Only signal MSI/MSI-X events on rising edges. So far we re-triggered the
interrupt sources even if the guest did no consumed the pending one,
easily causing interrupt storms.
Issue was observable with Linux 4.16 e1000e driver when MSI-X was used.
Vector 2 was causing interrupt storms after the driver activated the
device.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The spec does not justify clearing of any E1000_ICR_OTHER_CAUSES when
E1000_ICR_OTHER is set in EIAC. In fact, removing this code fixes the
issue the Linux driver runs into since 4aea7a5c5e94 ("e1000e: Avoid
receiver overrun interrupt bursts") and was worked around by
745d0bd3af99 ("e1000e: Remove Other from EIAC").
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
sum_needed and cptse flags are received from the guest within each
transmit data descriptor. They are not part of the offload context;
instead, they determine how to apply a previously received context to
the packet being transmitted:
- If cptse is set, perform both segmentation and checksum offload
using the parameters in the TSO context; otherwise just do checksum
offload. (Currently the e1000 device incorrectly stores only one
context, which will be fixed in a subsequent patch.)
- Depending on the bits set in sum_needed, possibly perform L4
checksum offload and/or IP checksum offload, using the parameters in
the appropriate context.
Move these flags out of struct e1000x_txd_props, which is otherwise
dedicated to storing values from a context descriptor, and into the
per-packet TX struct.
Signed-off-by: Ed Swierk <eswierk@skyportsystems.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
SunOS defines SEC in <sys/time.h> as 1 (commonly used time symbols).
This fixes build on SmartOS (Joyent).
Patch cherry-picked from pkgsrc by jperkin (Joyent).
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
Reviewed-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This commit fixes a bug which causes the guest to hang. The bug was
observed upon a "receive overrun" (bit #6 of the ICR register)
interrupt which could be triggered post migration in a heavy traffic
environment. Even though the "receive overrun" bit (#6) is masked out
by the IMS register (refer to the log below) the driver still receives
an interrupt as the "receive overrun" bit (#6) causes the "Other" -
bit #24 of the ICR register - bit to be set as documented below. The
driver handles the interrupt and clears the "Other" bit (#24) but
doesn't clear the "receive overrun" bit (#6) which leads to an
infinite loop. Apparently the Windows driver expects that the "receive
overrun" bit and other ones - documented below - to be cleared when
the "Other" bit (#24) is cleared.
So to sum that up:
1. Bit #6 of the ICR register is set by heavy traffic
2. As a results of setting bit #6, bit #24 is set
3. The driver receives an interrupt for bit 24 (it doesn't receieve an
interrupt for bit #6 as it is masked out by IMS)
4. The driver handles and clears the interrupt of bit #24
5. Bit #6 is still set.
6. 2 happens all over again
The Interrupt Cause Read - ICR register:
The ICR has the "Other" bit - bit #24 - that is set when one or more
of the following ICR register's bits are set:
LSC - bit #2, RXO - bit #6, MDAC - bit #9, SRPD - bit #16, ACK - bit
#17, MNG - bit #18
This bug can occur with any of these bits depending on the driver's
behaviour and the way it configures the device. However, trying to
reproduce it with any bit other than RX0 is challenging and came to
failure as the drivers don't implement most of these bits, trying to
reproduce it with LSC (Link Status Change - bit #2) bit didn't succeed
too as it seems that Windows handles this bit differently.
Log sample of the storm:
27563@1494850819.411877:e1000e_irq_pending_interrupts ICR PENDING: 0x1000000 (ICR: 0x815000c2, IMS: 0x1a00004)
27563@1494850819.411900:e1000e_irq_pending_interrupts ICR PENDING: 0x0 (ICR: 0x815000c2, IMS: 0xa00004)
27563@1494850819.411915:e1000e_irq_pending_interrupts ICR PENDING: 0x0 (ICR: 0x815000c2, IMS: 0xa00004)
27563@1494850819.412380:e1000e_irq_pending_interrupts ICR PENDING: 0x0 (ICR: 0x815000c2, IMS: 0xa00004)
27563@1494850819.412395:e1000e_irq_pending_interrupts ICR PENDING: 0x0 (ICR: 0x815000c2, IMS: 0xa00004)
27563@1494850819.412436:e1000e_irq_pending_interrupts ICR PENDING: 0x0 (ICR: 0x815000c2, IMS: 0xa00004)
27563@1494850819.412441:e1000e_irq_pending_interrupts ICR PENDING: 0x0 (ICR: 0x815000c2, IMS: 0xa00004)
27563@1494850819.412998:e1000e_irq_pending_interrupts ICR PENDING: 0x1000000 (ICR: 0x815000c2, IMS: 0x1a00004)
* This bug behaviour wasn't observed with the Linux driver.
