ENA admin command queue errors are not handled as part of ena_down().
As a result, in case of error admin queue transitions to non-running
state and aborts all subsequent commands including those coming from
ena_up(). Reset scheduled by the driver from the timer service
context would not proceed due to sharing rtnl with ena_up()/ena_down()
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Under certain conditions MSI-X interrupt might arrive right after it
was unmasked in ena_up(). There is a chance it would be processed by
the driver before device ENA_FLAG_DEV_UP flag is set. In such a case
the interrupt is ignored.
ENA device operates in auto-masked mode, therefore ignoring
interrupt leaves it masked for good.
Moving unmask of interrupt to be the last step in ena_up().
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In rare cases, ena driver would reset and re-start the device,
for example, in case of misbehaving application that causes
transmit timeout
The first step in the reset procedure is to stop the Tx traffic by
calling ena_carrier_off().
After the driver have just started the device reset procedure, device
happens to send an asynchronous notification (via AENQ) to the driver
than there was a link change (to link-up state).
This link change is mapped to a call to netif_carrier_on() which
re-activates the Tx queues, violating the assumption of no tx traffic
until device reset is completed, as the reset task might still be in
the process of queues initialization, leading to an access to
uninitialized memory.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Merge updates from Andrew Morton:
- a few misc bits
- ocfs2 updates
- almost all of MM
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (131 commits)
memory hotplug: fix comments when adding section
mm: make alloc_node_mem_map a void call if we don't have CONFIG_FLAT_NODE_MEM_MAP
mm: simplify nodemask printing
mm,oom_reaper: remove pointless kthread_run() error check
mm/page_ext.c: check if page_ext is not prepared
writeback: remove unused function parameter
mm: do not rely on preempt_count in print_vma_addr
mm, sparse: do not swamp log with huge vmemmap allocation failures
mm/hmm: remove redundant variable align_end
mm/list_lru.c: mark expected switch fall-through
mm/shmem.c: mark expected switch fall-through
mm/page_alloc.c: broken deferred calculation
mm: don't warn about allocations which stall for too long
fs: fuse: account fuse_inode slab memory as reclaimable
mm, page_alloc: fix potential false positive in __zone_watermark_ok
mm: mlock: remove lru_add_drain_all()
mm, sysctl: make NUMA stats configurable
shmem: convert shmem_init_inodecache() to void
Unify migrate_pages and move_pages access checks
mm, pagevec: rename pagevec drained field
...
As the page free path makes no distinction between cache hot and cold
pages, there is no real useful ordering of pages in the free list that
allocation requests can take advantage of. Juding from the users of
__GFP_COLD, it is likely that a number of them are the result of copying
other sites instead of actually measuring the impact. Remove the
__GFP_COLD parameter which simplifies a number of paths in the page
allocator.
This is potentially controversial but bear in mind that the size of the
per-cpu pagelists versus modern cache sizes means that the whole per-cpu
list can often fit in the L3 cache. Hence, there is only a potential
benefit for microbenchmarks that alloc/free pages in a tight loop. It's
even worse when THP is taken into account which has little or no chance
of getting a cache-hot page as the per-cpu list is bypassed and the
zeroing of multiple pages will thrash the cache anyway.
The truncate microbenchmarks are not shown as this patch affects the
allocation path and not the free path. A page fault microbenchmark was
tested but it showed no sigificant difference which is not surprising
given that the __GFP_COLD branches are a miniscule percentage of the
fault path.
Link: http://lkml.kernel.org/r/20171018075952.10627-9-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There were quite a few overlapping sets of changes here.
Daniel's bug fix for off-by-ones in the new BPF branch instructions,
along with the added allowances for "data_end > ptr + x" forms
collided with the metadata additions.
Along with those three changes came veritifer test cases, which in
their final form I tried to group together properly. If I had just
trimmed GIT's conflict tags as-is, this would have split up the
meta tests unnecessarily.
