OSDN Git Service

uclinux-h8/linux.git
3 years agonetfilter: nftables: counter hardware offload support
Pablo Neira Ayuso [Thu, 15 Apr 2021 18:10:18 +0000 (20:10 +0200)]
netfilter: nftables: counter hardware offload support

This patch adds the .offload_stats operation to synchronize hardware
stats with the expression data. Update the counter expression to use
this new interface. The hardware stats are retrieved from the netlink
dump path via FLOW_CLS_STATS command to the driver.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agoselftests: fib_tests: Add test cases for interaction with mangling
Ido Schimmel [Wed, 14 Apr 2021 08:20:33 +0000 (11:20 +0300)]
selftests: fib_tests: Add test cases for interaction with mangling

Test that packets are correctly routed when netfilter mangling rules are
present.

Without previous patch:

 # ./fib_tests.sh -t ipv4_mangle

 IPv4 mangling tests
     TEST:     Connection with correct parameters                        [ OK ]
     TEST:     Connection with incorrect parameters                      [ OK ]
     TEST:     Connection with correct parameters - mangling             [FAIL]
     TEST:     Connection with correct parameters - no mangling          [ OK ]
     TEST:     Connection check - server side                            [FAIL]

 Tests passed:   3
 Tests failed:   2

 # ./fib_tests.sh -t ipv6_mangle

 IPv6 mangling tests
     TEST:     Connection with correct parameters                        [ OK ]
     TEST:     Connection with incorrect parameters                      [ OK ]
     TEST:     Connection with correct parameters - mangling             [FAIL]
     TEST:     Connection with correct parameters - no mangling          [ OK ]
     TEST:     Connection check - server side                            [FAIL]

 Tests passed:   3
 Tests failed:   2

With previous patch:

 # ./fib_tests.sh -t ipv4_mangle

 IPv4 mangling tests
     TEST:     Connection with correct parameters                        [ OK ]
     TEST:     Connection with incorrect parameters                      [ OK ]
     TEST:     Connection with correct parameters - mangling             [ OK ]
     TEST:     Connection with correct parameters - no mangling          [ OK ]
     TEST:     Connection check - server side                            [ OK ]

 Tests passed:   5
 Tests failed:   0

 # ./fib_tests.sh -t ipv6_mangle

 IPv6 mangling tests
     TEST:     Connection with correct parameters                        [ OK ]
     TEST:     Connection with incorrect parameters                      [ OK ]
     TEST:     Connection with correct parameters - mangling             [ OK ]
     TEST:     Connection with correct parameters - no mangling          [ OK ]
     TEST:     Connection check - server side                            [ OK ]

 Tests passed:   5
 Tests failed:   0

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: Dissect flow after packet mangling
Ido Schimmel [Wed, 14 Apr 2021 08:20:32 +0000 (11:20 +0300)]
netfilter: Dissect flow after packet mangling

Netfilter tries to reroute mangled packets as a different route might
need to be used following the mangling. When this happens, netfilter
does not populate the IP protocol, the source port and the destination
port in the flow key. Therefore, FIB rules that match on these fields
are ignored and packets can be misrouted.

Solve this by dissecting the outer flow and populating the flow key
before rerouting the packet. Note that flow dissection only happens when
FIB rules that match on these fields are installed, so in the common
case there should not be a penalty.

Reported-by: Michal Soltys <msoltyspl@yandex.pl>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nftables_offload: special ethertype handling for VLAN
Pablo Neira Ayuso [Mon, 12 Apr 2021 12:20:55 +0000 (14:20 +0200)]
netfilter: nftables_offload: special ethertype handling for VLAN

The nftables offload parser sets FLOW_DISSECTOR_KEY_BASIC .n_proto to the
ethertype field in the ethertype frame. However:

- FLOW_DISSECTOR_KEY_BASIC .n_proto field always stores either IPv4 or IPv6
  ethertypes.
- FLOW_DISSECTOR_KEY_VLAN .vlan_tpid stores either the 802.1q and 802.1ad
  ethertypes. Same as for FLOW_DISSECTOR_KEY_CVLAN.

This function adjusts the flow dissector to handle two scenarios:

1) FLOW_DISSECTOR_KEY_VLAN .vlan_tpid is set to 802.1q or 802.1ad.
   Then, transfer:
   - the .n_proto field to FLOW_DISSECTOR_KEY_VLAN .tpid.
   - the original FLOW_DISSECTOR_KEY_VLAN .tpid to the
     FLOW_DISSECTOR_KEY_CVLAN .tpid
   - the original FLOW_DISSECTOR_KEY_CVLAN .tpid to the .n_proto field.

2) .n_proto is set to 802.1q or 802.1ad. Then, transfer:
   - the .n_proto field to FLOW_DISSECTOR_KEY_VLAN .tpid.
   - the original FLOW_DISSECTOR_KEY_VLAN .tpid to the .n_proto field.

Fixes: a82055af5959 ("netfilter: nft_payload: add VLAN offload support")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nftables_offload: VLAN id needs host byteorder in flow dissector
Pablo Neira Ayuso [Mon, 12 Apr 2021 12:20:15 +0000 (14:20 +0200)]
netfilter: nftables_offload: VLAN id needs host byteorder in flow dissector

The flow dissector representation expects the VLAN id in host byteorder.
Add the NFT_OFFLOAD_F_NETWORK2HOST flag to swap the bytes from nft_cmp.

Fixes: a82055af5959 ("netfilter: nft_payload: add VLAN offload support")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: nft_payload: fix C-VLAN offload support
Pablo Neira Ayuso [Mon, 12 Apr 2021 12:11:39 +0000 (14:11 +0200)]
netfilter: nft_payload: fix C-VLAN offload support

- add another struct flow_dissector_key_vlan for C-VLAN
- update layer 3 dependency to allow to match on IPv4/IPv6

Fixes: 89d8fd44abfb ("netfilter: nft_payload: add C-VLAN offload support")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: flowtable: Add FLOW_OFFLOAD_XMIT_UNSPEC xmit type
Roi Dayan [Tue, 13 Apr 2021 08:06:05 +0000 (11:06 +0300)]
netfilter: flowtable: Add FLOW_OFFLOAD_XMIT_UNSPEC xmit type

It could be xmit type was not set and would default to FLOW_OFFLOAD_XMIT_NEIGH
and in this type the gc expect to have a route info.
Fix that by adding FLOW_OFFLOAD_XMIT_UNSPEC which defaults to 0.

Fixes: 8b9229d15877 ("netfilter: flowtable: dst_check() from garbage collector path")
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: conntrack: convert sysctls to u8
Florian Westphal [Mon, 12 Apr 2021 19:55:44 +0000 (21:55 +0200)]
netfilter: conntrack: convert sysctls to u8

log_invalid sysctl allows values of 0 to 255 inclusive so we no longer
need a range check: the min/max values can be removed.

This also removes all member variables that were moved to net_generic
data in previous patches.

This reduces size of netns_ct struct by one cache line.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: conntrack: move ct counter to net_generic data
Florian Westphal [Mon, 12 Apr 2021 19:55:43 +0000 (21:55 +0200)]
netfilter: conntrack: move ct counter to net_generic data

Its only needed from slowpath (sysctl, ctnetlink, gc worker) and
when a new conntrack object is allocated.

Furthermore, each write dirties the otherwise read-mostly pernet
data in struct net.ct, which are accessed from packet path.

Move it to the net_generic data.  This makes struct netns_ct
read-mostly.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: conntrack: move expect counter to net_generic data
Florian Westphal [Mon, 12 Apr 2021 19:55:42 +0000 (21:55 +0200)]
netfilter: conntrack: move expect counter to net_generic data

Creation of a new conntrack entry isn't a frequent operation (compared
to 'ct entry already exists').  Creation of a new entry that is also an
expected (related) connection even less so.

Place this counter in net_generic data.

A followup patch will also move the conntrack count -- this will make
netns_ct a read-mostly structure.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: conntrack: move autoassign_helper sysctl to net_generic data
Florian Westphal [Mon, 12 Apr 2021 19:55:41 +0000 (21:55 +0200)]
netfilter: conntrack: move autoassign_helper sysctl to net_generic data

While at it, make it an u8, no need to use an integer for a boolean.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: conntrack: move autoassign warning member to net_generic data
Florian Westphal [Mon, 12 Apr 2021 19:55:40 +0000 (21:55 +0200)]
netfilter: conntrack: move autoassign warning member to net_generic data

Not accessed in fast path, place this is generic_net data instead.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: flowtable: add vlan pop action offload support
wenxu [Sat, 3 Apr 2021 13:59:43 +0000 (21:59 +0800)]
netfilter: flowtable: add vlan pop action offload support

This patch adds vlan pop action offload in the flowtable offload.

Signed-off-by: wenxu <wenxu@ucloud.cn>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonetfilter: flowtable: add vlan match offload support
wenxu [Sat, 3 Apr 2021 13:59:42 +0000 (21:59 +0800)]
netfilter: flowtable: add vlan match offload support

This patch adds support for vlan_id, vlan_priority and vlan_proto match
for flowtable offload.

Signed-off-by: wenxu <wenxu@ucloud.cn>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
3 years agonet: ethernet: ravb: Enable optional refclk
Adam Ford [Mon, 12 Apr 2021 13:26:19 +0000 (08:26 -0500)]
net: ethernet: ravb: Enable optional refclk

For devices that use a programmable clock for the AVB reference clock,
the driver may need to enable them.  Add code to find the optional clock
and enable it when available.

