Vladimir Oltean [Fri, 16 Apr 2021 23:42:21 +0000 (02:42 +0300)]
net: enetc: create a common enetc_pf_to_port helper
Even though ENETC interfaces are exposed as individual PCIe PFs with
their own driver instances, the ENETC is still fundamentally a
multi-port Ethernet controller, and some parts of the IP take a port
number (as can be seen in the PSFP implementation).
Create a common helper that can be used outside of the TSN code for
retrieving the ENETC port number based on the PF number. This is only
correct for LS1028A, the only Linux-capable instantiation of ENETC thus
far.
Note that ENETC port 3 is PF 6. The TSN code did not care about this
because ENETC port 3 does not support TSN, so the wrong mapping done by
enetc_get_port for PF 6 could have never been hit.
ethtool: ioctl: Fix out-of-bounds warning in store_link_ksettings_for_user()
Fix the following out-of-bounds warning:
net/ethtool/ioctl.c:492:2: warning: 'memcpy' offset [49, 84] from the object at 'link_usettings' is out of the bounds of referenced subobject 'base' with type 'struct ethtool_link_settings' at offset 0 [-Warray-bounds]
The problem is that the original code is trying to copy data into a
some struct members adjacent to each other in a single call to
memcpy(). This causes a legitimate compiler warning because memcpy()
overruns the length of &link_usettings.base. Fix this by directly
using &link_usettings and _from_ as destination and source addresses,
instead.
This helps with the ongoing efforts to globally enable -Warray-bounds
and get us closer to being able to tighten the FORTIFY_SOURCE routines
on memcpy().
David S. Miller [Mon, 19 Apr 2021 22:20:35 +0000 (15:20 -0700)]
Merge branch 'nh-flushing'
Ido Schimmel says:
====================
nexthop: Support large scale nexthop flushing
Patch #1 fixes a day-one bug in the nexthop code and allows "ip nexthop
flush" to work correctly with large number of nexthops that do not fit
in a single-part dump.
Patch #2 adds a test case.
Targeting at net-next since this use case never worked, the flow is
pretty obscure and such a large number of nexthops is unlikely to be
used in any real-world scenario.
nexthop: Restart nexthop dump based on last dumped nexthop identifier
Currently, a multi-part nexthop dump is restarted based on the number of
nexthops that have been dumped so far. This can result in a lot of
nexthops not being dumped when nexthops are simultaneously deleted:
# ip nexthop | wc -l
65536
# ip nexthop flush
Dump was interrupted and may be inconsistent.
Flushed 36040 nexthops
# ip nexthop | wc -l
29496
Instead, restart the dump based on the nexthop identifier (fixed number)
of the last successfully dumped nexthop:
# ip nexthop | wc -l
65536
# ip nexthop flush
Dump was interrupted and may be inconsistent.
Flushed 65536 nexthops
# ip nexthop | wc -l
0
Add CFO tracking, which stands for central frequency offset tracking, to
adjust oscillator to align central frequency of connected AP. Then, it can
yield better performance.
Johannes Berg [Thu, 15 Apr 2021 13:48:46 +0000 (16:48 +0300)]
iwlwifi: pcie: don't enable BHs with IRQs disabled
After the fix from Jiri that disabled local IRQs instead of
just BHs (necessary to fix an issue with submitting a command
with IRQs already disabled), there was still a situation in
which we could deep in there enable BHs, if the device config
sets the apmg_wake_up_wa configuration, which is true on all
7000 series devices.
To fix that, but not require reverting commit 1ed08f6fb5ae
("iwlwifi: remove flags argument for nic_access"), split up
nic access into a version with BH manipulation to use most
of the time, and without it for this specific case where the
local IRQs are already disabled.
In mwl8k_probe_hw, hw->priv->txq is freed at the first time by
dma_free_coherent() in the call chain:
if(!priv->ap_fw)->mwl8k_init_txqs(hw)->mwl8k_txq_init(hw, i).
Then in err_free_queues of mwl8k_probe_hw, hw->priv->txq is freed
at the second time by mwl8k_txq_deinit(hw, i)->dma_free_coherent().
My patch set txq->txd to NULL after the first free to avoid the
double free.
The problem arises because the value of group is 5 for channel 14. The trivial
increase in the dimension of bw40_base fails as this struct must match the layout of
efuse. The fix is to add the rate as an argument to rtw_get_channel_group() and set
the group for channel 14 to 4 if rate <= DESC_RATE11M.
This patch fixes commit fa6dfe6bff24 ("rtw88: resolve order of tx power setting routines")
Marek Vasut [Sat, 27 Mar 2021 23:59:32 +0000 (00:59 +0100)]
rsi: Use resume_noirq for SDIO
The rsi_resume() does access the bus to enable interrupts on the RSI
SDIO WiFi card, however when calling sdio_claim_host() in the resume
path, it is possible the bus is already claimed and sdio_claim_host()
spins indefinitelly. Enable the SDIO card interrupts in resume_noirq
instead to prevent anything else from claiming the SDIO bus first.
The opening comment mark '/**' is used for highlighting the beginning of
kernel-doc comments.
There are some files in drivers/net/wireless/rsi which follow this syntax
in their file headers, i.e. start with '/**' like comments, which causes
unexpected warnings from kernel-doc.
