It's done in probe so it should be undone here.
Fixes: 1d3bb99648 ("Device tree aware EMAC driver")
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20241008233050.9422-1-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
dcr_map is called in the previous if and therefore needs to be unmapped.
Fixes: 1ff0fcfcb1 ("ibm_newemac: Fix new MAL feature handling")
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Link: https://patch.msgid.link/20241007235711.5714-1-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Previously, the TX header requirement for standard frames was ignored.
This requirement is a bitstring sent from the VIOS which maps to the
type of header information needed during TX. If no header information,
is needed then send subcrq direct can be used (which can be more
performant).
This bitstring was previously ignored for standard packets (AKA non LSO,
non CSO) due to the belief that the bitstring was over-cautionary. It
turns out that there are some configurations where the backing device
does need header information for transmission of standard packets. If
the information is not supplied then this causes continuous "Adapter
error" transport events. Therefore, this bitstring should be respected
and observed before considering the use of send subcrq direct.
Fixes: 74839f7a82 ("ibmvnic: Introduce send sub-crq direct")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20241001163200.1802522-2-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
EPROBE_DEFER, which probably wasn't available when this driver was
written, can be used instead of waiting manually.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Link: https://patch.msgid.link/20240912024903.6201-9-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Cleans it up automatically. No need to handle manually.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20240912024903.6201-6-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Switching to devm management of mii_bus allows to remove
mdiobus_unregister calls and thus avoids needing a mii_bus global struct
member.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20240912024903.6201-5-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It's the last to go in remove. Safe to let devm handle it.
Also move request_irq to probe for clarity. It's removed in _remove not
close.
Use dev_err_probe instead of printk. Handles EPROBE_DEFER automatically.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20240912024903.6201-3-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allows to simplify the code slightly. This is safe to do as free_netdev
gets called last.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20240912024903.6201-2-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
dev->emacp contains an __iomem pointer and values derived
from it are used as __iomem pointers. So use this annotation
in the return type for helpers that derive pointers from dev->emacp.
Flagged by Sparse as:
.../core.c:444:36: warning: incorrect type in argument 1 (different address spaces)
.../core.c:444:36: expected unsigned int volatile [noderef] [usertype] __iomem *addr
.../core.c:444:36: got unsigned int [usertype] *
.../core.c: note: in included file:
.../core.h:416:25: warning: cast removes address space '__iomem' of expression
Compile tested only.
No functional change intended.
Signed-off-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20240906-emac-iomem-v1-1-207cc4f3fed0@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
for_each_child_of_node can help to iterate through the device_node,
and we don't need to use while loop. No functional change with this
conversion.
Signed-off-by: Zhang Zekun <zhangzekun11@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20240816015837.109627-1-zhangzekun11@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
During initialization with the vnic server, a bitstring is communicated
to the client regarding header info needed during CSO (See "VNIC
Capabilities" in PAPR). Most of the time, to be safe, vnic server
requests header info for CSO. When header info is needed, multiple TX
descriptors are required per skb; This limits the driver to use
send_subcrq_indirect instead of send_subcrq_direct.
Previously, the vnic server request for header info was ignored. This
allowed the use of send_sub_crq_direct. Transmissions were successful
because the bitstring returned by vnic server is broad and over
cautionary. It was observed that mlx backing devices could actually
transmit and handle CSO packets without the vnic server receiving
header info (despite the fact that the bitstring requested it).
There was a trust issue: The bitstring was overcautionary. This extra
precaution (requesting header info when the backing device may not use
it) comes at the cost of performance (using direct vs indirect hcalls
has a 30% delta in small packet RR transaction rate). So it has been
requested that the vnic server team tries to ensure that the bitstring
is more exact. In the meantime, disable CSO when it is possible to use
the skb in the send_subcrq_direct path. In other words, calculate the
checksum before handing the packet to FW when the packet is not
segmented and xmit_more is false.