Signed-off-by: Adam Ford <aford173@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodt-bindings: net: renesas,etheravb: Add additional clocks
Adam Ford [Mon, 12 Apr 2021 13:26:18 +0000 (08:26 -0500)]
dt-bindings: net: renesas,etheravb: Add additional clocks

The AVB driver assumes there is an external crystal, but it could
be clocked by other means.  In order to enable a programmable
clock, it needs to be added to the clocks list and enabled in the
driver.  Since there currently only one clock, there is no
clock-names list either.

Update bindings to add the additional optional clock, and explicitly
name both of them.

Signed-off-by: Adam Ford <aford173@gmail.com>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: Rob Herring <robh@kernel.org>
Reviewed-by: Sergei Shtylyov <sergei.shtylyov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'enetc-ptp'
David S. Miller [Mon, 12 Apr 2021 20:34:21 +0000 (13:34 -0700)]
Merge branch 'enetc-ptp'

Yangbo Lu says:

====================
enetc: support PTP Sync packet one-step timestamping

This patch-set is to add support for PTP Sync packet one-step timestamping.
Since ENETC single-step register has to be configured dynamically per
packet for correctionField offeset and UDP checksum update, current
one-step timestamping packet has to be sent only when the last one
completes transmitting on hardware. So, on the TX, this patch handles
one-step timestamping packet as below:

- Trasmit packet immediately if no other one in transfer, or queue to
  skb queue if there is already one in transfer.
  The test_and_set_bit_lock() is used here to lock and check state.
- Start a work when complete transfer on hardware, to release the bit
  lock and to send one skb in skb queue if has.

Changes for v2:
- Rebased.
- Fixed issues from patchwork checks.
- netif_tx_lock for one-step timestamping packet sending.
Changes for v3:
- Used system workqueue.
- Set bit lock when transmitted one-step packet, and scheduled
  work when completed. The worker cleared the bit lock, and
  transmitted one skb in skb queue if has, instead of a loop.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoenetc: support PTP Sync packet one-step timestamping
Yangbo Lu [Mon, 12 Apr 2021 09:03:27 +0000 (17:03 +0800)]
enetc: support PTP Sync packet one-step timestamping

This patch is to add support for PTP Sync packet one-step timestamping.
Since ENETC single-step register has to be configured dynamically per
packet for correctionField offeset and UDP checksum update, current
one-step timestamping packet has to be sent only when the last one
completes transmitting on hardware. So, on the TX, this patch handles
one-step timestamping packet as below:

- Trasmit packet immediately if no other one in transfer, or queue to
  skb queue if there is already one in transfer.
  The test_and_set_bit_lock() is used here to lock and check state.
- Start a work when complete transfer on hardware, to release the bit
  lock and to send one skb in skb queue if has.

And the configuration for one-step timestamping on ENETC before
transmitting is,

- Set one-step timestamping flag in extension BD.
- Write 30 bits current timestamp in tstamp field of extension BD.
- Update PTP Sync packet originTimestamp field with current timestamp.
- Configure single-step register for correctionField offeset and UDP
  checksum update.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoenetc: mark TX timestamp type per skb
Yangbo Lu [Mon, 12 Apr 2021 09:03:26 +0000 (17:03 +0800)]
enetc: mark TX timestamp type per skb

Mark TX timestamp type per skb on skb->cb[0], instead of
global variable for all skbs. This is a preparation for
one step timestamp support.

For one-step timestamping enablement, there will be both
one-step and two-step PTP messages to transfer. And a skb
queue is needed for one-step PTP messages making sure
start to send current message only after the last one
completed on hardware. (ENETC single-step register has to
be dynamically configured per message.) So, marking TX
timestamp type per skb is required.

Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'ibmvnic-errors'
David S. Miller [Mon, 12 Apr 2021 20:31:27 +0000 (13:31 -0700)]
Merge branch 'ibmvnic-errors'

Lijun Pan says:

====================
ibmvnic: improve error printing

Patch 1 prints reset reason as a string.
Patch 2 prints adapter state as a string.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: print adapter state as a string
Lijun Pan [Mon, 12 Apr 2021 07:41:28 +0000 (02:41 -0500)]
ibmvnic: print adapter state as a string

The adapter state can be added or deleted over different versions
of the source code. Print a string instead of a number.

Signed-off-by: Lijun Pan <lijunp213@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: print reset reason as a string
Lijun Pan [Mon, 12 Apr 2021 07:41:27 +0000 (02:41 -0500)]
ibmvnic: print reset reason as a string

The reset reason can be added or deleted over different versions
of the source code. Print a string instead of a number.

Signed-off-by: Lijun Pan <lijunp213@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoibmvnic: clean up the remaining debugfs data structures
Lijun Pan [Mon, 12 Apr 2021 07:40:59 +0000 (02:40 -0500)]
ibmvnic: clean up the remaining debugfs data structures

Commit e704f0434ea6 ("ibmvnic: Remove debugfs support") did not
clean up everything. Remove the remaining code.

Signed-off-by: Lijun Pan <lijunp213@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'netns-sysctl-isolation'
David S. Miller [Mon, 12 Apr 2021 20:27:11 +0000 (13:27 -0700)]
Merge branch 'netns-sysctl-isolation'

Jonathon Reinhart says:

====================
Ensuring net sysctl isolation

This patchset is the result of an audit of /proc/sys/net to prove that
it is safe to be mouted read-write in a container when a net namespace
is in use. See [1].

The first commit adds code to detect sysctls which are not netns-safe,
and can "leak" changes to other net namespaces.

My manual audit found, and the above feature confirmed, that there are
two nf_conntrack sysctls which are in fact not netns-safe.

I considered sending the latter to netfilter-devel, but I think it's
better to have both together on net-next: Adding only the former causes
undesirable warnings in the kernel log.

[1]: https://github.com/opencontainers/runc/issues/2826
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonetfilter: conntrack: Make global sysctls readonly in non-init netns
Jonathon Reinhart [Mon, 12 Apr 2021 04:24:53 +0000 (00:24 -0400)]
netfilter: conntrack: Make global sysctls readonly in non-init netns

These sysctls point to global variables:
- NF_SYSCTL_CT_MAX (&nf_conntrack_max)
- NF_SYSCTL_CT_EXPECT_MAX (&nf_ct_expect_max)
- NF_SYSCTL_CT_BUCKETS (&nf_conntrack_htable_size_user)

Because their data pointers are not updated to point to per-netns
structures, they must be marked read-only in a non-init_net ns.
Otherwise, changes in any net namespace are reflected in (leaked into)
all other net namespaces. This problem has existed since the
introduction of net namespaces.

The current logic marks them read-only only if the net namespace is
owned by an unprivileged user (other than init_user_ns).

Commit d0febd81ae77 ("netfilter: conntrack: re-visit sysctls in
unprivileged namespaces") "exposes all sysctls even if the namespace is
unpriviliged." Since we need to mark them readonly in any case, we can
forego the unprivileged user check altogether.

Fixes: d0febd81ae77 ("netfilter: conntrack: re-visit sysctls in unprivileged namespaces")
Signed-off-by: Jonathon Reinhart <Jonathon.Reinhart@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: Ensure net namespace isolation of sysctls
Jonathon Reinhart [Mon, 12 Apr 2021 04:24:52 +0000 (00:24 -0400)]
net: Ensure net namespace isolation of sysctls

This adds an ensure_safe_net_sysctl() check during register_net_sysctl()
to validate that sysctl table entries for a non-init_net netns are
sufficiently isolated. To be netns-safe, an entry must adhere to at
least (and usually exactly) one of these rules:

1. It is marked read-only inside the netns.
2. Its data pointer does not point to kernel/module global data.

An entry which fails both of these checks is indicative of a bug,
whereby a child netns can affect global net sysctl values.

If such an entry is found, this code will issue a warning to the kernel
log, and force the entry to be read-only to prevent a leak.

To test, simply create a new netns:

    $ sudo ip netns add dummy

As it sits now, this patch will WARN for two sysctls which will be
addressed in a subsequent patch:
- /proc/sys/net/netfilter/nf_conntrack_max
- /proc/sys/net/netfilter/nf_conntrack_expect_max

Signed-off-by: Jonathon Reinhart <Jonathon.Reinhart@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonfc: pn533: remove redundant assignment
wengjianfeng [Mon, 12 Apr 2021 02:20:06 +0000 (10:20 +0800)]
nfc: pn533: remove redundant assignment

In many places,first assign a value to a variable and then return
the variable. which is redundant, we should directly return the value.
in pn533_rf_field funciton,return rc also in the if statement, so we
use return 0 to replace the last return rc.

Signed-off-by: wengjianfeng <wengjianfeng@yulong.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'bnxt_en-error-recovery'
David S. Miller [Mon, 12 Apr 2021 20:20:38 +0000 (13:20 -0700)]
Merge branch 'bnxt_en-error-recovery'

Michael Chan says:

====================
bnxt_en: Error recovery fixes.

This series adds some fixes and enhancements to the error recovery
logic.  The health register logic is improved and we also add missing
code to free and re-create VF representors in the firmware after
error recovery.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Free and allocate VF-Reps during error recovery.
Sriharsha Basavapatna [Mon, 12 Apr 2021 00:18:15 +0000 (20:18 -0400)]
bnxt_en: Free and allocate VF-Reps during error recovery.