E.g., running scripts/kernel-doc -none on drivers/net/wireless/rsi/rsi_coex.h
causes this warning:
"warning: wrong kernel-doc identifier on line:
* Copyright (c) 2018 Redpine Signals Inc."
Similarly for other files too.
Provide a simple fix by replacing such occurrences with general comment
format, i.e., "/*", to prevent kernel-doc from parsing it.
r8169: keep pause settings on interface down/up cycle
Currently, if the user changes the pause settings, the default settings
will be restored after an interface down/up cycle, and also when
resuming from suspend. This doesn't seem to provide the best user
experience. Change this to keep user settings, and just ensure that in
jumbo mode pause is disabled.
Small drawback: When switching back mtu from jumbo to non-jumbo then
pause remains disabled (but user can enable it using ethtool).
I think that's a not too common scenario and acceptable.
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
- keep the ZC code, drop the code related to reinit
net/bridge/netfilter/ebtables.c
- fix build after move to net_generic
Arnd Bergmann [Tue, 23 Mar 2021 13:16:28 +0000 (14:16 +0100)]
airo: work around stack usage warning
gcc-11 with KASAN on 32-bit arm produces a warning about a function
that needs a lot of stack space:
drivers/net/wireless/cisco/airo.c: In function 'setup_card.constprop':
drivers/net/wireless/cisco/airo.c:3960:1: error: the frame size of 1512 bytes is larger than 1400 bytes [-Werror=frame-larger-than=]
Most of this is from a single large structure that could be dynamically
allocated or moved into the per-device structure. However, as the callers
all seem to have a fairly well bounded call chain, the easiest change
is to pull out the part of the function that needs the large variables
into a separate function and mark that as noinline_for_stack. This does
not reduce the total stack usage, but it gets rid of the warning and
requires minimal changes otherwise.
Arnd Bergmann [Tue, 23 Mar 2021 12:57:14 +0000 (13:57 +0100)]
wlcore: fix overlapping snprintf arguments in debugfs
gcc complains about undefined behavior in calling snprintf()
with the same buffer as input and output:
drivers/net/wireless/ti/wl18xx/debugfs.c: In function 'diversity_num_of_packets_per_ant_read':
drivers/net/wireless/ti/wl18xx/../wlcore/debugfs.h:86:3: error: 'snprintf' argument 4 overlaps destination object 'buf' [-Werror=restrict]
86 | snprintf(buf, sizeof(buf), "%s[%d] = %d\n", \
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
87 | buf, i, stats->sub.name[i]); \
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/wireless/ti/wl18xx/debugfs.c:24:2: note: in expansion of macro 'DEBUGFS_FWSTATS_FILE_ARRAY'
24 | DEBUGFS_FWSTATS_FILE_ARRAY(a, b, c, wl18xx_acx_statistics)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/net/wireless/ti/wl18xx/debugfs.c:159:1: note: in expansion of macro 'WL18XX_DEBUGFS_FWSTATS_FILE_ARRAY'
159 | WL18XX_DEBUGFS_FWSTATS_FILE_ARRAY(diversity, num_of_packets_per_ant,
There are probably other ways of handling the debugfs file, without
using on-stack buffers, but a simple workaround here is to remember the
current position in the buffer and just keep printing in there.
Arnd Bergmann [Mon, 22 Mar 2021 10:43:34 +0000 (11:43 +0100)]
libertas: avoid -Wempty-body warning
Building without mesh supports shows a couple of warnings with
'make W=1':
drivers/net/wireless/marvell/libertas/main.c: In function 'lbs_start_card':
drivers/net/wireless/marvell/libertas/main.c:1068:37: error: suggest braces around empty body in an 'if' statement [-Werror=empty-body]
1068 | lbs_start_mesh(priv);
Change the macros to use the usual "do { } while (0)" instead to shut up
the warnings and make the code a litte more robust.
The 'c2hcmd_lock' spinlock is only used to protect some __skb_queue_tail()
and __skb_dequeue() calls.
Use the lock provided in the skb itself and call skb_queue_tail() and
skb_dequeue(). These functions already include the correct locking.
Dan Carpenter [Fri, 19 Mar 2021 14:47:31 +0000 (17:47 +0300)]
wilc1000: fix a loop timeout condition
If the loop fails, the "while(trials--) {" loop will exit with "trials"
set to -1. The test for that expects it to end with "trials" set to 0
so the warning message will not be printed.
Fix this by changing from a post-op to a pre-op. This does mean that
we only make 99 attempts instead of 100 but that's okay.
In preparation to enable -Wimplicit-fallthrough for Clang, fix
multiple warnings by replacing /* fall through */ comments with
the new pseudo-keyword macro fallthrough; instead of letting the
code fall through to the next case.
Notice that Clang doesn't recognize /* fall through */ comments as
implicit fall-through markings.
wilc1000: Bring MAC address setting in line with typical Linux behavior
Linux network drivers normally disallow changing the MAC address when
the interface is up. This driver has been different in that it allows
to change the MAC address *only* when it's up. This patch brings
wilc1000 behavior more in line with other network drivers. We could
have replaced wilc_set_mac_addr() with eth_mac_addr() but that would
break existing documentation on how to change the MAC address.
Likewise, return -EADDRNOTAVAIL (not -EINVAL) when the specified MAC
address is invalid or unavailable.