Since the code path is only possible if the skb is non GSO and xmit_more
is false, the cost of doing checksum in the send_subcrq_direct path is
minimal. Any large segmented skb will have xmit_more set to true more
frequently and it is inexpensive to do checksumming on a small skb.
The worst-case workload would be a 9000 MTU TCP_RR test with close
to MTU sized packets (and TSO off). This allows xmit_more to be false
more frequently and open the code path up to use send_subcrq_direct.
Observing trace data (graph-time = 1) and packet rate with this workload
shows minimal performance degradation:
1. NIC does checksum w headers, safely use send_subcrq_indirect:
- Packet rate: 631k txs
- Trace data:
ibmvnic_xmit = 44344685.87 us / 6234576 hits = AVG 7.11 us
skb_checksum_help = 4.07 us / 2 hits = AVG 2.04 us
^ Notice hits, tracing this just for reassurance
ibmvnic_tx_scrq_flush = 33040649.69 us / 5638441 hits = AVG 5.86 us
send_subcrq_indirect = 37438922.24 us / 6030859 hits = AVG 6.21 us
2. NIC does checksum w/o headers, dangerously use send_subcrq_direct:
- Packet rate: 831k txs
- Trace data:
ibmvnic_xmit = 48940092.29 us / 8187630 hits = AVG 5.98 us
skb_checksum_help = 2.03 us / 1 hits = AVG 2.03
ibmvnic_tx_scrq_flush = 31141879.57 us / 7948960 hits = AVG 3.92 us
send_subcrq_indirect = 8412506.03 us / 728781 hits = AVG 11.54
^ notice hits is much lower b/c send_subcrq_direct was called
^ wasn't traceable
3. driver does checksum, safely use send_subcrq_direct (THIS PATCH):
- Packet rate: 829k txs
- Trace data:
ibmvnic_xmit = 56696077.63 us / 8066168 hits = AVG 7.03 us
skb_checksum_help = 8587456.16 us / 7526072 hits = AVG 1.14 us
ibmvnic_tx_scrq_flush = 30219545.55 us / 7782409 hits = AVG 3.88 us
send_subcrq_indirect = 8638326.44 us / 763693 hits = AVG 11.31 us
When the bitstring ever specifies that CSO does not require headers
(dependent on VIOS vnic server changes), then this patch should be
removed and replaced with one that investigates the bitstring before
using send_subcrq_direct.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://patch.msgid.link/20240807211809.1259563-8-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Byte Queue Limits depends on dql_completed being called once per tx
completion round in order to adjust its algorithm appropriately. The
dql->limit value is an approximation of the amount of bytes that the NIC
can consume per irq interval. If this approximation is too high then the
NIC will become over-saturated. Too low and the NIC will starve.
The dql->limit depends on dql->prev-* stats to calculate an optimal
value. If dql_completed() is called more than once per irq handler then
those prev-* values become unreliable (because they are not an accurate
representation of the previous state of the NIC) resulting in a
sub-optimal limit value.
Therefore, move the call to netdev_tx_completed_queue() to the end of
ibmvnic_complete_tx().
When performing 150 sessions of TCP rr (request-response 1 byte packets)
workloads, one could observe:
PREVIOUSLY: - limit and inflight values hovering around 130
- transaction rate of around 750k pps.
NOW: - limit rises and falls in response to inflight (130-900)
- transaction rate of around 1M pps (33% improvement)
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://patch.msgid.link/20240807211809.1259563-7-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Firmware supports two hcalls to send a sub-crq request:
H_SEND_SUB_CRQ_INDIRECT and H_SEND_SUB_CRQ. The indirect hcall allows
for submission of batched messages while the other hcall is limited to
only one message. This protocol is defined in PAPR section 17.2.3.3.
Previously, the ibmvnic xmit function only used the indirect hcall. This
allowed the driver to batch it's skbs. A single skb can occupy a few
entries per hcall depending on if FW requires skb header information or
not. The FW only needs header information if the packet is segmented.