During firmware recovery, VF-Rep configuration in the firmware is lost.
Fix it by freeing and (re)allocating VF-Reps in FW at relevant points
during the error recovery process.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Refactor __bnxt_vf_reps_destroy().
Michael Chan [Mon, 12 Apr 2021 00:18:14 +0000 (20:18 -0400)]
bnxt_en: Refactor __bnxt_vf_reps_destroy().

Add a new helper function __bnxt_free_one_vf_rep() to free one VF rep.
We also reintialize the VF rep fields to proper initial values so that
the function can be used without freeing the VF rep data structure.  This
will be used in subsequent patches to free and recreate VF reps after
error recovery.

Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Reviewed-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Refactor bnxt_vf_reps_create().
Sriharsha Basavapatna [Mon, 12 Apr 2021 00:18:13 +0000 (20:18 -0400)]
bnxt_en: Refactor bnxt_vf_reps_create().

Add a new function bnxt_alloc_vf_rep() to allocate a VF representor.
This function will be needed in subsequent patches to recreate the
VF reps after error recovery.

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Invalidate health register mapping at the end of probe.
Vasundhara Volam [Mon, 12 Apr 2021 00:18:12 +0000 (20:18 -0400)]
bnxt_en: Invalidate health register mapping at the end of probe.

After probe is successful, interface may not be bought up in all
the cases and health register mapping could be invalid if firmware
undergoes reset. Fix it by invalidating the health register at the
end of probe. It will be remapped during ifup.

Fixes: 43a440c4007b ("bnxt_en: Improve the status_reliable flag in bp->fw_health.")
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agobnxt_en: Treat health register value 0 as valid in bnxt_try_reover_fw().
Michael Chan [Mon, 12 Apr 2021 00:18:11 +0000 (20:18 -0400)]
bnxt_en: Treat health register value 0 as valid in bnxt_try_reover_fw().

The retry loop in bnxt_try_recover_fw() should not abort when the
health register value is 0.  It is a valid value that indicates the
firmware is booting up.

Fixes: 861aae786f2f ("bnxt_en: Enhance retry of the first message to the firmware.")
Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: seg6: trivial fix of a spelling mistake in comment
Andrea Mayer [Sat, 10 Apr 2021 17:46:14 +0000 (19:46 +0200)]
net: seg6: trivial fix of a spelling mistake in comment

There is a comment spelling mistake "interfarence" -> "interference" in
function parse_nla_action(). Fix it.

Signed-off-by: Andrea Mayer <andrea.mayer@uniroma2.it>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: hns3: Fix potential null pointer defererence of null ae_dev
Colin Ian King [Fri, 9 Apr 2021 16:37:26 +0000 (17:37 +0100)]
net: hns3: Fix potential null pointer defererence of null ae_dev

The reset_prepare and reset_done calls have a null pointer check
on ae_dev however ae_dev is being dereferenced via the call to
ns3_is_phys_func with the ae->pdev argument. Fix this by performing
a null pointer check on ae_dev and hence short-circuiting the
dereference to ae_dev on the call to ns3_is_phys_func.

Addresses-Coverity: ("Dereference before null check")
Fixes: 715c58e94f0d ("net: hns3: add suspend and resume pm_ops")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: thunderx: Fix unintentional sign extension issue
Colin Ian King [Fri, 9 Apr 2021 13:07:26 +0000 (14:07 +0100)]
net: thunderx: Fix unintentional sign extension issue

The shifting of the u8 integers rq->caching by 26 bits to
the left will be promoted to a 32 bit signed int and then
sign-extended to a u64. In the event that rq->caching is
greater than 0x1f then all then all the upper 32 bits of
the u64 end up as also being set because of the int
sign-extension. Fix this by casting the u8 values to a
u64 before the 26 bit left shift.

Addresses-Coverity: ("Unintended sign extension")
Fixes: 4863dea3fab0 ("net: Adding support for Cavium ThunderX network controller")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agocxgb4: Fix unintentional sign extension issues
Colin Ian King [Fri, 9 Apr 2021 11:08:57 +0000 (12:08 +0100)]
cxgb4: Fix unintentional sign extension issues

The shifting of the u8 integers f->fs.nat_lip[] by 24 bits to
the left will be promoted to a 32 bit signed int and then
sign-extended to a u64. In the event that the top bit of the u8
is set then all then all the upper 32 bits of the u64 end up as
also being set because of the sign-extension. Fix this by
casting the u8 values to a u64 before the 24 bit left shift.

Addresses-Coverity: ("Unintended sign extension")
Fixes: 12b276fbf6e0 ("cxgb4: add support to create hash filters")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'ipa-next'
David S. Miller [Sun, 11 Apr 2021 23:49:08 +0000 (16:49 -0700)]
Merge branch 'ipa-next'

Alex Elder says:

====================
net: ipa: support two more platforms

This series adds IPA support for two more Qualcomm SoCs.

The first patch updates the DT binding to add compatible strings.

The second temporarily disables checksum offload support for IPA
version 4.5 and above.  Changes are required to the RMNet driver
to support the "inline" checksum offload used for IPA v4.5+, and
once those are present this capability will be enabled for IPA.

The third and fourth patches add configuration data for IPA versions
4.5 (used for the SDX55 SoC) and 4.11 (used for the SD7280 SoC).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipa: add IPA v4.11 configuration data
Alex Elder [Fri, 9 Apr 2021 20:40:24 +0000 (15:40 -0500)]
net: ipa: add IPA v4.11 configuration data

Add support for the SC7280 SoC, which includes IPA version 4.11.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipa: add IPA v4.5 configuration data
Alex Elder [Fri, 9 Apr 2021 20:40:23 +0000 (15:40 -0500)]
net: ipa: add IPA v4.5 configuration data

Add support for the SDX55 SoC, which includes IPA version 4.5.

Starting with IPA v4.5, a few of the memory regions have a different
number of "canary" values; update comments in the where the region
identifers are defined to accurately reflect that.

I'll note three differences in SDX55 versus the other two existing
platforms (SDM845 and SC7180):
  - SDX55 uses a 32-bit Linux kernel
  - SDX55 has four interconnects rather than three
  - SDX55 uses IPA v4.5, which uses inline checksum offload

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ipa: disable checksum offload for IPA v4.5+
Alex Elder [Fri, 9 Apr 2021 20:40:22 +0000 (15:40 -0500)]
net: ipa: disable checksum offload for IPA v4.5+

Checksum offload for IPA v4.5+ is implemented differently, using
"inline" offload (which uses a common header format for both upload
and download offload).

The IPA hardware must be programmed to enable MAP checksum offload,
but the RMNet driver is responsible for interpreting checksum
metadata supplied with messages.

Currently, the RMNet driver does not support inline checksum offload.
This support is imminent, but until it is available, do not allow
newer versions of IPA to specify checksum offload for endpoints.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agodt-bindings: net: qcom,ipa: add some compatible strings
Alex Elder [Fri, 9 Apr 2021 20:40:21 +0000 (15:40 -0500)]
dt-bindings: net: qcom,ipa: add some compatible strings

Add existing supported platform "qcom,sc7180-ipa" to the set of IPA
compatible strings.  Also add newly-supported "qcom,sdx55-ipa",
"qcom,sc7280-ipa".

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoehea: add missing MODULE_DEVICE_TABLE
Qiheng Lin [Fri, 9 Apr 2021 11:09:11 +0000 (19:09 +0800)]
ehea: add missing MODULE_DEVICE_TABLE

This patch adds missing MODULE_DEVICE_TABLE definition which generates
correct modalias for automatic loading of this driver when it is built
as an external module.

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Qiheng Lin <linqiheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'veth-gro'
David S. Miller [Sun, 11 Apr 2021 23:39:28 +0000 (16:39 -0700)]
Merge branch 'veth-gro'

Paolo Abeni  says:

====================
veth: allow GRO even without XDP

This series allows the user-space to enable GRO/NAPI on a veth
device even without attaching an XDP program.

It does not change the default veth behavior (no NAPI, no GRO),
except that the GRO feature bit on top of this series will be
effectively off by default on veth devices. Note that currently
the GRO bit is on by default, but GRO never takes place in
absence of XDP.

On top of this series, setting the GRO feature bit enables NAPI
and allows the GRO to take place. The TSO features on the peer
device are preserved.

The main goal is improving UDP forwarding performances for
containers in a typical virtual network setup:

(container) veth -> veth peer -> bridge/ovs -> vxlan -> NIC

Enabling the NAPI threaded mode, GRO the NETIF_F_GRO_UDP_FWD
feature on the veth peer improves the UDP stream performance
with not void netfilter configuration by 2x factor with no
measurable overhead for TCP traffic: some heuristic ensures
that TCP will not go through the additional NAPI/GRO layer.

Some self-tests are added to check the expected behavior in
the default configuration, with XDP and with plain GRO enabled.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoself-tests: add veth tests
Paolo Abeni [Fri, 9 Apr 2021 11:04:40 +0000 (13:04 +0200)]
self-tests: add veth tests

Add some basic veth tests, that verify the expected flags and
aggregation with different setups (default, xdp, etc...)

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoveth: refine napi usage
Paolo Abeni [Fri, 9 Apr 2021 11:04:39 +0000 (13:04 +0200)]
veth: refine napi usage

After the previous patch, when enabling GRO, locally generated
TCP traffic experiences some measurable overhead, as it traverses
the GRO engine without any chance of aggregation.