The driver so far has always disabled CRC protection. This means any
data corruption that occurrs during the SPI transfers could go
undetected. This patch adds module parameters enable_crc7 and
enable_crc16 to selectively turn on CRC7 (for command transfers) and
CRC16 (for data transfers), respectively.
The default configuration remains unchanged, with both CRC7 and CRC16
off.
The performance impact of CRC was measured by running ttcp -t four
times in a row on a SAMA5 device:
CRC7 CRC16 Throughput: Standard deviation:
---- ----- ----------- -------------------
off off 1720 +/- 48 KB/s
on off 1658 +/- 58 KB/s
on on 1579 +/- 84 KB/s
wilc1000: Introduce symbolic names for SPI protocol register
The WILC1000 protocol control register has bits for enabling the CRCs
(CRC7 for commands and CRC16 for data) and to set the data packet
size. Define symbolic names for those so the code is more easily
understood.
For CMD_SINGLE_READ and CMD_INTERNAL_READ, WILC may insert one or more
zero bytes between the command response and the DATA Start tag (0xf3).
This behavior appears to be undocumented in "ATWILC1000 USER GUIDE"
(https://tinyurl.com/4hhshdts) but we have observed 1-4 zero bytes
when the SPI bus operates at 48MHz and none when it operates at 1MHz.
This code is derived from the equivalent code of the wilc driver in
the linux-at91 repository.
Brian Norris [Thu, 25 Feb 2021 02:44:54 +0000 (18:44 -0800)]
mwifiex: don't print SSID to logs
There are a few reasons not to dump SSIDs as-is in kernel logs:
1) they're not guaranteed to be any particular text encoding (UTF-8,
ASCII, ...) in general
2) it's somewhat redundant; the BSSID should be enough to uniquely
identify the AP/STA to which we're connecting
3) BSSIDs have an easily-recognized format, whereas SSIDs do not (they
are free-form)
4) other common drivers (e.g., everything based on mac80211) get along
just fine by only including BSSIDs when logging state transitions
Additional notes on reason #3: this is important for the
privacy-conscious, especially when providing tools that convey
kernel logs on behalf of a user -- e.g., when reporting bugs. So for
example, it's easy to automatically filter logs for MAC addresses, but
it's much harder to filter SSIDs out of unstructured text.
The 'c2hcmd_lock' spinlock is only used to protect some __skb_queue_tail()
and __skb_dequeue() calls.
Use the lock provided in the skb itself and call skb_queue_tail() and
skb_dequeue(). These functions already include the correct locking.
Yang Li [Wed, 31 Mar 2021 09:13:43 +0000 (17:13 +0800)]
rtlwifi: rtl8188ee: remove redundant assignment of variable rtlpriv->btcoexist.reg_bt_sco
Assigning value "3" to "rtlpriv->btcoexist.reg_bt_sco" here, but that
stored value is overwritten before it can be used.
Coverity reports this problem as
CWE563: A value assigned to a variable is never used.
drivers/net/wireless/realtek/rtlwifi/rtl8188ee/hw.c:
rtl8188ee_bt_reg_init
Colin Ian King [Sat, 27 Mar 2021 23:00:14 +0000 (23:00 +0000)]
rtlwifi: remove redundant assignment to variable err
Variable err is assigned -ENODEV followed by an error return path
via label error_out that does not access the variable and returns
with the -ENODEV error return code. The assignment to err is
redundant and can be removed.
Ping-Ke Shih [Fri, 19 Feb 2021 05:26:07 +0000 (13:26 +0800)]
rtlwifi: 8821ae: upgrade PHY and RF parameters
The signal strength of 5G is quite low, so user can't connect to an AP far
away. New parameters with new format and its parser are updated by the commit 84d26fda52e2 ("rtlwifi: Update 8821ae new phy parameters and its parser."), but
some parameters are missing. Use this commit to update to the novel parameters
that use new format.
Merge tag 'net-5.12-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Pull networking fixes from Jakub Kicinski:
"Networking fixes for 5.12-rc8, including fixes from netfilter, and
bpf. BPF verifier changes stand out, otherwise things have slowed
down.
Current release - regressions:
- gro: ensure frag0 meets IP header alignment
- Revert "net: stmmac: re-init rx buffers when mac resume back"
- ethernet: macb: fix the restore of cmp registers
Previous releases - regressions:
- ixgbe: Fix NULL pointer dereference in ethtool loopback test
- ixgbe: fix unbalanced device enable/disable in suspend/resume
- phy: marvell: fix detection of PHY on Topaz switches
- make tcp_allowed_congestion_control readonly in non-init netns
- xen-netback: Check for hotplug-status existence before watching
Previous releases - always broken:
- bpf: mitigate a speculative oob read of up to map value size by
tightening the masking window
- sctp: fix race condition in sctp_destroy_sock
- sit, ip6_tunnel: Unregister catch-all devices
- netfilter: nftables: clone set element expression template
- net: geneve: check skb is large enough for IPv4/IPv6 header
- netlink: don't call ->netlink_bind with table lock held"
* tag 'net-5.12-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (52 commits)
netlink: don't call ->netlink_bind with table lock held
MAINTAINERS: update my email
bpf: Update selftests to reflect new error states
bpf: Tighten speculative pointer arithmetic mask
bpf: Move sanitize_val_alu out of op switch
bpf: Refactor and streamline bounds check into helper
bpf: Improve verifier error messages for users
bpf: Rework ptr_limit into alu_limit and add common error path
bpf: Ensure off_reg has no mixed signed bounds for all types
bpf: Move off_reg into sanitize_ptr_alu
bpf: Use correct permission flag for mixed signed bounds arithmetic
ch_ktls: do not send snd_una update to TCB in middle
ch_ktls: tcb close causes tls connection failure
ch_ktls: fix device connection close
ch_ktls: Fix kernel panic
i40e: fix the panic when running bpf in xdpdrv mode
net/mlx5e: fix ingress_ifindex check in mlx5e_flower_parse_meta
net/mlx5e: Fix setting of RS FEC mode
net/mlx5: Fix setting of devlink traps in switchdev mode
Revert "net: stmmac: re-init rx buffers when mac resume back"
...
