platform/kernel/linux-starfive.git
3 years agoRDMA/hns: Encapsulate flushing CQE as a function
Wenpeng Liang [Fri, 18 Jun 2021 10:10:18 +0000 (18:10 +0800)]
RDMA/hns: Encapsulate flushing CQE as a function

The process of flushing CQE can be encapsultated into a function, which
can reduce duplicate code.

Link: https://lore.kernel.org/r/1624011020-16992-9-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Modify function return value type
Yangyang Li [Fri, 18 Jun 2021 10:10:17 +0000 (18:10 +0800)]
RDMA/hns: Modify function return value type

hns_roce_init_qp_table() will only return 0, because this function does
not need to return a value, so it is modified to void type.

Link: https://lore.kernel.org/r/1624011020-16992-8-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Clean definitions of EQC structure
Xi Wang [Fri, 18 Jun 2021 10:10:16 +0000 (18:10 +0800)]
RDMA/hns: Clean definitions of EQC structure

Remove unused members in EQ context structure.

Fixes: 782832f25404 ("RDMA/hns: Simplify the function config_eqc()")
Link: https://lore.kernel.org/r/1624011020-16992-7-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Delete unnecessary branch of hns_roce_v2_query_qp
Yangyang Li [Fri, 18 Jun 2021 10:10:15 +0000 (18:10 +0800)]
RDMA/hns: Delete unnecessary branch of hns_roce_v2_query_qp

When query_qp is called by userspace, max_send_wr and max_send_sge are set
to 0 by the kernel driver. However, the userspace does not use these two
return values from the kernel driver, but uses its own calculated values.
So there is no need for special treatment.

Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/1624011020-16992-6-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Add member assignments for qp_init_attr
Yangyang Li [Fri, 18 Jun 2021 10:10:14 +0000 (18:10 +0800)]
RDMA/hns: Add member assignments for qp_init_attr

Some kernel ULPs need to use the return value of qp_init_attr, so add
member assignments for qp_init_attr.

Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC")
Link: https://lore.kernel.org/r/1624011020-16992-5-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Fix some print issues
Yixing Liu [Fri, 18 Jun 2021 10:10:13 +0000 (18:10 +0800)]
RDMA/hns: Fix some print issues

Remove redundant print and fix a character type mismatch.

Fixes: 0e0ab04b5bbe ("RDMA/hns: Refactor the MTR creation flow")
Link: https://lore.kernel.org/r/1624011020-16992-4-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Fix uninitialized variable
Yixing Liu [Fri, 18 Jun 2021 10:10:12 +0000 (18:10 +0800)]
RDMA/hns: Fix uninitialized variable

A random value will be returned if the condition below is not met, so it
needs to be initialized.

Fixes: 9ea9a53ea93b ("RDMA/hns: Add mapped page count checking for MTR")
Link: https://lore.kernel.org/r/1624011020-16992-3-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Force rewrite inline flag of WQE
Lang Cheng [Fri, 18 Jun 2021 10:10:11 +0000 (18:10 +0800)]
RDMA/hns: Force rewrite inline flag of WQE

When a non-inline WR reuses a WQE that was used for inline last time, the
remaining inline flag should be cleared.

Fixes: 62490fd5a865 ("RDMA/hns: Avoid unnecessary memset on WQEs in post_send")
Link: https://lore.kernel.org/r/1624011020-16992-2-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoMerge branch 'mlx5_realtime_ts' into rdma.git for-next
Jason Gunthorpe [Tue, 22 Jun 2021 18:08:39 +0000 (15:08 -0300)]
Merge branch 'mlx5_realtime_ts' into rdma.git for-next

Aharon Landau says:

====================
In case device supports only real-time timestamp, the kernel will fail to
create QP despite rdma-core requested such timestamp type.

It is because device returns free-running timestamp, and the conversion
from free-running to real-time is performed in the user space.

This series fixes it, by returning real-time timestamp.
====================

* mlx5_realtime_ts:
  RDMA/mlx5: Support real-time timestamp directly from the device
  RDMA/mlx5: Refactor get_ts_format functions to simplify code

Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoMerge tag 'v5.13-rc7' into rdma.git for-next
Jason Gunthorpe [Tue, 22 Jun 2021 17:42:52 +0000 (14:42 -0300)]
Merge tag 'v5.13-rc7' into rdma.git for-next

Linux 5.13-rc7

Needed for dependencies in following patches. Merge conflict in rxe_cmop.c
resolved by compining both patches.

Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/mlx5: Support real-time timestamp directly from the device
Aharon Landau [Wed, 16 Jun 2021 07:57:39 +0000 (10:57 +0300)]
RDMA/mlx5: Support real-time timestamp directly from the device

Currently, if the user asks for a real-time timestamp, the device will
return a free-running one, and the timestamp will be translated to
real-time in the user-space.

When the device supports only real-time timestamp and not free-running,
the creation of the QP will fail even though the user needs supported the
real-time one. To prevent this, we will return the real-time timestamp
directly from the device.

Link: https://lore.kernel.org/r/c6cfc8e6f038575c5c2de6505830f7e74e4de80d.1623829775.git.leonro@nvidia.com
Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/core: Use flexible array for mad data
Kees Cook [Wed, 16 Jun 2021 20:26:15 +0000 (13:26 -0700)]
RDMA/core: Use flexible array for mad data

In preparation for FORTIFY_SOURCE performing compile-time and run-time
field bounds checking for memcpy(), memmove(), and memset(), avoid
intentionally reading across neighboring array fields.

Without a flexible array, this looks like an attempt to perform a memcpy()
read beyond the end of the packet->mad.data array:

drivers/infiniband/core/user_mad.c:
memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);

Switch from [0] to [] to use the appropriately handled type for trailing
bytes.

Link: https://lore.kernel.org/r/20210616202615.1247242-1-keescook@chromium.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/mlx5: Refactor get_ts_format functions to simplify code
Aharon Landau [Wed, 16 Jun 2021 07:57:38 +0000 (10:57 +0300)]
RDMA/mlx5: Refactor get_ts_format functions to simplify code

QPC, SQC and RQC timestamp formats and capabilities are always equal
because they represent general hardware support. So instead of code
duplication, let's merge them into general enum and logic.

Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Reviewed-by: Maor Gottlieb <maorg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
3 years agoRDMA/rxe: Don't overwrite errno from ib_umem_get()
Xiao Yang [Mon, 21 Jun 2021 07:14:56 +0000 (15:14 +0800)]
RDMA/rxe: Don't overwrite errno from ib_umem_get()

rxe_mr_init_user() always returns the fixed -EINVAL when ib_umem_get()
fails so it's hard for user to know which actual error happens in
ib_umem_get(). For example, ib_umem_get() will return -EOPNOTSUPP when
trying to pin pages on a DAX file.

Return actual error as mlx4/mlx5 does.

Link: https://lore.kernel.org/r/20210621071456.4259-1-ice_yangxiao@163.com
Signed-off-by: Xiao Yang <yangx.jy@fujitsu.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoIB/mlx4: Avoid field-overflowing memcpy()
Kees Cook [Wed, 16 Jun 2021 20:37:44 +0000 (13:37 -0700)]
IB/mlx4: Avoid field-overflowing memcpy()

In preparation for FORTIFY_SOURCE performing compile-time and run-time
field bounds checking for memcpy(), memmove(), and memset(), avoid
intentionally writing across neighboring array fields.

Use the ether_addr_copy() helper instead, as already done for smac.

Link: https://lore.kernel.org/r/20210616203744.1248551-1-keescook@chromium.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agornbd/rtrs-clt: Query and use max_segments from rtrs-clt.
Jack Wang [Mon, 21 Jun 2021 05:53:40 +0000 (07:53 +0200)]
rnbd/rtrs-clt: Query and use max_segments from rtrs-clt.

With fast memory registration on write request, rnbd-clt
can do bigger IO without split. rnbd-clt now can query
rtrs-clt to get the max_segments, instead of using
BMAX_SEGMENTS.

BMAX_SEGMENTS is not longer needed, so remove it.

Link: https://lore.kernel.org/r/20210621055340.11789-6-jinpu.wang@ionos.com
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs-clt: Raise MAX_SEGMENTS
Jack Wang [Mon, 21 Jun 2021 05:53:39 +0000 (07:53 +0200)]
RDMA/rtrs-clt: Raise MAX_SEGMENTS

As we can do fast memory registration on write, we can increase
the max_segments, default to 512K.

Link: https://lore.kernel.org/r/20210621055340.11789-5-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs_clt: Alloc less memory with write path fast memory registration
Jack Wang [Mon, 21 Jun 2021 05:53:38 +0000 (07:53 +0200)]
RDMA/rtrs_clt: Alloc less memory with write path fast memory registration

With write path fast memory registration, we need less memory for
each request.

With fast memory registration, we can reduce max_send_sge to save
memory usage.

Also convert the kmalloc_array to kcalloc.

Link: https://lore.kernel.org/r/20210621055340.11789-4-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs-clt: Write path fast memory registration
Jack Wang [Mon, 21 Jun 2021 05:53:37 +0000 (07:53 +0200)]
RDMA/rtrs-clt: Write path fast memory registration

With fast memory registration in write path, we can reduce
the memory consumption by using less max_send_sge, support IO bigger
than 116 KB (29 segments * 4 KB) without splitting, and it also
make the IO path more symmetric.

To avoid some times MR reg failed, waiting for the invalidation to finish
before the new mr reg. Introduce a refcount, only finish the request
when both local invalidation and io reply are there.

Link: https://lore.kernel.org/r/20210621055340.11789-3-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Md Haris Iqbal <haris.iqbal@ionos.com>
Signed-off-by: Dima Stepanov <dmitrii.stepanov@ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs: Introduce head/tail wr
Jack Wang [Mon, 21 Jun 2021 05:53:36 +0000 (07:53 +0200)]
RDMA/rtrs: Introduce head/tail wr

Introduce tail wr, we can send as the last wr, we want to send the local
invalidate wr after rdma wr in later patch.

While at it, also fix coding style issue.

Link: https://lore.kernel.org/r/20210621055340.11789-2-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/bnxt_re: Update ABI to pass wqe-mode to user space
Devesh Sharma [Wed, 16 Jun 2021 20:28:17 +0000 (01:58 +0530)]
RDMA/bnxt_re: Update ABI to pass wqe-mode to user space

Changing ucontext ABI response structure to pass wqe_mode to user library.
A flag in comp_mask has been set to indicate presence of wqe_mode.