By this logic, if an skb is not GSO then it can fit in one sub-crq
message and therefore is a candidate for H_SEND_SUB_CRQ.
Batching skb transmission is only useful when there are more packets
coming down the line (ie netdev_xmit_more is true).
As it turns out, H_SEND_SUB_CRQ induces less latency than
H_SEND_SUB_CRQ_INDIRECT. Therefore, use H_SEND_SUB_CRQ where
appropriate.
Small latency gains seen when doing TCP_RR_150 (request/response
workload). Ftrace results (graph-time=1):
Previous:
ibmvnic_xmit = 29618270.83 us / 8860058.0 hits = AVG 3.34
ibmvnic_tx_scrq_flush = 21972231.02 us / 6553972.0 hits = AVG 3.35
Now:
ibmvnic_xmit = 22153350.96 us / 8438942.0 hits = AVG 2.63
ibmvnic_tx_scrq_flush = 15858922.4 us / 6244076.0 hits = AVG 2.54
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://patch.msgid.link/20240807211809.1259563-6-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Previously when creating the header descriptors, the driver would:
1. allocate a temporary buffer on the stack (in build_hdr_descs_arr)
2. memcpy the header info into the temporary buffer (in build_hdr_data)
3. memcpy the temp buffer into a local variable (in create_hdr_descs)
4. copy the local variable into the return buffer (in create_hdr_descs)
Since, there is no opportunity for errors during this process, the temp
buffer is not needed and work can be done on the return buffer directly.
Repurpose build_hdr_data() to only calculate the header lengths. Rename
it to get_hdr_lens().
Edit create_hdr_descs() to read from the skb directly and copy directly
into the returned useful buffer.
The process now involves less memory and write operations while
also being more readable.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://patch.msgid.link/20240807211809.1259563-4-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the header length helper functions rather than trying to calculate
it within the driver. There are defined functions for mac and network
headers (skb_mac_header_len and skb_network_header_len) but no such
function exists for the transport header length.
Also, hdr_data was memset during allocation to all 0's so no need to
memset again.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://patch.msgid.link/20240807211809.1259563-3-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Previously, the driver would replenish the rx pool if the polling
function consumed less than the budget. The logic being that the driver
did not exhaust its budget so that must mean that the driver is not busy
and has cycles to spare for replenishing the pool.
So pool replenishment happens on every poll which did not consume
the budget. This can very costly during request-response tests.
In fact, an extra ~100pps can be seen in TCP_RR_150 tests when we remove
this conditional. Trace results (ftrace, graph-time=1) for the poll
function are below:
Previous results:
ibmvnic_poll = 64951846.0 us / 4167628.0 hits = AVG 15.58
replenish_rx_pool = 17602846.0 us / 4710437.0 hits = AVG 3.74
Now:
ibmvnic_poll = 57673941.0 us / 4791737.0 hits = AVG 12.04
replenish_rx_pool = 3938171.6 us / 4314.0 hits = AVG 912.88
While the replenish function takes longer, it is hit less frequently
meaning the ibmvnic_poll function, on average, is faster.
Furthermore, this change does not have a negative effect on
performance bandwidth/latency measurements.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://patch.msgid.link/20240807211809.1259563-2-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the length of a packet is under the rx_copybreak threshold, the
buffer is copied into a new skb and sent up the stack. This allows the
dma mapped memory to be recycled back to FW.
Previously, the reuse of the DMA space was handled immediately.
This means that further packet processing has to wait until
h_add_logical_lan finishes for this packet.
Therefore, when reusing a packet, offload the hcall to the replenish
function. As a result, much of the shared logic between the recycle and
replenish functions can be removed.