This commit solves:
https://bugzilla.redhat.com/show_bug.cgi?id=1447935https://bugzilla.redhat.com/show_bug.cgi?id=1449490
Cc: qemu-stable@nongnu.org
Signed-off-by: Sameeh Jubran <sjubran@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This issue is like the issue in e1000 network card addressed in
this commit:
e1000: eliminate infinite loops on out-of-bounds transfer start.
Signed-off-by: Li Qiang <liqiang6-s@360.cn>
Reviewed-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Because is_first is declared inside a loop, it is always true. The store
is dead, and so is the "else" branch of "if (is_first)". is_last is
okay though.
Reported by Coverity.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
The e1000e emulation zeroes out any used rx descriptor and then writes a
completely newly constructed value there. By doing this, it doesn't only
update the write-back area of the descriptors (as it's supposed to do),
but it also clears the buffer address, which real hardware doesn't do.
The spec explicitly mentions in chapter 7.1.8 that it is valid for a
driver to reuse a descriptor and only update the status field while
doing so, i.e. reusing the old buffer address:
If software statically allocates buffers, and uses memory read to
check for completed descriptors, it simply has to zero the status
byte in the descriptor to make it ready for reuse by hardware.
This patch fixes the behaviour to leave the buffer address in
descriptors unchanged even after the descriptor has been used.
Signed-off-by: Kevin Wolf <mail@kevin-wolf.de>
Reviewed-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This patch fixes 2 issues:
1. Bits set in EIAC register should be cleared
from IMS when EIAM is not used.
2. Only bit that corresonds to the interrupt being
raised should be cleared.
See spec. 10.2.4.7 Interrupt Auto Clear
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Do not raise ACK interrupts when
RFCTL.ACKDIS bit is set (see spec. 10.2.5.16).
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Interrupt mask for legacy OTHER causes should
not apply to MSI-X OTHER cause.
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
This patch fixes incorrect check for
interrypt type being used.
PBSCLR register is valid for MSI-X only.
See spec. 10.2.3.13 MSI—X PBA Clear
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
CTRL_EXT.EIAME bit controls clearing of IAM bits,
but current code clears IMS bits instead.
See spec. 10.2.2.5 Extended Device Control Register.
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Before this patch first netdev queue only was flushed.
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
hw/net/e1000e_core.c:56: warning: e1000e_set_interrupt_cause declared inline after being called
hw/net/e1000e_core.c:56: warning: previous declaration of e1000e_set_interrupt_cause was here
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
slave:~/.xie/qemu-colo # gcc --version
gcc (SUSE Linux) 4.3.4 [gcc-4_3-branch revision 152973]
slave:~/.xie/qemu-colo # make -j8
CC hw/net/e1000e_core.o
hw/net/e1000e_core.c:56: warning: ‘e1000e_set_interrupt_cause’ declared inline after being called
hw/net/e1000e_core.c:56: warning: previous declaration of ‘e1000e_set_interrupt_cause’ was here
LINK x86_64-softmmu/qemu-system-x86_64
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Reviewed-by: Dmitry Fleytman <dmitry@daynix.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
This is not dereferencing the pointer, and instead checking only
the value of the pointer.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Don't use *_to_cpup() to do byte-swapped loads; instead use
ld*_p() which correctly handle misaligned accesses.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Dmitry Fleytman <dmitry@daynix.com <mailto:dmitry@daynix.com>>
Message-id: 1466097446-981-6-git-send-email-peter.maydell@linaro.org
ust trace backend has limitation of maximum 10
arguments per event. Traces with more arguments
cannot be compiled for this backend.
Trace e1000e_rx_rss_ip6 introduced by previous
commits has 11 arguments and fails to compile with
ust trace backend.