Moved wqe-mode ABI to uapi/rdma/bnxt_re-abi.h

Link: https://lore.kernel.org/r/20210616202817.1185276-1-devesh.sharma@broadcom.com
Signed-off-by: Devesh Sharma <devesh.sharma@broadcom.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoIB/core: Shuffle locks in ib_port_data to save memory
Anand Khoje [Wed, 16 Jun 2021 15:45:08 +0000 (21:15 +0530)]
IB/core: Shuffle locks in ib_port_data to save memory

pahole shows two 4-byte holes in struct ib_port_data after pkey_list_lock
and netdev_lock respectively.

Shuffling the netdev_lock to be after pkey_list_lock, this shaves off
eight bytes from the struct.

Link: https://lore.kernel.org/r/20210616154509.1047-3-anand.a.khoje@oracle.com
Suggested-by: Haakon Bugge <haakon.bugge@oracle.com>
Signed-off-by: Anand Khoje <anand.a.khoje@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoIB/core: Removed port validity check from ib_get_cached_subnet_prefix
Anand Khoje [Wed, 16 Jun 2021 15:45:07 +0000 (21:15 +0530)]
IB/core: Removed port validity check from ib_get_cached_subnet_prefix

Removed port validity check from ib_get_cached_subnet_prefix() as this
check is not needed because "port_num" is valid.

Link: https://lore.kernel.org/r/20210616154509.1047-2-anand.a.khoje@oracle.com
Suggested-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Anand Khoje <anand.a.khoje@oracle.com>
Signed-off-by: Haakon Bugge <haakon.bugge@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA: Fix kernel-doc warnings about wrong comment
Leon Romanovsky [Wed, 16 Jun 2021 06:09:47 +0000 (09:09 +0300)]
RDMA: Fix kernel-doc warnings about wrong comment

Compilation with W=1 produces warnings similar to the below.

  drivers/infiniband/ulp/ipoib/ipoib_main.c:320: warning: This comment
starts with '/**', but isn't a kernel-doc comment. Refer
Documentation/doc-guide/kernel-doc.rst

All such occurrences were found with the following one line
 git grep -A 1 "\/\*\*" drivers/infiniband/

Link: https://lore.kernel.org/r/e57d5f4ddd08b7a19934635b44d6d632841b9ba7.1623823612.git.leonro@nvidia.com
Reviewed-by: Jack Wang <jinpu.wang@ionos.com> #rtrs
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use IDA interface to manage xrcd index
Yangyang Li [Thu, 10 Jun 2021 11:50:14 +0000 (19:50 +0800)]
RDMA/hns: Use IDA interface to manage xrcd index

Switch xrcd index allocation and release from hns own bitmap interface
to IDA interface.

Link: https://lore.kernel.org/r/1623325814-55737-7-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use IDA interface to manage pd index
Yangyang Li [Thu, 10 Jun 2021 11:50:13 +0000 (19:50 +0800)]
RDMA/hns: Use IDA interface to manage pd index

Switch pd index allocation and release from hns own bitmap interface
to IDA interface.

Link: https://lore.kernel.org/r/1623325814-55737-6-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use IDA interface to manage mtpt index
Yangyang Li [Thu, 10 Jun 2021 11:50:12 +0000 (19:50 +0800)]
RDMA/hns: Use IDA interface to manage mtpt index

Switch mtpt index allocation and release from hns own bitmap interface
to IDA interface.

Link: https://lore.kernel.org/r/1623325814-55737-5-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Remove unused RR mechanism
Yangyang Li [Thu, 10 Jun 2021 11:50:11 +0000 (19:50 +0800)]
RDMA/hns: Remove unused RR mechanism

Round-robin (RR) is no longer used in the allocation of the bitmap table,
and all the function input parameters that use this mechanism are
BITMAP_NO_RR. The code that defines and uses the RR needs to be deleted.

Link: https://lore.kernel.org/r/1623325814-55737-4-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Remove the unused hns_roce_bitmap_free_range function
Yangyang Li [Thu, 10 Jun 2021 11:50:10 +0000 (19:50 +0800)]
RDMA/hns: Remove the unused hns_roce_bitmap_free_range function

hns_roce_bitmap_free_range() is only called inside hns_roce_bitmap_free(),
and the input parameter "cnt" is set to a constant 1. In addition, the
driver does not use alloc_range scenarios, so free_range does not need to
exist.

Link: https://lore.kernel.org/r/1623325814-55737-3-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Remove the unused hns_roce_bitmap_alloc_range function
Yangyang Li [Thu, 10 Jun 2021 11:50:09 +0000 (19:50 +0800)]
RDMA/hns: Remove the unused hns_roce_bitmap_alloc_range function

The function is no longer used.

Link: https://lore.kernel.org/r/1623325814-55737-2-git-send-email-liweihang@huawei.com
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/core: Fix incorrect print format specifier
Wenpeng Liang [Thu, 10 Jun 2021 11:40:32 +0000 (19:40 +0800)]
RDMA/core: Fix incorrect print format specifier

There are some '%u' for 'int' and '%d' for 'unsigend int', they should be
fixed.

Link: https://lore.kernel.org/r/1623325232-30900-1-git-send-email-liweihang@huawei.com
Signed-off-by: Wenpeng Liang <liangwenpeng@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Clean SRQC structure definition
Xi Wang [Mon, 21 Jun 2021 08:00:43 +0000 (16:00 +0800)]
RDMA/hns: Clean SRQC structure definition

Remove unused members in srq context structure.

Link: https://lore.kernel.org/r/1624262443-24528-10-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use new interface to write DB related fields
Yixing Liu [Mon, 21 Jun 2021 08:00:42 +0000 (16:00 +0800)]
RDMA/hns: Use new interface to write DB related fields

Use hr_write_reg() instead of roce_set_field().

Link: https://lore.kernel.org/r/1624262443-24528-9-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use new interface to write FRMR fields
Yixing Liu [Mon, 21 Jun 2021 08:00:41 +0000 (16:00 +0800)]
RDMA/hns: Use new interface to write FRMR fields

Use "hr_reg_write" to replace "roce_set_filed".

Link: https://lore.kernel.org/r/1624262443-24528-8-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use new interface to get CQE fields
Lang Cheng [Mon, 21 Jun 2021 08:00:40 +0000 (16:00 +0800)]
RDMA/hns: Use new interface to get CQE fields

WQE_INDEX and OPCODE and QPN of CQE use redundant masks. Just remove them.

Link: https://lore.kernel.org/r/1624262443-24528-7-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use new interface to modify QP context
Lang Cheng [Mon, 21 Jun 2021 08:00:39 +0000 (16:00 +0800)]
RDMA/hns: Use new interface to modify QP context

Fill all QPC fileds with hr_reg_*() instead of roce_set_*(). SQPN is used
for HIP08 ES only, it should be removed.

Link: https://lore.kernel.org/r/1624262443-24528-6-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Use new interface to write CQ context.
Yixing Liu [Mon, 21 Jun 2021 08:00:38 +0000 (16:00 +0800)]
RDMA/hns: Use new interface to write CQ context.

Use hr_reg_*() to write CQ context, it's simpler than roce_set_*().

Link: https://lore.kernel.org/r/1624262443-24528-5-git-send-email-liweihang@huawei.com
Signed-off-by: Yixing Liu <liuyixing1@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Add hr_reg_write_bool()
Lang Cheng [Mon, 21 Jun 2021 08:00:37 +0000 (16:00 +0800)]
RDMA/hns: Add hr_reg_write_bool()

In order to avoid to do bitwise operations on a boolean value, add a new
register interface to avoid sparse comlaint about "dubious: x & !y" when
calling hr_reg_write(ctx, field, !!val).

Fixes: dc504774408b ("RDMA/hns: Use new interface to set MPT related fields")
Fixes: 495c24808ce7 ("RDMA/hns: Add XRC subtype in QPC and XRC type in SRQC")
Link: https://lore.kernel.org/r/1624262443-24528-4-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Add a check to ensure integer mtu is positive
Weihang Li [Mon, 21 Jun 2021 08:00:36 +0000 (16:00 +0800)]
RDMA/hns: Add a check to ensure integer mtu is positive

GCC may reports an running time assert error when a value calculated from
ib_mtu_enum_to_int() is using as 'val' in FIELD_PREDP:

include/linux/compiler_types.h:328:38: error: call to
'__compiletime_assert_1524' declared with attribute error: FIELD_PREP:
value too large for the field

So a check is added about whether integer mtu from ib_mtu_enum_to_int() is
negative to avoid this warning.

Link: https://lore.kernel.org/r/1624262443-24528-3-git-send-email-liweihang@huawei.com
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/hns: Do not use !! for values that are already bool when calling hr_reg_write()
Weihang Li [Mon, 21 Jun 2021 08:00:35 +0000 (16:00 +0800)]
RDMA/hns: Do not use !! for values that are already bool when calling hr_reg_write()

There is no need to use "!!" before "eq->eqe_size ==
HNS_ROCE_V3_EQE_SIZE", or sparse will complain about "dubious: x & !y".

Fixes: 782832f25404 ("RDMA/hns: Simplify the function config_eqc()")
Link: https://lore.kernel.org/r/1624262443-24528-2-git-send-email-liweihang@huawei.com
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/mlx5: Enable Relaxed Ordering by default for kernel ULPs
Avihai Horon [Wed, 9 Jun 2021 11:05:03 +0000 (14:05 +0300)]
RDMA/mlx5: Enable Relaxed Ordering by default for kernel ULPs

Relaxed Ordering is a capability that can only benefit users that support
it. All kernel ULPs should support Relaxed Ordering, as they are designed
to read data only after observing the CQE and use the DMA API correctly.

Hence, implicitly enable Relaxed Ordering by default for MR transfers in
kernel ULPs.