Merge tag 'libnvdimm-fixes-for-5.12-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm
Pull libnvdimm fixes from Dan Williams:
"The largest change is for a regression that landed during -rc1 for
block-device read-only handling. Vaibhav found a new use for the
ability (originally introduced by virtio_pmem) to call back to the
platform to flush data, but also found an original bug in that
implementation. Lastly, Arnd cleans up some compile warnings in dax.
This has all appeared in -next with no reported issues.
Summary:
- Fix a regression of read-only handling in the pmem driver
- Fix a compile warning
- Fix support for platform cache flush commands on powerpc/papr"
* tag 'libnvdimm-fixes-for-5.12-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
libnvdimm/region: Fix nvdimm_has_flush() to handle ND_REGION_ASYNC
libnvdimm: Notify disk drivers to revalidate region read-only
dax: avoid -Wempty-body warnings
Merge tag 'cxl-fixes-for-5.12-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl
Pull CXL memory class fixes from Dan Williams:
"A collection of fixes for the CXL memory class driver introduced in
this release cycle.
The driver was primarily developed on a work-in-progress QEMU
emulation of the interface and we have since found a couple places
where it hid spec compliance bugs in the driver, or had a spec
implementation bug itself.
The biggest change here is replacing a percpu_ref with an rwsem to
cleanup a couple bugs in the error unwind path during ioctl device
init. Lastly there were some minor cleanups to not export the
power-management sysfs-ABI for the ioctl device, use the proper sysfs
helper for emitting values, and prevent subtle bugs as new
administration commands are added to the supported list.
The bulk of it has appeared in -next save for the top commit which was
found today and validated on a fixed-up QEMU model.
Summary:
- Fix support for CXL memory devices with registers offset from the
BAR base.
- Fix the reporting of device capacity.
- Fix the driver commands list definition to be disconnected from the
UAPI command list.
- Replace percpu_ref with rwsem to fix initialization error path.
- Fix leaks in the driver initialization error path.
- Drop the power/ directory from CXL device sysfs.
- Use the recommended sysfs helper for attribute 'show'
implementations"
* tag 'cxl-fixes-for-5.12-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl:
cxl/mem: Fix memory device capacity probing
cxl/mem: Fix register block offset calculation
cxl/mem: Force array size of mem_commands[] to CXL_MEM_COMMAND_ID_MAX
cxl/mem: Disable cxl device power management
cxl/mem: Do not rely on device_add() side effects for dev_set_name() failures
cxl/mem: Fix synchronization mechanism for device removal vs ioctl operations
cxl/mem: Use sysfs_emit() for attribute show routines
Kalle Valo [Sat, 17 Apr 2021 08:38:01 +0000 (11:38 +0300)]
Merge tag 'iwlwifi-next-for-kalle-2021-04-12-v2' of git://git.kernel.org/pub/scm/linux/kernel/git/iwlwifi/iwlwifi-next
iwlwifi patches for v5.13
* Add support for new FTM FW APIs;
* Some CSA fixes;
* Support for new HW family and other HW detection fixes;
* Robustness improvement in the HW detection code;
* One fix in PMF;
* Some new regulatory features;
* Support for passive scan in 6GHz;
* Some improvements in the sync queue implementation;
* Support for new devices;
* Support for a new FW API command version;
* Some locking fixes;
* Bump the FW API version support for AX devices;
* Some other small fixes, clean-ups and improvements.
# gpg: Signature made Wed 14 Apr 2021 12:33:29 PM EEST using RSA key ID 1A3CC5FA
# gpg: Good signature from "Luciano Roth Coelho (Luca) <[email protected]>"
# gpg: aka "Luciano Roth Coelho (Intel) <[email protected]>"
Kalle Valo [Sat, 17 Apr 2021 08:34:43 +0000 (11:34 +0300)]
Merge tag 'mt76-for-kvalo-2021-04-12' of https://github.com/nbd168/wireless
mt76 patches for 5.13
* code cleanup
* mt7915/mt7615 decap offload support
* driver fixes
* mt7613 eeprom support
* MCU code unification
* threaded NAPI support
* new device IDs
* mt7921 device reset support
* rx timestamp support
# gpg: Signature made Tue 13 Apr 2021 12:11:25 AM EEST using DSA key ID 02A76EF5
# gpg: Good signature from "Felix Fietkau <[email protected]>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 75D1 1A7D 91A7 710F 4900 42EF D77D 141D 02A7 6EF5
Dan Williams [Sat, 17 Apr 2021 00:43:30 +0000 (17:43 -0700)]
cxl/mem: Fix memory device capacity probing
The CXL Identify Memory Device output payload emits capacity in 256MB
units. The driver is treating the capacity field as bytes. This was
missed because QEMU reports bytes when it should report bytes / 256MB.
