01 00 00 8b 57 34 d3 e2 48 01 f2 49 8b 0c 24 <48> 89 0a 49 8b 4c 24 08 48 89 4a 08 49 8b 4c 24 10 48 89 4a 10 49
[ 2123.049907] RSP: 0018:ffffbd85464f0800 EFLAGS: 00010082
[ 2123.112225] RAX: 0000000000000246 RBX: ffff9dad4fd3f800 RCX: 0000000000000000
[ 2123.197389] RDX: ffffbd8562275180 RSI: ffffbd8546273180 RDI: ffff9da2c3cee840
[ 2123.282555] RBP: ffffbd85464f0840 R08: ffffbd8546273000 R09: 0000000000000001
[ 2123.367722] R10: 0000000000000001 R11: 0000000000000001 R12: ffffbd85464f08b8
[ 2123.452886] R13: 0000000000000000 R14: ffff9dad4fd3f940 R15: ffff9da3002ac008
[ 2123.538053] FS:  0000000000000000(0000) GS:ffff9db94fa80000(0000) knlGS:0000000000000000
[ 2123.634642] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2123.703190] CR2: ffffbd8562275180 CR3: 00000013f340a005 CR4: 00000000000206e0
[ 2123.788357] Call Trace:
[ 2123.817443]  <IRQ>
[ 2123.841335]  rxe_responder+0x5d9/0x2490 [rdma_rxe]
[ 2123.898467]  ? native_apic_mem_write+0x10/0x10
[ 2123.951445]  ? native_apic_wait_icr_idle+0x22/0x30
[ 2124.008575]  ? arch_irq_work_raise+0x3a/0x40
[ 2124.059476]  ? __irq_work_queue_local+0x48/0x60
[ 2124.113486]  ? fib_table_lookup+0x21e/0x640
[ 2124.163348]  ? wake_up_klogd.part.31+0x34/0x40
[ 2124.216319]  rxe_do_task+0x94/0x110 [rdma_rxe]
[ 2124.269297]  rxe_run_task+0x2a/0x40 [rdma_rxe]
[ 2124.322275]  rxe_resp_queue_pkt+0x44/0x50 [rdma_rxe]
[ 2124.381485]  rxe_rcv+0x2eb/0x900 [rdma_rxe]
[ 2124.431347]  rxe_udp_encap_recv+0x6d/0xd0 [rdma_rxe]
[ 2124.490555]  ? rxe_enable_task+0x10/0x10 [rdma_rxe]
[ 2124.548725]  udp_queue_rcv_one_skb+0x1f2/0x500
[ 2124.601697]  udp_queue_rcv_skb+0x50/0x210
[ 2124.649475]  udp_unicast_rcv_skb.isra.67+0x78/0x90
[ 2124.706600]  __udp4_lib_rcv+0x57c/0xbe0
[ 2124.752303]  udp_rcv+0x1a/0x20
Quoted 128 lines · Zhu Yanjun
Quoted 4 lines · Shoaib Rao
Quoted 2 lines · Zhu Yanjun
Quoted 1 line · Jason Gunthorpe

On Tue, Jul 27, 2021 at 09:15:45AM -0700, Shoaib Rao wrote:

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines
Quoted 3 lines

Hi Jason et al,

Can I please get an up or down comment on my patch?

Bob and Zhu should check it

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

Can you paste the stack?