In the socketmap code, a set of preemption disabling changes
overlapped with the rename of bpf_compute_data_end() to
bpf_compute_data_pointers().
Changes were made to the mv88e6060.c driver set addr method
which got removed in net-next.
The hyperv transport socket layer had a locking change in 'net'
which overlapped with a change of socket state macro usage
in 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new statistic to ethtool stats that show the number of packets
without transmit acknowledgement from ENA device.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove ena_device_io_suspend/resume() methods
Those methods were intend to be used by the device to trigger
suspend/resume but eventually it was dropped.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ena admin commands timeout is in resolutions of 100ms.
Therefore, When the driver works in polling mode, it sleeps for 100ms
each time. The overall boot time of the ENA driver is ~1.5 sec.
To reduce the boot time, This change modifies the granularity of
the sleeps to 5ms.
This change improves the boot time to 220ms.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ethtool ena_get_channels() expose the max number of queues as the max
number of queues ENA supports (128 queues) and not the actual number
of created queues.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Decrease log level of checksum errors as these messages can be
triggered remotely by bad packets.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IS_ERR() already implies unlikely(), so it can be omitted.
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
rx drop counter is reported by the device in the keep-alive
event.
update the driver's counter with the device counter.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In ena_com_mem_addr_set(), use the above functions to split dma address
to the lower 32 bits and the higher 16 bits.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current driver tries to allocate msix vectors as the number of the
negotiated io queues. (with another msix vector for management).
If pci_alloc_irq_vectors() fails, the driver aborts the probe
and the ENA network device is never brought up.
With this patch, the driver's logic will reduce the number of IO
queues to the number of allocated msix vectors (minus one for management)
instead of failing probe().
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ENA driver post Rx buffers through the Rx submission queue
for the ENA device to fill them with receive packets.
Each Rx buffer is marked with req_id in the Rx descriptor.
Newer ENA devices could consume the posted Rx buffer in out of order,
and as result the corresponding Rx completion queue will have Rx
completion descriptors with non contiguous req_id(s)
In this change the driver holds two rings.
The first ring (called free_rx_ids) is a mapping ring.
It holds all the unused request ids.
The values in this ring are from 0 to ring_size -1.
When the driver wants to allocate a new Rx buffer it uses the head of
free_rx_ids and uses it's value as the index for rx_buffer_info ring.
The req_id is also written to the Rx descriptor
Upon Rx completion,
The driver took the req_id from the completion descriptor and uses it
as index in rx_buffer_info.
The req_id is then return to the free_rx_ids ring.
This patch also adds statistics to inform when the driver receive out
of range or unused req_id.
Note:
free_rx_ids is only accessible from the napi handler, so no locking is
required
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For each device reset, log to the device what is the cause
the reset occur.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of using:
memset(ptr, 0x0, sizeof(struct ...))
use:
memset(ptr, 0x0, sizeor(*ptr))
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With this patch, ENA device can update the ena driver about
the desired timeout values:
These values are part of the "hardware hints" which are transmitted
to the driver as Asynchronous event through ENA async
event notification queue.
In case the ENA device does not support this capability,
the driver will use its own default values.
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
check_for_missing_tx_completions() is called from a timer
task and looking for lost tx packets.
The old implementation accumulate all the lost tx packets
and did not check if those packets were retrieved on a later stage.
This cause to a situation where the driver reset
the device for no reason.
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the rare case where the device runs out of free rx buffer
descriptors (in case of pressure on kernel memory),
and the napi handler continuously fail to refill new Rx descriptors
until device rx queue totally runs out of all free rx buffers
to post incoming packet, leading to a deadlock:
* The device won't send interrupts since all the new
Rx packets will be dropped.
* The napi handler won't try to allocate new Rx descriptors
since allocation is part of NAPI that's not being invoked any more
The fix involves detecting this scenario and rescheduling NAPI
(to refill buffers) by the keepalive/watchdog task.
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch also change the mapping functions to devm_ functions
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bug:
"Completion context is occupied" error printout will be noticed in
dmesg.