David S. Miller [Sat, 17 Apr 2021 00:08:40 +0000 (17:08 -0700)]
Merge branch 'enetc-xdp-fixes'
Vladimir Oltean says:
====================
Fixups for XDP on NXP ENETC
After some more XDP testing on the NXP LS1028A, this is a set of 10 bug
fixes, simplifications and tweaks, ranging from addressing Toke's feedback
(the network stack can run concurrently with XDP on the same TX rings)
to fixing some OOM conditions seen under TX congestion.
====================
Vladimir Oltean [Fri, 16 Apr 2021 21:22:25 +0000 (00:22 +0300)]
net: enetc: apply the MDIO workaround for XDP_REDIRECT too
Described in fd5736bf9f23 ("enetc: Workaround for MDIO register access
issue") is a workaround for a hardware bug that requires a register
access of the MDIO controller to never happen concurrently with a
register access of a port PF. To avoid that, a mutual exclusion scheme
with rwlocks was implemented - the port PF accessors are the 'read'
side, and the MDIO accessors are the 'write' side.
When we do XDP_REDIRECT between two ENETC interfaces, all is fine
because the MDIO lock is already taken from the NAPI poll loop.
But when the ingress interface is not ENETC, just the egress is, the
MDIO lock is not taken, so we might access the port PF registers
concurrently with MDIO, which will make the link flap due to wrong
values returned from the PHY.
To avoid this, let's just slap an enetc_lock_mdio/enetc_unlock_mdio at
the beginning and ending of enetc_xdp_xmit. The fact that the MDIO lock
is designed as a rwlock is important here, because the read side is
reentrant (that is one of the main reasons why we chose it). Usually,
the way we benefit of its reentrancy is by running the data path
concurrently on both CPUs, but in this case, we benefit from the
reentrancy by taking the lock even when the lock is already taken
(and that's the situation where ENETC is both the ingress and the egress
interface for XDP_REDIRECT, which was fine before and still is fine now).
Vladimir Oltean [Fri, 16 Apr 2021 21:22:24 +0000 (00:22 +0300)]
net: enetc: fix buffer leaks with XDP_TX enqueue rejections
If the TX ring is congested, enetc_xdp_tx() returns false for the
current XDP frame (represented as an array of software BDs).
This array of software TX BDs is constructed in enetc_rx_swbd_to_xdp_tx_swbd
from software BDs freshly cleaned from the RX ring. The issue is that we
scrub the RX software BDs too soon, more precisely before we know that
we can enqueue the TX BDs successfully into the TX ring.
If we can't enqueue them (and enetc_xdp_tx returns false), we call
enetc_xdp_drop which attempts to recycle the buffers held by the RX
software BDs. But because we scrubbed those RX BDs already, two things
happen:
(a) we leak their memory
(b) we populate the RX software BD ring with an all-zero rx_swbd
structure, which makes the buffer refill path allocate more memory.
enetc_refill_rx_ring
-> if (unlikely(!rx_swbd->page))
-> enetc_new_page
That is a recipe for fast OOM.
Fixes: 7ed2bc80074e ("net: enetc: add support for XDP_TX") Signed-off-by: Vladimir Oltean <[email protected]> Signed-off-by: David S. Miller <[email protected]>
Vladimir Oltean [Fri, 16 Apr 2021 21:22:22 +0000 (00:22 +0300)]
net: enetc: use dedicated TX rings for XDP
It is possible for one CPU to perform TX hashing (see netdev_pick_tx)
between the 8 ENETC TX rings, and the TX hashing to select TX queue 1.
At the same time, it is possible for the other CPU to already use TX
ring 1 for XDP (either XDP_TX or XDP_REDIRECT). Since there is no mutual
exclusion between XDP and the network stack, we run into an issue
because the ENETC TX procedure is not reentrant.
The obvious approach would be to just make XDP take the lock of the
network stack's TX queue corresponding to the ring it's about to enqueue
in.
For XDP_REDIRECT, this is quite straightforward, a lock at the beginning
and end of enetc_xdp_xmit() should do the trick.
But for XDP_TX, it's a bit more complicated. For one, we do TX batching
all by ourselves for frames with the XDP_TX verdict. This is something
we would like to keep the way it is, for performance reasons. But
batching means that the network stack's lock should be kept from the
first enqueued XDP_TX frame and until we ring the doorbell. That is
mostly fine, except for cases when in the same NAPI loop we have mixed
XDP_TX and XDP_REDIRECT frames. So if enetc_xdp_xmit() gets called while
we are holding the lock from the RX NAPI, then bam, deadlock. The naive
answer could be 'just flush the XDP_TX frames first, then release the
network stack's TX queue lock, then call xdp_do_flush_map()'. But even
xdp_do_redirect() is capable of flushing the batched XDP_REDIRECT
frames, so unless we unlock/relock the TX queue around xdp_do_redirect(),
there simply isn't any clean way to protect XDP_TX from concurrent
network stack .ndo_start_xmit() on another CPU.