[  381.068203] rdma_rxe: qp#17 moved to error state
[  421.464485] BUG: unable to handle page fault for address: ffff9e5de298d180
[  421.464515] #PF: supervisor write access in kernel mode
[  421.464532] #PF: error_code(0x0002) - not-present page
[  421.464549] PGD 100c00067 P4D 100c00067 PUD 100dc1067 PMD 125e78067 PTE 0
[  421.464572] Oops: 0002 [#1] SMP PTI
[  421.464585] CPU: 25 PID: 0 Comm: swapper/25 Kdump: loaded Tainted:
G S      W  OE     5.13.1-rxe+ #17
[  421.464613] Hardware name: Intel Corporation S2600WFT/S2600WFT,
BIOS SE5C620.86B.02.01.0008.031920191559 03/19/2019
[  421.464642] RIP: 0010:rxe_cq_post+0x98/0x210 [rdma_rxe]
[  421.464667] Code: 8b b3 48 01 00 00 4d 8b 48 08 41 8b 48 28 49 8d
b9 80 01 00 00 85 f6 0f 84 78 01 00 00 41 8b 50 34 d3 e2 48 01 fa 48
8b 4d 00 <48> 89 0a 48 8b 4d 08 48 89 4a 08 48 8b 4d 10 48 89 4a 10 48
8b 4d
[  421.464718] RSP: 0018:ffff9e5dc6ce0918 EFLAGS: 00010082
[  421.464735] RAX: 0000000000000246 RBX: ffff8b200cabd800 RCX: 0000000000000000
[  421.464756] RDX: ffff9e5de298d180 RSI: 0000000000000001 RDI: ffff9e5dc698b180
[  421.464777] RBP: ffff9e5dc6ce09c0 R08: ffff8b2014d85a80 R09: ffff9e5dc698b000
[  421.464797] R10: ffffffff8bc90940 R11: 0000000000000001 R12: 0000000000000000
[  421.464817] R13: ffff8b200cabd940 R14: ffff8b206e014008 R15: 000000000000001a
[  421.464838] FS:  0000000000000000(0000) GS:ffff8b1fd1040000(0000)
knlGS:0000000000000000
[  421.464861] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  421.464879] CR2: ffff9e5de298d180 CR3: 0000000c4df4e006 CR4: 00000000007706e0
[  421.464899] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  421.464920] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  421.464941] PKRU: 55555554
[  421.464950] Call Trace:
[  421.464961]  <IRQ>
[  421.464971]  rxe_responder+0x621/0x2480 [rdma_rxe]
[  421.464993]  ? __fib_validate_source+0x2e9/0x450
[  421.465013]  rxe_do_task+0x89/0x100 [rdma_rxe]
[  421.465033]  rxe_rcv+0x2eb/0x900 [rdma_rxe]
[  421.465050]  ? __udp4_lib_lookup+0x2c8/0x440
[  421.465065]  rxe_udp_encap_recv+0x68/0xc0 [rdma_rxe]
[  421.465085]  ? rxe_enable_task+0x10/0x10 [rdma_rxe]
[  421.465104]  udp_queue_rcv_one_skb+0x1df/0x4e0
[  421.465120]  udp_unicast_rcv_skb.isra.67+0x74/0x90
[  421.465135]  __udp4_lib_rcv+0x555/0xb90
[  421.465150]  ? nf_ct_deliver_cached_events+0xc1/0x120 [nf_conntrack]
[  421.465181]  ip_protocol_deliver_rcu+0xe8/0x1b0
[  421.465199]  ip_local_deliver_finish+0x44/0x50
[  421.465215]  ip_local_deliver+0xf1/0x100
[  421.465229]  ? coalesce_fill_reply+0x2c1/0x480
[  421.465249]  ? ip_protocol_deliver_rcu+0x1b0/0x1b0
[  421.465265]  ip_sublist_rcv_finish+0x75/0x80
[  421.465281]  ip_sublist_rcv+0x196/0x220
[  421.465296]  ? ip_local_deliver+0x100/0x100
[  421.465312]  ip_list_rcv+0x137/0x160
[  421.465325]  __netif_receive_skb_list_core+0x29b/0x2c0
[  421.465344]  netif_receive_skb_list_internal+0x1c3/0x2f0
[  421.465361]  gro_normal_list.part.158+0x19/0x40
[  421.465376]  napi_complete_done+0x67/0x160
[  421.465391]  i40e_napi_poll+0x53b/0x840 [i40e]
[  421.465426]  __napi_poll+0x2b/0x120
[  421.466123]  net_rx_action+0x236/0x300
[  421.466783]  __do_softirq+0xc9/0x285
[  421.467440]  irq_exit_rcu+0xba/0xd0
[  421.468091]  common_interrupt+0x7f/0xa0
[  421.468737]  </IRQ>
[  421.469366]  asm_common_interrupt+0x1e/0x40
[  421.469990] RIP: 0010:cpuidle_enter_state+0xd6/0x350
[  421.470608] Code: 49 89 c4 0f 1f 44 00 00 31 ff e8 45 49 99 ff 45
84 ff 74 12 9c 58 f6 c4 02 0f 85 32 02 00 00 31 ff e8 ae c8 9f ff fb
45 85 f6 <0f> 88 e0 00 00 00 49 63 d6 4c 2b 24 24 48 8d 04 52 48 8d 04
82 49
[  421.471935] RSP: 0018:ffff9e5dc679fe80 EFLAGS: 00000202
[  421.472599] RAX: ffff8b1fd106bc40 RBX: 0000000000000002 RCX: 000000000000001f
[  421.473266] RDX: 00000062213d764d RSI: 000000003351fed6 RDI: 0000000000000000
[  421.473920] RBP: ffffbe51c1040000 R08: 0000000000000002 R09: 000000000002b480
[  421.474558] R10: 0000a82bea904be8 R11: ffff8b1fd106a984 R12: 00000062213d764d
[  421.475172] R13: ffffffff8c6c6d80 R14: 0000000000000002 R15: 0000000000000000
[  421.475763]  cpuidle_enter+0x29/0x40
[  421.476348]  do_idle+0x257/0x2a0
[  421.476926]  cpu_startup_entry+0x19/0x20
[  421.477497]  start_secondary+0x116/0x150
[  421.478067]  secondary_startup_64_no_verify+0xc2/0xcb
[  421.478640] Modules linked in: rdma_rxe(OE) ip6_udp_tunnel

