RDMA/bnxt_re: cancel all control path command waiters upon error

When an error is detected in FW, wake up all the waiters as the
all of them need to be completed with timeout. Add the device
error state also as a wait condition.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-15-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: consider timeout of destroy ah as success.

If destroy_ah is timed out, it is likely to be destroyed by firmware
but it is taking longer time due to temporary slowness
in processing the rcfw command. In worst case, there might be
AH resource leak in firmware.

Sending timeout return value can dump warning message from ib_core
which can be avoided if we map timeout of destroy_ah as success.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-14-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: post destroy_ah for delayed completion of AH creation

AH create may be called from interrpt context and driver has a special
timeout (8 sec) for this command. This is to avoid soft lockups when
the FW command takes more time. Driver returns -ETIMEOUT and fail
create AH, without waiting for actual completion from firmware.
When FW completion is received, use is_waiter_alive flag to avoid
a regular completion path.

If create_ah opcode is detected in completion path which does not have
waiter alive, driver will fetch ah_id from successful firmware
completion in the interrupt context and sends destroy_ah command
for same ah_id. This special post is done in quick manner using helper
function __send_message_no_waiter.

timeout_send is only used for debugging purposes.
If timeout_send value keeps incrementing, it indicates out of sync
active ah counter between driver and firmware. This is a limitation
but graceful handling is possible in future.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-13-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: Add firmware stall check detection

Every completion will update last_seen value in the unit of jiffies.
last_seen field will be used to know if firmware is alive and
is useful to detect firmware stall.

Non blocking interface __wait_for_resp will have logic to detect
firmware stall. After every 10 second interval if __wait_for_resp
has not received completion for a given command it will check for
firmware stall condition.

If current jiffies is greater than last_seen
jiffies + RCFW_FW_STALL_TIMEOUT_SEC * HZ, it is a firmware stall.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-12-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: handle command completions after driver detect a timedout

If calling context detect command timeout, associated memory stored on
stack will not be valid. If firmware complete the same command later,
this causes incorrect memory access by driver.

Added is_waiter_alive to handle delayed completion by firmware.
is_waiter_alive is set and reset under command queue lock.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-11-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: add helper function __poll_for_resp

This interface will be used if the driver has not enabled interrupt
and/or interrupt is disabled for a short period of time.
Completion is not possible from interrupt so this interface does
self-polling.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-10-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: Simplify the function that sends the FW commands

 - Use __send_message_basic_sanity helper function.
 - Do not retry posting same command if there is a queue full detection.
 - ENXIO is used to indicate controller recovery.
 - In the case of ERR_DEVICE_DETACHED state, the driver should not post
   commands to the firmware, but also return fabricated written code.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-9-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: use shadow qd while posting non blocking rcfw command

Whenever there is a fast path IO and create/destroy resources from
the slow path is happening in parallel, we may notice high latency
of slow path command completion.

Introduces a shadow queue depth to prevent the outstanding requests
to the FW. Driver will not allow more than #RCFW_CMD_NON_BLOCKING_SHADOW_QD
non-blocking commands to the Firmware.

Shadow queue depth is a soft limit only for non-blocking
commands. Blocking commands will be posted to the firmware
as long as there is a free slot.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-8-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: Avoid the command wait if firmware is inactive

Add a check to avoid waiting if driver already detects a
FW timeout. Return success for resource destroy in case
the device is detached. Add helper function to map timeout
error code to success.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-7-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: Enhance the existing functions that wait for FW responses

Use jiffies based timewait instead of counting iteration for
commands that block for FW response.

Also add a poll routine for control path commands. This is for
polling completion if the waiting commands timeout. This avoids cases
where the driver misses completion interrupts.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-6-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: set fixed command queue depth

There is no need of setting max command queue entries based on
firmware version check.

Removing deperecated code.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-5-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: remove virt_func check while creating RoCE FW channel

There is a common FW communication offset for both PF and VF.
Removed code around virt_fn check while creating FW channel.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-4-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: Avoid calling wake_up threads from spin_lock context

bnxt_qplib_service_creq can be called from interrupt or tasklet or
process context. So the function take irq variant  of spin_lock.
But when wake_up is invoked with the lock held, it is putting the
calling context to sleep.

[exception RIP: __wake_up_common+190]
RIP: ffffffffb7539d7e  RSP: ffffa73300207ad8  RFLAGS: 00000083
RAX: 0000000000000001  RBX: ffff91fa295f69b8  RCX: dead000000000200
RDX: ffffa733344af940  RSI: ffffa73336527940  RDI: ffffa73336527940
RBP: 000000000000001c   R8: 0000000000000002   R9: 00000000000299c0
R10: 0000017230de82c5  R11: 0000000000000002  R12: ffffa73300207b28
R13: 0000000000000000  R14: ffffa733341bf928  R15: 0000000000000000
ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018

Call the wakeup after releasing the lock.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-3-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/bnxt_re: wraparound mbox producer index

Driver is not handling the wraparound of the mbox producer index correctly.
Currently the wraparound happens once u32 max is reached.

Bit 31 of the producer index register is special and should be set
only once for the first command. Because the producer index overflow
setting bit31 after a long time, FW goes to initialization sequence
and this causes FW hang.

Fix is to wraparound the mbox producer index once it reaches u16 max.

Fixes: cee0c7bba486 ("RDMA/bnxt_re: Refactor command queue management code")
Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Link: https://lore.kernel.org/r/1686308514-11996-2-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/erdma: Refactor the original doorbell allocation mechanism

The original doorbell allocation mechanism is complex and does not meet
the isolation requirement. So we introduce a new doorbell mechanism and the
original mechanism (only be used with CAP_SYS_RAWIO if hardware does not
support the new mechanism) needs to be kept as simple as possible for
compatibility.

Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230606055005.80729-5-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/erdma: Associate QPs/CQs with doorbells for authorization

For the isolation requirement, each QP/CQ can only issue doorbells from the
allocated mmio space. Configure the relationship between QPs/CQs and
mmio doorbell spaces to hardware in create_qp/create_cq interfaces.

Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230606055005.80729-4-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/erdma: Allocate doorbell resources from hardware

Each ucontext will try to allocate doorbell resources in the extended bar
space from hardware. For compatibility, we change nothing for the original
bar space, and it will be used only for applications with CAP_SYS_RAWIO
authority in the older HW/FW environments.

Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230606055005.80729-3-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/erdma: Configure PAGE_SIZE to hardware

Add a new CMDQ message to configure hardware. Initially the page size (in
the format of shift) will be passed to hardware, so that hardware can
organize the mmio space properly. It's called only if hardware supports it.

Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com>
Link: https://lore.kernel.org/r/20230606055005.80729-2-chengyou@linux.alibaba.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/mlx5: Return the firmware result upon destroying QP/RQ

Previously when destroying a QP/RQ, the result of the firmware
destruction function was ignored and upper layers weren't informed
about the failure.
Which in turn could lead to various problems since when upper layer
isn't aware of the failure it continues its operation thinking that the
related QP/RQ was successfully destroyed while it actually wasn't,
which could lead to the below kernel WARN.

Currently, we return the correct firmware destruction status to upper
layers which in case of the RQ would be mlx5_ib_destroy_wq() which
was already capable of handling RQ destruction failure or in case of
a QP to destroy_qp_common(), which now would actually warn upon qp
destruction failure.