This change refine the NAPI receive path admission test, to avoid
unnecessary GRO overhead in most scenarios, when GRO is enabled
on a veth peer.

Only skbs that are eligible for aggregation enter the GRO layer,
the others will go through the traditional receive path.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoveth: allow enabling NAPI even without XDP
Paolo Abeni [Fri, 9 Apr 2021 11:04:38 +0000 (13:04 +0200)]
veth: allow enabling NAPI even without XDP

Currently the veth device has the GRO feature bit set, even if
no GRO aggregation is possible with the default configuration,
as the veth device does not hook into the GRO engine.

Flipping the GRO feature bit from user-space is a no-op, unless
XDP is enabled. In such scenario GRO could actually take place, but
TSO is forced to off on the peer device.

This change allow user-space to really control the GRO feature, with
no need for an XDP program.

The GRO feature bit is now cleared by default - so that there are no
user-visible behavior changes with the default configuration.

When the GRO bit is set, the per-queue NAPI instances are initialized
and registered. On xmit, when napi instances are available, we try
to use them.

Some additional checks are in place to ensure we initialize/delete NAPIs
only when needed in case of overlapping XDP and GRO configuration
changes.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoveth: use skb_orphan_partial instead of skb_orphan
Paolo Abeni [Fri, 9 Apr 2021 11:04:37 +0000 (13:04 +0200)]
veth: use skb_orphan_partial instead of skb_orphan

As described by commit 9c4c325252c5 ("skbuff: preserve sock
reference when scrubbing the skb."), orphaning a skb
in the TX path will cause OoO.

Let's use skb_orphan_partial() instead of skb_orphan(), so
that we keep the sk around for queue's selection sake and we
still avoid the problem fixed with commit 4bf9ffa0fb57 ("veth:
Orphan skb before GRO")

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'ethtool-eeprom'
David S. Miller [Sun, 11 Apr 2021 23:34:56 +0000 (16:34 -0700)]
Merge branch 'ethtool-eeprom'

Moshe Shemesh says:

====================
ethtool: Extend module EEPROM dump API

Ethtool supports module EEPROM dumps via the `ethtool -m <dev>` command.
But in current state its functionality is limited - offset and length
parameters, which are used to specify a linear desired region of EEPROM
data to dump, is not enough, considering emergence of complex module
EEPROM layouts such as CMIS 4.0.
Moreover, CMIS 4.0 extends the amount of pages that may be accessible by
introducing another parameter for page addressing - banks.

Besides, currently module EEPROM is represented as a chunk of
concatenated pages, where lower 128 bytes of all pages, except page 00h,
are omitted. Offset and length are used to address parts of this fake
linear memory. But in practice drivers, which implement
get_module_info() and get_module_eeprom() ethtool ops still calculate
page number and set I2C address on their own.

This series tackles these issues by adding ethtool op, which allows to
pass page number, bank number and I2C address in addition to offset and
length parameters to the driver, adds corresponding netlink
infrastructure and implements the new interface in mlx5 driver.

This allows to extend userspace 'ethtool -m' CLI by adding new
parameters - page, bank and i2c. New command line format:
 ethtool -m <dev> [hex on|off] [raw on|off] [offset N] [length N] [page N] [bank N] [i2c N]

The consequence of this series is a possibility to dump arbitrary EEPROM
page at a time, in contrast to dumps of concatenated pages. Therefore,
offset and length change their semantics and may be used only to specify
a part of data within half page boundary, which size is currently limited
to 128 bytes.

As for drivers that support legacy get_module_info() and
get_module_eeprom() pair, the series addresses it by implementing a
fallback mechanism. As mentioned earlier, such drivers derive a page
number from 'global' offset, so this can be done vice versa without
their involvement thanks to standardization. If kernel netlink handler
of 'ethtool -m' command detects that new ethtool op is not supported by
the driver, it calculates offset from given page number and page offset
and calls old ndos, if they are available.
====================

\Signed-off-by: David S. Miller <davem@davemloft.net>

3 years agoethtool: wire in generic SFP module access
Andrew Lunn [Fri, 9 Apr 2021 08:06:41 +0000 (11:06 +0300)]
ethtool: wire in generic SFP module access

If the device has a sfp bus attached, call its
sfp_get_module_eeprom_by_page() function, otherwise use the ethtool op
for the device. This follows how the IOCTL works.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agophy: sfp: add netlink SFP support to generic SFP code
Andrew Lunn [Fri, 9 Apr 2021 08:06:40 +0000 (11:06 +0300)]
phy: sfp: add netlink SFP support to generic SFP code

The new netlink API for reading SFP data requires a new op to be
implemented. The idea of the new netlink SFP code is that userspace is
responsible to parsing the EEPROM data and requesting pages, rather
than have the kernel decide what pages are interesting and returning
them. This allows greater flexibility for newer formats.

Currently the generic SFP code only supports simple SFPs. Allow i2c
address 0x50 and 0x51 to be accessed with page and bank must always be
0. This interface will later be extended when for example QSFP support
is added.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoethtool: Add fallback to get_module_eeprom from netlink command
Vladyslav Tarasiuk [Fri, 9 Apr 2021 08:06:39 +0000 (11:06 +0300)]
ethtool: Add fallback to get_module_eeprom from netlink command

In case netlink get_module_eeprom_by_page() callback is not implemented
by the driver, try to call old get_module_info() and get_module_eeprom()
pair. Recalculate parameters to get_module_eeprom() offset and len using
page number and their sizes. Return error if this can't be done.

Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet: ethtool: Export helpers for getting EEPROM info
Andrew Lunn [Fri, 9 Apr 2021 08:06:38 +0000 (11:06 +0300)]
net: ethtool: Export helpers for getting EEPROM info

There are two ways to retrieve information from SFP EEPROMs.  Many
devices make use of the common code, and assign the sfp_bus pointer in
the netdev to point to the bus holding the SFP device. Some MAC
drivers directly implement ops in there ethool structure.

Export within net/ethtool the two helpers used to call these methods,
so that they can also be used in the new netlink code.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx5: Add support for DSFP module EEPROM dumps
Vladyslav Tarasiuk [Fri, 9 Apr 2021 08:06:37 +0000 (11:06 +0300)]
net/mlx5: Add support for DSFP module EEPROM dumps

Allow the driver to recognise DSFP transceiver module ID and therefore
allow its EEPROM dumps using ethtool.

Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx5: Implement get_module_eeprom_by_page()
Vladyslav Tarasiuk [Fri, 9 Apr 2021 08:06:36 +0000 (11:06 +0300)]
net/mlx5: Implement get_module_eeprom_by_page()

Implement ethtool_ops::get_module_eeprom_by_page() to enable
support of new SFP standards.

Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agonet/mlx5: Refactor module EEPROM query
Vladyslav Tarasiuk [Fri, 9 Apr 2021 08:06:35 +0000 (11:06 +0300)]
net/mlx5: Refactor module EEPROM query

Prepare for ethtool_ops::get_module_eeprom_data() implementation by
extracting common part of mlx5_query_module_eeprom() into a separate
function.

Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoethtool: Allow network drivers to dump arbitrary EEPROM data
Vladyslav Tarasiuk [Fri, 9 Apr 2021 08:06:34 +0000 (11:06 +0300)]
ethtool: Allow network drivers to dump arbitrary EEPROM data

Define get_module_eeprom_by_page() ethtool callback and implement
netlink infrastructure.

get_module_eeprom_by_page() allows network drivers to dump a part of
module's EEPROM specified by page and bank numbers along with offset and
length. It is effectively a netlink replacement for get_module_info()
and get_module_eeprom() pair, which is needed due to emergence of
complex non-linear EEPROM layouts.

Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
3 years agoMerge branch 'net-ipa-a-few-small-fixes'
Jakub Kicinski [Sat, 10 Apr 2021 03:57:28 +0000 (20:57 -0700)]
Merge branch 'net-ipa-a-few-small-fixes'

Alex Elder says:

====================
net: ipa: a few small fixes

This series implements some minor bug fixes or improvements.

The first patch removes an apparently unnecessary restriction, which
results in an error on a 32-bit ARM build.

The second makes a definition used for SDM845 match what is used in
the downstream code.

The third just ensures two netdev pointers are only non-null when
valid.

The fourth simplifies a little code, knowing that a called function
never returns an error.

The fifth and sixth just remove some empty/place holder functions.

And the last patch fixes a comment, makes a function private, and
removes an unnecessary double-negation of a Boolean variable.  This
patch produces a warning from checkpatch, indicating that a pair of
parentheses is unnecessary.  I agree with that advice, but it
conflicts with a suggestion from the compiler.  I left the "problem"
in place to avoid the compiler warning.
====================

Link: https://lore.kernel.org/r/20210409180722.1176868-1-elder@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: three small fixes
Alex Elder [Fri, 9 Apr 2021 18:07:22 +0000 (13:07 -0500)]
net: ipa: three small fixes

Some time ago changes were made to stop referring to clearing the
hardware pipeline as a "tag process."  Fix a comment to use the
newer terminology.

Get rid of a pointless double-negation of the Boolean toward_ipa
flag in ipa_endpoint_config().

make ipa_endpoint_exit_one() private; it's only referenced inside
"ipa_endpoint.c".