Link: https://lore.kernel.org/r/b7e820aab7402b8efa63605f4ea465831b3b1e5e.1623236426.git.leonro@nvidia.com
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoLinux 5.13-rc7
Linus Torvalds [Sun, 20 Jun 2021 22:03:15 +0000 (15:03 -0700)]
Linux 5.13-rc7

3 years agoMerge tag 'sched_urgent_for_v5.13_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Jun 2021 16:44:52 +0000 (09:44 -0700)]
Merge tag 'sched_urgent_for_v5.13_rc6' of git://git./linux/kernel/git/tip/tip

Pull scheduler fix from Borislav Petkov:
 "A single fix to restore fairness between control groups with equal
  priority"

* tag 'sched_urgent_for_v5.13_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/fair: Correctly insert cfs_rq's to list on unthrottle

3 years agoMerge tag 'irq_urgent_for_v5.13_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Jun 2021 16:38:14 +0000 (09:38 -0700)]
Merge tag 'irq_urgent_for_v5.13_rc6' of git://git./linux/kernel/git/tip/tip

Pull irq fix from Borislav Petkov:
 "A single fix for GICv3 to not take an interrupt in an NMI context"

* tag 'irq_urgent_for_v5.13_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/gic-v3: Workaround inconsistent PMR setting on NMI entry

3 years agoMerge tag 'x86_urgent_for_v5.13_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Jun 2021 16:09:58 +0000 (09:09 -0700)]
Merge tag 'x86_urgent_for_v5.13_rc6' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Borislav Petkov:
 "A first set of urgent fixes to the FPU/XSTATE handling mess^W code.
  (There's a lot more in the pipe):

   - Prevent corruption of the XSTATE buffer in signal handling by
     validating what is being copied from userspace first.

   - Invalidate other task's preserved FPU registers on XRSTOR failure
     (#PF) because latter can still modify some of them.

   - Restore the proper PKRU value in case userspace modified it

   - Reset FPU state when signal restoring fails

  Other:

   - Map EFI boot services data memory as encrypted in a SEV guest so
     that the guest can access it and actually boot properly

   - Two SGX correctness fixes: proper resources freeing and a NUMA fix"

* tag 'x86_urgent_for_v5.13_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mm: Avoid truncating memblocks for SGX memory
  x86/sgx: Add missing xa_destroy() when virtual EPC is destroyed
  x86/fpu: Reset state for all signal restore failures
  x86/pkru: Write hardware init value to PKRU when xstate is init
  x86/process: Check PF_KTHREAD and not current->mm for kernel threads
  x86/fpu: Invalidate FPU state after a failed XRSTOR from a user buffer
  x86/fpu: Prevent state corruption in __fpu__restore_sig()
  x86/ioremap: Map EFI-reserved memory as encrypted for SEV

3 years agoMerge tag 'powerpc-5.13-6' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sat, 19 Jun 2021 23:50:23 +0000 (16:50 -0700)]
Merge tag 'powerpc-5.13-6' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "Fix initrd corruption caused by our recent change to use relative jump
  labels.

  Fix a crash using perf record on systems without a hardware PMU
  backend.

  Rework our 64-bit signal handling slighty to make it more closely
  match the old behaviour, after the recent change to use unsafe user
  accessors.

  Thanks to Anastasia Kovaleva, Athira Rajeev, Christophe Leroy, Daniel
  Axtens, Greg Kurz, and Roman Bolshakov"

* tag 'powerpc-5.13-6' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/perf: Fix crash in perf_instruction_pointer() when ppmu is not set
  powerpc: Fix initrd corruption with relative jump labels
  powerpc/signal64: Copy siginfo before changing regs->nip
  powerpc/mem: Add back missing header to fix 'no previous prototype' error

3 years agoMerge tag 'perf-tools-fixes-for-v5.13-2021-06-19' of git://git.kernel.org/pub/scm...
Linus Torvalds [Sat, 19 Jun 2021 21:50:43 +0000 (14:50 -0700)]
Merge tag 'perf-tools-fixes-for-v5.13-2021-06-19' of git://git./linux/kernel/git/acme/linux

Pull perf tools fixes from Arnaldo Carvalho de Melo:

 - Fix refcount usage when processing PERF_RECORD_KSYMBOL.

 - 'perf stat' metric group fixes.

 - Fix 'perf test' non-bash issue with stat bpf counters.

 - Update unistd, in.h and socket.h with the kernel sources, silencing
   perf build warnings.

* tag 'perf-tools-fixes-for-v5.13-2021-06-19' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
  tools headers UAPI: Sync linux/in.h copy with the kernel sources
  tools headers UAPI: Sync asm-generic/unistd.h with the kernel original
  perf beauty: Update copy of linux/socket.h with the kernel sources
  perf test: Fix non-bash issue with stat bpf counters
  perf machine: Fix refcount usage when processing PERF_RECORD_KSYMBOL
  perf metricgroup: Return error code from metricgroup__add_metric_sys_event_iter()
  perf metricgroup: Fix find_evsel_group() event selector

3 years agoMerge tag 'riscv-for-linus-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 19 Jun 2021 15:45:34 +0000 (08:45 -0700)]
Merge tag 'riscv-for-linus-5.13-rc7' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Palmer Dabbelt:

 - A build fix to always build modules with the 'medany' code model, as
   the module loader doesn't support 'medlow'.

 - A Kconfig warning fix for the SiFive errata.

 - A pair of fixes that for regressions to the recent memory layout
   changes.

 - A fix for the FU740 device tree.

* tag 'riscv-for-linus-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: dts: fu740: fix cache-controller interrupts
  riscv: Ensure BPF_JIT_REGION_START aligned with PMD size
  riscv: kasan: Fix MODULES_VADDR evaluation due to local variables' name
  riscv: sifive: fix Kconfig errata warning
  riscv32: Use medany C model for modules

3 years agoMerge tag 's390-5.13-4' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Sat, 19 Jun 2021 15:39:13 +0000 (08:39 -0700)]
Merge tag 's390-5.13-4' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Vasily Gorbik:

 - Fix zcrypt ioctl hang due to AP queue msg counter dropping below 0
   when pending requests are purged.

 - Two fixes for the machine check handler in the entry code.

* tag 's390-5.13-4' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/ap: Fix hanging ioctl caused by wrong msg counter
  s390/mcck: fix invalid KVM guest condition check
  s390/mcck: fix calculation of SIE critical section size

3 years agotools headers UAPI: Sync linux/in.h copy with the kernel sources
Arnaldo Carvalho de Melo [Sat, 19 Jun 2021 13:15:22 +0000 (10:15 -0300)]
tools headers UAPI: Sync linux/in.h copy with the kernel sources

To pick the changes in:

  321827477360934d ("icmp: don't send out ICMP messages with a source address of 0.0.0.0")

That don't result in any change in tooling, as INADDR_ are not used to
generate id->string tables used by 'perf trace'.

This addresses this build warning:

  Warning: Kernel ABI header at 'tools/include/uapi/linux/in.h' differs from latest version at 'include/uapi/linux/in.h'
  diff -u tools/include/uapi/linux/in.h include/uapi/linux/in.h

Cc: David S. Miller <davem@davemloft.net>
Cc: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agotools headers UAPI: Sync asm-generic/unistd.h with the kernel original
Arnaldo Carvalho de Melo [Sat, 19 Jun 2021 13:11:46 +0000 (10:11 -0300)]
tools headers UAPI: Sync asm-generic/unistd.h with the kernel original

To pick the changes in:

  8b1462b67f23da54 ("quota: finish disable quotactl_path syscall")

Those headers are used in some arches to generate the syscall table used
in 'perf trace' to translate syscall numbers into strings.

This addresses this perf build warning:

  Warning: Kernel ABI header at 'tools/include/uapi/asm-generic/unistd.h' differs from latest version at 'include/uapi/asm-generic/unistd.h'
  diff -u tools/include/uapi/asm-generic/unistd.h include/uapi/asm-generic/unistd.h

Cc: Jan Kara <jack@suse.cz>
Cc: Marcin Juszkiewicz <marcin@juszkiewicz.com.pl>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf beauty: Update copy of linux/socket.h with the kernel sources
Arnaldo Carvalho de Melo [Sat, 19 Jun 2021 13:09:08 +0000 (10:09 -0300)]
perf beauty: Update copy of linux/socket.h with the kernel sources

To pick the changes in:

  ea6932d70e223e02 ("net: make get_net_ns return error if NET_NS is disabled")

That don't result in any changes in the tables generated from that
header.

This silences this perf build warning:

  Warning: Kernel ABI header at 'tools/perf/trace/beauty/include/linux/socket.h' differs from latest version at 'include/linux/socket.h'
  diff -u tools/perf/trace/beauty/include/linux/socket.h include/linux/socket.h

Cc: Changbin Du <changbin.du@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf test: Fix non-bash issue with stat bpf counters
Ian Rogers [Thu, 17 Jun 2021 18:42:13 +0000 (11:42 -0700)]
perf test: Fix non-bash issue with stat bpf counters

$(( .. )) is a bash feature but the test's interpreter is !/bin/sh,
switch the code to use expr.

Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Cc: bpf@vger.kernel.org
Link: http://lore.kernel.org/lkml/20210617184216.2075588-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf machine: Fix refcount usage when processing PERF_RECORD_KSYMBOL
Riccardo Mancini [Sat, 12 Jun 2021 17:37:48 +0000 (19:37 +0200)]
perf machine: Fix refcount usage when processing PERF_RECORD_KSYMBOL

ASan reported a memory leak of BPF-related ksymbols map and dso. The
leak is caused by refount never reaching 0, due to missing __put calls
in the function machine__process_ksymbol_register.

Once the dso is inserted in the map, dso__put() should be called
(map__new2() increases the refcount to 2).

The same thing applies for the map when it's inserted into maps
(maps__insert() increases the refcount to 2).

  $ sudo ./perf record -- sleep 5
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.025 MB perf.data (8 samples) ]

  =================================================================
  ==297735==ERROR: LeakSanitizer: detected memory leaks

  Direct leak of 6992 byte(s) in 19 object(s) allocated from:
      #0 0x4f43c7 in calloc (/home/user/linux/tools/perf/perf+0x4f43c7)
      #1 0x8e4e53 in map__new2 /home/user/linux/tools/perf/util/map.c:216:20
      #2 0x8cf68c in machine__process_ksymbol_register /home/user/linux/tools/perf/util/machine.c:778:10
      [...]

  Indirect leak of 8702 byte(s) in 19 object(s) allocated from:
      #0 0x4f43c7 in calloc (/home/user/linux/tools/perf/perf+0x4f43c7)
      #1 0x8728d7 in dso__new_id /home/user/linux/tools/perf/util/dso.c:1256:20
      #2 0x872015 in dso__new /home/user/linux/tools/perf/util/dso.c:1295:9
      #3 0x8cf623 in machine__process_ksymbol_register /home/user/linux/tools/perf/util/machine.c:774:21
      [...]

  Indirect leak of 1520 byte(s) in 19 object(s) allocated from:
      #0 0x4f43c7 in calloc (/home/user/linux/tools/perf/perf+0x4f43c7)
      #1 0x87b3da in symbol__new /home/user/linux/tools/perf/util/symbol.c:269:23
      #2 0x888954 in map__process_kallsym_symbol /home/user/linux/tools/perf/util/symbol.c:710:8
      [...]