WARNING: CPU: 3 PID: 995 at drivers/infiniband/core/rdma_core.c:940 uverbs_destroy_ufile_hw+0xcb/0xe0 [ib_uverbs]
Modules linked in: xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter rpcrdma rdma_ucm ib_iser libiscsi scsi_transport_iscsi rdma_cm ib_umad ib_ipoib iw_cm ib_cm mlx5_ib ib_uverbs ib_core overlay mlx5_core fuse
CPU: 3 PID: 995 Comm: python3 Not tainted 5.16.0-rc5+ #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
RIP: 0010:uverbs_destroy_ufile_hw+0xcb/0xe0 [ib_uverbs]
Code: 41 5c 41 5d 41 5e e9 44 34 f0 e0 48 89 df e8 4c 77 ff ff 49 8b 86 10 01 00 00 48 85 c0 74 a1 4c 89 e7 ff d0 eb 9a 0f 0b eb c1 <0f> 0b be 04 00 00 00 48 89 df e8 b6 f6 ff ff e9 75 ff ff ff 90 0f
RSP: 0018:ffff8881533e3e78 EFLAGS: 00010287
RAX: ffff88811b2cf3e0 RBX: ffff888106209700 RCX: 0000000000000000
RDX: ffff888106209780 RSI: ffff8881533e3d30 RDI: ffff888109b101a0
RBP: 0000000000000001 R08: ffff888127cb381c R09: 0de9890000000009
R10: ffff888127cb3800 R11: 0000000000000000 R12: ffff888106209780
R13: ffff888106209750 R14: ffff888100f20660 R15: 0000000000000000
FS:  00007f8be353b740(0000) GS:ffff88852c980000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f8bd5b117c0 CR3: 000000012cd8a004 CR4: 0000000000370ea0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
 ib_uverbs_close+0x1a/0x90 [ib_uverbs]
 __fput+0x82/0x230
 task_work_run+0x59/0x90
 exit_to_user_mode_prepare+0x138/0x140
 syscall_exit_to_user_mode+0x1d/0x50
 ? __x64_sys_close+0xe/0x40
 do_syscall_64+0x4a/0x90
 entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7f8be3ae0abb
Code: 03 00 00 00 0f 05 48 3d 00 f0 ff ff 77 41 c3 48 83 ec 18 89 7c 24 0c e8 83 43 f9 ff 8b 7c 24 0c 41 89 c0 b8 03 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 35 44 89 c7 89 44 24 0c e8 c1 43 f9 ff 8b 44
RSP: 002b:00007ffdb51909c0 EFLAGS: 00000293 ORIG_RAX: 0000000000000003
RAX: 0000000000000000 RBX: 0000557bb7f7c020 RCX: 00007f8be3ae0abb
RDX: 0000557bb7c74010 RSI: 0000557bb7f14ca0 RDI: 0000000000000005
RBP: 0000557bb7fbd598 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000293 R12: 0000557bb7fbd5b8
R13: 0000557bb7fbd5a8 R14: 0000000000001000 R15: 0000557bb7f7c020
 </TASK>

Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Link: https://lore.kernel.org/r/c6df677f931d18090bafbe7f7dbb9524047b7d9b.1685953497.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/mlx5: Handle DCT QP logic separately from low level QP interface

Previously when destroying a DCT, if the firmware function for the
destruction failed, the common resource would have been destroyed
either way, since it was destroyed before the firmware object.
Which leads to kernel warning "refcount_t: underflow" which indicates
possible use-after-free.
Which is triggered when we try to destroy the common resource for the
second time and execute refcount_dec_and_test(&common->refcount).

So, let's fix the destruction order by factoring out the DCT QP logic
to be in separate XArray database.

refcount_t: underflow; use-after-free.
WARNING: CPU: 8 PID: 1002 at lib/refcount.c:28 refcount_warn_saturate+0xd8/0xe0
Modules linked in: xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter rpcrdma rdma_ucm ib_iser libiscsi scsi_transport_iscsi ib_umad rdma_cm ib_ipoib iw_cm ib_cm mlx5_ib ib_uverbs ib_core overlay mlx5_core fuse
CPU: 8 PID: 1002 Comm: python3 Not tainted 5.16.0-rc5+ #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
RIP: 0010:refcount_warn_saturate+0xd8/0xe0
Code: ff 48 c7 c7 18 f5 23 82 c6 05 60 70 ff 00 01 e8 d0 0a 45 00 0f 0b c3 48 c7 c7 c0 f4 23 82 c6 05 4c 70 ff 00 01 e8 ba 0a 45 00 <0f> 0b c3 0f 1f 44 00 00 8b 07 3d 00 00 00 c0 74 12 83 f8 01 74 13
RSP: 0018:ffff8881221d3aa8 EFLAGS: 00010286
RAX: 0000000000000000 RBX: ffff8881313e8d40 RCX: ffff88852cc1b5c8
RDX: 00000000ffffffd8 RSI: 0000000000000027 RDI: ffff88852cc1b5c0
RBP: ffff888100f70000 R08: ffff88853ffd1ba8 R09: 0000000000000003
R10: 00000000fffff000 R11: 3fffffffffffffff R12: 0000000000000246
R13: ffff888100f71fa0 R14: ffff8881221d3c68 R15: 0000000000000020
FS:  00007efebbb13740(0000) GS:ffff88852cc00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00005611aac29f80 CR3: 00000001313de004 CR4: 0000000000370ea0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <TASK>
 destroy_resource_common+0x6e/0x95 [mlx5_ib]
 mlx5_core_destroy_rq_tracked+0x38/0xbe [mlx5_ib]
 mlx5_ib_destroy_wq+0x22/0x80 [mlx5_ib]
 ib_destroy_wq_user+0x1f/0x40 [ib_core]
 uverbs_free_wq+0x19/0x40 [ib_uverbs]
 destroy_hw_idr_uobject+0x18/0x50 [ib_uverbs]
 uverbs_destroy_uobject+0x2f/0x190 [ib_uverbs]
 uobj_destroy+0x3c/0x80 [ib_uverbs]
 ib_uverbs_cmd_verbs+0x3e4/0xb80 [ib_uverbs]
 ? uverbs_free_wq+0x40/0x40 [ib_uverbs]
 ? ip_list_rcv+0xf7/0x120
 ? netif_receive_skb_list_internal+0x1b6/0x2d0
 ? task_tick_fair+0xbf/0x450
 ? __handle_mm_fault+0x11fc/0x1450
 ib_uverbs_ioctl+0xa4/0x110 [ib_uverbs]
 __x64_sys_ioctl+0x3e4/0x8e0
 ? handle_mm_fault+0xb9/0x210
 do_syscall_64+0x3d/0x90
 entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7efebc0be17b
Code: 0f 1e fa 48 8b 05 1d ad 0c 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d ed ac 0c 00 f7 d8 64 89 01 48
RSP: 002b:00007ffe71813e78 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 00007ffe71813fb8 RCX: 00007efebc0be17b
RDX: 00007ffe71813fa0 RSI: 00000000c0181b01 RDI: 0000000000000005
RBP: 00007ffe71813f80 R08: 00005611aae96020 R09: 000000000000004f
R10: 00007efebbf9ffa0 R11: 0000000000000246 R12: 00007ffe71813f80
R13: 00007ffe71813f4c R14: 00005611aae2eca0 R15: 00007efeae6c89d0
 </TASK>

Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/4470888466c8a898edc9833286967529cc5f3c0d.1685953497.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/mlx5: Reduce QP table exposure

driver.h is common header to whole mlx5 code base, but struct
mlx5_qp_table is used in mlx5_ib driver only. So move that struct
to be under sole responsibility of mlx5_ib.

Link: https://lore.kernel.org/r/bec0dc1158e795813b135d1143147977f26bf668.1685953497.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

net/mlx5: Nullify qp->dbg pointer post destruction

Nullifying qp->dbg is a preparation for the next patches
from the series in which mlx5_core_destroy_qp() could actually fail,
and then it can be called again which causes a kernel crash, since
qp->dbg was not nullified in previous call.

Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Link: https://lore.kernel.org/r/1677e52bb642fd8d6062d73a5aa69083c0283dc9.1685953497.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>

RDMA/vmw_pvrdma: Remove unnecessary check on wr->opcode

wr->opcode is unsigned; checking if it is negative is unnecessary. Fix
this issue by removing the check.

Fixes: 29c8d9eba550 ("IB: Add vmw_pvrdma driver")
Link: https://lore.kernel.org/r/20230605183728.47021-1-bryantan@vmware.com
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/r/07c48eee-0aca-48ee-897a-38588c341c41@kili.mountain
Signed-off-by: Bryan Tan <bryantan@vmware.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Send last wqe reached event on qp cleanup

The IBA requires:
o11-5.2.5: If the HCA supports SRQ, for RC and UD service,
the CI shall generate a Last WQE Reached Affiliated Asynchronous
Event on a QP that is in the Error State and is associated with
an SRQ when either:
• a CQE is generated for the last WQE, or
• the QP gets in the Error State and there are no more
  WQEs on the RQ.

This patch implements this behavior in flush_recv_queue() which is called
as a result of rxe_qp_error() being called whenever the qp is put into the
error state. The rxe responder executes SRQ WQEs directly from the SRQ so
there are never more WQES on the RQ.

Link: https://lore.kernel.org/r/20230602164229.9277-1-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Implement rereg_user_mr

Implement the two easy cases of ib_rereg_user_mr.

Link: https://lore.kernel.org/r/20230530221334.89432-7-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Let rkey == lkey for local access

In order to conform to other drivers stop using rkey == 0 as an indication
that there are no remote access flags set.  Set rkey == lkey by default
for all MRs.