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: get rid of empty GSI functions
Alex Elder [Fri, 9 Apr 2021 18:07:21 +0000 (13:07 -0500)]
net: ipa: get rid of empty GSI functions

There are place holder functions in the GSI code that do nothing.
Remove these, knowing we can add something back in their place if
they're really needed someday.

Some of these are inverse functions (such as teardown to match setup).
Explicitly comment that there is no inverse in these cases.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: get rid of empty IPA functions
Alex Elder [Fri, 9 Apr 2021 18:07:20 +0000 (13:07 -0500)]
net: ipa: get rid of empty IPA functions

There are place holder functions in the IPA code that do nothing.
For the most part these are inverse functions, for example, once the
routing or filter tables are set up there is no need to perform any
matching teardown activity at shutdown, or in the case of an error.

These can be safely removed, resulting in some code simplification.
Add comments in these spots making it explicit that there is no
inverse.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: ipa_stop() does not return an error
Alex Elder [Fri, 9 Apr 2021 18:07:19 +0000 (13:07 -0500)]
net: ipa: ipa_stop() does not return an error

In ipa_modem_stop(), if the modem netdev pointer is non-null we call
ipa_stop().  We check for an error and if one is returned we handle
it.  But ipa_stop() never returns an error, so this extra handling
is unnecessary.  Simplify the code in ipa_modem_stop() based on the
knowledge no error handling is needed at this spot.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: only set endpoint netdev pointer when in use
Alex Elder [Fri, 9 Apr 2021 18:07:18 +0000 (13:07 -0500)]
net: ipa: only set endpoint netdev pointer when in use

In ipa_modem_start(), we set endpoint netdev pointers before the
network device is registered.  If registration fails, we don't undo
those assignments.  Instead, wait to assign the netdev pointer until
after registration succeeds.

Set these endpoint netdev pointers to NULL in ipa_modem_stop()
before unregistering the network device.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: update sequence type for modem TX endpoint
Alex Elder [Fri, 9 Apr 2021 18:07:17 +0000 (13:07 -0500)]
net: ipa: update sequence type for modem TX endpoint

On IPA v3.5.1, the sequencer type for the modem TX endpoint does not
define the replication portion in the same way the downstream code
does.  This difference doesn't affect the behavior of the upstream
code, but I'd prefer the two code bases use the same configuration
value here.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: ipa: relax pool entry size requirement
Alex Elder [Fri, 9 Apr 2021 18:07:16 +0000 (13:07 -0500)]
net: ipa: relax pool entry size requirement

I no longer know why a validation check ensured the size of an entry
passed to gsi_trans_pool_init() was restricted to be a multiple of 8.
For 32-bit builds, this condition doesn't always hold, and for DMA
pools, the size is rounded up to a power of 2 anyway.

Remove this restriction.

Signed-off-by: Alex Elder <elder@linaro.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Jakub Kicinski [Sat, 10 Apr 2021 03:46:01 +0000 (20:46 -0700)]
Merge git://git./linux/kernel/git/netdev/net

Conflicts:

MAINTAINERS
 - keep Chandrasekar
drivers/net/ethernet/mellanox/mlx5/core/en_main.c
 - simple fix + trust the code re-added to param.c in -next is fine
include/linux/bpf.h
 - trivial
include/linux/ethtool.h
 - trivial, fix kdoc while at it
include/linux/skmsg.h
 - move to relevant place in tcp.c, comment re-wrapped
net/core/skmsg.c
 - add the sk = sk // sk = NULL around calls
net/tipc/crypto.c
 - trivial

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoenetc: Use generic rule to map Tx rings to interrupt vectors
Claudiu Manoil [Fri, 9 Apr 2021 07:16:13 +0000 (10:16 +0300)]
enetc: Use generic rule to map Tx rings to interrupt vectors

Even if the current mapping is correct for the 1 CPU and 2 CPU cases
(currently enetc is included in SoCs with up to 2 CPUs only), better
use a generic rule for the mapping to cover all possible cases.
The number of CPUs is the same as the number of interrupt vectors:

Per device Tx rings -
device_tx_ring[idx], where idx = 0..n_rings_total-1

Per interrupt vector Tx rings -
int_vector[i].ring[j], where i = 0..n_int_vects-1
     j = 0..n_rings_per_v-1

Mapping rule -
n_rings_per_v = n_rings_total / n_int_vects
for i = 0..n_int_vects - 1:
for j = 0..n_rings_per_v - 1:
idx = n_int_vects * j + i
int_vector[i].ring[j] <- device_tx_ring[idx]

Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20210409071613.28912-1-claudiu.manoil@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: enetc: fix TX ring interrupt storm
Vladimir Oltean [Fri, 9 Apr 2021 19:27:59 +0000 (22:27 +0300)]
net: enetc: fix TX ring interrupt storm

The blamed commit introduced a bit in the TX software buffer descriptor
structure for determining whether a BD is final or not; we rearm the TX
interrupt vector for every frame (hence final BD) transmitted.

But there is a problem with the patch: it replaced a condition whose
expression is a bool which was evaluated at the beginning of the "while"
loop with a bool expression that is evaluated on the spot: tx_swbd->is_eof.

The problem with the latter expression is that the tx_swbd has already
been incremented at that stage, so the tx_swbd->is_eof check is in fact
with the _next_ software BD. Which is _not_ final.

The effect is that the CPU is in 100% load with ksoftirqd because it
does not acknowledge the TX interrupt, so the handler keeps getting
called again and again.

The fix is to restore the code structure, and keep the local bool is_eof
variable, just to assign it the tx_swbd->is_eof value instead of
!!tx_swbd->skb.

Fixes: d504498d2eb3 ("net: enetc: add a dedicated is_eof bit in the TX software BD")
Reported-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Link: https://lore.kernel.org/r/20210409192759.3895104-1-olteanv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox...
Jakub Kicinski [Sat, 10 Apr 2021 01:07:20 +0000 (18:07 -0700)]
Merge branch 'mlx5-next' of git://git./linux/kernel/git/mellanox/linux

Saeed Mahameed says:

====================
mlx5-next 2021-04-09

This pr contains changes from  mlx5-next branch,
already reviewed on netdev and rdma mailing lists, links below.

1) From Leon, Dynamically assign MSI-X vectors count
Already Acked by Bjorn Helgaas.
https://patchwork.kernel.org/project/netdevbpf/cover/20210314124256.70253-1-leon@kernel.org/

2) Cleanup series:
https://patchwork.kernel.org/project/netdevbpf/cover/20210311070915.321814-1-saeed@kernel.org/

From Mark, E-Switch cleanups and refactoring, and the addition
of single FDB mode needed HW bits.

From Mikhael, Remove unused struct field

From Saeed, Cleanup W=1 prototype warning

From Zheng, Esw related cleanup

From Tariq, User order-0 page allocation for EQs

* 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
  net/mlx5: Implement sriov_get_vf_total_msix/count() callbacks
  net/mlx5: Dynamically assign MSI-X vectors count
  net/mlx5: Add dynamic MSI-X capabilities bits
  PCI/IOV: Add sysfs MSI-X vector assignment interface
  net/mlx5: Use order-0 allocations for EQs
  net/mlx5: Add IFC bits needed for single FDB mode
  net/mlx5: E-Switch, Refactor send to vport to be more generic
  RDMA/mlx5: Use representor E-Switch when getting netdev and metadata
  net/mlx5: E-Switch, Add eswitch pointer to each representor
  net/mlx5: E-Switch, Add match on vhca id to default send rules
  net/mlx5: Remove unused mlx5_core_health member recover_work
  net/mlx5: simplify the return expression of mlx5_esw_offloads_pair()
  net/mlx5: Cleanup prototype warning
====================

Link: https://lore.kernel.org/r/20210409200704.10886-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: enetc: fix array underflow in error handling code
Dan Carpenter [Fri, 9 Apr 2021 12:24:28 +0000 (15:24 +0300)]
net: enetc: fix array underflow in error handling code

This loop will try to unmap enetc_unmap_tx_buff[-1] and crash.

Fixes: 9d2b68cc108d ("net: enetc: add support for XDP_REDIRECT")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/YHBHfCY/yv3EnM9z@mwanda
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agocxgb4: remove unneeded if-null-free check
Qiheng Lin [Fri, 9 Apr 2021 11:53:39 +0000 (19:53 +0800)]
cxgb4: remove unneeded if-null-free check

Eliminate the following coccicheck warning:

drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c:529:3-9: WARNING:
 NULL check before some freeing functions is not needed.
drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_u32.c:533:2-8: WARNING:
 NULL check before some freeing functions is not needed.
drivers/net/ethernet/chelsio/cxgb4/cxgb4_cudbg.c:161:2-7: WARNING:
 NULL check before some freeing functions is not needed.
drivers/net/ethernet/chelsio/cxgb4/clip_tbl.c:327:3-9: WARNING:
 NULL check before some freeing functions is not needed.

Signed-off-by: Qiheng Lin <linqiheng@huawei.com>
Link: https://lore.kernel.org/r/20210409115339.4598-1-linqiheng@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge branch 'net-make-phy-pm-ops-a-no-op-if-mac-driver-manages-phy-pm'
Jakub Kicinski [Fri, 9 Apr 2021 23:37:08 +0000 (16:37 -0700)]
Merge branch 'net-make-phy-pm-ops-a-no-op-if-mac-driver-manages-phy-pm'

Heiner Kallweit says:

====================
net: make PHY PM ops a no-op if MAC driver manages PHY PM

Resume callback of the PHY driver is called after the one for the MAC
driver. The PHY driver resume callback calls phy_init_hw(), and this is
potentially problematic if the MAC driver calls phy_start() in its resume
callback. One issue was reported with the fec driver and a KSZ8081 PHY
which seems to become unstable if a soft reset is triggered during aneg.