  Indirect leak of 1406 byte(s) in 19 object(s) allocated from:
      #0 0x4f43c7 in calloc (/home/user/linux/tools/perf/perf+0x4f43c7)
      #1 0x87b3da in symbol__new /home/user/linux/tools/perf/util/symbol.c:269:23
      #2 0x8cfbd8 in machine__process_ksymbol_register /home/user/linux/tools/perf/util/machine.c:803:8
      [...]

Signed-off-by: Riccardo Mancini <rickyman7@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tommi Rantala <tommi.t.rantala@nokia.com>
Link: http://lore.kernel.org/lkml/20210612173751.188582-1-rickyman7@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf metricgroup: Return error code from metricgroup__add_metric_sys_event_iter()
John Garry [Thu, 10 Jun 2021 14:33:00 +0000 (22:33 +0800)]
perf metricgroup: Return error code from metricgroup__add_metric_sys_event_iter()

The error code is not set at all in the sys event iter function.

This may lead to an uninitialized value of "ret" in
metricgroup__add_metric() when no CPU metric is added.

Fix by properly setting the error code.

It is not necessary to init "ret" to 0 in metricgroup__add_metric(), as
if we have no CPU or sys event metric matching, then "has_match" should
be 0 and "ret" is set to -EINVAL.

However gcc cannot detect that it may not have been set after the
map_for_each_metric() loop for CPU metrics, which is strange.

Fixes: be335ec28efa8 ("perf metricgroup: Support adding metrics for system PMUs")
Signed-off-by: John Garry <john.garry@huawei.com>
Acked-by: Ian Rogers <irogers@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kajol Jain <kjain@linux.ibm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/1623335580-187317-3-git-send-email-john.garry@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf metricgroup: Fix find_evsel_group() event selector
John Garry [Thu, 10 Jun 2021 14:32:59 +0000 (22:32 +0800)]
perf metricgroup: Fix find_evsel_group() event selector

The following command segfaults on my x86 broadwell:

  $ ./perf stat  -M frontend_bound,retiring,backend_bound,bad_speculation sleep 1
  WARNING: grouped events cpus do not match, disabling group:
    anon group { raw 0x10e }
    anon group { raw 0x10e }
  perf: util/evsel.c:1596: get_group_fd: Assertion `!(!leader->core.fd)' failed.
  Aborted (core dumped)

The issue shows itself as a use-after-free in evlist__check_cpu_maps(),
whereby the leader of an event selector (evsel) has been deleted (yet we
still attempt to verify for an evsel).

Fundamentally the problem comes from metricgroup__setup_events() ->
find_evsel_group(), and has developed from the previous fix attempt in
commit 9c880c24cb0d ("perf metricgroup: Fix for metrics containing
duration_time").

The problem now is that the logic in checking if an evsel is in the same
group is subtly broken for the "cycles" event. For the "cycles" event,
the pmu_name is NULL; however the logic in find_evsel_group() may set an
event matched against "cycles" as used, when it should not be.

This leads to a condition where an evsel is set, yet its leader is not.

Fix the check for evsel pmu_name by not matching evsels when either has a
NULL pmu_name.

There is still a pre-existing metric issue whereby the ordering of the
metrics may break the 'stat' function, as discussed at:
https://lore.kernel.org/lkml/49c6fccb-b716-1bf0-18a6-cace1cdb66b9@huawei.com/

Fixes: 9c880c24cb0d ("perf metricgroup: Fix for metrics containing duration_time")
Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> # On a Thinkpad T450S
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kajol Jain <kjain@linux.ibm.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/1623335580-187317-2-git-send-email-john.garry@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoriscv: dts: fu740: fix cache-controller interrupts
David Abdurachmanov [Sun, 13 Jun 2021 00:43:57 +0000 (17:43 -0700)]
riscv: dts: fu740: fix cache-controller interrupts

The order of interrupt numbers is incorrect.

The order for FU740 is: DirError, DataError, DataFail, DirFail

From SiFive FU740-C000 Manual:
19 - L2 Cache DirError
20 - L2 Cache DirFail
21 - L2 Cache DataError
22 - L2 Cache DataFail

Signed-off-by: David Abdurachmanov <david.abdurachmanov@sifive.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoriscv: Ensure BPF_JIT_REGION_START aligned with PMD size
Jisheng Zhang [Fri, 18 Jun 2021 14:09:13 +0000 (22:09 +0800)]
riscv: Ensure BPF_JIT_REGION_START aligned with PMD size

Andreas reported commit fc8504765ec5 ("riscv: bpf: Avoid breaking W^X")
breaks booting with one kind of defconfig, I reproduced a kernel panic
with the defconfig:

[    0.138553] Unable to handle kernel paging request at virtual address ffffffff81201220
[    0.139159] Oops [#1]
[    0.139303] Modules linked in:
[    0.139601] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.13.0-rc5-default+ #1
[    0.139934] Hardware name: riscv-virtio,qemu (DT)
[    0.140193] epc : __memset+0xc4/0xfc
[    0.140416]  ra : skb_flow_dissector_init+0x1e/0x82
[    0.140609] epc : ffffffff8029806c ra : ffffffff8033be78 sp : ffffffe001647da0
[    0.140878]  gp : ffffffff81134b08 tp : ffffffe001654380 t0 : ffffffff81201158
[    0.141156]  t1 : 0000000000000002 t2 : 0000000000000154 s0 : ffffffe001647dd0
[    0.141424]  s1 : ffffffff80a43250 a0 : ffffffff81201220 a1 : 0000000000000000
[    0.141654]  a2 : 000000000000003c a3 : ffffffff81201258 a4 : 0000000000000064
[    0.141893]  a5 : ffffffff8029806c a6 : 0000000000000040 a7 : ffffffffffffffff
[    0.142126]  s2 : ffffffff81201220 s3 : 0000000000000009 s4 : ffffffff81135088
[    0.142353]  s5 : ffffffff81135038 s6 : ffffffff8080ce80 s7 : ffffffff80800438
[    0.142584]  s8 : ffffffff80bc6578 s9 : 0000000000000008 s10: ffffffff806000ac
[    0.142810]  s11: 0000000000000000 t3 : fffffffffffffffc t4 : 0000000000000000
[    0.143042]  t5 : 0000000000000155 t6 : 00000000000003ff
[    0.143220] status: 0000000000000120 badaddr: ffffffff81201220 cause: 000000000000000f
[    0.143560] [<ffffffff8029806c>] __memset+0xc4/0xfc
[    0.143859] [<ffffffff8061e984>] init_default_flow_dissectors+0x22/0x60
[    0.144092] [<ffffffff800010fc>] do_one_initcall+0x3e/0x168
[    0.144278] [<ffffffff80600df0>] kernel_init_freeable+0x1c8/0x224
[    0.144479] [<ffffffff804868a8>] kernel_init+0x12/0x110
[    0.144658] [<ffffffff800022de>] ret_from_exception+0x0/0xc
[    0.145124] ---[ end trace f1e9643daa46d591 ]---

After some investigation, I think I found the root cause: commit
2bfc6cd81bd ("move kernel mapping outside of linear mapping") moves
BPF JIT region after the kernel:

| #define BPF_JIT_REGION_START PFN_ALIGN((unsigned long)&_end)

The &_end is unlikely aligned with PMD size, so the front bpf jit
region sits with part of kernel .data section in one PMD size mapping.
But kernel is mapped in PMD SIZE, when bpf_jit_binary_lock_ro() is
called to make the first bpf jit prog ROX, we will make part of kernel
.data section RO too, so when we write to, for example memset the
.data section, MMU will trigger a store page fault.

To fix the issue, we need to ensure the BPF JIT region is PMD size
aligned. This patch acchieve this goal by restoring the BPF JIT region
to original position, I.E the 128MB before kernel .text section. The
modification to kasan_init.c is inspired by Alexandre.

Fixes: fc8504765ec5 ("riscv: bpf: Avoid breaking W^X")
Reported-by: Andreas Schwab <schwab@linux-m68k.org>
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoriscv: kasan: Fix MODULES_VADDR evaluation due to local variables' name
Jisheng Zhang [Fri, 18 Jun 2021 14:01:36 +0000 (22:01 +0800)]
riscv: kasan: Fix MODULES_VADDR evaluation due to local variables' name

commit 2bfc6cd81bd1 ("riscv: Move kernel mapping outside of linear
mapping") makes use of MODULES_VADDR to populate kernel, BPF, modules
mapping. Currently, MODULES_VADDR is defined as below for RV64:

| #define MODULES_VADDR   (PFN_ALIGN((unsigned long)&_end) - SZ_2G)

But kasan_init() has two local variables which are also named as _start,
_end, so MODULES_VADDR is evaluated with the local variable _end
rather than the global "_end" as we expected. Fix this issue by
renaming the two local variables.

Fixes: 2bfc6cd81bd1 ("riscv: Move kernel mapping outside of linear mapping")
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoMerge tag 'net-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Sat, 19 Jun 2021 01:55:29 +0000 (18:55 -0700)]
Merge tag 'net-5.13-rc7' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Jakub Kicinski:
 "Networking fixes for 5.13-rc7, including fixes from wireless, bpf,
  bluetooth, netfilter and can.

