On resume e1000e_vm_state_change() always calls e1000e_autoneg_resume()
that sets link_down to false, and thus activates the link even
if we have disabled it.
The problem can be reproduced starting qemu in paused state (-S) and
then set the link to down. When we resume the machine the link appears
to be up.
Reproducer:
# qemu-system-x86_64 ... -device e1000e,netdev=netdev0,id=net0 -S
{"execute": "qmp_capabilities" }
{"execute": "set_link", "arguments": {"name": "net0", "up": false}}
{"execute": "cont" }
To fix the problem, merge the content of e1000e_vm_state_change()
into e1000e_core_post_load() as e1000 does.
Buglink: https://issues.redhat.com/browse/RHEL-21867
Fixes: 6f3fbe4ed0 ("net: Introduce e1000e device emulation")
Suggested-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Rename e1000e_ba_state according and e1000e_write_hdr_to_rx_buffers for
consistency with IGB.
Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Tested-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Refactoring is done in preparation for support of multiple advanced
descriptors RX modes, especially packet-split modes.
Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Tested-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Rename E1000E_RingInfo_st and E1000E_RingInfo according to qemu typdefs guide.
Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Tested-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Passing additional parameters (USOv4 and USOv6 offloads) when
setting TAP offloads
Signed-off-by: Yuri Benditovich <yuri.benditovich@daynix.com>
Signed-off-by: Andrew Melnychenko <andrew@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The datasheet does not say what happens when interrupt was asserted
(ICR.INT_ASSERT=1) and auto mask is *not* active.
However, section of 13.3.27 the PCIe* GbE Controllers Open Source
Software Developer’s Manual, which were written for older devices,
namely 631xESB/632xESB, 82563EB/82564EB, 82571EB/82572EI &
82573E/82573V/82573L, does say:
> If IMS = 0b, then the ICR register is always clear-on-read. If IMS is
> not 0b, but some ICR bit is set where the corresponding IMS bit is not
> set, then a read does not clear the ICR register. For example, if
> IMS = 10101010b and ICR = 01010101b, then a read to the ICR register
> does not clear it. If IMS = 10101010b and ICR = 0101011b, then a read
> to the ICR register clears it entirely (ICR.INT_ASSERTED = 1b).
Linux does no longer activate auto mask since commit
0a8047ac68e50e4ccbadcfc6b6b070805b976885 and the real hardware clears
ICR even in such a case so we also should do so.
Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1707441
Signed-off-by: Andrew Melnychenko <andrew@daynix.com>
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
In MSI-X mode, if there are interrupts already notified but not cleared
and a new interrupt arrives, e1000e incorrectly notifies the notified
ones again along with the new one.
To fix this issue, replace e1000e_update_interrupt_state() with
two new functions: e1000e_raise_interrupts() and
e1000e_lower_interrupts(). These functions don't only raise or lower
interrupts, but it also performs register writes which updates the
interrupt state. Before it performs a register write, these function
determines the interrupts already raised, and compares with the
interrupts raised after the register write to determine the interrupts
to notify.
The introduction of these functions made tracepoints which assumes that
the caller of e1000e_update_interrupt_state() performs register writes
obsolete. These tracepoints are now removed, and alternative ones are
added to the new functions.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Rename variable "n" to "causes", which properly represents the content
of the variable.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Keeping Tx packet state after the transmit queue is emptied has some
problems:
- The datasheet says the descriptors can be reused after the transmit
queue is emptied, but the Tx packet state may keep references to them.
- The Tx packet state cannot be migrated so it can be reset anytime the
migration happens.
Always reset Tx packet state always after the queue is emptied.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Without this change, the status flags may not be traced e.g. if checksum
offloading is disabled.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
e1000e and igb employs NetPktRssIpV6TcpEx for RSS hash if TcpIpv6 MRQC
bit is set. Moreover, igb also has a MRQC bit for NetPktRssIpV6Tcp
though it is not implemented yet. Rename it to TcpIpv6Ex to avoid
confusion.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Sriram Yagnaraman <sriram.yagnaraman@est.tech>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This saves some code and enables tracepoint for e1000's VLAN filtering.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Sriram Yagnaraman <sriram.yagnaraman@est.tech>
Signed-off-by: Jason Wang <jasowang@redhat.com>
e1000e_receive_internal() used to check the iov length to determine
copy the iovs to a contiguous buffer, but the check is flawed in two
ways:
- It does not ensure that iovcnt > 0.
- It does not take virtio-net header into consideration.
The size of this copy is just 18 octets, which can be even less than
the code size required for checks. This (wrong) optimization is probably
not worth so just remove it.
Fixes: 6f3fbe4ed0 ("net: Introduce e1000e device emulation")
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Before this change, e1000 and the common code updated BPRC and MPRC
depending on the matched filter, but e1000e and igb decided to update
those counters by deriving the packet type independently. This
inconsistency caused a multicast packet to be counted twice.