The new flag allows MAC drivers to indicate that they take care of
suspending/resuming the PHY. Then the MAC PM callbacks can handle
any dependency between MAC and PHY PM.
====================

Link: https://lore.kernel.org/r/9e695411-ab1d-34fe-8b90-3e8192ab84f6@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agor8169: use mac-managed PHY PM
Heiner Kallweit [Wed, 7 Apr 2021 15:53:52 +0000 (17:53 +0200)]
r8169: use mac-managed PHY PM

Use the new mac_managed_pm flag to indicate that the driver takes care
of PHY power management.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: fec: use mac-managed PHY PM
Heiner Kallweit [Wed, 7 Apr 2021 15:52:45 +0000 (17:52 +0200)]
net: fec: use mac-managed PHY PM

Use the new mac_managed_pm flag to work around an issue with KSZ8081 PHY
that becomes unstable when a soft reset is triggered during aneg.

Reported-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Tested-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: phy: make PHY PM ops a no-op if MAC driver manages PHY PM
Heiner Kallweit [Wed, 7 Apr 2021 15:51:56 +0000 (17:51 +0200)]
net: phy: make PHY PM ops a no-op if MAC driver manages PHY PM

Resume callback of the PHY driver is called after the one for the MAC
driver. The PHY driver resume callback calls phy_init_hw(), and this is
potentially problematic if the MAC driver calls phy_start() in its resume
callback. One issue was reported with the fec driver and a KSZ8081 PHY
which seems to become unstable if a soft reset is triggered during aneg.

The new flag allows MAC drivers to indicate that they take care of
suspending/resuming the PHY. Then the MAC PM callbacks can handle
any dependency between MAC and PHY PM.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoRevert "tcp: Reset tcp connections in SYN-SENT state"
Eric Dumazet [Fri, 9 Apr 2021 17:02:37 +0000 (10:02 -0700)]
Revert "tcp: Reset tcp connections in SYN-SENT state"

This reverts commit e880f8b3a24a73704731a7227ed5fee14bd90192.

1) Patch has not been properly tested, and is wrong [1]
2) Patch submission did not include TCP maintainer (this is me)

[1]
divide error: 0000 [#1] PREEMPT SMP KASAN
CPU: 0 PID: 8426 Comm: syz-executor478 Not tainted 5.12.0-rc4-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:__tcp_select_window+0x56d/0xad0 net/ipv4/tcp_output.c:3015
Code: 44 89 ff e8 d5 cd f0 f9 45 39 e7 0f 8d 20 ff ff ff e8 f7 c7 f0 f9 44 89 e3 e9 13 ff ff ff e8 ea c7 f0 f9 44 89 e0 44 89 e3 99 <f7> 7c 24 04 29 d3 e9 fc fe ff ff e8 d3 c7 f0 f9 41 f7 dc bf 1f 00
RSP: 0018:ffffc9000184fac0 EFLAGS: 00010293
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: ffffffff87832e76 RDI: 0000000000000003
RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
R10: ffffffff87832e14 R11: 0000000000000000 R12: 0000000000000000
R13: 1ffff92000309f5c R14: 0000000000000000 R15: 0000000000000000
FS:  00000000023eb300(0000) GS:ffff8880b9c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fc2b5f426c0 CR3: 000000001c5cf000 CR4: 00000000001506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 tcp_select_window net/ipv4/tcp_output.c:264 [inline]
 __tcp_transmit_skb+0xa82/0x38f0 net/ipv4/tcp_output.c:1351
 tcp_transmit_skb net/ipv4/tcp_output.c:1423 [inline]
 tcp_send_active_reset+0x475/0x8e0 net/ipv4/tcp_output.c:3449
 tcp_disconnect+0x15a9/0x1e60 net/ipv4/tcp.c:2955
 inet_shutdown+0x260/0x430 net/ipv4/af_inet.c:905
 __sys_shutdown_sock net/socket.c:2189 [inline]
 __sys_shutdown_sock net/socket.c:2183 [inline]
 __sys_shutdown+0xf1/0x1b0 net/socket.c:2201
 __do_sys_shutdown net/socket.c:2209 [inline]
 __se_sys_shutdown net/socket.c:2207 [inline]
 __x64_sys_shutdown+0x50/0x70 net/socket.c:2207
 do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Fixes: e880f8b3a24a ("tcp: Reset tcp connections in SYN-SENT state")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Manoj Basapathi <manojbm@codeaurora.org>
Cc: Sauvik Saha <ssaha@codeaurora.org>
Link: https://lore.kernel.org/r/20210409170237.274904-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: dccp: use net_generic storage
Florian Westphal [Thu, 8 Apr 2021 17:45:02 +0000 (19:45 +0200)]
net: dccp: use net_generic storage

DCCP is virtually never used, so no need to use space in struct net for it.

Put the pernet ipv4/v6 socket in the dccp ipv4/ipv6 modules instead.

Signed-off-by: Florian Westphal <fw@strlen.de>
Link: https://lore.kernel.org/r/20210408174502.1625-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoMerge tag 'net-5.12-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Fri, 9 Apr 2021 22:26:51 +0000 (15:26 -0700)]
Merge tag 'net-5.12-rc7' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Jakub Kicinski:
 "Networking fixes for 5.12-rc7, including fixes from can, ipsec,
  mac80211, wireless, and bpf trees.

  No scary regressions here or in the works, but small fixes for 5.12
  changes keep coming.

  Current release - regressions:

   - virtio: do not pull payload in skb->head

   - virtio: ensure mac header is set in virtio_net_hdr_to_skb()

   - Revert "net: correct sk_acceptq_is_full()"

   - mptcp: revert "mptcp: provide subflow aware release function"

   - ethernet: lan743x: fix ethernet frame cutoff issue

   - dsa: fix type was not set for devlink port

   - ethtool: remove link_mode param and derive link params from driver

   - sched: htb: fix null pointer dereference on a null new_q

   - wireless: iwlwifi: Fix softirq/hardirq disabling in
     iwl_pcie_enqueue_hcmd()

   - wireless: iwlwifi: fw: fix notification wait locking

   - wireless: brcmfmac: p2p: Fix deadlock introduced by avoiding the
     rtnl dependency

  Current release - new code bugs:

   - napi: fix hangup on napi_disable for threaded napi

   - bpf: take module reference for trampoline in module

   - wireless: mt76: mt7921: fix airtime reporting and related tx hangs

   - wireless: iwlwifi: mvm: rfi: don't lock mvm->mutex when sending
     config command

  Previous releases - regressions:

   - rfkill: revert back to old userspace API by default

   - nfc: fix infinite loop, refcount & memory leaks in LLCP sockets

   - let skb_orphan_partial wake-up waiters

   - xfrm/compat: Cleanup WARN()s that can be user-triggered

   - vxlan, geneve: do not modify the shared tunnel info when PMTU
     triggers an ICMP reply

   - can: fix msg_namelen values depending on CAN_REQUIRED_SIZE

   - can: uapi: mark union inside struct can_frame packed

   - sched: cls: fix action overwrite reference counting

   - sched: cls: fix err handler in tcf_action_init()

   - ethernet: mlxsw: fix ECN marking in tunnel decapsulation

   - ethernet: nfp: Fix a use after free in nfp_bpf_ctrl_msg_rx

   - ethernet: i40e: fix receiving of single packets in xsk zero-copy
     mode

   - ethernet: cxgb4: avoid collecting SGE_QBASE regs during traffic

  Previous releases - always broken:

   - bpf: Refuse non-O_RDWR flags in BPF_OBJ_GET

   - bpf: Refcount task stack in bpf_get_task_stack

   - bpf, x86: Validate computation of branch displacements

   - ieee802154: fix many similar syzbot-found bugs
       - fix NULL dereferences in netlink attribute handling
       - reject unsupported operations on monitor interfaces
       - fix error handling in llsec_key_alloc()

   - xfrm: make ipv4 pmtu check honor ip header df

   - xfrm: make hash generation lock per network namespace

   - xfrm: esp: delete NETIF_F_SCTP_CRC bit from features for esp
     offload

   - ethtool: fix incorrect datatype in set_eee ops

   - xdp: fix xdp_return_frame() kernel BUG throw for page_pool memory
     model

   - openvswitch: fix send of uninitialized stack memory in ct limit
     reply

  Misc:

   - udp: add get handling for UDP_GRO sockopt"

* tag 'net-5.12-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (182 commits)
  net: fix hangup on napi_disable for threaded napi
  net: hns3: Trivial spell fix in hns3 driver
  lan743x: fix ethernet frame cutoff issue
  net: ipv6: check for validity before dereferencing cfg->fc_nlinfo.nlh
  net: dsa: lantiq_gswip: Configure all remaining GSWIP_MII_CFG bits
  net: dsa: lantiq_gswip: Don't use PHY auto polling
  net: sched: sch_teql: fix null-pointer dereference
  ipv6: report errors for iftoken via netlink extack
  net: sched: fix err handler in tcf_action_init()
  net: sched: fix action overwrite reference counting
  Revert "net: sched: bump refcount for new action in ACT replace mode"
  ice: fix memory leak of aRFS after resuming from suspend
  i40e: Fix sparse warning: missing error code 'err'
  i40e: Fix sparse error: 'vsi->netdev' could be null
  i40e: Fix sparse error: uninitialized symbol 'ring'
  i40e: Fix sparse errors in i40e_txrx.c
  i40e: Fix parameters in aq_get_phy_register()
  nl80211: fix beacon head validation
  bpf, x86: Validate computation of branch displacements for x86-32
  bpf, x86: Validate computation of branch displacements for x86-64
  ...