Link: https://lore.kernel.org/r/20230530221334.89432-6-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Introduce rxe access supported flags

Introduce supported bit masks for setting the access attributes of MWs,
MRs, and QPs. Check these when attributes are set.

Link: https://lore.kernel.org/r/20230530221334.89432-5-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix access checks in rxe_check_bind_mw

The subroutine rxe_check_bind_mw() in rxe_mw.c performs checks on the mw
access flags before they are set so they always succeed.  This patch
instead checks the access flags passed in the send wqe.

Fixes: 32a577b4c3a9 ("RDMA/rxe: Add support for bind MW work requests")
Link: https://lore.kernel.org/r/20230530221334.89432-4-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA//rxe: Optimize send path in rxe_resp.c

Bypass calling check_rkey() in rxe_resp.c for non-rdma messages.

Link: https://lore.kernel.org/r/20230530221334.89432-3-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Rename IB_ACCESS_REMOTE

Rename IB_ACCESS_REMOTE to RXE_ACCESS_REMOTE and move to rxe_verbs.h as an
enum instead of a #define. Shouldn't use IB_xxx for rxe symbols.

Link: https://lore.kernel.org/r/20230530221334.89432-2-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Add clear_hem return value to log

Log return value of clear_hem() to help diagnose.

Link: https://lore.kernel.org/r/20230523121641.3132102-4-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Fix hns_roce_table_get return value

The return value of set_hem has been fixed to ENODEV, which will lead a
diagnostic information missing.

Fixes: 9a4435375cd1 ("IB/hns: Add driver files for hns RoCE driver")
Link: https://lore.kernel.org/r/20230523121641.3132102-3-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/hns: Remove unnecessary QP type checks

It is not necessary to check the type of the queue on IO path because
unsupported QP type cannot be created.

Link: https://lore.kernel.org/r/20230523121641.3132102-2-huangjunxian6@hisilicon.com
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/hfi1: Remove unused struct mmu_rb_ops fields .insert, .invalidate

The struct mmu_rb_ops function pointers .insert, .invalidate were only
used to increment and decrement struct sdma_mmu_node.refcount.

With the deletion of struct sdma_mmu_node.refcount and the addition of
struct mmu_rb_node.refcount  these function pointers are not called and
there are no implementations of them. So it is safe to delete these from
struct mmu_rb_ops.

Link: https://lore.kernel.org/r/168451526508.3702129.8677714753157495310.stgit@awfm-02.cornelisnetworks.com
Signed-off-by: Brendan Cunningham <bcunningham@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/hfi1: Add mmu_rb_node refcount to hfi1_mmu_rb_template tracepoints

Add kref_read() of mmu_rb_node.refcount in hfi1_mmu_rb_template-type
tracepoint output.

Change hfi1_mmu_rb_template tracepoint to take a struct mmu_rb_node* and
record the values it needs from that. This makes the trace_hfi1_mmu*()
calls shorter and easier to read.

Add hfi1_mmu_release_node() tracepoint before all
mmu_rb_node->handler->ops->remove() calls.

Make hfi1_mmu_rb_search() tracepoint its own tracepoint type separate
from hfi1_mmu_rb_template since hfi1_mmu_rb_search() does not take a
struct mmu_rb_node*.

Link: https://lore.kernel.org/r/168451525987.3702129.12824880387615916700.stgit@awfm-02.cornelisnetworks.com
Reviewed-by: Dean Luick <dean.luick@cornelisnetworks.com>
Signed-off-by: Brendan Cunningham <bcunningham@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

IB/hfi1: Fix wrong mmu_node used for user SDMA packet after invalidate

The hfi1 user SDMA pinned-page cache will leave a stale cache entry when
the cache-entry's virtual address range is invalidated but that cache
entry is in-use by an outstanding SDMA request.

Subsequent user SDMA requests with buffers in or spanning the virtual
address range of the stale cache entry will result in packets constructed
from the wrong memory, the physical pages pointed to by the stale cache
entry.

To fix this, remove mmu_rb_node cache entries from the mmu_rb_handler
cache independent of the cache entry's refcount. Add 'struct kref
refcount' to struct mmu_rb_node and manage mmu_rb_node lifetime with
kref_get() and kref_put().

mmu_rb_node.refcount makes sdma_mmu_node.refcount redundant. Remove
'atomic_t refcount' from struct sdma_mmu_node and change sdma_mmu_node
code to use mmu_rb_node.refcount.

Move the mmu_rb_handler destructor call after a
wait-for-SDMA-request-completion call so mmu_rb_nodes that need
mmu_rb_handler's workqueue to queue themselves up for destruction from an
interrupt context may do so.

Fixes: f48ad614c100 ("IB/hfi1: Move driver out of staging")
Fixes: 00cbce5cbf88 ("IB/hfi1: Fix bugs with non-PAGE_SIZE-end multi-iovec user SDMA requests")
Link: https://lore.kernel.org/r/168451393605.3700681.13493776139032178861.stgit@awfm-02.cornelisnetworks.com
Reviewed-by: Dean Luick <dean.luick@cornelisnetworks.com>
Signed-off-by: Brendan Cunningham <bcunningham@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rtrs: Remove duplicate cq_num assignment

line 1701 and 1713 are duplicate:
> 1701         cq_num = max_send_wr + max_recv_wr;
  1702         /* alloc iu to recv new rkey reply when server reports flags set */
  1703         if (clt_path->flags & RTRS_MSG_NEW_RKEY_F || con->c.cid == 0) {
  1704                 con->rsp_ius = rtrs_iu_alloc(cq_num, sizeof(*rsp),
  1705                                               GFP_KERNEL,
  1706                                               clt_path->s.dev->ib_dev,
  1707                                               DMA_FROM_DEVICE,
  1708                                               rtrs_clt_rdma_done);
  1709                 if (!con->rsp_ius)
  1710                         return -ENOMEM;
  1711                 con->queue_num = cq_num;
  1712         }
> 1713         cq_num = max_send_wr + max_recv_wr;

Remove the duplicate.

Link: https://lore.kernel.org/r/1682384563-2-2-git-send-email-lizhijian@fujitsu.com
Acked-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Remove dangling declaration of rxe_cq_disable()

rxe_cq_disable() has been removed but not its declaration.

Fixes: 78b26a335310 ("RDMA/rxe: Remove tasklet call from rxe_cq.c")
Link: https://lore.kernel.org/r/4f20ffc5-b2c4-0c11-2883-a835caf01a94@suse.com
Signed-off-by: Nicolas Morey <nmorey@suse.com>
Acked-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: avoid fortify-string warning in irdma_clr_wqes

Commit df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3") triggers a
warning for fortified memset():

In function 'fortify_memset_chk',
    inlined from 'irdma_clr_wqes' at drivers/infiniband/hw/irdma/uk.c:103:4:
include/linux/fortify-string.h:493:25: error: call to '__write_overflow_field' declared with attribute warning: detected write beyond size of field (1st parameter); maybe use struct_group()? [-Werror=attribute-warning]
  493 |                         __write_overflow_field(p_size_field, size);
      |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The problem here isthat the inner array only has four 8-byte elements, so
clearing 4096 bytes overflows that. As this structure is part of an outer
array, change the code to pass a pointer to the irdma_qp_quanta instead,
and change the size argument for readability, matching the comment above
it.

Fixes: 551c46edc769 ("RDMA/irdma: Add user/kernel shared libraries")
Link: https://lore.kernel.org/r/20230523111859.2197825-1-arnd@kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/mana_ib: Use v2 version of cfg_rx_steer_req to enable RX coalescing

With RX coalescing, one CQE entry can be used to indicate multiple packets
on the receive queue. This saves processing time and PCI bandwidth over
the CQ.

The MANA Ethernet driver also uses the v2 version of the protocol. It
doesn't use RX coalescing and its behavior is not changed.

Link: https://lore.kernel.org/r/1684045095-31228-1-git-send-email-longli@linuxonhyperv.com
Signed-off-by: Long Li <longli@microsoft.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Remove unnecessary checks

The NULL check inside bnxt_qplib_del_sgid() and bnxt_qplib_add_sgid()
always return false as the "sgid_tbl" inside "rdev->qplib_res" is a static
memory.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-8-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Return directly without goto jumps

When there is no cleanup to be done, return directly.  This will help
eliminating unnecessary local variables and goto labels.  This patch fixes
such occurrences in qplib_fp.c file.