This error will cause the admin command to fail, which will lead to
an ena_probe() failure or a watchdog reset (depends on which admin
command failed).
Root cause:
__ena_com_submit_admin_cmd() is the function that submits new entries to
the admin queue.
The function have a check that makes sure the queue is not full and the
function does not override any outstanding command.
It uses head and tail indexes for this check.
The head is increased by ena_com_handle_admin_completion() which runs
from interrupt context, and the tail index is increased by the submit
function (the function is running under ->q_lock, so there is no risk
of multithread increment).
Each command is associated with a completion context. This context
allocated before call to __ena_com_submit_admin_cmd() and freed by
ena_com_wait_and_process_admin_cq_interrupts(), right after the command
was completed.
This can lead to a state where the head was increased, the check passed,
but the completion context is still in use.
Solution:
Use the atomic variable ->outstanding_cmds instead of using the head and
the tail indexes.
This variable is safe for use since it is bumped in get_comp_ctx() in
__ena_com_submit_admin_cmd() and is freed by comp_ctxt_release()
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixing a bug that the driver does not unmask the IO interrupts
in ndo_open():
occasionally, the MSI-X interrupt (for one or more IO queues)
can be masked when ndo_close() was called.
If that is followed by ndo open(),
then the MSI-X will be still masked so no interrupt
will be received by the driver.
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current flow to detect admin completion is:
while (command_not_completed) {
if (timeout)
error
check_for_completion()
sleep()
}
So in case the sleep took more than the timeout
(in case the thread/workqueue was not scheduled due to higher priority
task or prolonged VMexit), the driver can detect a stall even if
the completion is present.
The fix changes the order of this function to first check for
completion and only after that check if the timeout expired.
Fixes: 1738cd3ed3 ("Add a driver for Amazon Elastic Network Adapters (ENA)")
Signed-off-by: Netanel Belgazal <netanel@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the deprecated pci_enable_msix API in favour of its successor.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The check in ena_remove() for the pci driver data not being NULL is not
needed, since it is always set in the probe() function. Remove the
superfluous check.
Signed-off-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the host info config to be the first admin command that is executed.
This change require the driver to remove the 'feature check'
from host info configuration flow.
The check is removed since the supported features bitmask field
is retrieved only after calling ENA_ADMIN_DEVICE_ATTRIBUTES admin command.
If set host info is not supported an error will be returned by the device.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The timeouts were too agressive and sometimes cause false alarms.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Completion descriptors are accessed from the driver and from the device.
To avoid reading the old value, use READ_ONCE macro.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Do not unamsk interrupts if we are in busy poll mode.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the ena driver detects that the device is not behave as expected,
it tries to reset the device.
The reset flow calls ena_down, which will frees all the resources
the driver allocates and then it will reset the device.
This flow can cause memory corruption if the device is still writes
to the driver's memory space.
To overcome this potential race, move the reset before the device
resources are freed.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ndo_get_stat64() can be called from atomic context, but the current
implementation sends an admin command to retrieve the statistics from
the device. This admin command can sleep.
This patch re-factors the implementation of ena_get_stats64() to use
the {rx,tx}bytes/count from the driver's inner counters, and to obtain
the rx drop counter from the asynchronous keep alive (heart bit)
event.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ENA default hash configures IPv4_frag hash twice instead of
configure non-IP packets.
The bug caused IPv4 fragmented packets to be calculated based on
L2 source and destination address instead of L3 source and destination.
IPv4 packets can reach to the wrong Rx queue.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ena_flow_data_to_flow_hash and ena_flow_hash_to_flow_type
treat the ena_flow_hash_to_flow_type enum as power of two values.
Change the values of ena_admin_flow_hash_fields to be power of two values.
This bug effect the ethtool set/get rxnfc.
ethtool will report wrong values hash fields for get and will
configure wrong hash fields in set.
Signed-off-by: Netanel Belgazal <netanel@annapurnalabs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netanel Belgazal