This change increases TCP_RR packet rate by another 15% (370k to 430k
txns). We can see the ftrace data supports this:
PREV: ibmveth_poll = 8078553.0 us / 190999.0 hits = AVG 42.3 us
NEW: ibmveth_poll = 7632787.0 us / 224060.0 hits = AVG 34.07 us
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://patch.msgid.link/20240801211215.128101-3-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the ibmveth driver processes less than the budget, it must call
napi_complete_done() to release the instance. This function will
return false if the driver should avoid rearming interrupts.
Previously, the driver was ignoring the return code of
napi_complete_done(). As a result, there were unnecessary calls to
enable the veth irq.
Therefore, use the return code napi_complete_done() to determine if
irq rearm is necessary.
Additionally, in the event that new data is received immediately after
rearming interrupts, rather than just rescheduling napi, also jump
back to the poll processing loop since we are already in the poll
function (and know that we did not expense all of budget).
This slight tweak results in a 15% increase in TCP_RR transaction rate
(320k to 370k txns). We can see the ftrace data supports this:
PREV: ibmveth_poll = 8818014.0 us / 182802.0 hits = AVG 48.24
NEW: ibmveth_poll = 8082398.0 us / 191413.0 hits = AVG 42.22
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://patch.msgid.link/20240801211215.128101-2-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Below is a summary of how the driver stores a reference to an skb during
transmit:
tx_buff[free_map[consumer_index]]->skb = new_skb;
free_map[consumer_index] = IBMVNIC_INVALID_MAP;
consumer_index ++;
Where variable data looks like this:
free_map == [4, IBMVNIC_INVALID_MAP, IBMVNIC_INVALID_MAP, 0, 3]
consumer_index^
tx_buff == [skb=null, skb=<ptr>, skb=<ptr>, skb=null, skb=null]
The driver has checks to ensure that free_map[consumer_index] pointed to
a valid index but there was no check to ensure that this index pointed
to an unused/null skb address. So, if, by some chance, our free_map and
tx_buff lists become out of sync then we were previously risking an
skb memory leak. This could then cause tcp congestion control to stop
sending packets, eventually leading to ETIMEDOUT.
Therefore, add a conditional to ensure that the skb address is null. If
not then warn the user (because this is still a bug that should be
patched) and free the old pointer to prevent memleak/tcp problems.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
There are 2 types of outstanding tx skb's:
Type 1: Packets that are sitting in the drivers ind_buff that are
waiting to be batch sent to the NIC. During a device reset, these are
freed with a call to ibmvnic_tx_scrq_clean_buffer()
Type 2: Packets that have been sent to the NIC and are awaiting a TX
completion IRQ. These are free'd during a reset with a call to
clean_tx_pools()
During any reset which requires us to free the tx irq, ensure that the
Type 2 skb references are freed. Since the irq is released, it is
impossible for the NIC to inform of any completions.
Furthermore, later in the reset process is a call to init_tx_pools()
which marks every entry in the tx pool as free (ie not outstanding).
So if the driver is to make a call to init_tx_pools(), it must first
be sure that the tx pool is empty of skb references.
This issue was discovered by observing the following in the logs during
EEH testing:
TX free map points to untracked skb (tso_pool 0 idx=4)
TX free map points to untracked skb (tso_pool 0 idx=5)
TX free map points to untracked skb (tso_pool 1 idx=36)
Fixes: 65d6470d13 ("ibmvnic: clean pending indirect buffs during reset")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Simon reported that ndo_change_mtu() methods were never
updated to use WRITE_ONCE(dev->mtu, new_mtu) as hinted
in commit 501a90c945 ("inet: protect against too small
mtu values.")
We read dev->mtu without holding RTNL in many places,
with READ_ONCE() annotations.
It is time to take care of ndo_change_mtu() methods
to use corresponding WRITE_ONCE()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Simon Horman <horms@kernel.org>
Closes: https://lore.kernel.org/netdev/20240505144608.GB67882@kernel.org/
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Acked-by: Shannon Nelson <shannon.nelson@amd.com>
Link: https://lore.kernel.org/r/20240506102812.3025432-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Embedding net_device into structures prohibits the usage of flexible
arrays in the net_device structure. For more details, see the discussion
at [1].