Fixes: 37cb11acf1f7 ("RDMA/bnxt_re: Add SRQ support for Broadcom adapters")
Fixes: 159fb4ceacd7 ("RDMA/bnxt_re: introduce a function to allocate swq")
Link: https://lore.kernel.org/r/1684478897-12247-7-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Fix to remove an unnecessary log

During destroy_qp, driver sets the qp handle in the existing CQEs
belonging to the QP being destroyed to NULL. As a result, a poll_cq after
destroy_qp can report unnecessary messages.  Remove this noise from system
logs.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-6-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Remove a redundant check inside bnxt_re_update_gid

The NULL check inside bnxt_re_update_gid() always return false.  If
sgid_tbl->tbl is not allocated, then dev_init would have failed.

Fixes: 5fac5b1b297f ("RDMA/bnxt_re: Add vlan tag for untagged RoCE traffic when PFC is configured")
Link: https://lore.kernel.org/r/1684478897-12247-5-git-send-email-selvin.xavier@broadcom.com
Reviewed-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Reviewed-by: Damodharam Ammepalli <damodharam.ammepalli@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Use unique names while registering interrupts

bnxt_re currently uses the names "bnxt_qplib_creq" and "bnxt_qplib_nq-0"
while registering IRQs. There is no way to distinguish the IRQs of
different device ports when there are multiple IB devices registered.
This could make the scenarios worse where one want to pin IRQs of a device
port to certain CPUs.

Fixed the code to use unique names which has PCI BDF information while
registering interrupts like: "bnxt_re-nq-0@pci:0000:65:00.0" and
"bnxt_re-creq@pci:0000:65:00.1".

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-4-git-send-email-selvin.xavier@broadcom.com
Reviewed-by: Bhargava Chenna Marreddy <bhargava.marreddy@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Fix to remove unnecessary return labels

If there is no cleanup needed then just return directly.  This cleans up
the code and improve readability.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-3-git-send-email-selvin.xavier@broadcom.com
Reviewed-by: Kashyap Desai <kashyap.desai@broadcom.com>
Reviewed-by: Saravanan Vajravel <saravanan.vajravel@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/bnxt_re: Disable/kill tasklet only if it is enabled

When the ulp hook to start the IRQ fails because the rings are not
available, tasklets are not enabled. In this case when the driver is
unloaded, driver calls CREQ tasklet_kill. This causes an indefinite hang
as the tasklet is not enabled.

Driver shouldn't call tasklet_kill if it is not enabled. So using the
creq->requested and nq->requested flags to identify if both tasklets/irqs
are registered. Checking this flag while scheduling the tasklet from
ISR. Also, added a cleanup for disabling tasklet, in case request_irq
fails during start_irq.

Check for return value for bnxt_qplib_rcfw_start_irq and in case the
bnxt_qplib_rcfw_start_irq fails, return bnxt_re_start_irq without
attempting to start NQ IRQs.

Fixes: 1ac5a4047975 ("RDMA/bnxt_re: Add bnxt_re RoCE driver")
Link: https://lore.kernel.org/r/1684478897-12247-2-git-send-email-selvin.xavier@broadcom.com
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Fix comments about removed tasklets

The commit 9b4b7c1f9f54 ("RDMA/rxe: Add workqueue support for rxe tasks")
removed tasklets and replaced them with a workqueue, but relevant comments
are still remaining in the source code.

Fixes: 9b4b7c1f9f54 ("RDMA/rxe: Add workqueue support for rxe tasks")
Link: https://lore.kernel.org/r/20230518070027.942715-1-matsuda-daisuke@fujitsu.com
Signed-off-by: Daisuke Matsuda <matsuda-daisuke@fujitsu.com>
Reviewed-by: Bob Pearson <rpearsonhpe@gmail.com>
Acked-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Move iw device ops initialization

Move the initialization of the iw device ops to be under the declaration
of the irdma_iw_dev_ops.

Link: https://lore.kernel.org/r/20230515191142.413633-4-kheib@redhat.com
Signed-off-by: Kamal Heib <kheib@redhat.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Return void from irdma_init_rdma_device()

The return value from irdma_init_rdma_device() is always 0 - change it to
be void.

Link: https://lore.kernel.org/r/20230515191142.413633-3-kheib@redhat.com
Signed-off-by: Kamal Heib <kheib@redhat.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/irdma: Return void from irdma_init_iw_device()

The return value from irdma_init_iw_device() is always 0 - change it to be
void.

Link: https://lore.kernel.org/r/20230515191142.413633-2-kheib@redhat.com
Signed-off-by: Kamal Heib <kheib@redhat.com>
Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

RDMA/rxe: Add workqueue support for rxe tasks

Replace tasklets by work queues for the three main rxe tasklets:
rxe_requester, rxe_completer and rxe_responder.

work queues are a more modern way to process work from an IRQ and provide
more control over how that work is run for future patches.

Link: https://lore.kernel.org/r/20230428171321.5774-1-rpearsonhpe@gmail.com
Signed-off-by: Ian Ziemba <ian.ziemba@hpe.com>
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Reviewed-by: Daisuke Matsuda <matsuda-daisuke@fujitsu.com>
Tested-by: Daisuke Matsuda <matsuda-daisuke@fujitsu.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Linux 6.4-rc2

Merge tag 'cxl-fixes-6.4-rc2' of git://git./linux/kernel/git/cxl/cxl

Pull compute express link fixes from Dan Williams:

 - Fix a compilation issue with DEFINE_STATIC_SRCU() in the unit tests

 - Fix leaking kernel memory to a root-only sysfs attribute

* tag 'cxl-fixes-6.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl:
  cxl: Add missing return to cdat read error path
  tools/testing/cxl: Use DEFINE_STATIC_SRCU()

Merge tag 'parisc-for-6.4-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc architecture fixes from Helge Deller:

 - Fix encoding of swp_entry due to added SWP_EXCLUSIVE flag

 - Include reboot.h to avoid gcc-12 compiler warning

* tag 'parisc-for-6.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix encoding of swp_entry due to added SWP_EXCLUSIVE flag
  parisc: kexec: include reboot.h

Merge tag 'for-linus' of git://git.armlinux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:

 - fix unwinder for uleb128 case

 - fix kernel-doc warnings for HP Jornada 7xx

 - fix unbalanced stack on vfp success path

* tag 'for-linus' of git://git.armlinux.org.uk/~rmk/linux-arm:
  ARM: 9297/1: vfp: avoid unbalanced stack on 'success' return path
  ARM: 9296/1: HP Jornada 7XX: fix kernel-doc warnings
  ARM: 9295/1: unwind:fix unwind abort for uleb128 case

Merge tag 'locking_urgent_for_v6.4_rc2' of git://git./linux/kernel/git/tip/tip

Pull locking fix from Borislav Petkov:

 - Make sure __down_read_common() is always inlined so that the callers'
   names land in traceevents output and thus the blocked function can be
   identified

* tag 'locking_urgent_for_v6.4_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/rwsem: Add __always_inline annotation to __down_read_common() and inlined callers

Merge tag 'perf_urgent_for_v6.4_rc2' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Borislav Petkov:

 - Make sure the PEBS buffer is flushed before reprogramming the
   hardware so that the correct record sizes are used

 - Update the sample size for AMD BRS events

 - Fix a confusion with using the same on-stack struct with different
   events in the event processing path

* tag 'perf_urgent_for_v6.4_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/x86/intel/ds: Flush PEBS DS when changing PEBS_DATA_CFG
  perf/x86: Fix missing sample size update on AMD BRS
  perf/core: Fix perf_sample_data not properly initialized for different swevents in perf_tp_event()

Merge tag 'sched_urgent_for_v6.4_rc2' of git://git./linux/kernel/git/tip/tip

Pull scheduler fix from Borislav Petkov:

 - Fix a couple of kernel-doc warnings

* tag 'sched_urgent_for_v6.4_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched: fix cid_lock kernel-doc warnings

Merge tag 'x86_urgent_for_v6.4_rc2' of git://git./linux/kernel/git/tip/tip

Pull x86 fix from Borislav Petkov:

 - Add the required PCI IDs so that the generic SMN accesses provided by
   amd_nb.c work for drivers which switch to them. Add a PCI device ID
   to k10temp's table so that latter is loaded on such systems too

* tag 'x86_urgent_for_v6.4_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  hwmon: (k10temp) Add PCI ID for family 19, model 78h
  x86/amd_nb: Add PCI ID for family 19h model 78h

Merge tag 'timers_urgent_for_v6.4_rc2' of git://git./linux/kernel/git/tip/tip

Pull timer fix from Borislav Petkov:

 - Prevent CPU state corruption when an active clockevent broadcast
   device is replaced while the system is already in oneshot mode