3 years agoMerge tag 'io_uring-5.12-2021-04-09' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 9 Apr 2021 22:06:52 +0000 (15:06 -0700)]
Merge tag 'io_uring-5.12-2021-04-09' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "Two minor fixups for the reissue logic, and one for making sure that
  unbounded work is canceled on io-wq exit"

* tag 'io_uring-5.12-2021-04-09' of git://git.kernel.dk/linux-block:
  io-wq: cancel unbounded works on io-wq destroy
  io_uring: fix rw req completion
  io_uring: clear F_REISSUE right after getting it

3 years agoMerge tag 'devicetree-fixes-for-5.12-2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 9 Apr 2021 20:01:48 +0000 (13:01 -0700)]
Merge tag 'devicetree-fixes-for-5.12-2' of git://git./linux/kernel/git/robh/linux

Pull devicetree fixes from Rob Herring:

 - Fix fw_devlink failure with ".*,nr-gpios" properties

 - Doc link reference fixes from Mauro

 - Fixes for unaligned FDT handling found on OpenRisc. First, avoid
   crash with better error handling when unflattening an unaligned FDT.
   Second, fix memory allocations for FDTs to ensure alignment.

* tag 'devicetree-fixes-for-5.12-2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  of: property: fw_devlink: do not link ".*,nr-gpios"
  dt-bindings:iio:adc: update motorola,cpcap-adc.yaml reference
  dt-bindings: fix references for iio-bindings.txt
  dt-bindings: don't use ../dir for doc references
  of: unittest: overlay: ensure proper alignment of copied FDT
  of: properly check for error returned by fdt_get_name()

3 years agoMerge tag 'drm-fixes-2021-04-10' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 9 Apr 2021 19:56:10 +0000 (12:56 -0700)]
Merge tag 'drm-fixes-2021-04-10' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Was relatively quiet this week, but still a few pulls came in, pretty
  much small fixes across the board, a couple of regression fixes in the
  amdgpu/radeon code, msm has a few minor fixes across the board, a
  panel regression fix also.

  amdgpu:
   - DCN3 fix
   - Fix CAC setting regression for TOPAZ
   - Fix ttm regression

  radeon:
   - Fix ttm regression

  msm:
   - a5xx/a6xx timestamp fix
   - microcode version check
   - fail path fix
   - block programming fix
   - error removal fix

  i915:
   - Fix invalid access to ACPI _DSM objects

  xen:
   - Fix use-after-free in xen
   - minor duplicate defintion cleanup

  vc4:
   - Reduce fifo threshold on hvs4 to fix a fifo full error
   - minor redunantant assignment cleanup

  panel:
   - Disable TE support for Droid4 and N950"

* tag 'drm-fixes-2021-04-10' of git://anongit.freedesktop.org/drm/drm:
  drm/vc4: crtc: Reduce PV fifo threshold on hvs4
  drm/vc4: plane: Remove redundant assignment
  drm/amdgpu/smu7: fix CAC setting on TOPAZ
  drm/radeon: Fix size overflow
  drm/amdgpu: Fix size overflow
  drm/i915: Fix invalid access to ACPI _DSM objects
  drm/amd/display: Add missing mask for DCN3
  drm/panel: panel-dsi-cm: disable TE for now
  drm/msm/disp/dpu1: program 3d_merge only if block is attached
  drm/msm: a6xx: fix version check for the A650 SQE microcode
  drm/msm: Fix a5xx/a6xx timestamps
  drm/msm: Fix removal of valid error case when checking speed_bin
  drm/msm: Set drvdata to NULL when msm_drm_init() fails
  drivers: gpu: drm: xen_drm_front_drm_info is declared twice
  gpu/xen: Fix a use after free in xen_drm_drv_init

3 years agonet: fix hangup on napi_disable for threaded napi
Paolo Abeni [Fri, 9 Apr 2021 15:24:17 +0000 (17:24 +0200)]
net: fix hangup on napi_disable for threaded napi

napi_disable() is subject to an hangup, when the threaded
mode is enabled and the napi is under heavy traffic.

If the relevant napi has been scheduled and the napi_disable()
kicks in before the next napi_threaded_wait() completes - so
that the latter quits due to the napi_disable_pending() condition,
the existing code leaves the NAPI_STATE_SCHED bit set and the
napi_disable() loop waiting for such bit will hang.

This patch addresses the issue by dropping the NAPI_STATE_DISABLE
bit test in napi_thread_wait(). The later napi_threaded_poll()
iteration will take care of clearing the NAPI_STATE_SCHED.

This also addresses a related problem reported by Jakub:
before this patch a napi_disable()/napi_enable() pair killed
the napi thread, effectively disabling the threaded mode.
On the patched kernel napi_disable() simply stops scheduling
the relevant thread.

v1 -> v2:
  - let the main napi_thread_poll() loop clear the SCHED bit

Reported-by: Jakub Kicinski <kuba@kernel.org>
Fixes: 29863d41bb6e ("net: implement threaded-able napi poll loop support")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/883923fa22745a9589e8610962b7dc59df09fb1f.1617981844.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agonet: hns3: Trivial spell fix in hns3 driver
Salil Mehta [Fri, 9 Apr 2021 07:42:23 +0000 (08:42 +0100)]
net: hns3: Trivial spell fix in hns3 driver

Some trivial spelling mistakes which caught my eye during the
review of the code.

Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Link: https://lore.kernel.org/r/20210409074223.32480-1-salil.mehta@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agolan743x: fix ethernet frame cutoff issue
Sven Van Asbroeck [Fri, 9 Apr 2021 00:39:04 +0000 (20:39 -0400)]
lan743x: fix ethernet frame cutoff issue

The ethernet frame length is calculated incorrectly. Depending on
the value of RX_HEAD_PADDING, this may result in ethernet frames
that are too short (cut off at the end), or too long (garbage added
to the end).

Fix by calculating the ethernet frame length correctly. For added
clarity, use the ETH_FCS_LEN constant in the calculation.

Many thanks to Heiner Kallweit for suggesting this solution.

Suggested-by: Heiner Kallweit <hkallweit1@gmail.com>
Fixes: 3e21a10fdea3 ("lan743x: trim all 4 bytes of the FCS; not just 2")
Link: https://lore.kernel.org/lkml/20210408172353.21143-1-TheSven73@gmail.com/
Signed-off-by: Sven Van Asbroeck <thesven73@gmail.com>
Reviewed-by: George McCollister <george.mccollister@gmail.com>
Tested-by: George McCollister <george.mccollister@gmail.com>
Link: https://lore.kernel.org/r/20210409003904.8957-1-TheSven73@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
3 years agoof: property: fw_devlink: do not link ".*,nr-gpios"
Ilya Lipnitskiy [Mon, 5 Apr 2021 22:25:40 +0000 (15:25 -0700)]
of: property: fw_devlink: do not link ".*,nr-gpios"

[<vendor>,]nr-gpios property is used by some GPIO drivers[0] to indicate
the number of GPIOs present on a system, not define a GPIO. nr-gpios is
not configured by #gpio-cells and can't be parsed along with other
"*-gpios" properties.

nr-gpios without the "<vendor>," prefix is not allowed by the DT
spec[1], so only add exception for the ",nr-gpios" suffix and let the
error message continue being printed for non-compliant implementations.

[0] nr-gpios is referenced in Documentation/devicetree/bindings/gpio:
 - gpio-adnp.txt
 - gpio-xgene-sb.txt
 - gpio-xlp.txt
 - snps,dw-apb-gpio.yaml

Link: https://github.com/devicetree-org/dt-schema/blob/cb53a16a1eb3e2169ce170c071e47940845ec26e/schemas/gpio/gpio-consumer.yaml#L20
Fixes errors such as:
  OF: /palmbus@300000/gpio@600: could not find phandle

Fixes: 7f00be96f125 ("of: property: Add device link support for interrupt-parent, dmas and -gpio(s)")
Signed-off-by: Ilya Lipnitskiy <ilya.lipnitskiy@gmail.com>
Cc: Saravana Kannan <saravanak@google.com>
Cc: stable@vger.kernel.org # v5.5+
Link: https://lore.kernel.org/r/20210405222540.18145-1-ilya.lipnitskiy@gmail.com
Signed-off-by: Rob Herring <robh@kernel.org>
3 years agodt-bindings:iio:adc: update motorola,cpcap-adc.yaml reference
Mauro Carvalho Chehab [Fri, 9 Apr 2021 12:47:47 +0000 (14:47 +0200)]
dt-bindings:iio:adc: update motorola,cpcap-adc.yaml reference

Changeset 1ca9d1b1342d ("dt-bindings:iio:adc:motorola,cpcap-adc yaml conversion")
renamed: Documentation/devicetree/bindings/iio/adc/cpcap-adc.txt
to: Documentation/devicetree/bindings/iio/adc/motorola,cpcap-adc.yaml.

