powerpc/pseries/msi: Avoid reading PCI device registers in reduced power states
PATCH 1/1 powerpc/pseries/msi: Avoid reading PCI device registers in reduced power states
When a system is being suspended to RAM, the PCI devices are also suspended and the PPC code ends up calling pseries_msi_compose_msg() and this triggers the BUG_ON() in __pci_read_msi_msg() because the device at this point is in reduced power state. In reduced power state, the memory mapped registers of the PCI device are not accessible.
To replicate the bug:
1. Make sure deep sleep is selected
# cat /sys/power/mem_sleep s2idle [deep]
2. Make sure console is not suspended (so that dmesg logs are visible) echo N > /sys/module/printk/parameters/console_suspend
3. Suspend the system
echo mem > /sys/power/state
To fix this behaviour, read the cached msi message of the device when the device is not in PCI_D0 power state instead of touching the hardware.
--- arch/powerpc/platforms/pseries/msi.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
arch/powerpc/platforms/pseries/msi.c +6 -1
@@ -525,7 +525,12 @@ static struct msi_domain_info pseries_msi_domain_info = {static void pseries_msi_compose_msg (struct irq_data *data,struct msi_msg *msg) { -__pci_read_msi_msg (irq_data_get_msi_desc (data), msg); +struct pci_dev *dev =msi_desc_to_pci_dev (irq_data_get_msi_desc (data)); + +if (dev->current_state ==PCI_D0 ) +__pci_read_msi_msg (irq_data_get_msi_desc (data), msg); +else +get_cached_msi_msg (data->irq, msg); }static struct irq_chip pseries_msi_irq_chip = {
Signature · 1 line
2.47.0
Quoted 17 lines · Gautam Menghani
When a system is being suspended to RAM, the PCI devices are also suspended and the PPC code ends up calling
pseries_msi_compose_msg()and this triggers theBUG_ON()in__pci_read_msi_msg()because the device at this point is in reduced power state. In reduced power state, the memory mapped registers of the PCI device are not accessible.To replicate the bug:
1. Make sure deep sleep is selected# cat /sys/power/mem_sleep s2idle [deep]2. Make sure console is not suspended (so that dmesg logs are visible) echo N >
/sys/module/printk/parameters/console_suspend3. Suspend the system
echo mem >/sys/power/state
Quoted 6 lines
To fix this behaviour, read the cached msi message of the device when the device is not in
PCI_D0power state instead of touching the hardware.Fixes a5f3d2c17b07 ("powerpc/pseries/pci: Add MSI domains")Cc [email protected] # v5.15+Signed-off-by Gautam Menghani <[email protected]>
LGTM. Hence
Signature · 2 lines
Cheers ~ Vaibhav
Venkat Rao Bagalkote
Tested-byQuoted 19 lines · Vaibhav Jain
Quoted 17 lines · Gautam Menghani
When a system is being suspended to RAM, the PCI devices are also suspended and the PPC code ends up calling
pseries_msi_compose_msg()and this triggers theBUG_ON()in__pci_read_msi_msg()because the device at this point is in reduced power state. In reduced power state, the memory mapped registers of the PCI device are not accessible.To replicate the bug:
1. Make sure deep sleep is selected# cat /sys/power/mem_sleep s2idle [deep]2. Make sure console is not suspended (so that dmesg logs are visible) echo N >
/sys/module/printk/parameters/console_suspend3. Suspend the system
echo mem >/sys/power/state
Quoted 6 lines
Quoted 6 lines
To fix this behaviour, read the cached msi message of the device when the device is not in
PCI_D0power state instead of touching the hardware.Fixes a5f3d2c17b07 ("powerpc/pseries/pci: Add MSI domains")Cc [email protected] # v5.15+Signed-off-by Gautam Menghani <[email protected]>
I am able to reporduce this issue without this patch and with this pacth, there is no BUG_ON() in __pci_read_msi_msg(), but did see kernel warnings. not sure if its side effect of this patch or a seperate issue.