* tag 'timers_urgent_for_v6.4_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  tick/broadcast: Make broadcast device replacement work correctly

Merge tag 'ext4_for_linus_stable' of git://git./linux/kernel/git/tytso/ext4

Pull ext4 fixes from Ted Ts'o:
 "Some ext4 bug fixes (mostly to address Syzbot reports)"

* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: bail out of ext4_xattr_ibody_get() fails for any reason
  ext4: add bounds checking in get_max_inline_xattr_value_size()
  ext4: add indication of ro vs r/w mounts in the mount message
  ext4: fix deadlock when converting an inline directory in nojournal mode
  ext4: improve error recovery code paths in __ext4_remount()
  ext4: improve error handling from ext4_dirhash()
  ext4: don't clear SB_RDONLY when remounting r/w until quota is re-enabled
  ext4: check iomap type only if ext4_iomap_begin() does not fail
  ext4: avoid a potential slab-out-of-bounds in ext4_group_desc_csum
  ext4: fix data races when using cached status extents
  ext4: avoid deadlock in fs reclaim with page writeback
  ext4: fix invalid free tracking in ext4_xattr_move_to_block()
  ext4: remove a BUG_ON in ext4_mb_release_group_pa()
  ext4: allow ext4_get_group_info() to fail
  ext4: fix lockdep warning when enabling MMP
  ext4: fix WARNING in mb_find_extent

Merge tag 'fbdev-for-6.4-rc2' of git://git./linux/kernel/git/deller/linux-fbdev

Pull fbdev fixes from Helge Deller:

 - use after free fix in imsttfb (Zheng Wang)

 - fix error handling in arcfb (Zongjie Li)

 - lots of whitespace cleanups (Thomas Zimmermann)

 - add 1920x1080 modedb entry (me)

* tag 'fbdev-for-6.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/linux-fbdev:
  fbdev: stifb: Fix info entry in sti_struct on error path
  fbdev: modedb: Add 1920x1080 at 60 Hz video mode
  fbdev: imsttfb: Fix use after free bug in imsttfb_probe
  fbdev: vfb: Remove trailing whitespaces
  fbdev: valkyriefb: Remove trailing whitespaces
  fbdev: stifb: Remove trailing whitespaces
  fbdev: sa1100fb: Remove trailing whitespaces
  fbdev: platinumfb: Remove trailing whitespaces
  fbdev: p9100: Remove trailing whitespaces
  fbdev: maxinefb: Remove trailing whitespaces
  fbdev: macfb: Remove trailing whitespaces
  fbdev: hpfb: Remove trailing whitespaces
  fbdev: hgafb: Remove trailing whitespaces
  fbdev: g364fb: Remove trailing whitespaces
  fbdev: controlfb: Remove trailing whitespaces
  fbdev: cg14: Remove trailing whitespaces
  fbdev: atmel_lcdfb: Remove trailing whitespaces
  fbdev: 68328fb: Remove trailing whitespaces
  fbdev: arcfb: Fix error handling in arcfb_probe()

Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fix from James Bottomley:
 "A single small fix for the UFS driver to fix a power management
  failure"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: ufs: core: Fix I/O hang that occurs when BKOPS fails in W-LUN suspend

parisc: Fix encoding of swp_entry due to added SWP_EXCLUSIVE flag

Fix the __swp_offset() and __swp_entry() macros due to commit 6d239fc78c0b
("parisc/mm: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE") which introduced the
SWP_EXCLUSIVE flag by reusing the _PAGE_ACCESSED flag.

Reported-by: Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de>
Tested-by: Christoph Biedl <linux-kernel.bfrz@manchmal.in-ulm.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Helge Deller <deller@gmx.de>
Fixes: 6d239fc78c0b ("parisc/mm: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE")
Cc: <stable@vger.kernel.org> # v6.3+

ext4: bail out of ext4_xattr_ibody_get() fails for any reason

In ext4_update_inline_data(), if ext4_xattr_ibody_get() fails for any
reason, it's best if we just fail as opposed to stumbling on,
especially if the failure is EFSCORRUPTED.

Cc: stable@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: add bounds checking in get_max_inline_xattr_value_size()

Normally the extended attributes in the inode body would have been
checked when the inode is first opened, but if someone is writing to
the block device while the file system is mounted, it's possible for
the inode table to get corrupted.  Add bounds checking to avoid
reading beyond the end of allocated memory if this happens.

Reported-by: syzbot+1966db24521e5f6e23f7@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?extid=1966db24521e5f6e23f7
Cc: stable@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: add indication of ro vs r/w mounts in the mount message

Whether the file system is mounted read-only or read/write is more
important than the quota mode, which we are already printing.  Add the
ro vs r/w indication since this can be helpful in debugging problems
from the console log.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: fix deadlock when converting an inline directory in nojournal mode

In no journal mode, ext4_finish_convert_inline_dir() can self-deadlock
by calling ext4_handle_dirty_dirblock() when it already has taken the
directory lock.  There is a similar self-deadlock in
ext4_incvert_inline_data_nolock() for data files which we'll fix at
the same time.

A simple reproducer demonstrating the problem:

    mke2fs -Fq -t ext2 -O inline_data -b 4k /dev/vdc 64
    mount -t ext4 -o dirsync /dev/vdc /vdc
    cd /vdc
    mkdir file0
    cd file0
    touch file0
    touch file1
    attr -s BurnSpaceInEA -V abcde .
    touch supercalifragilisticexpialidocious

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230507021608.1290720-1-tytso@mit.edu
Reported-by: syzbot+91dccab7c64e2850a4e5@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=ba84cc80a9491d65416bc7877e1650c87530fe8a
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: improve error recovery code paths in __ext4_remount()

If there are failures while changing the mount options in
__ext4_remount(), we need to restore the old mount options.

This commit fixes two problem.  The first is there is a chance that we
will free the old quota file names before a potential failure leading
to a use-after-free.  The second problem addressed in this commit is
if there is a failed read/write to read-only transition, if the quota
has already been suspended, we need to renable quota handling.

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230506142419.984260-2-tytso@mit.edu
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: improve error handling from ext4_dirhash()

The ext4_dirhash() will *almost* never fail, especially when the hash
tree feature was first introduced.  However, with the addition of
support of encrypted, casefolded file names, that function can most
certainly fail today.

So make sure the callers of ext4_dirhash() properly check for
failures, and reflect the errors back up to their callers.

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230506142419.984260-1-tytso@mit.edu
Reported-by: syzbot+394aa8a792cb99dbc837@syzkaller.appspotmail.com
Reported-by: syzbot+344aaa8697ebd232bfc8@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=db56459ea4ac4a676ae4b4678f633e55da005a9b
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: don't clear SB_RDONLY when remounting r/w until quota is re-enabled

When a file system currently mounted read/only is remounted
read/write, if we clear the SB_RDONLY flag too early, before the quota
is initialized, and there is another process/thread constantly
attempting to create a directory, it's possible to trigger the

WARN_ON_ONCE(dquot_initialize_needed(inode));

in ext4_xattr_block_set(), with the following stack trace:

   WARNING: CPU: 0 PID: 5338 at fs/ext4/xattr.c:2141 ext4_xattr_block_set+0x2ef2/0x3680
   RIP: 0010:ext4_xattr_block_set+0x2ef2/0x3680 fs/ext4/xattr.c:2141
   Call Trace:
    ext4_xattr_set_handle+0xcd4/0x15c0 fs/ext4/xattr.c:2458
    ext4_initxattrs+0xa3/0x110 fs/ext4/xattr_security.c:44
    security_inode_init_security+0x2df/0x3f0 security/security.c:1147
    __ext4_new_inode+0x347e/0x43d0 fs/ext4/ialloc.c:1324
    ext4_mkdir+0x425/0xce0 fs/ext4/namei.c:2992
    vfs_mkdir+0x29d/0x450 fs/namei.c:4038
    do_mkdirat+0x264/0x520 fs/namei.c:4061
    __do_sys_mkdirat fs/namei.c:4076 [inline]
    __se_sys_mkdirat fs/namei.c:4074 [inline]
    __x64_sys_mkdirat+0x89/0xa0 fs/namei.c:4074

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230506142419.984260-1-tytso@mit.edu
Reported-by: syzbot+6385d7d3065524c5ca6d@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=6513f6cb5cd6b5fc9f37e3bb70d273b94be9c34c
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: check iomap type only if ext4_iomap_begin() does not fail

When ext4_iomap_overwrite_begin() calls ext4_iomap_begin() map blocks may
fail for some reason (e.g. memory allocation failure, bare disk write), and
later because "iomap->type ! = IOMAP_MAPPED" triggers WARN_ON(). When ext4
iomap_begin() returns an error, it is normal that the type of iomap->type
may not match the expectation. Therefore, we only determine if iomap->type
is as expected when ext4_iomap_begin() is executed successfully.