Updating BPRC and MPRC depending on are fundamentally flawed anyway as
a filter can be used for different types of packets. For example, it is
possible to filter broadcast packets with MTA.
Always determine what counters to update by inspecting the packets.
Fixes: 3b27430177 ("e1000: Implementing various counters")
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Sriram Yagnaraman <sriram.yagnaraman@est.tech>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This allows to use the network packet abstractions even if PCI is not
used.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The bytes and packets counter registers are cleared on read.
Copying the "total counter" registers to the "good counter" registers has
side effects.
If the "total" register is never read by the OS, it only gets incremented.
This leads to exponential growth of the "good" register.
This commit increments the counters individually to avoid this.
Signed-off-by: Timothée Cocault <timothee.cocault@gmail.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
igb used to specify the PF as DMA requester when reading Tx packets.
This made Tx requests from VFs to be performed on the address space of
the PF, defeating the purpose of SR-IOV. Add some logic to change the
requester depending on the queue, which can be assigned to a VF.
Fixes: 3a977deebe ("Intrdocue igb device emulation")
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Some definitions in the header files are invalid for igb so extract
them to new header files to keep igb from referring to them.
Signed-off-by: Gal Hammer <gal.hammer@sap.com>
Signed-off-by: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
igb, a new network device emulation, will need SCTP checksum offloading.
Currently eth_get_protocols() has a bool parameter for each protocol
currently it supports, but there will be a bit too many parameters if
we add yet another protocol.
Introduce an enum type, EthL4HdrProto to represent all L4 protocols
eth_get_protocols() support with one parameter.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The system clock is necessary to implement PTP features. While we are
not implementing PTP features for e1000e yet, we do have a plan to
implement them for igb, a new network device derived from e1000e,
so add system clock to the common base first.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The values returned by eth_get_protocols() are used to perform RSS,
checksumming and segmentation. Even when a packet signals the use of the
protocols which these operations can be applied to, the headers for them
may not be present because of too short packet or fragmentation, for
example. In such a case, the operations cannot be applied safely.
Report the presence of headers instead of whether the use of the
protocols are indicated with eth_get_protocols(). This also makes
corresponding changes to the callers of eth_get_protocols() to match
with its new signature and to remove redundant checks for fragmentation.
Fixes: 75020a7021 ("Common definitions for VMWARE devices")
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The datasheet 8.19.29 "Good Packets Transmitted Count - GPTC (0x04080;
RC)" says:
> This register counts the number of good (no errors) packets
> transmitted. A good transmit packet is considered one that is 64 or
> more bytes in length (from <Destination Address> through <CRC>,
> inclusively) in length.
It also says similar for the other Tx statistics registers. Add the
number of bytes for CRC to those registers.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Whether a packet will be written back to the guest depends on the
remaining space of the queue. Therefore, e1000e_rx_written_to_guest and
e1000e_rx_not_written_to_guest should log the index of the queue instead
of generated interrupts. This also removes the need of
e1000e_rx_rss_dispatched_to_queue, which logs the queue index.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Assertions will fail if MSI-X gets disabled while a timer for MSI-X
interrupts is running so remove them to avoid abortions. Fortunately,
nothing bad happens even if the assertions won't trigger as
msix_notify(), called by timer handlers, does nothing when MSI-X is
disabled.
This bug was found by Alexander Bulekov when fuzzing igb, a new
device implementation derived from e1000e:
https://patchew.org/QEMU/20230129053316.1071513-1-alxndr@bu.edu/
The fixed test case is:
fuzz/crash_aea040166819193cf9fedb810c6d100221da721a
Fixes: 6f3fbe4ed0 ("net: Introduce e1000e device emulation")
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
e1000e didn't perform software segmentation for loopback if virtio-net
header is enabled, which is wrong.
To fix the problem, introduce net_tx_pkt_send_custom(), which allows the
caller to specify whether offloading should be assumed or not.
net_tx_pkt_send_custom() also allows the caller to provide a custom
sending function. Packets with virtio-net headers and ones without
virtio-net headers will be provided at the same time so the function
can choose the preferred version. In case of e1000e loopback, it prefers
to have virtio-net headers as they allows to skip the checksum
verification if VIRTIO_NET_HDR_F_DATA_VALID is set.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
When virtio-net header is not set, net_rx_pkt_get_vhdr() returns
zero-filled virtio_net_hdr, which is actually valid. In fact, tap device
uses zero-filled virtio_net_hdr when virtio-net header is not provided
by the peer. Therefore, we can just remove net_rx_pkt_has_virt_hdr() and
always assume NetTxPkt has a valid virtio-net header.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The new function qemu_get_using_vnet_hdr() allows to automatically
determine if virtio-net header is used.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
net_tx_pkt_build_vheader() inspects TCP header but had no check for
the header size, resulting in an undefined behavior. Check the header
size and drop the packet if the header is too small.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
e1000e_write_packet_to_guest() passes the reference of variable ba as a
pointer to an array, and that pointer indirection is just unnecessary;
all functions which uses the passed reference performs no pointer
operation on the pointer and they simply dereference the passed
pointer. Remove the extra pointer indirection.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This keeps Windows driver 12.18.9.23 from generating an event with ID
30. The description of the event is as follows:
> Intel(R) 82574L Gigabit Network Connection
> PROBLEM: The network adapter is configured for auto-negotiation but
> the link partner is not. This may result in a duplex mismatch.