Un-embed the net_device from the private struct by converting it
into a pointer. Then use the leverage the new alloc_netdev_dummy()
helper to allocate and initialize dummy devices.
[1] https://lore.kernel.org/all/20240229225910.79e224cf@kernel.org/
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In ibmvnic_xmit() if ibmvnic_tx_scrq_flush() returns H_CLOSED then
it will inform upper level networking functions to disable tx
queues. H_CLOSED signals that the connection with the vnic server is
down and a transport event is expected to recover the device.
Previously, ibmvnic_tx_scrq_flush() was hard-coded to return success.
Therefore, the queues would remain active until ibmvnic_cleanup() is
called within do_reset().
The problem is that do_reset() depends on the RTNL lock. If several
ibmvnic devices are resetting then there can be a long wait time until
the last device can grab the lock. During this time the tx/rx queues
still appear active to upper level functions.
FYI, we do make a call to netif_carrier_off() outside the RTNL lock but
its calls to dev_deactivate() are also dependent on the RTNL lock.
As a result, large amounts of retransmissions were observed in a short
period of time, eventually leading to ETIMEOUT. This was specifically
seen with HNV devices, likely because of even more RTNL dependencies.
Therefore, ensure the return code of ibmvnic_tx_scrq_flush() is
propagated to the xmit function to allow for an earlier (and lock-less)
response to a transport event.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://lore.kernel.org/r/20240416164128.387920-1-nnac123@linux.ibm.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
commit 23baf831a3 ("mm, treewide: redefine MAX_ORDER sanely") has
changed the definition of MAX_ORDER to be inclusive. This has caused
issues with code that was not yet upstream and depended on the previous
definition.
To draw attention to the altered meaning of the define, rename MAX_ORDER
to MAX_PAGE_ORDER.
Link: https://lkml.kernel.org/r/20231228144704.14033-2-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
`strncpy` is deprecated for use on NUL-terminated destination strings
[1] and as such we should prefer more robust and less ambiguous string
interfaces.
NUL-padding is not required as the buffer is already memset to 0:
| memset(adapter->fw_version, 0, 32);
Note that another usage of strscpy exists on the same buffer:
| strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
Considering the above, a suitable replacement is `strscpy` [2] due to
the fact that it guarantees NUL-termination on the destination buffer
without unnecessarily NUL-padding.
Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#strncpy-on-nul-terminated-strings [1]
Link: https://manpages.debian.org/testing/linux-manual-4.8/strscpy.9.en.html [2]
Link: https://github.com/KSPP/linux/issues/90
Cc: linux-hardening@vger.kernel.org
Signed-off-by: Justin Stitt <justinstitt@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that napi_schedule return a bool, we can drop napi_reschedule that
does the same exact function. The function comes from a very old commit
bfe13f54f5 ("ibm_emac: Convert to use napi_struct independent of struct
net_device") and the purpose is actually deprecated in favour of
different logic.
Convert every user of napi_reschedule to napi_schedule.
Signed-off-by: Christian Marangi <ansuelsmth@gmail.com>
Acked-by: Jeff Johnson <quic_jjohnson@quicinc.com> # ath10k
Acked-by: Nick Child <nnac123@linux.ibm.com> # ibm
Acked-by: Marc Kleine-Budde <mkl@pengutronix.de> # for can/dev/rx-offload.c
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://lore.kernel.org/r/20231009133754.9834-3-ansuelsmth@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In some OVS environments the TCP pseudo header checksum may need to be
recomputed. Currently this is only done when the interface instance is
configured for "Trunk Mode". We found the issue also occurs in some
Kubernetes environments, these environments do not use "Trunk Mode",
therefor the condition is removed.