Cc: stable@kernel.org
Reported-by: syzbot+08106c4b7d60702dbc14@syzkaller.appspotmail.com
Link: https://lore.kernel.org/all/00000000000015760b05f9b4eee9@google.com
Signed-off-by: Baokun Li <libaokun1@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230505132429.714648-1-libaokun1@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: avoid a potential slab-out-of-bounds in ext4_group_desc_csum

When modifying the block device while it is mounted by the filesystem,
syzbot reported the following:

BUG: KASAN: slab-out-of-bounds in crc16+0x206/0x280 lib/crc16.c:58
Read of size 1 at addr ffff888075f5c0a8 by task syz-executor.2/15586

CPU: 1 PID: 15586 Comm: syz-executor.2 Not tainted 6.2.0-rc5-syzkaller-00205-gc96618275234 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/12/2023
Call Trace:
 <TASK>
 __dump_stack lib/dump_stack.c:88 [inline]
 dump_stack_lvl+0x1b1/0x290 lib/dump_stack.c:106
 print_address_description+0x74/0x340 mm/kasan/report.c:306
 print_report+0x107/0x1f0 mm/kasan/report.c:417
 kasan_report+0xcd/0x100 mm/kasan/report.c:517
 crc16+0x206/0x280 lib/crc16.c:58
 ext4_group_desc_csum+0x81b/0xb20 fs/ext4/super.c:3187
 ext4_group_desc_csum_set+0x195/0x230 fs/ext4/super.c:3210
 ext4_mb_clear_bb fs/ext4/mballoc.c:6027 [inline]
 ext4_free_blocks+0x191a/0x2810 fs/ext4/mballoc.c:6173
 ext4_remove_blocks fs/ext4/extents.c:2527 [inline]
 ext4_ext_rm_leaf fs/ext4/extents.c:2710 [inline]
 ext4_ext_remove_space+0x24ef/0x46a0 fs/ext4/extents.c:2958
 ext4_ext_truncate+0x177/0x220 fs/ext4/extents.c:4416
 ext4_truncate+0xa6a/0xea0 fs/ext4/inode.c:4342
 ext4_setattr+0x10c8/0x1930 fs/ext4/inode.c:5622
 notify_change+0xe50/0x1100 fs/attr.c:482
 do_truncate+0x200/0x2f0 fs/open.c:65
 handle_truncate fs/namei.c:3216 [inline]
 do_open fs/namei.c:3561 [inline]
 path_openat+0x272b/0x2dd0 fs/namei.c:3714
 do_filp_open+0x264/0x4f0 fs/namei.c:3741
 do_sys_openat2+0x124/0x4e0 fs/open.c:1310
 do_sys_open fs/open.c:1326 [inline]
 __do_sys_creat fs/open.c:1402 [inline]
 __se_sys_creat fs/open.c:1396 [inline]
 __x64_sys_creat+0x11f/0x160 fs/open.c:1396
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x3d/0xb0 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f72f8a8c0c9
Code: 28 00 00 00 75 05 48 83 c4 28 c3 e8 f1 19 00 00 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 b8 ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007f72f97e3168 EFLAGS: 00000246 ORIG_RAX: 0000000000000055
RAX: ffffffffffffffda RBX: 00007f72f8bac050 RCX: 00007f72f8a8c0c9
RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000020000280
RBP: 00007f72f8ae7ae9 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
R13: 00007ffd165348bf R14: 00007f72f97e3300 R15: 0000000000022000

Replace
le16_to_cpu(sbi->s_es->s_desc_size)
with
sbi->s_desc_size

It reduces ext4's compiled text size, and makes the code more efficient
(we remove an extra indirect reference and a potential byte
swap on big endian systems), and there is no downside. It also avoids the
potential KASAN / syzkaller failure, as a bonus.

Reported-by: syzbot+fc51227e7100c9294894@syzkaller.appspotmail.com
Reported-by: syzbot+8785e41224a3afd04321@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=70d28d11ab14bd7938f3e088365252aa923cff42
Link: https://syzkaller.appspot.com/bug?id=b85721b38583ecc6b5e72ff524c67302abbc30f3
Link: https://lore.kernel.org/all/000000000000ece18705f3b20934@google.com/
Fixes: 717d50e4971b ("Ext4: Uninitialized Block Groups")
Cc: stable@vger.kernel.org
Signed-off-by: Tudor Ambarus <tudor.ambarus@linaro.org>
Link: https://lore.kernel.org/r/20230504121525.3275886-1-tudor.ambarus@linaro.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: fix data races when using cached status extents

When using cached extent stored in extent status tree in tree->cache_es
another process holding ei->i_es_lock for reading can be racing with us
setting new value of tree->cache_es. If the compiler would decide to
refetch tree->cache_es at an unfortunate moment, it could result in a
bogus in_range() check. Fix the possible race by using READ_ONCE() when
using tree->cache_es only under ei->i_es_lock for reading.

Cc: stable@kernel.org
Reported-by: syzbot+4a03518df1e31b537066@syzkaller.appspotmail.com
Link: https://lore.kernel.org/all/000000000000d3b33905fa0fd4a6@google.com
Suggested-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230504125524.10802-1-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: avoid deadlock in fs reclaim with page writeback

Ext4 has a filesystem wide lock protecting ext4_writepages() calls to
avoid races with switching of journalled data flag or inode format. This
lock can however cause a deadlock like:

CPU0                            CPU1

ext4_writepages()
  percpu_down_read(sbi->s_writepages_rwsem);
                                ext4_change_inode_journal_flag()
                                  percpu_down_write(sbi->s_writepages_rwsem);
                                    - blocks, all readers block from now on
  ext4_do_writepages()
    ext4_init_io_end()
      kmem_cache_zalloc(io_end_cachep, GFP_KERNEL)
        fs_reclaim frees dentry...
          dentry_unlink_inode()
            iput() - last ref =>
              iput_final() - inode dirty =>
                write_inode_now()...
                  ext4_writepages() tries to acquire sbi->s_writepages_rwsem
                    and blocks forever

Make sure we cannot recurse into filesystem reclaim from writeback code
to avoid the deadlock.

Reported-by: syzbot+6898da502aef574c5f8a@syzkaller.appspotmail.com
Link: https://lore.kernel.org/all/0000000000004c66b405fa108e27@google.com
Fixes: c8585c6fcaf2 ("ext4: fix races between changing inode journal mode and ext4_writepages")
CC: stable@vger.kernel.org
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230504124723.20205-1-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: fix invalid free tracking in ext4_xattr_move_to_block()

In ext4_xattr_move_to_block(), the value of the extended attribute
which we need to move to an external block may be allocated by
kvmalloc() if the value is stored in an external inode.  So at the end
of the function the code tried to check if this was the case by
testing entry->e_value_inum.

However, at this point, the pointer to the xattr entry is no longer
valid, because it was removed from the original location where it had
been stored.  So we could end up calling kvfree() on a pointer which
was not allocated by kvmalloc(); or we could also potentially leak
memory by not freeing the buffer when it should be freed.  Fix this by
storing whether it should be freed in a separate variable.

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230430160426.581366-1-tytso@mit.edu
Link: https://syzkaller.appspot.com/bug?id=5c2aee8256e30b55ccf57312c16d88417adbd5e1
Link: https://syzkaller.appspot.com/bug?id=41a6b5d4917c0412eb3b3c3c604965bed7d7420b
Reported-by: syzbot+64b645917ce07d89bde5@syzkaller.appspotmail.com
Reported-by: syzbot+0d042627c4f2ad332195@syzkaller.appspotmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: remove a BUG_ON in ext4_mb_release_group_pa()

If a malicious fuzzer overwrites the ext4 superblock while it is
mounted such that the s_first_data_block is set to a very large
number, the calculation of the block group can underflow, and trigger
a BUG_ON check.  Change this to be an ext4_warning so that we don't
crash the kernel.