  Current release - regressions:

   - mlxsw: spectrum_qdisc: Pass handle, not band number to find_class()
     to fix modifying offloaded qdiscs

   - lantiq: net: fix duplicated skb in rx descriptor ring

   - rtnetlink: fix regression in bridge VLAN configuration, empty info
     is not an error, bot-generated "fix" was not needed

   - libbpf: s/rx/tx/ typo on umem->rx_ring_setup_done to fix umem
     creation

  Current release - new code bugs:

   - ethtool: fix NULL pointer dereference during module EEPROM dump via
     the new netlink API

   - mlx5e: don't update netdev RQs with PTP-RQ, the special purpose
     queue should not be visible to the stack

   - mlx5e: select special PTP queue only for SKBTX_HW_TSTAMP skbs

   - mlx5e: verify dev is present in get devlink port ndo, avoid a panic

  Previous releases - regressions:

   - neighbour: allow NUD_NOARP entries to be force GCed

   - further fixes for fallout from reorg of WiFi locking (staging:
     rtl8723bs, mac80211, cfg80211)

   - skbuff: fix incorrect msg_zerocopy copy notifications

   - mac80211: fix NULL ptr deref for injected rate info

   - Revert "net/mlx5: Arm only EQs with EQEs" it may cause missed IRQs

  Previous releases - always broken:

   - bpf: more speculative execution fixes

   - netfilter: nft_fib_ipv6: skip ipv6 packets from any to link-local

   - udp: fix race between close() and udp_abort() resulting in a panic

   - fix out of bounds when parsing TCP options before packets are
     validated (in netfilter: synproxy, tc: sch_cake and mptcp)

   - mptcp: improve operation under memory pressure, add missing
     wake-ups

   - mptcp: fix double-lock/soft lookup in subflow_error_report()

   - bridge: fix races (null pointer deref and UAF) in vlan tunnel
     egress

   - ena: fix DMA mapping function issues in XDP

   - rds: fix memory leak in rds_recvmsg

  Misc:

   - vrf: allow larger MTUs

   - icmp: don't send out ICMP messages with a source address of 0.0.0.0

   - cdc_ncm: switch to eth%d interface naming"

* tag 'net-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (139 commits)
  net: ethernet: fix potential use-after-free in ec_bhf_remove
  selftests/net: Add icmp.sh for testing ICMP dummy address responses
  icmp: don't send out ICMP messages with a source address of 0.0.0.0
  net: ll_temac: Avoid ndo_start_xmit returning NETDEV_TX_BUSY
  net: ll_temac: Fix TX BD buffer overwrite
  net: ll_temac: Add memory-barriers for TX BD access
  net: ll_temac: Make sure to free skb when it is completely used
  MAINTAINERS: add Guvenc as SMC maintainer
  bnxt_en: Call bnxt_ethtool_free() in bnxt_init_one() error path
  bnxt_en: Fix TQM fastpath ring backing store computation
  bnxt_en: Rediscover PHY capabilities after firmware reset
  cxgb4: fix wrong shift.
  mac80211: handle various extensible elements correctly
  mac80211: reset profile_periodicity/ema_ap
  cfg80211: avoid double free of PMSR request
  cfg80211: make certificate generation more robust
  mac80211: minstrel_ht: fix sample time check
  net: qed: Fix memcpy() overflow of qed_dcbx_params()
  net: cdc_eem: fix tx fixup skb leak
  net: hamradio: fix memory leak in mkiss_close
  ...

3 years agoMerge tag 'for-5.13-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
Linus Torvalds [Fri, 18 Jun 2021 23:39:03 +0000 (16:39 -0700)]
Merge tag 'for-5.13-rc6-tag' of git://git./linux/kernel/git/kdave/linux

Pull btrfs fix from David Sterba:
 "One more fix, for a space accounting bug in zoned mode. It happens
  when a block group is switched back rw->ro and unusable bytes (due to
  zoned constraints) are subtracted twice.

  It has user visible effects so I consider it important enough for late
  -rc inclusion and backport to stable"

* tag 'for-5.13-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: zoned: fix negative space_info->bytes_readonly

3 years agoMerge tag 'pci-v5.13-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
Linus Torvalds [Fri, 18 Jun 2021 20:54:11 +0000 (13:54 -0700)]
Merge tag 'pci-v5.13-fixes-2' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fixes from Bjorn Helgaas:

 - Clear 64-bit flag for host bridge windows below 4GB to fix a resource
   allocation regression added in -rc1 (Punit Agrawal)

 - Fix tegra194 MCFG quirk build regressions added in -rc1 (Jon Hunter)

 - Avoid secondary bus resets on TI KeyStone C667X devices (Antti
   Järvinen)

 - Avoid secondary bus resets on some NVIDIA GPUs (Shanker Donthineni)

 - Work around FLR erratum on Huawei Intelligent NIC VF (Chiqijun)

 - Avoid broken ATS on AMD Navi14 GPU (Evan Quan)

 - Trust Broadcom BCM57414 NIC to isolate functions even though it
   doesn't advertise ACS support (Sriharsha Basavapatna)

 - Work around AMD RS690 BIOSes that don't configure DMA above 4GB
   (Mikel Rychliski)

 - Fix panic during PIO transfer on Aardvark controller (Pali Rohár)

* tag 'pci-v5.13-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI: aardvark: Fix kernel panic during PIO transfer
  PCI: Add AMD RS690 quirk to enable 64-bit DMA
  PCI: Add ACS quirk for Broadcom BCM57414 NIC
  PCI: Mark AMD Navi14 GPU ATS as broken
  PCI: Work around Huawei Intelligent NIC VF FLR erratum
  PCI: Mark some NVIDIA GPUs to avoid bus reset
  PCI: Mark TI C667X to avoid bus reset
  PCI: tegra194: Fix MCFG quirk build regressions
  PCI: of: Clear 64-bit flag for non-prefetchable memory below 4GB

3 years agoafs: Re-enable freezing once a page fault is interrupted
Matthew Wilcox (Oracle) [Wed, 16 Jun 2021 21:22:28 +0000 (22:22 +0100)]
afs: Re-enable freezing once a page fault is interrupted

If a task is killed during a page fault, it does not currently call
sb_end_pagefault(), which means that the filesystem cannot be frozen
at any time thereafter.  This may be reported by lockdep like this:

====================================
WARNING: fsstress/10757 still has locks held!
5.13.0-rc4-build4+ #91 Not tainted
------------------------------------
1 lock held by fsstress/10757:
 #0: ffff888104eac530
 (
sb_pagefaults

as filesystem freezing is modelled as a lock.

Fix this by removing all the direct returns from within the function,
and using 'ret' to indicate whether we were interrupted or successful.

Fixes: 1cf7a1518aef ("afs: Implement shared-writeable mmap")
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Link: https://lore.kernel.org/r/20210616154900.1958373-1-willy@infradead.org/
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agonet: ethernet: fix potential use-after-free in ec_bhf_remove
Pavel Skripkin [Fri, 18 Jun 2021 13:49:02 +0000 (16:49 +0300)]
net: ethernet: fix potential use-after-free in ec_bhf_remove

static void ec_bhf_remove(struct pci_dev *dev)
{
...
struct ec_bhf_priv *priv = netdev_priv(net_dev);

unregister_netdev(net_dev);
free_netdev(net_dev);

pci_iounmap(dev, priv->dma_io);
pci_iounmap(dev, priv->io);
...
}

priv is netdev private data, but it is used
after free_netdev(). It can cause use-after-free when accessing priv
pointer. So, fix it by moving free_netdev() after pci_iounmap()
calls.

Fixes: 6af55ff52b02 ("Driver for Beckhoff CX5020 EtherCAT master module.")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge tag 'mac80211-for-net-2021-06-18' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Fri, 18 Jun 2021 19:22:55 +0000 (12:22 -0700)]
Merge tag 'mac80211-for-net-2021-06-18' of git://git./linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
A couple of straggler fixes:
 * a minstrel HT sample check fix
 * peer measurement could double-free on races
 * certificate file generation at build time could
   sometimes hang
 * some parameters weren't reset between connections
   in mac80211
 * some extensible elements were treated as non-
   extensible, possibly causuing bad connections
   (or failures) if the AP adds data
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoselftests/net: Add icmp.sh for testing ICMP dummy address responses
Toke Høiland-Jørgensen [Fri, 18 Jun 2021 11:04:36 +0000 (13:04 +0200)]
selftests/net: Add icmp.sh for testing ICMP dummy address responses

This adds a new icmp.sh selftest for testing that the kernel will respond
correctly with an ICMP unreachable message with the dummy (192.0.0.8)
source address when there are no IPv4 addresses configured to use as source
addresses.

Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoicmp: don't send out ICMP messages with a source address of 0.0.0.0
Toke Høiland-Jørgensen [Fri, 18 Jun 2021 11:04:35 +0000 (13:04 +0200)]
icmp: don't send out ICMP messages with a source address of 0.0.0.0

When constructing ICMP response messages, the kernel will try to pick a
suitable source address for the outgoing packet. However, if no IPv4
addresses are configured on the system at all, this will fail and we end up
producing an ICMP message with a source address of 0.0.0.0. This can happen
on a box routing IPv4 traffic via v6 nexthops, for instance.

Since 0.0.0.0 is not generally routable on the internet, there's a good
chance that such ICMP messages will never make it back to the sender of the
original packet that the ICMP message was sent in response to. This, in
turn, can create connectivity and PMTUd problems for senders. Fortunately,
RFC7600 reserves a dummy address to be used as a source for ICMP
messages (192.0.0.8/32), so let's teach the kernel to substitute that
address as a last resort if the regular source address selection procedure
fails.

Below is a quick example reproducing this issue with network namespaces:

ip netns add ns0
ip l add type veth peer netns ns0
ip l set dev veth0 up
ip a add 10.0.0.1/24 dev veth0
ip a add fc00:dead:cafe:42::1/64 dev veth0
ip r add 10.1.0.0/24 via inet6 fc00:dead:cafe:42::2
ip -n ns0 l set dev veth0 up
ip -n ns0 a add fc00:dead:cafe:42::2/64 dev veth0
ip -n ns0 r add 10.0.0.0/24 via inet6 fc00:dead:cafe:42::1
ip netns exec ns0 sysctl -w net.ipv4.icmp_ratelimit=0
ip netns exec ns0 sysctl -w net.ipv4.ip_forward=1
tcpdump -tpni veth0 -c 2 icmp &
ping -w 1 10.1.0.1 > /dev/null
tcpdump: verbose output suppressed, use -v[v]... for full protocol decode
listening on veth0, link-type EN10MB (Ethernet), snapshot length 262144 bytes
IP 10.0.0.1 > 10.1.0.1: ICMP echo request, id 29, seq 1, length 64
IP 0.0.0.0 > 10.0.0.1: ICMP net 10.1.0.1 unreachable, length 92
2 packets captured
2 packets received by filter
0 packets dropped by kernel

With this patch the above capture changes to:
IP 10.0.0.1 > 10.1.0.1: ICMP echo request, id 31127, seq 1, length 64
IP 192.0.0.8 > 10.0.0.1: ICMP net 10.1.0.1 unreachable, length 92

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Juliusz Chroboczek <jch@irif.fr>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ll_temac: Avoid ndo_start_xmit returning NETDEV_TX_BUSY
Esben Haabendal [Fri, 18 Jun 2021 10:52:38 +0000 (12:52 +0200)]
net: ll_temac: Avoid ndo_start_xmit returning NETDEV_TX_BUSY

As documented in Documentation/networking/driver.rst, the ndo_start_xmit
method must not return NETDEV_TX_BUSY under any normal circumstances, and
as recommended, we simply stop the tx queue in advance, when there is a
risk that the next xmit would cause a NETDEV_TX_BUSY return.

Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ll_temac: Fix TX BD buffer overwrite
Esben Haabendal [Fri, 18 Jun 2021 10:52:33 +0000 (12:52 +0200)]
net: ll_temac: Fix TX BD buffer overwrite

Just as the initial check, we need to ensure num_frag+1 buffers available,
as that is the number of buffers we are going to use.

This fixes a buffer overflow, which might be seen during heavy network
load. Complete lockup of TEMAC was reproducible within about 10 minutes of
a particular load.

Fixes: 84823ff80f74 ("net: ll_temac: Fix race condition causing TX hang")
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ll_temac: Add memory-barriers for TX BD access
Esben Haabendal [Fri, 18 Jun 2021 10:52:28 +0000 (12:52 +0200)]
net: ll_temac: Add memory-barriers for TX BD access

Add a couple of memory-barriers to ensure correct ordering of read/write
access to TX BDs.

In xmit_done, we should ensure that reading the additional BD fields are
only done after STS_CTRL_APP0_CMPLT bit is set.

When xmit_done marks the BD as free by setting APP0=0, we need to ensure
that the other BD fields are reset first, so we avoid racing with the xmit
path, which writes to the same fields.

Finally, making sure to read APP0 of next BD after the current BD, ensures
that we see all available buffers.

Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ll_temac: Make sure to free skb when it is completely used
Esben Haabendal [Fri, 18 Jun 2021 10:52:23 +0000 (12:52 +0200)]
net: ll_temac: Make sure to free skb when it is completely used

With the skb pointer piggy-backed on the TX BD, we have a simple and
efficient way to free the skb buffer when the frame has been transmitted.
But in order to avoid freeing the skb while there are still fragments from
the skb in use, we need to piggy-back on the TX BD of the skb, not the
first.

Without this, we are doing use-after-free on the DMA side, when the first
BD of a multi TX BD packet is seen as completed in xmit_done, and the
remaining BDs are still being processed.

Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Esben Haabendal <esben@geanix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMAINTAINERS: add Guvenc as SMC maintainer
Karsten Graul [Fri, 18 Jun 2021 07:00:30 +0000 (09:00 +0200)]
MAINTAINERS: add Guvenc as SMC maintainer

Add Guvenc as maintainer for Shared Memory Communications (SMC)
Sockets.

Cc: Julian Wiedmann <jwi@linux.ibm.com>
Acked-by: Guvenc Gulce <guvenc@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'bnxt_en-fixes'
David S. Miller [Fri, 18 Jun 2021 19:00:27 +0000 (12:00 -0700)]
Merge branch 'bnxt_en-fixes'

Michael Chan says:

====================
bnxt_en: Bug fixes

This patchset includes 3 small bug fixes to reinitialize PHY capabilities
after firmware reset, setup the chip's internal TQM fastpath ring
backing memory properly for RoCE traffic, and to free ethtool related
memory if driver probe fails.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Call bnxt_ethtool_free() in bnxt_init_one() error path
Somnath Kotur [Fri, 18 Jun 2021 06:07:27 +0000 (02:07 -0400)]
bnxt_en: Call bnxt_ethtool_free() in bnxt_init_one() error path

bnxt_ethtool_init() may have allocated some memory and we need to
call bnxt_ethtool_free() to properly unwind if bnxt_init_one()
fails.

Fixes: 7c3809181468 ("bnxt_en: Refactor bnxt_init_one() and turn on TPA support on 57500 chips.")
Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Fix TQM fastpath ring backing store computation
Rukhsana Ansari [Fri, 18 Jun 2021 06:07:26 +0000 (02:07 -0400)]
bnxt_en: Fix TQM fastpath ring backing store computation

TQM fastpath ring needs to be sized to store both the requester
and responder side of RoCE QPs in TQM for supporting bi-directional
tests.  Fix bnxt_alloc_ctx_mem() to multiply the RoCE QPs by a factor of
2 when computing the number of entries for TQM fastpath ring.  This
fixes an RX pipeline stall issue when running bi-directional max
RoCE QP tests.

Fixes: c7dd7ab4b204 ("bnxt_en: Improve TQM ring context memory sizing formulas.")
Signed-off-by: Rukhsana Ansari <rukhsana.ansari@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Rediscover PHY capabilities after firmware reset
Michael Chan [Fri, 18 Jun 2021 06:07:25 +0000 (02:07 -0400)]
bnxt_en: Rediscover PHY capabilities after firmware reset

There is a missing bnxt_probe_phy() call in bnxt_fw_init_one() to
rediscover the PHY capabilities after a firmware reset.  This can cause
some PHY related functionalities to fail after a firmware reset.  For
example, in multi-host, the ability for any host to configure the PHY
settings may be lost after a firmware reset.

Fixes: ec5d31e3c15d ("bnxt_en: Handle firmware reset status during IF_UP.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocxgb4: fix wrong shift.
Pavel Machek [Fri, 18 Jun 2021 09:29:48 +0000 (11:29 +0200)]
cxgb4: fix wrong shift.

While fixing coverity warning, commit dd2c79677375 introduced typo in
shift value. Fix that.

Signed-off-by: Pavel Machek (CIP) <pavel@denx.de>
Fixes: dd2c79677375 ("cxgb4: Fix unintentional sign extension issues")
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge tag 'arc-5.13-rc7-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupt...
Linus Torvalds [Fri, 18 Jun 2021 18:09:23 +0000 (11:09 -0700)]
Merge tag 'arc-5.13-rc7-fixes' of git://git./linux/kernel/git/vgupta/arc

Pull ARC fixes from Vineet Gupta:

 - ARCv2 userspace ABI not populating a few registers

 - Unbork CONFIG_HARDENED_USERCOPY for ARC

* tag 'arc-5.13-rc7-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  ARC: fix CONFIG_HARDENED_USERCOPY
  ARCv2: save ABI registers across signal handling

3 years agoMerge tag 'trace-v5.13-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
Linus Torvalds [Fri, 18 Jun 2021 17:57:09 +0000 (10:57 -0700)]
Merge tag 'trace-v5.13-rc6' of git://git./linux/kernel/git/rostedt/linux-trace

Pull tracing fixes from Steven Rostedt:

 - Have recordmcount check for valid st_shndx otherwise some archs may
   have invalid references for the mcount location.

 - Two fixes done for mapping pids to task names. Traces were not
   showing the names of tasks when they should have.

 - Fix to trace_clock_global() to prevent it from going backwards

* tag 'trace-v5.13-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  tracing: Do no increment trace_clock_global() by one
  tracing: Do not stop recording comms if the trace file is being read
  tracing: Do not stop recording cmdlines when tracing is off
  recordmcount: Correct st_shndx handling

3 years agoMerge tag 'printk-for-5.13-fixup' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 18 Jun 2021 17:50:41 +0000 (10:50 -0700)]
Merge tag 'printk-for-5.13-fixup' of git://git./linux/kernel/git/printk/linux

Pull printk fixup from Petr Mladek:
 "Fix misplaced EXPORT_SYMBOL(vsprintf)"

* tag 'printk-for-5.13-fixup' of git://git.kernel.org/pub/scm/linux/kernel/git/printk/linux:
  printk: Move EXPORT_SYMBOL() closer to vprintk definition

3 years agoMerge tag 'pm-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Fri, 18 Jun 2021 17:42:36 +0000 (10:42 -0700)]
Merge tag 'pm-5.13-rc7' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fix from Rafael Wysocki:
 "Remove recently added frequency invariance support from the CPPC
  cpufreq driver, because it has turned out to be problematic and it
  cannot be fixed properly on time for 5.13 (Viresh Kumar)"

* tag 'pm-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  Revert "cpufreq: CPPC: Add support for frequency invariance"

3 years agoMerge tag 'usb-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Fri, 18 Jun 2021 17:39:32 +0000 (10:39 -0700)]
Merge tag 'usb-5.13-rc7' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are three small USB fixes for reported problems for 5.13-rc7.
  They include:

   - disable autosuspend for a cypress USB hub

   - fix the battery charger detection for the chipidea driver

   - fix a kernel panic in the dwc3 driver due to a previous change in
     5.13-rc1.

  All have been in linux-next with no reported problems"

* tag 'usb-5.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  usb: core: hub: Disable autosuspend for Cypress CY7C65632
  usb: chipidea: imx: Fix Battery Charger 1.2 CDP detection
  usb: dwc3: core: fix kernel panic when do reboot

3 years agox86/mm: Avoid truncating memblocks for SGX memory
Fan Du [Thu, 17 Jun 2021 19:46:57 +0000 (12:46 -0700)]
x86/mm: Avoid truncating memblocks for SGX memory

tl;dr:

Several SGX users reported seeing the following message on NUMA systems:

  sgx: [Firmware Bug]: Unable to map EPC section to online node. Fallback to the NUMA node 0.

This turned out to be the memblock code mistakenly throwing away SGX
memory.

=== Full Changelog ===

The 'max_pfn' variable represents the highest known RAM address.  It can
be used, for instance, to quickly determine for which physical addresses
there is mem_map[] space allocated.  The numa_meminfo code makes an
effort to throw out ("trim") all memory blocks which are above 'max_pfn'.

SGX memory is not considered RAM (it is marked as "Reserved" in the
e820) and is not taken into account by max_pfn. Despite this, SGX memory
areas have NUMA affinity and are enumerated in the ACPI SRAT table. The
existing SGX code uses the numa_meminfo mechanism to look up the NUMA
affinity for its memory areas.

In cases where SGX memory was above max_pfn (usually just the one EPC
section in the last highest NUMA node), the numa_memblock is truncated
at 'max_pfn', which is below the SGX memory.  When the SGX code tries to
look up the affinity of this memory, it fails and produces an error message:

  sgx: [Firmware Bug]: Unable to map EPC section to online node. Fallback to the NUMA node 0.

and assigns the memory to NUMA node 0.

Instead of silently truncating the memory block at 'max_pfn' and
dropping the SGX memory, add the truncated portion to
'numa_reserved_meminfo'.  This allows the SGX code to later determine
the NUMA affinity of its 'Reserved' area.

Before, numa_meminfo looked like this (from 'crash'):

  blk = { start =          0x0, end = 0x2080000000, nid = 0x0 }
        { start = 0x2080000000, end = 0x4000000000, nid = 0x1 }

numa_reserved_meminfo is empty.