Performance tests with this change show only a fractional decrease in
throughput (< 0.2%).
Fixes: 7525de2516 ("ibmveth: Set CHECKSUM_PARTIAL if NULL TCP CSUM.")
Signed-off-by: David Wilder <dwilder@us.ibm.com>
Reviewed-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The .remove() callback for a platform driver returns an int which makes
many driver authors wrongly assume it's possible to do error handling by
returning an error code. However the value returned is ignored (apart
from emitting a warning) and this typically results in resource leaks.
To improve here there is a quest to make the remove callback return
void. In the first step of this quest all drivers are converted to
.remove_new() which already returns void. Eventually after all drivers
are converted, .remove_new() is renamed to .remove().
Trivially convert these drivers from always returning zero in the remove
callback to the void returning variant.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
When building for power4, newer binutils don't recognise the "dcbfl"
extended mnemonic.
dcbfl RA, RB is equivalent to dcbf RA, RB, 1.
Switch to "dcbf" to avoid the build error.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a login request fails, the recovery process should be protected
against parallel resets. It is a known issue that freeing and
registering CRQ's in quick succession can result in a failover CRQ from
the VIOS. Processing a failover during login recovery is dangerous for
two reasons:
1. This will result in two parallel initialization processes, this can
cause serious issues during login.
2. It is possible that the failover CRQ is received but never executed.
We get notified of a pending failover through a transport event CRQ.
The reset is not performed until a INIT CRQ request is received.
Previously, if CRQ init fails during login recovery, then the ibmvnic
irq is freed and the login process returned error. If failover_pending
is true (a transport event was received), then the ibmvnic device
would never be able to process the reset since it cannot receive the
CRQ_INIT request due to the irq being freed. This leaved the device
in a inoperable state.
Therefore, the login failure recovery process must be hardened against
these possible issues. Possible failovers (due to quick CRQ free and
init) must be avoided and any issues during re-initialization should be
dealt with instead of being propagated up the stack. This logic is
similar to that of ibmvnic_probe().
Fixes: dff515a3e7 ("ibmvnic: Harden device login requests")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230809221038.51296-5-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Perform a partial reset before sending a login request if any of the
following are true:
1. If a previous request times out. This can be dangerous because the
VIOS could still receive the old login request at any point after
the timeout. Therefore, it is best to re-register the CRQ's and
sub-CRQ's before retrying.
2. If the previous request returns an error that is not described in
PAPR. PAPR provides procedures if the login returns with partial
success or aborted return codes (section L.5.1) but other values
do not have a defined procedure. Previously, these conditions
just returned error from the login function rather than trying
to resolve the issue.
This can cause further issues since most callers of the login
function are not prepared to handle an error when logging in. This
improper cleanup can lead to the device being permanently DOWN'd.
For example, if the VIOS believes that the device is already logged
in then it will return INVALID_STATE (-7). If we never re-register
CRQ's then it will always think that the device is already logged
in. This leaves the device inoperable.
The partial reset involves freeing the sub-CRQs, freeing the CRQ then
registering and initializing a new CRQ and sub-CRQs. This essentially
restarts all communication with VIOS to allow for a fresh login attempt
that will be unhindered by any previous failed attempts.
Fixes: dff515a3e7 ("ibmvnic: Harden device login requests")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230809221038.51296-4-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Rather than leaving the DMA unmapping of the login buffers to the
login response handler, move this work into the login release functions.
Previously, these functions were only used for freeing the allocated
buffers. This could lead to issues if there are more than one
outstanding login buffer requests, which is possible if a login request
times out.
If a login request times out, then there is another call to send login.
The send login function makes a call to the login buffer release
function. In the past, this freed the buffers but did not DMA unmap.
Therefore, the VIOS could still write to the old login (now freed)
buffer. It is for this reason that it is a good idea to leave the DMA
unmap call to the login buffers release function.