udp_tunnel xt_CHECKSUM xt_MASQUERADE xt_conntrack ipt_REJECT nf_reject_ipv4 nft_compat nft_counter nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables nfnetlink tun bridge stp llc nls_utf8 isofs cdrom loop rfkill ib_isert iscsi_target_mod ib_srpt ext4 target_core_mod ib_srp scsi_transport_srp mbcache jbd2 rpcrdma sunrpc intel_rapl_msr intel_rapl_common rdma_ucm isst_if_common ib_iser ib_umad rdma_cm ib_ipoib iw_cm skx_edac libiscsi ib_cm nfit libnvdimm scsi_transport_iscsi x86_pkg_temp_thermal intel_powerclamp mlx5_ib coretemp crct10dif_pclmul crc32_pclmul iTCO_wdt iTCO_vendor_support ib_uverbs ghash_clmulni_intel rapl ipmi_ssif intel_cstate ib_core mei_me acpi_ipmi i2c_i801 joydev intel_uncore pcspkr mei i2c_smbus lpc_ich ioatdma ipmi_si intel_pch_thermal dca ipmi_devintf ipmi_msghandler acpi_pad acpi_power_meter ip_tables xfs libcrc32c sd_mod t10_pi sg mlx5_core ast i2c_algo_bit drm_vram_helper

[  421.478702]  drm_kms_helper syscopyarea sysfillrect sysimgblt
fb_sys_fops drm_ttm_helper ttm mlxfw ahci libahci pci_hyperv_intf ice
drm i40e tls crc32c_intel libata psample wmi dm_mirror dm_region_hash
dm_log dm_mod fuse [last unloaded: ip6_udp_tunnel]
[  421.483665] CR2: ffff9e5de298d180
Quoted 14 lines
Quoted 4 lines

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

That is strange because all the large values do is initialize the pool with large values. Nothing else. So unless large values are used there should be no issues. Is it possible that the issue is with 5.14-rc3. Do things work between 5.12-stable systems. Anyways, please post the stack trace and also information on the setup and rping commands used.

Shoaib

Quoted 2 lines

Zhu Yanjun

Quoted 1 line

Jason

Quoted 9 lines · Jason Gunthorpe
Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

I agree with what you are saying and I plan to help address the issue.

For the record, I just tested my changes on a 5.13.6 kernel and they work.