With this, numa_meminfo looks like this:

  blk = { start =          0x0, end = 0x2080000000, nid = 0x0 }
        { start = 0x2080000000, end = 0x4000000000, nid = 0x1 }

and numa_reserved_meminfo has an entry for node 1's SGX memory:

  blk =  { start = 0x4000000000, end = 0x4080000000, nid = 0x1 }

 [ daveh: completely rewrote/reworked changelog ]

Fixes: 5d30f92e7631 ("x86/NUMA: Provide a range-to-target_node lookup facility")
Reported-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: Fan Du <fan.du@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Dave Hansen <dave.hansen@intel.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20210617194657.0A99CB22@viggo.jf.intel.com
3 years agoMerge tag 'drm-fixes-2021-06-18' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 18 Jun 2021 17:36:18 +0000 (10:36 -0700)]
Merge tag 'drm-fixes-2021-06-18' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Not much happening in fixes land this week only one PR for two amdgpu
  powergating fixes was waiting for me, maybe something will show up
  over the weekend, maybe not.

  amdgpu:

   - GFX9 and 10 powergating fixes"

* tag 'drm-fixes-2021-06-18' of git://anongit.freedesktop.org/drm/drm:
  drm/amdgpu/gfx10: enlarge CP_MEC_DOORBELL_RANGE_UPPER to cover full doorbell.
  drm/amdgpu/gfx9: fix the doorbell missing when in CGPG issue.

3 years agoRDMA/hns: Clear extended doorbell info before using
Xi Wang [Fri, 11 Jun 2021 06:14:49 +0000 (14:14 +0800)]
RDMA/hns: Clear extended doorbell info before using

Both of HIP08 and HIP09 require the extended doorbell information to be
cleared before being used.

Fixes: 6b63597d3540 ("RDMA/hns: Add TSQ link table support")
Link: https://lore.kernel.org/r/1623392089-35639-1-git-send-email-liweihang@huawei.com
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs: Check device max_qp_wr limit when create QP
Jack Wang [Mon, 14 Jun 2021 09:03:37 +0000 (11:03 +0200)]
RDMA/rtrs: Check device max_qp_wr limit when create QP

Currently we only check device max_qp_wr limit for IO connection, but not
for service connection. We should check for both.

So save the max_qp_wr device limit in wr_limit, and use it for both IO
connections and service connections.

While at it, also remove an outdated comments.

Link: https://lore.kernel.org/r/20210614090337.29557-6-jinpu.wang@ionos.com
Suggested-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs: Rename cq_size/queue_size to cq_num/queue_num
Guoqing Jiang [Mon, 14 Jun 2021 09:03:36 +0000 (11:03 +0200)]
RDMA/rtrs: Rename cq_size/queue_size to cq_num/queue_num

Those variables are passed to create_cq, create_qp, rtrs_iu_alloc and
rtrs_iu_free, so these *_size means the num of unit. And cq_size also
means number of cq element.

Also move the setting of cq_num to common path.

Link: https://lore.kernel.org/r/20210614090337.29557-5-jinpu.wang@ionos.com
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs: RDMA_RXE requires more number of WR
Md Haris Iqbal [Mon, 14 Jun 2021 09:03:35 +0000 (11:03 +0200)]
RDMA/rtrs: RDMA_RXE requires more number of WR

When using rdma_rxe, post_one_recv() returns ENOMEM error due to the full
recv queue.  This patch increase the number of WR for receive queue to
support all devices.

Link: https://lore.kernel.org/r/20210614090337.29557-4-jinpu.wang@ionos.com
Signed-off-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs-clt: Use minimal max_send_sge when create qp
Jack Wang [Mon, 14 Jun 2021 09:03:34 +0000 (11:03 +0200)]
RDMA/rtrs-clt: Use minimal max_send_sge when create qp

We use device limit max_send_sge, which is suboptimal for memory usage.
We don't need that much for User Con, 1 is enough. And for IO con,
sess->max_segments + 1 is enough

Link: https://lore.kernel.org/r/20210614090337.29557-3-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoRDMA/rtrs-srv: Set minimal max_send_wr and max_recv_wr
Jack Wang [Mon, 14 Jun 2021 09:03:33 +0000 (11:03 +0200)]
RDMA/rtrs-srv: Set minimal max_send_wr and max_recv_wr

Currently rtrs when create_qp use a coarse numbers (bigger in general),
which leads to hardware create more resources which only waste memory with
no benefits.

For max_send_wr, we don't really need alway max_qp_wr size when creating
qp, reduce it to cq_size.

For max_recv_wr,  cq_size is enough.

With the patch when sess_queue_depth=128, per session (2 paths) memory
consumption reduced from 188 MB to 65MB

When always_invalidate is enabled, we need send more wr, so treat it
special.

Fixes: 9cb837480424e ("RDMA/rtrs: server: main functionality")
Link: https://lore.kernel.org/r/20210614090337.29557-2-jinpu.wang@ionos.com
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@ionos.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
3 years agoPCI: aardvark: Fix kernel panic during PIO transfer
Pali Rohár [Tue, 8 Jun 2021 20:36:55 +0000 (22:36 +0200)]
PCI: aardvark: Fix kernel panic during PIO transfer

Trying to start a new PIO transfer by writing value 0 in PIO_START register
when previous transfer has not yet completed (which is indicated by value 1
in PIO_START) causes an External Abort on CPU, which results in kernel
panic:

    SError Interrupt on CPU0, code 0xbf000002 -- SError
    Kernel panic - not syncing: Asynchronous SError Interrupt

To prevent kernel panic, it is required to reject a new PIO transfer when
previous one has not finished yet.

If previous PIO transfer is not finished yet, the kernel may issue a new
PIO request only if the previous PIO transfer timed out.

In the past the root cause of this issue was incorrectly identified (as it
often happens during link retraining or after link down event) and special
hack was implemented in Trusted Firmware to catch all SError events in EL3,
to ignore errors with code 0xbf000002 and not forwarding any other errors
to kernel and instead throw panic from EL3 Trusted Firmware handler.

Links to discussion and patches about this issue:
https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/commit/?id=3c7dcdac5c50
https://lore.kernel.org/linux-pci/20190316161243.29517-1-repk@triplefau.lt/
https://lore.kernel.org/linux-pci/971be151d24312cc533989a64bd454b4@www.loen.fr/
https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/1541

But the real cause was the fact that during link retraining or after link
down event the PIO transfer may take longer time, up to the 1.44s until it
times out. This increased probability that a new PIO transfer would be
issued by kernel while previous one has not finished yet.

After applying this change into the kernel, it is possible to revert the
mentioned TF-A hack and SError events do not have to be caught in TF-A EL3.

Link: https://lore.kernel.org/r/20210608203655.31228-1-pali@kernel.org
Signed-off-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Marek Behún <kabel@kernel.org>
Cc: stable@vger.kernel.org # 7fbcb5da811b ("PCI: aardvark: Don't rely on jiffies while holding spinlock")
3 years agoPCI: Add AMD RS690 quirk to enable 64-bit DMA
Mikel Rychliski [Fri, 11 Jun 2021 21:48:23 +0000 (17:48 -0400)]
PCI: Add AMD RS690 quirk to enable 64-bit DMA

Although the AMD RS690 chipset has 64-bit DMA support, BIOS implementations
sometimes fail to configure the memory limit registers correctly.

The Acer F690GVM mainboard uses this chipset and a Marvell 88E8056 NIC. The
sky2 driver programs the NIC to use 64-bit DMA, which will not work:

  sky2 0000:02:00.0: error interrupt status=0x8
  sky2 0000:02:00.0 eth0: tx timeout
  sky2 0000:02:00.0 eth0: transmit ring 0 .. 22 report=0 done=0

Other drivers required by this mainboard either don't support 64-bit DMA,
or have it disabled using driver specific quirks. For example, the ahci
driver has quirks to enable or disable 64-bit DMA depending on the BIOS
version (see ahci_sb600_enable_64bit() in ahci.c). This ahci quirk matches
against the SB600 SATA controller, but the real issue is almost certainly
with the RS690 PCI host that it was commonly attached to.

To avoid this issue in all drivers with 64-bit DMA support, fix the
configuration of the PCI host. If the kernel is aware of physical memory
above 4GB, but the BIOS never configured the PCI host with this
information, update the registers with our values.

[bhelgaas: drop PCI_DEVICE_ID_ATI_RS690 definition]
Link: https://lore.kernel.org/r/20210611214823.4898-1-mikel@mikelr.com
Signed-off-by: Mikel Rychliski <mikel@mikelr.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
3 years agoPCI: Add ACS quirk for Broadcom BCM57414 NIC
Sriharsha Basavapatna [Sat, 22 May 2021 01:13:17 +0000 (21:13 -0400)]
PCI: Add ACS quirk for Broadcom BCM57414 NIC

The Broadcom BCM57414 NIC may be a multi-function device.  While it does
not advertise an ACS capability, peer-to-peer transactions are not possible
between the individual functions, so it is safe to treat them as fully
isolated.

Add an ACS quirk for this device so the functions can be in independent
IOMMU groups and attached individually to userspace applications using
VFIO.

[bhelgaas: commit log]
Link: https://lore.kernel.org/r/1621645997-16251-1-git-send-email-michael.chan@broadcom.com
Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: stable@vger.kernel.org
3 years agoPCI: Mark AMD Navi14 GPU ATS as broken
Evan Quan [Wed, 2 Jun 2021 02:12:55 +0000 (10:12 +0800)]
PCI: Mark AMD Navi14 GPU ATS as broken

Observed unexpected GPU hang during runpm stress test on 0x7341 rev 0x00.
Further debugging shows broken ATS is related.

Disable ATS on this part.  Similar issues on other devices:

  a2da5d8cc0b0 ("PCI: Mark AMD Raven iGPU ATS as broken in some platforms")
  45beb31d3afb ("PCI: Mark AMD Navi10 GPU rev 0x00 ATS as broken")
  5e89cd303e3a ("PCI: Mark AMD Navi14 GPU rev 0xc5 ATS as broken")

Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://lore.kernel.org/r/20210602021255.939090-1-evan.quan@amd.com
Signed-off-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Krzysztof Wilczyński <kw@linux.com>
Cc: stable@vger.kernel.org
3 years agoPCI: Work around Huawei Intelligent NIC VF FLR erratum
Chiqijun [Mon, 24 May 2021 22:44:07 +0000 (17:44 -0500)]
PCI: Work around Huawei Intelligent NIC VF FLR erratum

pcie_flr() starts a Function Level Reset (FLR), waits 100ms (the maximum
time allowed for FLR completion by PCIe r5.0, sec 6.6.2), and waits for the
FLR to complete.  It assumes the FLR is complete when a config read returns
valid data.