Since the login buffer release functions now handle DMA unmapping,
remove the duplicate DMA unmapping in handle_login_rsp().
Fixes: dff515a3e7 ("ibmvnic: Harden device login requests")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230809221038.51296-3-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If the LOGIN CRQ fails to send then we must DMA unmap the response
buffer. Previously, if the CRQ failed then the memory was freed without
DMA unmapping.
Fixes: c98d9cc417 ("ibmvnic: send_login should check for crq errors")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230809221038.51296-2-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Ensure that all offsets in a login response buffer are within the size
of the allocated response buffer. Any offsets or lengths that surpass
the allocation are likely the result of an incomplete response buffer.
In these cases, a full reset is necessary.
When attempting to login, the ibmvnic device will allocate a response
buffer and pass a reference to the VIOS. The VIOS will then send the
ibmvnic device a LOGIN_RSP CRQ to signal that the buffer has been filled
with data. If the ibmvnic device does not get a response in 20 seconds,
the old buffer is freed and a new login request is sent. With 2
outstanding requests, any LOGIN_RSP CRQ's could be for the older
login request. If this is the case then the login response buffer (which
is for the newer login request) could be incomplete and contain invalid
data. Therefore, we must enforce strict sanity checks on the response
buffer values.
Testing has shown that the `off_rxadd_buff_size` value is filled in last
by the VIOS and will be the smoking gun for these circumstances.
Until VIOS can implement a mechanism for tracking outstanding response
buffers and a method for mapping a LOGIN_RSP CRQ to a particular login
response buffer, the best ibmvnic can do in this situation is perform a
full reset.
Fixes: dff515a3e7 ("ibmvnic: Harden device login requests")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://lore.kernel.org/r/20230809221038.51296-1-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
gcc with W=1 reports
drivers/net/ethernet/ibm/ibmvnic.c:194:13: warning: variable 'rc' set but not used [-Wunused-but-set-variable]
^
This variable is not used so remove it.
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202308040609.zQsSXWXI-lkp@intel.com/
Signed-off-by: Yu Liao <liaoyu15@huawei.com>
Reviewed-by: Nick Child <nnac123@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DT of_device.h and of_platform.h date back to the separate
of_platform_bus_type before it as merged into the regular platform bus.
As part of that merge prepping Arm DT support 13 years ago, they
"temporarily" include each other. They also include platform_device.h
and of.h. As a result, there's a pretty much random mix of those include
files used throughout the tree. In order to detangle these headers and
replace the implicit includes with struct declarations, users need to
explicitly include the correct includes.
Acked-by: Alex Elder <elder@linaro.org>
Reviewed-by: Bhupesh Sharma <bhupesh.sharma@linaro.org>
Reviewed-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230727014944.3972546-1-robh@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
All ibmvnic resets, make a call to netdev_tx_reset_queue() when
re-opening the device. netdev_tx_reset_queue() resets the num_queued
and num_completed byte counters. These stats are used in Byte Queue
Limit (BQL) algorithms. The difference between these two stats tracks
the number of bytes currently sitting on the physical NIC. ibmvnic
increases the number of queued bytes though calls to
netdev_tx_sent_queue() in the drivers xmit function. When, VIOS reports
that it is done transmitting bytes, the ibmvnic device increases the
number of completed bytes through calls to netdev_tx_completed_queue().
It is important to note that the driver batches its transmit calls and
num_queued is increased every time that an skb is added to the next
batch, not necessarily when the batch is sent to VIOS for transmission.
Unlike other reset types, a NON FATAL reset will not flush the sub crq
tx buffers. Therefore, it is possible for the batched skb array to be
partially full. So if there is call to netdev_tx_reset_queue() when
re-opening the device, the value of num_queued (0) would not account
for the skb's that are currently batched. Eventually, when the batch
is sent to VIOS, the call to netdev_tx_completed_queue() would increase
num_completed to a value greater than the num_queued. This causes a
BUG_ON crash:
ibmvnic 30000002: Firmware reports error, cause: adapter problem.