Without this patch: [ 96.888399] ------------[ cut here ]------------ [ 96.888402] kernel BUG at drivers/pci/msi/msi.c:158! [ 96.888407] Oops: Exception in kernel mode, sig: 5 [#1] [ 96.888410] LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=8192 NUMA pSeries [ 96.888414] Modules linked in: nft_compat nf_tables nfnetlink bonding tls rfkill binfmt_misc kmem device_dax pseries_rng vmx_crypto dax_pmem drm drm_panel_orientation_quirks xfs dm_service_time sd_mod sg nd_pmem ibmvfc nd_btt ibmvscsi scsi_transport_fc ibmveth scsi_transport_srp papr_scm libnvdimm tg3 dm_multipath dm_mirror dm_region_hash dm_log dm_mod fuse [ 96.888473] CPU: 14 UID: 0 PID: 89 Comm: migration/14 Kdump: loaded Not tainted 6.14.0-auto #3 [ 96.888479] Hardware name: IBM,9009-42A POWER9 (architected) 0x4e0202 0xf000005 of:IBM,FW950.A0 (VL950_141) hv:phyp pSeries [ 96.888481] Stopper: multi_cpu_stop+0x0/0x22c <- __stop_cpus.constprop.0+0x68/0xc0 [ 96.888494] NIP: c000000000995aec LR: c00000000010ec20 CTR: c00000000010ebf8 [ 96.888498] REGS: c00000000680f830 TRAP: 0700 Not tainted (6.14.0-auto) [ 96.888501] MSR: 8000000002823033 <SF,VEC,VSX,FP,ME,IR,DR,RI,LE> CR: 44004208 XER: 00000000 [ 96.888520] CFAR: c00000000010ec1c IRQMASK: 3 GPR00: c00000000010ec20 c00000000680fad0 c000000001668100 c000000006c537e0 GPR04: c00000000680fb80 0000000000000000 0000000000000000 c009ffffff8325f0 GPR08: 0000000000000001 0000000000000001 0000000000000003 0000000000001003 GPR12: c00000000010ebf8 c00000000f7beb00 c0000000001acbe0 c000000004056d40 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000003 000000000000001d c000000002cfaa88 GPR24: c00000006545b800 c000000002cfc080 c000000001126bb0 0000000000000000 GPR28: 0000000000000010 c00000000ce790c8 c00000000680fb80 c000000006c537e0 [ 96.888586] NIP [c000000000995aec] __pci_read_msi_msg+0x48/0x278 [ 96.888592] LR [c00000000010ec20] pseries_msi_compose_msg+0x28/0x3c [ 96.888599] Call Trace: [ 96.888600] [c00000000680fad0] [000000000000001d] 0x1d (unreliable) [ 96.888608] [c00000000680fb20] [c00000006545b820] 0xc00000006545b820 [ 96.888613] [c00000000680fb40] [c00000000023b41c] irq_chip_compose_msi_msg+0x5c/0x90 [ 96.888620] [c00000000680fb60] [c000000000242aec] msi_domain_set_affinity+0xb8/0xf4 [ 96.888627] [c00000000680fbb0] [c000000000234634] irq_do_set_affinity+0x14c/0x25c [ 96.888633] [c00000000680fc10] [c000000000234870] irq_set_affinity_locked+0x12c/0x1c4 [ 96.888639] [c00000000680fc60] [c000000000234a84] irq_set_affinity+0x64/0xa0 [ 96.888644] [c00000000680fca0] [c0000000000c9d40] xics_migrate_irqs_away+0x27c/0x30c [ 96.888650] [c00000000680fd60] [c000000000111834] pseries_cpu_disable+0xc8/0xf0 [ 96.888657] [c00000000680fd90] [c0000000000611e0] __cpu_disable+0x54/0xb0 [ 96.888662] [c00000000680fdc0] [c0000000001715e8] take_cpu_down+0x4c/0xcc [ 96.888669] [c00000000680fe10] [c0000000002ebbc4] multi_cpu_stop+0xd8/0x22c [ 96.888676] [c00000000680fe80] [c0000000002eb898] cpu_stopper_thread+0x158/0x24c [ 96.888683] [c00000000680ff30] [c0000000001b7a0c] smpboot_thread_fn+0x1ec/0x25c [ 96.888691] [c00000000680ff90] [c0000000001acd04] kthread+0x12c/0x14c [ 96.888697] [c00000000680ffe0] [c00000000000df98] start_kernel_thread+0x14/0x18 [ 96.888703] Code: fba1ffe8 39200001 f821ffb1 7c7f1b78 7c9e2378 e94d0c78 f9410028 39400000 eba30008 815dffd8 2c0a0000 7d20489e <0b090000> a123004c 712a0001 41820168 [ 96.888730] ---[ end trace 0000000000000000 ]---
With this patch: No System crash observed. But below warnings were observed.