[root@ca-dev14 ~]# rping -s  -a 10.129.135.23
server DISCONNECT EVENT...
wait for RDMA_READ_ADV state 10
[root@ca-dev14 ~]# uname -a
Linux ca-dev14.us.oracle.com 5.13.0-rc6_rxe_defaults+ #1 SMP Thu Jul 29 12:48:41 PDT 2021 x86_64 x86_64 x86_64 GNU/Linux

[root@ca-dev13 ~]# rping -c -a 10.129.135.23 -C 10 -v
ping data: rdma-ping-0: ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqr
ping data: rdma-ping-1: BCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrs
ping data: rdma-ping-2: CDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrst
ping data: rdma-ping-3: DEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstu
ping data: rdma-ping-4: EFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuv
ping data: rdma-ping-5: FGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvw
ping data: rdma-ping-6: GHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwx
ping data: rdma-ping-7: HIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxy
ping data: rdma-ping-8: IJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz
ping data: rdma-ping-9: JKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyzA
[root@ca-dev13 ~]# uname -a
Linux ca-dev13.us.oracle.com 5.13.0-rc6_rxe_defaults+ #1 SMP Thu Jul 29 12:48:41 PDT 2021 x86_64 x86_64 x86_64 GNU/Linux

Shoaib

Quoted 2 lines

Jason

maintainer
#15 lore raw
Quoted 25 lines · Shoaib Rao
Quoted 10 lines · Zhu Yanjun
Quoted 3 lines · Shoaib Rao

Hi Zhu,

Any update on your testing after applying Bob's fixes

Do you read my problem carefully?
I mean that before your commit, the whole rxe can work well. After your commit, the rxe can not work well. Please reproduce this problem in your host and fix it.

Zhu Yanjun

You posted

Quoted 9 lines

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

Quoted 7 lines
Quoted 1 line

Zhu Yanjun

I am not sure how you made rxe to work because it did not work for me and neither for Bob. Since then, Bob has posted patches for the issue. I also posted that my changes work on 5.13.6 kernel. emails attached.

Even if rxe in 5.14 is working for you some how, please apply Bob's patches and then mine and test.

I have already applied this commit
https://patchwork.kernel.org/project/linux-rdma/patch/[email protected]/.

And with your commit, rxe can not work well.

Zhu Yanjun

Quoted 41 lines

Thanks,

Shoaib

Quoted 35 lines
Quoted 35 lines

Shoaib

Quoted 31 lines · Shoaib Rao

Thanks Bob.

Zhu can you please apply those patches and test.

Shoaib

Quoted 24 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 4 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

author
#16 lore raw
Quoted 29 lines · Zhu Yanjun
Quoted 28 lines · Shoaib Rao
Quoted 10 lines · Zhu Yanjun
Quoted 3 lines · Shoaib Rao

Hi Zhu,

Any update on your testing after applying Bob's fixes

Do you read my problem carefully?
I mean that before your commit, the whole rxe can work well. After your commit, the rxe can not work well. Please reproduce this problem in your host and fix it.

Zhu Yanjun

You posted

Quoted 10 lines

In my daily tests, I found that one host 5.12-stable, the other host is 5.14.-rc3 + this commit.
rping can not work. Sometimes crash will occur.

It seems that changing maximum values breaks backward compatibility.

But without this commit, that is, 5.12-stable <-------> 5.14-rc3, rping can work well.

Zhu Yanjun

I am not sure how you made rxe to work because it did not work for me and neither for Bob. Since then, Bob has posted patches for the issue. I also posted that my changes work on 5.13.6 kernel. emails attached.

Quoted 8 lines
Quoted 2 lines

Even if rxe in 5.14 is working for you some how, please apply Bob's patches and then mine and test.

I have already applied this commit

https://urldefense.com/v3/__https://patchwork.kernel.org/project/linux-rdma/patch/[email protected]/__;!!ACWV5N9M2RV99hQ!b2c47MGvP_kCr0tkQgySPZaB3QX3DMeh4l_iwAS3IQHh9R589oF9BWrcgftcidGA$ .

And with your commit, rxe can not work well.