Starting recovery...
ibmvnic 30000002: tx error 600
ibmvnic 30000002: tx error 600
ibmvnic 30000002: tx error 600
ibmvnic 30000002: tx error 600
------------[ cut here ]------------
kernel BUG at lib/dynamic_queue_limits.c:27!
Oops: Exception in kernel mode, sig: 5
[....]
NIP dql_completed+0x28/0x1c0
LR ibmvnic_complete_tx.isra.0+0x23c/0x420 [ibmvnic]
Call Trace:
ibmvnic_complete_tx.isra.0+0x3f8/0x420 [ibmvnic] (unreliable)
ibmvnic_interrupt_tx+0x40/0x70 [ibmvnic]
__handle_irq_event_percpu+0x98/0x270
---[ end trace ]---
Therefore, do not reset the dql stats when performing a NON_FATAL reset.
Fixes: 0d97338818 ("ibmvnic: Introduce xmit_more support using batched subCRQ hcalls")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It is preferred to use typed property access functions (i.e.
of_property_read_<type> functions) rather than low-level
of_get_property/of_find_property functions for reading properties.
Convert reading boolean properties to of_property_read_bool().
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Acked-by: Marc Kleine-Budde <mkl@pengutronix.de> # for net/can
Acked-by: Kalle Valo <kvalo@kernel.org>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
Reviewed-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
When setting the XPS map value for TX queues, use the index of the
transmit queue.
Previously, the function was passing the index of the loop that iterates
over all queues (RX and TX). This was causing invalid XPS map values.
Fixes: 6831582937 ("ibmvnic: Toggle between queue types in affinity mapping")
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://lore.kernel.org/r/20230223153944.44969-1-nnac123@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Previously, ibmvnic IRQs were assigned to CPU numbers by assigning all
the IRQs for transmit queues then assigning all the IRQs for receive
queues. With multi-threaded processors, in a heavy RX or TX environment,
physical cores would either be overloaded or underutilized (due to the
IRQ assignment algorithm). This approach is sub-optimal because IRQs for
the same subprocess (RX or TX) would be bound to adjacent CPU numbers,
meaning they were more likely to be contending for the same core.
For example, in a system with 64 CPU's and 32 queues, the IRQs would
be bound to CPU in the following pattern:
IRQ type | CPU number
-----------------------
TX0 | 0-1
TX1 | 2-3
<etc>
RX0 | 32-33
RX1 | 34-35
<etc>
Observe that in SMT-8, the first 4 tx queues would be sharing the
same core.
A more optimal algorithm would balance the number RX and TX IRQ's across
the physical cores. Therefore, to increase performance, distribute RX and
TX IRQs across cores by alternating between assigning IRQs for RX and TX
queues to CPUs.
With a system with 64 CPUs and 32 queues, this results in the following
pattern:
IRQ type | CPU number
-----------------------
TX0 | 0-1
RX0 | 2-3
TX1 | 4-5
RX1 | 6-7
<etc>
Observe that in SMT-8, there is equal distribution of RX and TX IRQs
per core. In the above case, each core handles 2 TX and 2 RX IRQ's.
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Haren Myneni <haren@linux.ibm.com>
Link: https://lore.kernel.org/r/20230127214358.318152-1-nnac123@linux.ibm.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Transmit Packet Steering (XPS) maps cpu numbers to transmit
queues. By running the same connection on the same set of cpu's,
contention for the queue and cache miss rate can be minimized.
When assigning a cpu mask for a tranmit queues irq number, assign
the same cpu mask as the set of cpu's that XPS should use for that
queue.
Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Signed-off-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Reviewed-by: Rick Lindsley <ricklind@linux.ibm.com>
Reviewed-by: Haren Myneni <haren@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