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230430154311.579720-3-tytso@mit.edu
Reported-by: syzbot+e2efa3efc15a1c9e95c3@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=69b28112e098b070f639efb356393af3ffec4220
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

ext4: allow ext4_get_group_info() to fail

Previously, ext4_get_group_info() would treat an invalid group number
as BUG(), since in theory it should never happen.  However, if a
malicious attaker (or fuzzer) modifies the superblock via the block
device while it is the file system is mounted, it is possible for
s_first_data_block to get set to a very large number.  In that case,
when calculating the block group of some block number (such as the
starting block of a preallocation region), could result in an
underflow and very large block group number.  Then the BUG_ON check in
ext4_get_group_info() would fire, resutling in a denial of service
attack that can be triggered by root or someone with write access to
the block device.

For a quality of implementation perspective, it's best that even if
the system administrator does something that they shouldn't, that it
will not trigger a BUG.  So instead of BUG'ing, ext4_get_group_info()
will call ext4_error and return NULL.  We also add fallback code in
all of the callers of ext4_get_group_info() that it might NULL.

Also, since ext4_get_group_info() was already borderline to be an
inline function, un-inline it.  The results in a next reduction of the
compiled text size of ext4 by roughly 2k.

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230430154311.579720-2-tytso@mit.edu
Reported-by: syzbot+e2efa3efc15a1c9e95c3@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=69b28112e098b070f639efb356393af3ffec4220
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>

Merge tag 'block-6.4-2023-05-13' of git://git.kernel.dk/linux

Pull block fixes from Jens Axboe:
 "Just a few minor fixes for drivers, and a deletion of a file that is
  woefully out-of-date these days"

* tag 'block-6.4-2023-05-13' of git://git.kernel.dk/linux:
  Documentation/block: drop the request.rst file
  ublk: fix command op code check
  block/rnbd: replace REQ_OP_FLUSH with REQ_OP_WRITE
  nbd: Fix debugfs_create_dir error checking

cxl: Add missing return to cdat read error path

Add a return to the error path when cxl_cdat_read_table() fails. Current
code continues with the table pointer points to freed memory.

Fixes: 7a877c923995 ("cxl/pci: Simplify CDAT retrieval error path")
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
Link: https://lore.kernel.org/r/168382793506.3510737.4792518576623749076.stgit@djiang5-mobl3
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

tools/testing/cxl: Use DEFINE_STATIC_SRCU()

Starting with commit:

95433f726301 ("srcu: Begin offloading srcu_struct fields to srcu_update")

...it is no longer possible to do:

static DEFINE_SRCU(x)

Switch to DEFINE_STATIC_SRCU(x) to fix:

tools/testing/cxl/test/mock.c:22:1: error: duplicate ‘static’
   22 | static DEFINE_SRCU(cxl_mock_srcu);
      | ^~~~~~

Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Link: https://lore.kernel.org/r/168392709546.1135523.10424917245934547117.stgit@dwillia2-xfh.jf.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

x86/retbleed: Fix return thunk alignment

SYM_FUNC_START_LOCAL_NOALIGN() adds an endbr leading to this layout
(leaving only the last 2 bytes of the address):

  3bff <zen_untrain_ret>:
  3bff:       f3 0f 1e fa             endbr64
  3c03:       f6                      test   $0xcc,%bl

  3c04 <__x86_return_thunk>:
  3c04:       c3                      ret
  3c05:       cc                      int3
  3c06:       0f ae e8                lfence

However, "the RET at __x86_return_thunk must be on a 64 byte boundary,
for alignment within the BTB."

Use SYM_START instead.

Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: <stable@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge tag 'for-6.4-rc1-tag' of git://git./linux/kernel/git/kdave/linux

Pull more btrfs fixes from David Sterba:

 - fix incorrect number of bitmap entries for space cache if loading is
   interrupted by some error

 - fix backref walking, this breaks a mode of LOGICAL_INO_V2 ioctl that
   is used in deduplication tools

 - zoned mode fixes:
      - properly finish zone reserved for relocation
      - correctly calculate super block zone end on ZNS
      - properly initialize new extent buffer for redirty

 - make mount option clear_cache work with block-group-tree, to rebuild
   free-space-tree instead of temporarily disabling it that would lead
   to a forced read-only mount

 - fix alignment check for offset when printing extent item

* tag 'for-6.4-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: make clear_cache mount option to rebuild FST without disabling it
  btrfs: zero the buffer before marking it dirty in btrfs_redirty_list_add
  btrfs: zoned: fix full zone super block reading on ZNS
  btrfs: zoned: zone finish data relocation BG with last IO
  btrfs: fix backref walking not returning all inode refs
  btrfs: fix space cache inconsistency after error loading it from disk
  btrfs: print-tree: parent bytenr must be aligned to sector size

Merge tag '6.4-rc1-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs client fixes from Steve French:

 - fix for copy_file_range bug for very large files that are multiples
   of rsize

 - do not ignore "isolated transport" flag if set on share

 - set rasize default better

 - three fixes related to shutdown and freezing (fixes 4 xfstests, and
   closes deferred handles faster in some places that were missed)

* tag '6.4-rc1-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: release leases for deferred close handles when freezing
  smb3: fix problem remounting a share after shutdown
  SMB3: force unmount was failing to close deferred close files
  smb3: improve parallel reads of large files
  do not reuse connection if share marked as isolated
  cifs: fix pcchunk length type in smb2_copychunk_range

Merge tag 'vfs/v6.4-rc1/pipe' of gitolite.pub/scm/linux/kernel/git/vfs/vfs

Pull vfs fix from Christian Brauner:
 "During the pipe nonblock rework the check for both O_NONBLOCK and
  IOCB_NOWAIT was dropped. Both checks need to be performed to ensure
  that files without O_NONBLOCK but IOCB_NOWAIT don't block when writing
  to or reading from a pipe.

  This just contains the fix adding the check for IOCB_NOWAIT back in"

* tag 'vfs/v6.4-rc1/pipe' of gitolite.kernel.org:pub/scm/linux/kernel/git/vfs/vfs:
  pipe: check for IOCB_NOWAIT alongside O_NONBLOCK

Merge tag 'io_uring-6.4-2023-05-12' of git://git.kernel.dk/linux

Pull io_uring fix from Jens Axboe:
 "Just a single fix making io_uring_sqe_cmd() available regardless of
  CONFIG_IO_URING, fixing a regression introduced during the merge
  window if nvme was selected but io_uring was not"

* tag 'io_uring-6.4-2023-05-12' of git://git.kernel.dk/linux:
  io_uring: make io_uring_sqe_cmd() unconditionally available

Merge tag 'riscv-for-linus-6.4-rc2' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fix from Palmer Dabbelt:
 "Just a single fix this week for a build issue. That'd usually be a
  good sign, but we've started to get some reports of boot failures on
  some hardware/bootloader configurations. Nothing concrete yet, but
  I've got a funny feeling that's where much of the bug hunting is going
  right now.

  Nothing's reproducing on my end, though, and this fixes some pretty
  concrete issues so I figured there's no reason to delay it:

   - a fix to the linker script to avoid orpahaned sections in
     kernel/pi"

* tag 'riscv-for-linus-6.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: Fix orphan section warnings caused by kernel/pi

Documentation/block: drop the request.rst file

Documentation/block/request.rst is outdated and should be removed.
Also delete its entry in the block/index.rst file.

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-doc@vger.kernel.org
Link: https://lore.kernel.org/r/20230507182606.12647-1-rdunlap@infradead.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

pipe: check for IOCB_NOWAIT alongside O_NONBLOCK

Pipe reads or writes need to enable nonblocking attempts, if either
O_NONBLOCK is set on the file, or IOCB_NOWAIT is set in the iocb being
passed in. The latter isn't currently true, ensure we check for both
before waiting on data or space.

Fixes: afed6271f5b0 ("pipe: set FMODE_NOWAIT on pipes")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Message-Id: <e5946d67-4e5e-b056-ba80-656bab12d9f6@kernel.dk>
Signed-off-by: Christian Brauner <brauner@kernel.org>

ublk: fix command op code check

In case of CONFIG_BLKDEV_UBLK_LEGACY_OPCODES, type of cmd opcode could
be 0 or 'u'; and type can only be 'u' if CONFIG_BLKDEV_UBLK_LEGACY_OPCODES
isn't set.

So fix the wrong check.

Fixes: 2d786e66c966 ("block: ublk: switch to ioctl command encoding")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230505153142.1258336-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/rnbd: replace REQ_OP_FLUSH with REQ_OP_WRITE

Since flush bios are implemented as writes with no data and
the preflush flag per Christoph's comment [1].

And we need to change it in rnbd accordingly. Otherwise, I
got splatting when create fs from rnbd client.