When we do an FLR on several Huawei Intelligent NIC VFs at the same time,
firmware on the NIC processes them serially.  The VF may respond to config
reads before the firmware has completed its reset processing.  If we bind a
driver to the VF (e.g., by assigning the VF to a virtual machine) in the
interval between the successful config read and completion of the firmware
reset processing, the NIC VF driver may fail to load.

Prevent this driver failure by waiting for the NIC firmware to complete its
reset processing.  Not all NIC firmware supports this feature.

[bhelgaas: commit log]
Link: https://support.huawei.com/enterprise/en/doc/EDOC1100063073/87950645/vm-oss-occasionally-fail-to-load-the-in200-driver-when-the-vf-performs-flr
Link: https://lore.kernel.org/r/20210414132301.1793-1-chiqijun@huawei.com
Signed-off-by: Chiqijun <chiqijun@huawei.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Cc: stable@vger.kernel.org
3 years agoPCI: Mark some NVIDIA GPUs to avoid bus reset
Shanker Donthineni [Tue, 8 Jun 2021 05:48:56 +0000 (11:18 +0530)]
PCI: Mark some NVIDIA GPUs to avoid bus reset

Some NVIDIA GPU devices do not work with SBR.  Triggering SBR leaves the
device inoperable for the current system boot. It requires a system
hard-reboot to get the GPU device back to normal operating condition
post-SBR. For the affected devices, enable NO_BUS_RESET quirk to avoid the
issue.

This issue will be fixed in the next generation of hardware.

Link: https://lore.kernel.org/r/20210608054857.18963-8-ameynarkhede03@gmail.com
Signed-off-by: Shanker Donthineni <sdonthineni@nvidia.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Cc: stable@vger.kernel.org
3 years agoPCI: Mark TI C667X to avoid bus reset
Antti Järvinen [Mon, 15 Mar 2021 10:26:06 +0000 (10:26 +0000)]
PCI: Mark TI C667X to avoid bus reset

Some TI KeyStone C667X devices do not support bus/hot reset.  The PCIESS
automatically disables LTSSM when Secondary Bus Reset is received and
device stops working.  Prevent bus reset for these devices.  With this
change, the device can be assigned to VMs with VFIO, but it will leak state
between VMs.

Reference: https://e2e.ti.com/support/processors/f/791/t/954382
Link: https://lore.kernel.org/r/20210315102606.17153-1-antti.jarvinen@gmail.com
Signed-off-by: Antti Järvinen <antti.jarvinen@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Kishon Vijay Abraham I <kishon@ti.com>
Cc: stable@vger.kernel.org
3 years agoPCI: tegra194: Fix MCFG quirk build regressions
Jon Hunter [Thu, 10 Jun 2021 06:41:34 +0000 (07:41 +0100)]
PCI: tegra194: Fix MCFG quirk build regressions

7f100744749e ("PCI: tegra: Add Tegra194 MCFG quirks for ECAM errata")
caused a few build regressions:

  - 7f100744749e removed the Makefile rule for CONFIG_PCIE_TEGRA194, so
    pcie-tegra.c can no longer be built as a module.  Restore that rule.

  - 7f100744749e added "#ifdef CONFIG_PCIE_TEGRA194" around the native
    driver, but that's only set when the driver is built-in (for a module,
    CONFIG_PCIE_TEGRA194_MODULE is defined).

    The ACPI quirk is completely independent of the rest of the native
    driver, so move the quirk to its own file and remove the #ifdef in the
    native driver.

  - 7f100744749e added symbols that are always defined but used only when
    CONFIG_PCIEASPM, which causes warnings when CONFIG_PCIEASPM is not set:

      drivers/pci/controller/dwc/pcie-tegra194.c:259:18: warning: ‘event_cntr_data_offset’ defined but not used [-Wunused-const-variable=]
      drivers/pci/controller/dwc/pcie-tegra194.c:250:18: warning: ‘event_cntr_ctrl_offset’ defined but not used [-Wunused-const-variable=]
      drivers/pci/controller/dwc/pcie-tegra194.c:243:27: warning: ‘pcie_gen_freq’ defined but not used [-Wunused-const-variable=]

Fixes: 7f100744749e ("PCI: tegra: Add Tegra194 MCFG quirks for ECAM errata")
Link: https://lore.kernel.org/r/20210610064134.336781-1-jonathanh@nvidia.com
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
3 years agoPCI: of: Clear 64-bit flag for non-prefetchable memory below 4GB
Punit Agrawal [Mon, 14 Jun 2021 23:04:57 +0000 (08:04 +0900)]
PCI: of: Clear 64-bit flag for non-prefetchable memory below 4GB

Alexandru and Qu reported this resource allocation failure on ROCKPro64 v2
and ROCK Pi 4B, both based on the RK3399:

  pci_bus 0000:00: root bus resource [mem 0xfa000000-0xfbdfffff 64bit]
  pci 0000:00:00.0: PCI bridge to [bus 01]
  pci 0000:00:00.0: BAR 14: no space for [mem size 0x00100000]
  pci 0000:01:00.0: reg 0x10: [mem 0x00000000-0x00003fff 64bit]

"BAR 14" is the PCI bridge's 32-bit non-prefetchable window, and our PCI
allocation code isn't smart enough to allocate it in a host bridge window
marked as 64-bit, even though this should work fine.

A DT host bridge description includes the windows from the CPU address
space to the PCI bus space.  On a few architectures (microblaze, powerpc,
sparc), the DT may also describe PCI devices themselves, including their
BARs.

Before 9d57e61bf723 ("of/pci: Add IORESOURCE_MEM_64 to resource flags for
64-bit memory addresses"), of_bus_pci_get_flags() ignored the fact that
some DT addresses described 64-bit windows and BARs.  That was a problem
because the virtio virtual NIC has a 32-bit BAR and a 64-bit BAR, and the
driver couldn't distinguish them.

9d57e61bf723 set IORESOURCE_MEM_64 for those 64-bit DT ranges, which fixed
the virtio driver.  But it also set IORESOURCE_MEM_64 for host bridge
windows, which exposed the fact that the PCI allocator isn't smart enough
to put 32-bit resources in those 64-bit windows.

Clear IORESOURCE_MEM_64 from host bridge windows since we don't need that
information.

Suggested-by: Bjorn Helgaas <bhelgaas@google.com>
Fixes: 9d57e61bf723 ("of/pci: Add IORESOURCE_MEM_64 to resource flags for 64-bit memory addresses")
Link: https://lore.kernel.org/r/20210614230457.752811-1-punitagrawal@gmail.com
Reported-at: https://lore.kernel.org/lkml/7a1e2ebc-f7d8-8431-d844-41a9c36a8911@arm.com/
Reported-at: https://lore.kernel.org/lkml/YMyTUv7Jsd89PGci@m4/T/#u
Reported-by: Alexandru Elisei <alexandru.elisei@arm.com>
Reported-by: Qu Wenruo <wqu@suse.com>
Tested-by: Alexandru Elisei <alexandru.elisei@arm.com>
Tested-by: Domenico Andreoli <domenico.andreoli@linux.com>
Signed-off-by: Punit Agrawal <punitagrawal@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
3 years agotracing: Do no increment trace_clock_global() by one
Steven Rostedt (VMware) [Thu, 17 Jun 2021 21:12:35 +0000 (17:12 -0400)]
tracing: Do no increment trace_clock_global() by one

The trace_clock_global() tries to make sure the events between CPUs is
somewhat in order. A global value is used and updated by the latest read
of a clock. If one CPU is ahead by a little, and is read by another CPU, a
lock is taken, and if the timestamp of the other CPU is behind, it will
simply use the other CPUs timestamp.

The lock is also only taken with a "trylock" due to tracing, and strange
recursions can happen. The lock is not taken at all in NMI context.

In the case where the lock is not able to be taken, the non synced
timestamp is returned. But it will not be less than the saved global
timestamp.

The problem arises because when the time goes "backwards" the time
returned is the saved timestamp plus 1. If the lock is not taken, and the
plus one to the timestamp is returned, there's a small race that can cause
the time to go backwards!

CPU0 CPU1
---- ----
trace_clock_global() {
    ts = clock() [ 1000 ]
    trylock(clock_lock) [ success ]
    global_ts = ts; [ 1000 ]

    <interrupted by NMI>
 trace_clock_global() {
    ts = clock() [ 999 ]
    if (ts < global_ts)
ts = global_ts + 1 [ 1001 ]

    trylock(clock_lock) [ fail ]

    return ts [ 1001]
 }
    unlock(clock_lock);
    return ts; [ 1000 ]
}

 trace_clock_global() {
    ts = clock() [ 1000 ]
    if (ts < global_ts) [ false 1000 == 1000 ]

    trylock(clock_lock) [ success ]
    global_ts = ts; [ 1000 ]
    unlock(clock_lock)

    return ts; [ 1000 ]
 }

The above case shows to reads of trace_clock_global() on the same CPU, but
the second read returns one less than the first read. That is, time when
backwards, and this is not what is allowed by trace_clock_global().

This was triggered by heavy tracing and the ring buffer checker that tests
for the clock going backwards:

 Ring buffer clock went backwards: 20613921464 -> 20613921463
 ------------[ cut here ]------------
 WARNING: CPU: 2 PID: 0 at kernel/trace/ring_buffer.c:3412 check_buffer+0x1b9/0x1c0
 Modules linked in:
 [..]
 [CPU: 2]TIME DOES NOT MATCH expected:20620711698 actual:20620711697 delta:6790234 before:20613921463 after:20613921463
   [20613915818] PAGE TIME STAMP
   [20613915818] delta:0
   [20613915819] delta:1
   [20613916035] delta:216
   [20613916465] delta:430
   [20613916575] delta:110
   [20613916749] delta:174
   [20613917248] delta:499
   [20613917333] delta:85
   [20613917775] delta:442
   [20613917921] delta:146
   [20613918321] delta:400
   [20613918568] delta:247
   [20613918768] delta:200
   [20613919306] delta:538
   [20613919353] delta:47
   [20613919980] delta:627
   [20613920296] delta:316
   [20613920571] delta:275
   [20613920862] delta:291
   [20613921152] delta:290
   [20613921464] delta:312
   [20613921464] delta:0 TIME EXTEND
   [20613921464] delta:0

This happened more than once, and always for an off by one result. It also
started happening after commit aafe104aa9096 was added.

Cc: stable@vger.kernel.org
Fixes: aafe104aa9096 ("tracing: Restructure trace_clock_global() to never block")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>