Zhu Yanjun

I am not sure how anyone can claim that the code works without my changes. Rxe in Linux 5.14-rc4 is broken due to following change made to rxe_cq_post() and will cause panic or corruption guaranteed.

addr = producer_addr(cq->queue, QUEUE_TYPE_TO_CLIENT);

It should be

addr = producer_addr(cq->queue, QUEUE_TYPE_FROM_CLIENT);

The following function also seems wrong

Quoted 15 lines

static inline void *producer_addr(struct rxe_queue *q, enum queue_type type) {
        u32 prod;

        switch (type) {
        case QUEUE_TYPE_FROM_CLIENT:
                /* protect user space index */
                prod = smp_load_acquire(&q->buf->producer_index); prod &= q->index_mask;
                break;
        case QUEUE_TYPE_TO_CLIENT:
                prod = q->index;
                break;
        }

Quoted 2 lines

        return q->buf->data + (prod << q->log2_elem_size); }

index should be returned as it is.

The code has changed again in v5.14-rc4-22-g251a1524293d, so now I have to try again.

Can we please make sure that the code is working after the application of each patch or else it is a moving target.

BTW I liked the old code as it distinctly said what was being returned.

Shoaib

Quoted 39 lines
Quoted 39 lines

Thanks,

Shoaib

Quoted 34 lines
Quoted 34 lines

Shoaib

Quoted 31 lines · Shoaib Rao

Thanks Bob.

Zhu can you please apply those patches and test.

Shoaib

Quoted 24 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 4 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

author
#18 lore raw
In reply to Pearson, Robert B

Bob,

Your third patch has an issue.

In rxe_cq_post()

addr = producer_addr(cq->queue, QUEUE_TYPE_TO_CLIENT);

It should be

addr = producer_addr(cq->queue, QUEUE_TYPE_FROM_CLIENT);

After making this change, I have tested my patch and rping works.

Bob can you please point me to the discussion which lead to the current changes, particularly the need for user barrier.

Zhu can you apply Bob's 3 patches + the change above + my patch and report back. In my testing it works.

Regards,

Shoaib

Quoted 19 lines · Robert B

I found another rxe bug (for SRQ) and sent three bug fixes in a set including the one you mention. They should all be applied.

-----Original Message-----
From: Jason Gunthorpe <[email protected]>
Sent: Thursday, July 29, 2021 2:51 PM
To: Shoaib Rao <[email protected]>
Cc: Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 3 lines · Shoaib Rao

Can we please accept my initial patch where I bumped up the values of a few parameters. We have extensively tested with those values. I will try to resolve CRC errors and panic and make changes to other tuneables later?

I think Bob posted something for the icrc issues already

Please try to work in a sane fashion, rxe shouldn't be left broken with so many people apparently interested in it??

Jason

#19 lore raw
Quoted 13 lines · Shoaib Rao

Bob,

Your third patch has an issue.

In rxe_cq_post()

addr = producer_addr(cq->queue, QUEUE_TYPE_TO_CLIENT);

It should be

addr = producer_addr(cq->queue, QUEUE_TYPE_FROM_CLIENT);

Quoted 7 lines

After making this change, I have tested my patch and rping works.

Bob can you please point me to the discussion which lead to the current changes, particularly the need for user barrier.

Zhu can you apply Bob's 3 patches + the change above + my patch and report back. In my testing it works.

I'll expect Bob to resend

[for-next,v2,3/3] RDMA/rxe: Add memory barriers to kernel queues

Jason

author
#20 lore raw
Quoted 19 lines · Jason Gunthorpe
Quoted 18 lines · Shoaib Rao

Bob,

Your third patch has an issue.

In rxe_cq_post()

addr = producer_addr(cq->queue, QUEUE_TYPE_TO_CLIENT);

It should be

addr = producer_addr(cq->queue, QUEUE_TYPE_FROM_CLIENT);

After making this change, I have tested my patch and rping works.

Bob can you please point me to the discussion which lead to the current changes, particularly the need for user barrier.

Quoted 7 lines
Quoted 2 lines

Zhu can you apply Bob's 3 patches + the change above + my patch and report back. In my testing it works.