Update its cross-reference accordingly.

Fixes: 1ca9d1b1342d ("dt-bindings:iio:adc:motorola,cpcap-adc yaml conversion")
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Link: https://lore.kernel.org/r/3e205e5fa701e4bc15d39d6ac1f57717df2bb4c6.1617972339.git.mchehab+huawei@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
3 years agodt-bindings: fix references for iio-bindings.txt
Mauro Carvalho Chehab [Fri, 9 Apr 2021 12:47:46 +0000 (14:47 +0200)]
dt-bindings: fix references for iio-bindings.txt

The iio-bindings.txt was converted into two files and merged
at the dt-schema git tree at:

https://github.com/devicetree-org/dt-schema

Yet, some documents still refer to the old file. Fix their
references, in order to point to the right URL.

Fixes: dba91f82d580 ("dt-bindings:iio:iio-binding.txt Drop file as content now in dt-schema")
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Link: https://lore.kernel.org/r/4efd81eca266ca0875d3bf9d1672097444146c69.1617972339.git.mchehab+huawei@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
3 years agodt-bindings: don't use ../dir for doc references
Mauro Carvalho Chehab [Fri, 9 Apr 2021 12:47:45 +0000 (14:47 +0200)]
dt-bindings: don't use ../dir for doc references

As documents have been renamed and moved around, their
references will break, but this will be unnoticed, as the
script which checks for it won't handle "../" references.

So, replace them by the full patch.

Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Link: https://lore.kernel.org/r/68d3a1244119d1f2829c375b0ef554cf348bc89f.1617972339.git.mchehab+huawei@kernel.org
Signed-off-by: Rob Herring <robh@kernel.org>
3 years agoMerge tag 'drm-intel-fixes-2021-04-09' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Fri, 9 Apr 2021 19:18:31 +0000 (05:18 +1000)]
Merge tag 'drm-intel-fixes-2021-04-09' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

- Fix invalid access to ACPI _DSM objects (Takashi)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/YHAW6NInrybUoat6@intel.com
3 years agoMerge tag 'drm-misc-fixes-2021-04-09' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Fri, 9 Apr 2021 19:15:35 +0000 (05:15 +1000)]
Merge tag 'drm-misc-fixes-2021-04-09' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

drm-misc-fixes for v5.12-rc7:
- Fix use-after-free in xen.
- Reduce fifo threshold on hvs4 to fix a fifo full error.
- Disable TE support for Droid4 and N950.
- Small compiler fixes.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/e7647dd9-60c3-9dfd-a377-89d717212e13@linux.intel.com
3 years agoMerge tag 'selinux-pr-20210409' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 9 Apr 2021 18:51:06 +0000 (11:51 -0700)]
Merge tag 'selinux-pr-20210409' of git://git./linux/kernel/git/pcmoore/selinux

Pull selinux fixes from Paul Moore:
 "Three SELinux fixes.

  These fix known problems relating to (re)loading SELinux policy or
  changing the policy booleans, and pass our test suite without problem"

* tag 'selinux-pr-20210409' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/selinux:
  selinux: fix race between old and new sidtab
  selinux: fix cond_list corruption when changing booleans
  selinux: make nslot handling in avtab more robust

3 years agoMerge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
Linus Torvalds [Fri, 9 Apr 2021 17:09:51 +0000 (10:09 -0700)]
Merge tag 'for_linus' of git://git./linux/kernel/git/mst/vhost

Pull vdpa/mlx5 fixes from Michael Tsirkin:
 "Last minute fixes.

  These all look like something we are better off having
  than not ..."

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
  vdpa/mlx5: Fix suspend/resume index restoration
  vdpa/mlx5: Fix wrong use of bit numbers
  vdpa/mlx5: Retrieve BAR address suitable any function
  vdpa/mlx5: Use the correct dma device when registering memory
  vdpa/mlx5: should exclude header length and fcs from mtu

3 years agoMerge tag 'rproc-v5.12-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/anders...
Linus Torvalds [Fri, 9 Apr 2021 17:05:25 +0000 (10:05 -0700)]
Merge tag 'rproc-v5.12-fixes' of git://git./linux/kernel/git/andersson/remoteproc

Pull remoteproc fixes from Bjorn Andersson:
 "This fixes an issue with firmware loading on the TI K3 PRU, fixes
  compatibility with GNU binutils for the same and resolves link error
  due to a 64-bit division in the Qualcomm PIL info.

  It also recognizes Mathieu Poirier as co-maintainer of the remoteproc
  and rpmsg subsystems"

* tag 'rproc-v5.12-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/andersson/remoteproc:
  remoteproc: pru: Fix firmware loading crashes on K3 SoCs
  remoteproc: pru: Fix loading of GNU Binutils ELF
  MAINTAINERS: Add co-maintainer for remoteproc/RPMSG subsystems
  remoteproc: qcom: pil_info: avoid 64-bit division

3 years agoMerge tag 'for-linus-5.12b-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 9 Apr 2021 16:58:42 +0000 (09:58 -0700)]
Merge tag 'for-linus-5.12b-rc7-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fix from Juergen Gross:
 "A single fix of a 5.12 patch for the rather uncommon problem of
  running as a Xen guest with a real time kernel config"

* tag 'for-linus-5.12b-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/evtchn: Change irq_info lock to raw_spinlock_t

3 years agoMerge tag 'acpi-5.12-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Fri, 9 Apr 2021 16:25:31 +0000 (09:25 -0700)]
Merge tag 'acpi-5.12-rc7' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fix from Rafael Wysocki:
 "Fix a build issue introduced by a previous fix in the ACPI processor
  driver (Vitaly Kuznetsov)"

* tag 'acpi-5.12-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: processor: Fix build when CONFIG_ACPI_PROCESSOR=m

3 years agovdpa/mlx5: Fix suspend/resume index restoration
Eli Cohen [Thu, 8 Apr 2021 09:10:47 +0000 (12:10 +0300)]
vdpa/mlx5: Fix suspend/resume index restoration

When we suspend the VM, the VDPA interface will be reset. When the VM is
resumed again, clear_virtqueues() will clear the available and used
indices resulting in hardware virqtqueue objects becoming out of sync.
We can avoid this function alltogether since qemu will clear them if
required, e.g. when the VM went through a reboot.

Moreover, since the hw available and used indices should always be
identical on query and should be restored to the same value same value
for virtqueues that complete in order, we set the single value provided
by set_vq_state(). In get_vq_state() we return the value of hardware
used index.

Fixes: b35ccebe3ef7 ("vdpa/mlx5: Restore the hardware used index after change map")
Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
Signed-off-by: Eli Cohen <elic@nvidia.com>
Link: https://lore.kernel.org/r/20210408091047.4269-6-elic@nvidia.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
3 years agovdpa/mlx5: Fix wrong use of bit numbers
Eli Cohen [Thu, 8 Apr 2021 09:10:46 +0000 (12:10 +0300)]
vdpa/mlx5: Fix wrong use of bit numbers

VIRTIO_F_VERSION_1 is a bit number. Use BIT_ULL() with mask
conditionals.

Also, in mlx5_vdpa_is_little_endian() use BIT_ULL for consistency with
the rest of the code.

Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
Signed-off-by: Eli Cohen <elic@nvidia.com>
Link: https://lore.kernel.org/r/20210408091047.4269-5-elic@nvidia.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
3 years agovdpa/mlx5: Retrieve BAR address suitable any function
Eli Cohen [Thu, 8 Apr 2021 09:10:45 +0000 (12:10 +0300)]
vdpa/mlx5: Retrieve BAR address suitable any function

struct mlx5_core_dev has a bar_addr field that contains the correct bar
address for the function regardless of whether it is pci function or sub
function. Use it.

Fixes: 1958fc2f0712 ("net/mlx5: SF, Add auxiliary device driver")
Signed-off-by: Eli Cohen <elic@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Link: https://lore.kernel.org/r/20210408091047.4269-4-elic@nvidia.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
3 years agovdpa/mlx5: Use the correct dma device when registering memory
Eli Cohen [Thu, 8 Apr 2021 09:10:44 +0000 (12:10 +0300)]
vdpa/mlx5: Use the correct dma device when registering memory

In cases where the vdpa instance uses a SF (sub function), the DMA
device is the parent device. Use a function to retrieve the correct DMA
device.

Fixes: 1958fc2f0712 ("net/mlx5: SF, Add auxiliary device driver")
Signed-off-by: Eli Cohen <elic@nvidia.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Link: https://lore.kernel.org/r/20210408091047.4269-3-elic@nvidia.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
3 years agovdpa/mlx5: should exclude header length and fcs from mtu
Si-Wei Liu [Thu, 8 Apr 2021 09:10:43 +0000 (12:10 +0300)]
vdpa/mlx5: should exclude header length and fcs from mtu

When feature VIRTIO_NET_F_MTU is negotiated on mlx5_vdpa,
22 extra bytes worth of MTU length is shown in guest.
This is because the mlx5_query_port_max_mtu API returns
the "hardware" MTU value, which does not just contain the
 Ethernet payload, but includes extra lengths starting
from the Ethernet header up to the FCS altogether.

Fix the MTU so packets won't get dropped silently.

Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
Signed-off-by: Si-Wei Liu <si-wei.liu@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Eli Cohen <elic@nvidia.com>
Link: https://lore.kernel.org/r/20210408091047.4269-2-elic@nvidia.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>