[ 99.450644] ------------[ cut here ]------------ [ 99.450648] WARNING: CPU: 0 PID: 17 at arch/powerpc/sysdev/xics/icp-hv.c:55 icp_hv_eoi+0xc4/0x120 [ 99.450659] Modules linked in: nft_compat nf_tables nfnetlink bonding tls rfkill binfmt_misc kmem device_dax pseries_rng vmx_crypto dax_pmem drm drm_panel_orientation_quirks xfs dm_service_time sd_mod sg nd_pmem ibmvfc nd_btt ibmvscsi scsi_transport_fc ibmveth scsi_transport_srp papr_scm libnvdimm tg3 dm_multipath dm_mirror dm_region_hash dm_log dm_mod fuse [ 99.450704] CPU: 0 UID: 0 PID: 17 Comm: ksoftirqd/0 Kdump: loaded Not tainted 6.14.0-auto-00001-g03419579f433 #4 [ 99.450712] Hardware name: IBM,9009-42A POWER9 (architected) 0x4e0202 0xf000005 of:IBM,FW950.A0 (VL950_141) hv:phyp pSeries [ 99.450717] NIP: c0000000000cadd4 LR: c0000000000cadd0 CTR: 00000000007088ec [ 99.450722] REGS: c000000004a2fa20 TRAP: 0700 Not tainted (6.14.0-auto-00001-g03419579f433) [ 99.450727] MSR: 800000000282b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 2804424f XER: 00000010 [ 99.450743] CFAR: c000000000224da8 IRQMASK: 1 GPR00: c0000000000cadd0 c000000004a2fcc0 c000000001668100 000000000000003f GPR04: c0000007fd907c88 c0000007fd916000 c000000004a2fb08 00000007fb6a0000 GPR08: 0000000000000027 0000000000000000 0000000000000000 0000000000000001 GPR12: c000000002a37d48 c000000003000000 c0000000001acc60 c000000004052080 GPR16: 0000000000000006 0000000000000040 0000000000000006 0000000000000100 GPR20: 0000000004208040 0000000000000000 0000000000000001 c0000000002382c0 GPR24: 0000000000000001 0000000000000000 0000000000000006 0000000000000002 GPR28: c0000007fd9078b8 0000000000000000 c0000000010e69e8 00000000050a0002 [ 99.450802] NIP [c0000000000cadd4] icp_hv_eoi+0xc4/0x120 [ 99.450808] LR [c0000000000cadd0] icp_hv_eoi+0xc0/0x120 [ 99.450814] Call Trace: [ 99.450816] [c000000004a2fcc0] [c0000000000cadd0] icp_hv_eoi+0xc0/0x120 (unreliable) [ 99.450824] [c000000004a2fd30] [c000000000239eac] handle_fasteoi_irq+0x16c/0x344 [ 99.450832] [c000000004a2fd70] [c000000000238380] resend_irqs+0xc0/0x188 [ 99.450838] [c000000004a2fdb0] [c00000000017b054] tasklet_action_common+0x154/0x418 [ 99.450845] [c000000004a2fe20] [c00000000017a788] handle_softirqs+0x148/0x3b4 [ 99.450852] [c000000004a2ff10] [c00000000017aa58] run_ksoftirqd+0x64/0xa0 [ 99.450858] [c000000004a2ff30] [c0000000001b7a8c] smpboot_thread_fn+0x1ec/0x25c [ 99.450866] [c000000004a2ff90] [c0000000001acd84] kthread+0x12c/0x14c [ 99.450873] [c000000004a2ffe0] [c00000000000df98] start_kernel_thread+0x14/0x18 [ 99.450879] Code: ebe1fff8 7c0803a6 4e800020 3c82ffa8 3c62ffdc 7fc5f378 3fc2ffa8 3884e908 38632268 3bdee8e8 48159f95 60000000 <0fe00000> 7bff4622 38600068 7fe4fb78 [ 99.450900] ---[ end trace 0000000000000000 ]---
Please add below tag:
Regards,
Venkat.
Quoted 3 lines
LGTM. Hence
Reviewed-by Vaibhav Jain <[email protected]>
Hi Venkat,
Thanks for the report. I looked into this and found that the new warning you reported can be observed even on current distro kernels, and is not caused by the patch I've posted.