I'll expect Bob to resend

[for-next,v2,3/3] RDMA/rxe: Add memory barriers to kernel queues

Jason

I have not seen a reply to this email thread. Has the issue been resolved and I missed it?

Shoaib

#21 lore raw

Sorry. I totally missed this. I will look at it in the morning.

Bob

-----Original Message-----
From: Shoaib Rao <[email protected]>
Sent: Sunday, September 12, 2021 7:50 PM
To: Jason Gunthorpe <[email protected]>
Cc: Pearson, Robert B <[email protected]>; Zhu Yanjun <[email protected]>; RDMA mailing list <[email protected]>
Subject: Re: [PATCH v3 0/1] RDMA/rxe: Bump up default maximum values used via uverbs

Quoted 19 lines · Jason Gunthorpe
Quoted 18 lines · Shoaib Rao

Bob,

Your third patch has an issue.

In rxe_cq_post()

addr = producer_addr(cq->queue, QUEUE_TYPE_TO_CLIENT);

It should be

addr = producer_addr(cq->queue, QUEUE_TYPE_FROM_CLIENT);

After making this change, I have tested my patch and rping works.

Bob can you please point me to the discussion which lead to the current changes, particularly the need for user barrier.

Quoted 7 lines
Quoted 2 lines

Zhu can you apply Bob's 3 patches + the change above + my patch and report back. In my testing it works.

I'll expect Bob to resend

[for-next,v2,3/3] RDMA/rxe: Add memory barriers to kernel queues

Jason

I have not seen a reply to this email thread. Has the issue been resolved and I missed it?

Shoaib

#22 lore raw
In reply to Shoaib Rao
Quoted 25 lines · Shoaib Rao
Quoted 23 lines · Jason Gunthorpe
Quoted 13 lines · Shoaib Rao

Bob,

Your third patch has an issue.

In rxe_cq_post()

addr = producer_addr(cq->queue, QUEUE_TYPE_TO_CLIENT);

It should be

addr = producer_addr(cq->queue, QUEUE_TYPE_FROM_CLIENT);

Quoted 7 lines

After making this change, I have tested my patch and rping works.

Bob can you please point me to the discussion which lead to the current changes, particularly the need for user barrier.

Zhu can you apply Bob's 3 patches + the change above + my patch and report back. In my testing it works.

I'll expect Bob to resend

Quoted 8 lines
Quoted 3 lines

    [for-next,v2,3/3] RDMA/rxe: Add memory barriers to kernel queues

Jason

I have not seen a reply to this email thread. Has the issue been resolved and I missed it?

Shoaib

Shoaib,

Thanks for this. I think I figured out what was causing the problem you tried to fix by changing _TO_CLIENT to _FROM_CLIENT. That change isn't the whole solution.

The inline functions in rxe_queue.h all take a type parameter to let the compiler remove the switch statement since the case is known at compile time. The types currently refer to the direction of data flow in the queues from the point of view of the internals of the rxe driver. I.e. for WQs data flows from the CLIENT to the DRIVER and for CQs the data flows to the CLIENT from the DRIVER. This lets the routines in rxe_queue.h selectively use smp_load_acquire or smp_store_release to 'protect' the queue indices owned by the client and private q->index for the indices owned by the internals of the driver which is then copied to q->buf->producer/consumer_index. The reason for this it that the driver can't trust the client in user space to not touch its data.

In rxe_cq.c where you made the change data is flowing to the client and the original type is the correct one. I believe the problem lies in rxe_verbs.c where verbs code manipulates the 'client' end of the queues. This occurs in post_one_recv(), post_one_send(), rxe_poll_cq(), and rxe_peek_cq().
This code was using the same APIs as the driver internals which had two problems. First it had
the direction wrong in terms of which indices needed protection and worse it used the private
copies of the indices that should not be visible to clients. You put memory barriers in rxe_cq
that had the same effect ias putting the correct barriers in rxe_verbs.c.

I have fixed this by adding two new types for these verbs routines to use that put the correct memory barriers on the correct indices. It is going to be resent as v4 of the patch series shortly. I would like you to try it on rping and see of it cleans up what you were seeing.

Bob