> ACTION: Configure the link partner for auto-negotiation.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Before this change, e1000e_write_packet_to_guest() allocated the
receive descriptor buffer as an array of uint8_t. This does not ensure
the buffer is sufficiently aligned.
Introduce e1000_rx_desc_union type, a union type of all receive
descriptor types to correct this.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This change makes e1000e reset more things when software reset was
triggered. Some registers are exempted from software reset in the
datasheet and this change also implements the behavior accordingly.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
They are duplicate of running throttling timer flags and incomplete as
the flags are not cleared when the interrupts are fired or the device is
reset.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Use memcpy instead of memmove to initialize registers. The initial
register templates and register table instances will never overlap.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
When a register has effective bits fewer than their width, the old code
inconsistently masked when writing or reading. Make the code consistent
by always masking when writing, and remove some code duplication.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
e1000e_set_16bit and e1000e_set_12bit look so similar so define a
generic macro.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
hw/net/mii.h provides common definitions for MII.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
igb implementation first starts off by copying e1000e code. Correct the
code style before that.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Like commit 034d00d485 ("e1000: set RX descriptor status in
a separate operation"), there is also same issue in e1000e, which
would cause lost packets or stop sending packets to VM with DPDK.
Do similar fix in e1000e.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/402
Signed-off-by: Ding Hui <dinghui@sangfor.com.cn>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The compiler isn't clever enough to figure 'min_buf_size'
is a constant, so help it by using a definitions instead.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20220819153931.3147384-6-peter.maydell@linaro.org
Commit "e1000e: Prevent MSI/MSI-X storms" introduced msi_causes_pending
to prevent interrupt storms problem. It was tested with MSI-X.
In case of MSI, the guest can rely solely on interrupts to clear ICR.
Upon clearing all pending interrupts, msi_causes_pending gets cleared.
However, when e1000e_itr_should_postpone() in e1000e_send_msi() returns
true, MSI never gets fired by e1000e_intrmgr_on_throttling_timer()
because msi_causes_pending is still set. This results in interrupt loss.
To prevent this, we need to clear msi_causes_pending when MSI is going
to get fired by the throttling timer. The guest can then receive
interrupts eventually.
Signed-off-by: Ake Koomsin <ake@igel.co.jp>
Signed-off-by: Jason Wang <jasowang@redhat.com>
In section 7.4.3 of the 82574 datasheet it states that
"In systems that do not support MSI-X, reading the ICR
register clears it's bits..."
Some OSes rely on this.
Signed-off-by: Nick Hudson <skrll@netbsd.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
In the legacy RX descriptor mode, VLAN tag was saved to d->special
by e1000e_build_rx_metadata() in e1000e_write_lgcy_rx_descr(), but
it was then zeroed out again at the end of the call, which is wrong.
Fixes: c89d416a2b ("e1000e: Don't zero out buffer address in rx descriptor")
Reported-by: Markus Carlstedt <markus.carlstedt@windriver.com>
Signed-off-by: Christina Wang <christina.wang@windriver.com>
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The initial value of VLAN Ether Type (VET) register is 0x8100, as per
the manual and real hardware.
While Linux e1000e driver always writes VET register to 0x8100, it is
not always the case for everyone. Drivers relying on the reset value
of VET won't be able to transmit and receive VLAN frames in QEMU.
Unlike e1000 in QEMU, e1000e uses a field 'vet' in "struct E1000Core"
to cache the value of VET register, but the cache only gets updated
when VET register is written. To always get a consistent VET value
no matter VET is written or remains its reset value, drop the 'vet'
field and use 'core->mac[VET]' directly.
Reported-by: Markus Carlstedt <markus.carlstedt@windriver.com>
Signed-off-by: Christina Wang <christina.wang@windriver.com>
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The 'running' argument from VMChangeStateHandler does not require
other value than 0 / 1. Make it a plain boolean.
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <20210111152020.1422021-3-philmd@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This commit is the result of running the timer-del-timer-free.cocci
script on the whole source tree.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Corey Minyard <cminyard@mvista.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20201215154107.3255-4-peter.maydell@linaro.org
While receiving packets via e1000e_write_packet_to_guest() routine,
'desc_offset' is advanced only when RX descriptor is processed. And
RX descriptor is not processed if it has NULL buffer address.
This may lead to an infinite loop condition. Increament 'desc_offset'
to process next descriptor in the ring to avoid infinite loop.
Reported-by: Cheol-woo Myung <330cjfdn@gmail.com>
Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
Signed-off-by: Jason Wang <jasowang@redhat.com>