So we need to take a different approach, and that is to reserve two
rings for the sole use of XDP. We leave TX rings
0..ndev->real_num_tx_queues-1 to be handled by the network stack, and we
pick them from the end of the priv->tx_ring array.
We make an effort to keep the mapping done by enetc_alloc_msix() which
decides which CPU handles the TX completions of which TX ring in its
NAPI poll. So the XDP TX ring of CPU 0 is handled by TX ring 6, and the
XDP TX ring of CPU 1 is handled by TX ring 7.
Vladimir Oltean [Fri, 16 Apr 2021 21:22:21 +0000 (00:22 +0300)]
net: enetc: increase TX ring size
Now that commit d6a2829e82cf ("net: enetc: increase RX ring default
size") has increased the RX ring size, it is quite easy to congest the
TX rings when the traffic is predominantly XDP_TX, as the RX ring is
quite a bit larger than the TX one.
Since we bit the bullet and did the expensive thing already (larger RX
rings consume more memory pages), it seems quite foolish to keep the TX
rings small. So make them equally sized with TX.
enetc_clean_rx_ring_xdp will 'break', but that 'break' instruction isn't
strong enough to actually break the NAPI poll loop, just the switch/case
statement for XDP actions. So we increment rx_frm_cnt and go to the next
frames minding our own business.
Instead let's do what the skb NAPI poll function does, and break the
loop now, waiting for the memory pressure to go away. Otherwise the next
calls to build_skb() are likely to fail too.
Vladimir Oltean [Fri, 16 Apr 2021 21:22:17 +0000 (00:22 +0300)]
net: enetc: rename the buffer reuse helpers
enetc_put_xdp_buff has nothing to do with XDP, frankly, it is just a
helper to populate the recycle end of the shadow RX BD ring
(next_to_alloc) with a given buffer.
On the other hand, enetc_put_rx_buff plays more tricks than its name
would suggest.
So let's rename enetc_put_rx_buff into enetc_flip_rx_buff to reflect the
half-page buffer reuse tricks that it employs, and enetc_put_xdp_buff
into enetc_put_rx_buff which suggests a more garden-variety operation.
David S. Miller [Sat, 17 Apr 2021 00:06:14 +0000 (17:06 -0700)]
Merge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue
Tony Nguyen says:
====================
1GbE Intel Wired LAN Driver Updates 2021-04-16
This series contains updates to igb and igc drivers.
Ederson adjusts Tx buffer distributions in Qav mode to improve
TSN-aware traffic for igb. He also enable PPS support and auxiliary PHC
functions for igc.
Grzegorz checks that the MTA register was properly written and
retries if not for igb.
Sasha adds reporting of EEE low power idle counters to ethtool and fixes
a return value being overwritten through looping for igc.
====================
flow_dissector: Fix out-of-bounds warning in __skb_flow_bpf_to_target()
Fix the following out-of-bounds warning:
net/core/flow_dissector.c:835:3: warning: 'memcpy' offset [33, 48] from the object at 'flow_keys' is out of the bounds of referenced subobject 'ipv6_src' with type '__u32[4]' {aka 'unsigned int[4]'} at offset 16 [-Warray-bounds]
The problem is that the original code is trying to copy data into a
couple of struct members adjacent to each other in a single call to
memcpy(). So, the compiler legitimately complains about it. As these
are just a couple of members, fix this by copying each one of them in
separate calls to memcpy().
This helps with the ongoing efforts to globally enable -Warray-bounds
and get us closer to being able to tighten the FORTIFY_SOURCE routines
on memcpy().
netlink: don't call ->netlink_bind with table lock held
When I added support to allow generic netlink multicast groups to be
restricted to subscribers with CAP_NET_ADMIN I was unaware that a
genl_bind implementation already existed in the past.
It was reverted due to ABBA deadlock:
1. ->netlink_bind gets called with the table lock held.
2. genetlink bind callback is invoked, it grabs the genl lock.
But when a new genl subsystem is (un)registered, these two locks are
taken in reverse order.
One solution would be to revert again and add a comment in genl
referring 1e82a62fec613, "genetlink: remove genl_bind").
This would need a second change in mptcp to not expose the raw token
value anymore, e.g. by hashing the token with a secret key so userspace
can still associate subflow events with the correct mptcp connection.
However, Paolo Abeni reminded me to double-check why the netlink table is
locked in the first place.
I can't find one. netlink_bind() is already called without this lock
when userspace joins a group via NETLINK_ADD_MEMBERSHIP setsockopt.
Same holds for the netlink_unbind operation.
Digging through the history, commit f773608026ee1
("netlink: access nlk groups safely in netlink bind and getname")
expanded the lock scope.
commit 3a20773beeeeade ("net: netlink: cap max groups which will be considered in netlink_bind()")
... removed the nlk->ngroups access that the lock scope
extension was all about.
Reduce the lock scope again and always call ->netlink_bind without
the table lock.
The Fixes tag should be vs. the patch mentioned in the link below,
but that one got squash-merged into the patch that came earlier in the
series.
David S. Miller [Sat, 17 Apr 2021 00:00:03 +0000 (17:00 -0700)]
Merge branch 'ethtool-stats'
Jakub Kicinski says:
====================
ethtool: add uAPI for reading standard stats
Continuing the effort of providing a unified access method
to standard stats, and explicitly tying the definitions to
the standards this series adds an API for general stats
which do no fit into more targeted control APIs.