[  464.028545] ------------[ cut here ]------------
[  464.028553] WARNING: CPU: 0 PID: 65 at block/blk-core.c:751 submit_bio_noacct+0x32c/0x5d0
[ ... ]
[  464.028668] CPU: 0 PID: 65 Comm: kworker/0:1H Tainted: G           OE      6.4.0-rc1 #9
[  464.028671] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.15.0-0-g2dd4b9b-rebuilt.opensuse.org 04/01/2014
[  464.028673] Workqueue: ib-comp-wq ib_cq_poll_work [ib_core]
[  464.028717] RIP: 0010:submit_bio_noacct+0x32c/0x5d0
[  464.028720] Code: 03 0f 85 51 fe ff ff 48 8b 43 18 8b 88 04 03 00 00 85 c9 0f 85 3f fe ff ff e9 be fd ff ff 0f b6 d0 3c 0d 74 26 83 fa 01 74 21 <0f> 0b b8 0a 00 00 00 e9 56 fd ff ff 4c 89 e7 e8 70 a1 03 00 84 c0
[  464.028722] RSP: 0018:ffffaf3680b57c68 EFLAGS: 00010202
[  464.028724] RAX: 0000000000060802 RBX: ffffa09dcc18bf00 RCX: 0000000000000000
[  464.028726] RDX: 0000000000000002 RSI: 0000000000000000 RDI: ffffa09dde081d00
[  464.028727] RBP: ffffaf3680b57c98 R08: ffffa09dde081d00 R09: ffffa09e38327200
[  464.028729] R10: 0000000000000000 R11: 0000000000000000 R12: ffffa09dde081d00
[  464.028730] R13: ffffa09dcb06e1e8 R14: 0000000000000000 R15: 0000000000200000
[  464.028733] FS:  0000000000000000(0000) GS:ffffa09e3bc00000(0000) knlGS:0000000000000000
[  464.028735] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  464.028736] CR2: 000055a4e8206c40 CR3: 0000000119f06000 CR4: 00000000003506f0
[  464.028738] Call Trace:
[  464.028740]  <TASK>
[  464.028746]  submit_bio+0x1b/0x80
[  464.028748]  rnbd_srv_rdma_ev+0x50d/0x10c0 [rnbd_server]
[  464.028754]  ? percpu_ref_get_many.constprop.0+0x55/0x140 [rtrs_server]
[  464.028760]  ? __this_cpu_preempt_check+0x13/0x20
[  464.028769]  process_io_req+0x1dc/0x450 [rtrs_server]
[  464.028775]  rtrs_srv_inv_rkey_done+0x67/0xb0 [rtrs_server]
[  464.028780]  __ib_process_cq+0xbc/0x1f0 [ib_core]
[  464.028793]  ib_cq_poll_work+0x2b/0xa0 [ib_core]
[  464.028804]  process_one_work+0x2a9/0x580

[1]. https://lore.kernel.org/all/ZFHgefWofVt24tRl@infradead.org/

Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://lore.kernel.org/r/20230512034631.28686-1-guoqing.jiang@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nbd: Fix debugfs_create_dir error checking

The debugfs_create_dir function returns ERR_PTR in case of error, and the
only correct way to check if an error occurred is 'IS_ERR' inline function.
This patch will replace the null-comparison with IS_ERR.

Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
Link: https://lore.kernel.org/r/20230512130533.98709-1-ivan.orlov0322@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'firewire-fixes-6.4-rc2' of git://git./linux/kernel/git/ieee1394/linux1394

Pull firewire fix from Takashi Sakamoto:

 - fix early release of request packet

* tag 'firewire-fixes-6.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394:
  firewire: net: fix unexpected release of object for asynchronous request packet

fbdev: stifb: Fix info entry in sti_struct on error path

Minor fix to reset the info field to NULL in case of error.

Signed-off-by: Helge Deller <deller@gmx.de>

fbdev: modedb: Add 1920x1080 at 60 Hz video mode

Add typical resolution for Full-HD monitors.

Signed-off-by: Helge Deller <deller@gmx.de>

Merge tag 'drm-fixes-2023-05-12' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "About the usual for this stage, bunch of amdgpu, a few i915 and a
  scattering of fixes across the board"

  dsc:
   - macro fixes

  simplefb:
   - fix VESA format

  scheduler:
   - timeout handling fix

  fbdev:
   - avoid potential out-of-bounds access in generic fbdev emulation

  ast:
   - improve AST2500+ compat on ARM

  mipi-dsi:
   - small mipi-dsi fix

  amdgpu:
   - VCN3 fixes
   - APUs always support PCI atomics
   - legacy power management fixes
   - DCN 3.1.4 fix
   - DCFCLK fix
   - fix several RAS irq refcount mismatches
   - GPU Reset fix
   - GFX 11.0.4 fix

  i915:
   - taint kernel when force_probe is used
   - NULL deref and div-by-zero fixes for display
   - GuC error capture fix for Xe devices"

* tag 'drm-fixes-2023-05-12' of git://anongit.freedesktop.org/drm/drm: (24 commits)
  drm/amdgpu: change gfx 11.0.4 external_id range
  drm/amdgpu/jpeg: Remove harvest checking for JPEG3
  drm/amdgpu/gfx: disable gfx9 cp_ecc_error_irq only when enabling legacy gfx ras
  drm/amd/pm: avoid potential UBSAN issue on legacy asics
  drm/i915: taint kernel when force probing unsupported devices
  drm/i915/dp: prevent potential div-by-zero
  drm/i915: Fix NULL ptr deref by checking new_crtc_state
  drm/i915/guc: Don't capture Gen8 regs on Xe devices
  drm/amdgpu: disable sdma ecc irq only when sdma RAS is enabled in suspend
  drm/amdgpu: Fix vram recover doesn't work after whole GPU reset (v2)
  drm/amdgpu: drop gfx_v11_0_cp_ecc_error_irq_funcs
  drm/amd/display: Enforce 60us prefetch for 200Mhz DCFCLK modes
  drm/amd/display: Add symclk workaround during disable link output
  drm/amd/pm: parse pp_handle under appropriate conditions
  drm/amdgpu: set gfx9 onwards APU atomics support to be true
  drm/amdgpu/nv: update VCN 3 max HEVC encoding resolution
  drm/sched: Check scheduler work queue before calling timeout handling
  drm/mipi-dsi: Set the fwnode for mipi_dsi_device
  drm/nouveau/disp: More DP_RECEIVER_CAP_SIZE array fixes
  drm/dsc: fix DP_DSC_MAX_BPP_DELTA_* macro values
  ...

Merge tag 'xfs-6.4-rc1-fixes' of git://git./fs/xfs/xfs-linux

Pull xfs bug fixes from Dave Chinner:
 "Largely minor bug fixes and cleanups, th emost important of which are
  probably the fixes for regressions in the extent allocation code:

   - fixes for inode garbage collection shutdown racing with work queue
     updates

   - ensure inodegc workers run on the CPU they are supposed to

   - disable counter scrubbing until we can exclusively freeze the
     filesystem from the kernel

   - regression fixes for new allocation related bugs

   - a couple of minor cleanups"

* tag 'xfs-6.4-rc1-fixes' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
  xfs: fix xfs_inodegc_stop racing with mod_delayed_work
  xfs: disable reaping in fscounters scrub
  xfs: check that per-cpu inodegc workers actually run on that cpu
  xfs: explicitly specify cpu when forcing inodegc delayed work to run immediately
  xfs: fix negative array access in xfs_getbmap
  xfs: don't allocate into the data fork for an unshare request
  xfs: flush dirty data and drain directios before scrubbing cow fork
  xfs: set bnobt/cntbt numrecs correctly when formatting new AGs
  xfs: don't unconditionally null args->pag in xfs_bmap_btalloc_at_eof

fbdev: imsttfb: Fix use after free bug in imsttfb_probe

A use-after-free bug may occur if init_imstt invokes framebuffer_release
and free the info ptr. The caller, imsttfb_probe didn't notice that and
still keep the ptr as private data in pdev.

If we remove the driver which will call imsttfb_remove to make cleanup,
UAF happens.

Fix it by return error code if bad case happens in init_imstt.

Signed-off-by: Zheng Wang <zyytlz.wz@163.com>
Signed-off-by: Helge Deller <deller@gmx.de>

Merge tag 'amd-drm-fixes-6.4-2023-05-11' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amdgpu:
- VCN3 fixes
- APUs always support PCI atomics
- Legacy power management fixes
- DCN 3.1.4 fix
- DCFCLK fix
- Fix several RAS irq refcount mismatches
- GPU Reset fix
- GFX 11.0.4 fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230511141755.7896-1-alexander.deucher@amd.com