I was able to observe the same warning with fedora distro kernel 6.13.7-200.fc41
[ 70.294478] icp_hv_set_xirr: bad return code eoi xirr=0x50a0002 returned -4 [ 70.294521] ------------[ cut here ]------------ [ 70.294546] WARNING: CPU: 7 PID: 54 at arch/powerpc/sysdev/xics/icp-hv.c:55 icp_hv_eoi+0xf8/0x120 [ 70.294599] Modules linked in: xt_conntrack xt_MASQUERADE bridge stp llc ip6table_nat ip6table_filter ip6_tables xt_set ip_set iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 xt_addrtype iptable_filter ip_tables kvm rpcrdma rdma_cm iw_cm ib_cm ib_core bonding overlay rfkill binfmt_misc vmx_crypto pseries_rng nfsd auth_rpcgss nfs_acl loop dm_multipath lockd grace nfs_localio nfnetlink vsock_loopback vmw_vsock_virtio_transport_common vsock xfs nvme_tcp nvme_fabrics nvme_keyring nvme_core nvme_auth ibmvscsi ibmveth scsi_transport_srp crct10dif_vpmsum crc32c_vpmsum pseries_wdt sunrpc be2iscsi bnx2i cnic uio cxgb4i cxgb4 tls cxgb3i cxgb3 mdio libcxgbi libcxgb qla4xxx iscsi_boot_sysfs iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi scsi_dh_rdac scsi_dh_emc scsi_dh_alua fuse aes_gcm_p10_crypto crypto_simd cryptd [ 70.295015] CPU: 7 UID: 0 PID: 54 Comm: ksoftirqd/7 Kdump: loaded Not tainted 6.13.7-200.fc41.ppc64le #1 [ 70.295064] Hardware name: IBM,9080-HEX POWER8 (architected) 0x800200 0xf000004 of:IBM,FW1060.00 (NH1060_022) hv:phyp pSeries [ 70.295120] NIP: c000000000197c98 LR: c000000000197c94 CTR: 0000000000000000 [ 70.295157] REGS: c000000007dd3a20 TRAP: 0700 Not tainted (6.13.7-200.fc41.ppc64le) [ 70.295197] MSR: 8000000002029033 <SF,VEC,EE,ME,IR,DR,RI,LE> CR: 24004202 XER: 00000001 [ 70.295247] CFAR: c00000000032731c IRQMASK: 1 [ 70.295247] GPR00: c000000000197c94 c000000007dd3cc0 c0000000024daa00 000000000000003f [ 70.295247] GPR04: 00000000ffff7fff 00000000ffff7fff c000000007dd3ae8 00000007ec8e0000 [ 70.295247] GPR08: 0000000000000027 0000000000000000 0000000000000000 0000000000004000 [ 70.295247] GPR12: 0000000000000000 c00000000ffc6f00 c000000000287ef8 c000000004a51080 [ 70.295247] GPR16: 0000000000000000 0000000004208040 c000000003d62c80 c0000000031faf80 [ 70.295247] GPR20: 00000000ffffa63b 000000000000000a c0000000031e6990 c000000000335f10 [ 70.295247] GPR24: 0000000000000001 0000000000000000 0000000000000006 0000000000000002 [ 70.295247] GPR28: c0000007efac68b8 0000000000000000 00000000050a0002 00000000050a0002 [ 70.295603] NIP [c000000000197c98] icp_hv_eoi+0xf8/0x120 [ 70.295633] LR [c000000000197c94] icp_hv_eoi+0xf4/0x120 [ 70.295661] Call Trace: [ 70.295675] [c000000007dd3cc0] [c000000000197c94] icp_hv_eoi+0xf4/0x120 (unreliable) [ 70.295717] [c000000007dd3d40] [c000000000337a5c] handle_fasteoi_irq+0x16c/0x350 [ 70.295757] [c000000007dd3d70] [c000000000335fd0] resend_irqs+0xc0/0x190 [ 70.295793] [c000000007dd3db0] [c000000000254064] tasklet_action_common+0x154/0x440 [ 70.295833] [c000000007dd3e20] [c000000000253458] handle_softirqs+0x168/0x4f0 [ 70.295871] [c000000007dd3f10] [c000000000253848] run_ksoftirqd+0x68/0xb0 [ 70.295912] [c000000007dd3f30] [c000000000292f20] smpboot_thread_fn+0x1d0/0x240 [ 70.295951] [c000000007dd3f90] [c000000000288020] kthread+0x130/0x140 [ 70.295984] [c000000007dd3fe0] [c00000000000ded8] start_kernel_thread+0x14/0x18 [ 70.296022] Code: 48c84251 60000000 e9210068 4bffff98 7c661b78 3c82ff31 3c62ff7d 7fc5f378 38842b40 38639bf8 4818f649 60000000 <0fe00000> 38210080 7be34622 e8010010 [ 70.296104] ---[ end trace 0000000000000000 ]--- [ 70.297273] PM: resume devices took 0.000 seconds [ 70.297415] OOM killer enabled. [ 70.297433] Restarting tasks ... done. [ 70.298959] random: crng reseeded on system resumption [ 70.299106] PM: suspend exit
This can be tracked as a separate bug, as it is unrelated to the patch.
Thanks,
Gautam
Quoted 6 lines · Gautam Menghani
When a system is being suspended to RAM, the PCI devices are also suspended and the PPC code ends up calling
pseries_msi_compose_msg()and this triggers theBUG_ON()in__pci_read_msi_msg()because the device at this point is in reduced power state. In reduced power state, the memory mapped registers of the PCI device are not accessible.
Quoted 6 lines
To replicate the bug:
1. Make sure deep sleep is selected# cat /sys/power/mem_sleep s2idle [deep][...]
Applied to powerpc/next.
[1/1] powerpc/pseries/msi: Avoid reading PCI device registers in reduced power states
https://git.kernel.org/powerpc/c/9cc0eafd28c7faef300822992bb08d79cab2a36c
Thanks