There is nothing clever here, just a netlink API for dumping
statistics defined by standards and RFCs which today end up
in ethtool -S under infinite variations of names.
This series adds basic IEEE stats (for PHY, MAC, Ctrl frames)
and RMON stats. AFAICT other RFCs only duplicate the IEEE
stats.
This series does _not_ add a netlink API to read driver-defined
stats. There seems to be little to gain from moving that part
to netlink.
The netlink message format is very simple, and aims to allow
adding stats and groups with no changes to user tooling (which
IIUC is expected for ethtool).
On user space side we can re-use -S, and make it dump
standard stats if --groups are defined.
Compared to the RFC I went ahead with wrapping the stats into
a 1:1 nest. Now IDs of stats can start from 0, at a cost of
slightly "careful" u64 alignment handling.
Jakub Kicinski [Fri, 16 Apr 2021 19:27:42 +0000 (12:27 -0700)]
ethtool: add interface to read RMON stats
Most devices maintain RMON (RFC 2819) stats - particularly
the "histogram" of packets received by size. Unlike other
RFCs which duplicate IEEE stats, the short/oversized frame
counters in RMON don't seem to match IEEE stats 1-to-1 either,
so expose those, too. Do not expose basic packet, CRC errors
etc - those are already otherwise covered.
Because standard defines packet ranges only up to 1518, and
everything above that should theoretically be "oversized"
- devices often create their own ranges.
Going beyond what the RFC defines - expose the "histogram"
in the Tx direction (assume for now that the ranges will
be the same).
Jakub Kicinski [Fri, 16 Apr 2021 19:27:40 +0000 (12:27 -0700)]
ethtool: add interface to read standard MAC stats
Most of the MAC statistics are included in
struct rtnl_link_stats64, but some fields
are aggregated. Besides it's good to expose
these clearly hardware stats separately.
sctp: Fix out-of-bounds warning in sctp_process_asconf_param()
Fix the following out-of-bounds warning:
net/sctp/sm_make_chunk.c:3150:4: warning: 'memcpy' offset [17, 28] from the object at 'addr' is out of the bounds of referenced subobject 'v4' with type 'struct sockaddr_in' at offset 0 [-Warray-bounds]
This helps with the ongoing efforts to globally enable -Warray-bounds
and get us closer to being able to tighten the FORTIFY_SOURCE routines
on memcpy().
Fix various kernel-doc warnings in lib/ due to missing or erroneous
function names.
Add kernel-doc for some function parameters that was missing. Use
kernel-doc "Return:" notation in earlycpio.c.
Quietens the following warnings:
lib/earlycpio.c:61: warning: expecting prototype for cpio_data find_cpio_data(). Prototype was for find_cpio_data() instead
lib/lru_cache.c:640: warning: expecting prototype for lc_dump(). Prototype was for lc_seq_dump_details() instead
lru_cache.c:90: warning: Function parameter or member 'cache' not described in 'lc_create'
lib/parman.c:368: warning: expecting prototype for parman_item_del(). Prototype was for parman_item_remove() instead
parman.c:309: warning: Excess function parameter 'prority' description in 'parman_prio_init'
lib/radix-tree.c:703: warning: expecting prototype for __radix_tree_insert(). Prototype was for radix_tree_insert() instead
radix-tree.c:180: warning: Excess function parameter 'addr' description in 'radix_tree_find_next_bit'
radix-tree.c:180: warning: Excess function parameter 'size' description in 'radix_tree_find_next_bit'
radix-tree.c:931: warning: Function parameter or member 'iter' not described in 'radix_tree_iter_replace'
Johannes Berg [Fri, 16 Apr 2021 22:46:23 +0000 (15:46 -0700)]
gcov: clang: fix clang-11+ build
With clang-11+, the code is broken due to my kvmalloc() conversion
(which predated the clang-11 support code) leaving one vmalloc() in
place. Fix that.
READ_ONCE() cannot be used for reading PTEs. Use ptep_get() instead, to
avoid the following errors:
CC mm/ptdump.o
In file included from <command-line>:
mm/ptdump.c: In function 'ptdump_pte_entry':
include/linux/compiler_types.h:320:38: error: call to '__compiletime_assert_207' declared with attribute error: Unsupported access size for {READ,WRITE}_ONCE().
320 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
| ^
include/linux/compiler_types.h:301:4: note: in definition of macro '__compiletime_assert'
301 | prefix ## suffix(); \
| ^~~~~~
include/linux/compiler_types.h:320:2: note: in expansion of macro '_compiletime_assert'
320 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
| ^~~~~~~~~~~~~~~~~~~
include/asm-generic/rwonce.h:36:2: note: in expansion of macro 'compiletime_assert'
36 | compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long), \
| ^~~~~~~~~~~~~~~~~~
include/asm-generic/rwonce.h:49:2: note: in expansion of macro 'compiletime_assert_rwonce_type'
49 | compiletime_assert_rwonce_type(x); \
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
mm/ptdump.c:114:14: note: in expansion of macro 'READ_ONCE'
114 | pte_t val = READ_ONCE(*pte);
| ^~~~~~~~~
make[2]: *** [mm/ptdump.o] Error 1
See commit 481e980a7c19 ("mm: Allow arches to provide ptep_get()") and
commit c0e1c8c22beb ("powerpc/8xx: Provide ptep_get() with 16k pages")
for details.
Mapping dirty helpers have, so far, been only used on X86, but a port of
vmwgfx to ARM64 exposed a problem which results in a compilation error
on ARM64 systems:
mm/mapping_dirty_helpers.c: In function `wp_clean_pud_entry':
mm/mapping_dirty_helpers.c:172:32: error: implicit declaration of function `pud_dirty'; did you mean `pmd_dirty'? [-Werror=implicit-function-declaration]
This is due to the fact that mapping_dirty_helpers code assumes that
pud_dirty is always defined, which is not the case for architectures
that don't define CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD.
ARM64 arch is a little inconsistent when it comes to PUD hugepage
helpers, e.g. it defines pud_young but not pud_dirty but regardless of
that the core kernel code shouldn't assume that any of the PUD hugepage
helpers are available unless CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
is defined. This prevents compilation errors whenever one of the
drivers is ported to new architectures.
ia64: tools: remove duplicate definition of ia64_mf() on ia64
The ia64_mf() macro defined in tools/arch/ia64/include/asm/barrier.h is
already defined in <asm/gcc_intrin.h> on ia64 which causes libbpf
failing to build:
CC /usr/src/linux/tools/bpf/bpftool//libbpf/staticobjs/libbpf.o
In file included from /usr/src/linux/tools/include/asm/barrier.h:24,
from /usr/src/linux/tools/include/linux/ring_buffer.h:4,
from libbpf.c:37:
/usr/src/linux/tools/include/asm/../../arch/ia64/include/asm/barrier.h:43: error: "ia64_mf" redefined [-Werror]
43 | #define ia64_mf() asm volatile ("mf" ::: "memory")
|
In file included from /usr/include/ia64-linux-gnu/asm/intrinsics.h:20,
from /usr/include/ia64-linux-gnu/asm/swab.h:11,
from /usr/include/linux/swab.h:8,
from /usr/include/linux/byteorder/little_endian.h:13,
from /usr/include/ia64-linux-gnu/asm/byteorder.h:5,
from /usr/src/linux/tools/include/uapi/linux/perf_event.h:20,
from libbpf.c:36:
/usr/include/ia64-linux-gnu/asm/gcc_intrin.h:382: note: this is the location of the previous definition
382 | #define ia64_mf() __asm__ volatile ("mf" ::: "memory")
|
cc1: all warnings being treated as errors
Thus, remove the definition from tools/arch/ia64/include/asm/barrier.h.
ia64: tools: remove inclusion of ia64-specific version of errno.h header
There is no longer an ia64-specific version of the errno.h header below
arch/ia64/include/uapi/asm/, so trying to build tools/bpf fails with:
CC /usr/src/linux/tools/bpf/bpftool/btf_dumper.o
In file included from /usr/src/linux/tools/include/linux/err.h:8,
from btf_dumper.c:11:
/usr/src/linux/tools/include/uapi/asm/errno.h:13:10: fatal error: ../../../arch/ia64/include/uapi/asm/errno.h: No such file or directory
13 | #include "../../../arch/ia64/include/uapi/asm/errno.h"
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
Thus, just remove the inclusion of the ia64-specific errno.h so that the
build will use the generic errno.h header on this target which was used
there anyway as the ia64-specific errno.h was just a wrapper for the
generic header.
Fixes: c25f867ddd00 ("ia64: remove unneeded uapi asm-generic wrappers") Signed-off-by: John Paul Adrian Glaubitz <[email protected]> Signed-off-by: Andrew Morton <[email protected]> Signed-off-by: Linus Torvalds <[email protected]>
Randy Dunlap [Fri, 16 Apr 2021 22:46:09 +0000 (15:46 -0700)]
ia64: fix discontig.c section mismatches
Fix IA64 discontig.c Section mismatch warnings.
When CONFIG_SPARSEMEM=y and CONFIG_MEMORY_HOTPLUG=y, the functions
computer_pernodesize() and scatter_node_data() should not be marked as
__meminit because they are needed after init, on any memory hotplug
event. Also, early_nr_cpus_node() is called by compute_pernodesize(),
so early_nr_cpus_node() cannot be __meminit either.
WARNING: modpost: vmlinux.o(.text.unlikely+0x1612): Section mismatch in reference from the function arch_alloc_nodedata() to the function .meminit.text:compute_pernodesize()
The function arch_alloc_nodedata() references the function __meminit compute_pernodesize().
This is often because arch_alloc_nodedata lacks a __meminit annotation or the annotation of compute_pernodesize is wrong.
WARNING: modpost: vmlinux.o(.text.unlikely+0x1692): Section mismatch in reference from the function arch_refresh_nodedata() to the function .meminit.text:scatter_node_data()
The function arch_refresh_nodedata() references the function __meminit scatter_node_data().
This is often because arch_refresh_nodedata lacks a __meminit annotation or the annotation of scatter_node_data is wrong.
WARNING: modpost: vmlinux.o(.text.unlikely+0x1502): Section mismatch in reference from the function compute_pernodesize() to the function .meminit.text:early_nr_cpus_node()
The function compute_pernodesize() references the function __meminit early_nr_cpus_node().
This is often because compute_pernodesize lacks a __meminit annotation or the annotation of early_nr_cpus_node is wrong.
projects / linux.git / log