This patch fixes the problem by splitting this
tracepoint into two successive tracepoints with
smaller number of arguments.
For more information see comment regarding TP_ARGS
in lttng/tracepoint.h:
/*
* TP_ARGS takes tuples of type, argument separated by a comma.
* It can take up to 10 tuples (which means that less than 10 tuples is
* fine too).
* Each tuple is also separated by a comma.
*/
Build log generated by this problem:
In file included from ./trace/generated-tracers.h:9:0,
from /home/travis/build/qemu/qemu/include/trace.h:4,
from util/oslib-posix.c:36:
./trace/generated-ust-provider.h:16556:3: error: unknown type name ‘_TP_EXPROTO_Bool’
In file included from /home/travis/build/qemu/qemu/include/trace.h:4:0,
from util/oslib-posix.c:36:
./trace/generated-tracers.h: In function ‘trace_e1000e_rx_rss_ip6’:
./trace/generated-tracers.h:8379:431: error: expected string literal before ‘_SDT_ASM_OPERANDS_ipv6_enabled’
./trace/generated-tracers.h:8379:431: error: implicit declaration of function ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=implicit-function-declaration]
./trace/generated-tracers.h:8379:431: error: nested extern declaration of ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=nested-externs]
cc1: all warnings being treated as errors
make: *** [util/oslib-posix.o] Error 1
make: *** Waiting for unfinished jobs....
In file included from ./trace/generated-tracers.h:9:0,
from /home/travis/build/qemu/qemu/include/trace.h:4,
from util/hbitmap.c:16:
./trace/generated-ust-provider.h:16556:3: error: unknown type name ‘_TP_EXPROTO_Bool’
In file included from /home/travis/build/qemu/qemu/include/trace.h:4:0,
from util/hbitmap.c:16:
./trace/generated-tracers.h: In function ‘trace_e1000e_rx_rss_ip6’:
./trace/generated-tracers.h:8379:431: error: expected string literal before ‘_SDT_ASM_OPERANDS_ipv6_enabled’
./trace/generated-tracers.h:8379:431: error: implicit declaration of function ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=implicit-function-declaration]
./trace/generated-tracers.h:8379:431: error: nested extern declaration of ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=nested-externs]
cc1: all warnings being treated as errors
make: *** [util/hbitmap.o] Error 1
Signed-off-by: Dmitry Fleytman <dmitry@daynix.com>
Message-id: 1464894748-27803-1-git-send-email-dmitry@daynix.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This patch introduces emulation for the Intel 82574 adapter, AKA e1000e.
This implementation is derived from the e1000 emulation code, and
utilizes the TX/RX packet abstractions that were initially developed for
the vmxnet3 device. Although some parts of the introduced code may be
shared with e1000, the differences are substantial enough so that the
only shared resources for the two devices are the definitions in
hw/net/e1000_regs.h.
Similarly to vmxnet3, the new device uses virtio headers for task
offloads (for backends that support virtio extensions). Usage of
virtio headers may be forcibly disabled via a boolean device property
"vnet" (which is enabled by default). In such case task offloads
will be performed in software, in the same way it is done on
backends that do not support virtio headers.
The device code is split into two parts:
1. hw/net/e1000e.c: QEMU-specific code for a network device;
2. hw/net/e1000e_core.[hc]: Device emulation according to the spec.
The new device name is e1000e.
Intel specifications for the 82574 controller are available at:
http://www.intel.com/content/dam/doc/datasheet/82574l-gbe-controller-datasheet.pdf
Throughput measurement results (iperf2):
Fedora 22 guest, TCP, RX
4 ++------------------------------------------+
| |
| X X X X X
3.5 ++ X X X X |
| X |
| |
3 ++ |
G | X |
b | |
/ 2.5 ++ |
s | |
| |
2 ++ |
| |
| |
1.5 X+ |
| |
+ + + + + + + + + + + +
1 ++--+---+---+---+---+---+---+---+---+---+---+
32 64 128 256 512 1 2 4 8 16 32 64
B B B B B KB KB KB KB KB KB KB
Buffer size
Fedora 22 guest, TCP, TX
18 ++-------------------------------------------+
| X |
16 ++ X X X X X
| X |
14 ++ |
| |
12 ++ |
G | X |
b 10 ++ |
/ | |
s 8 ++ |
| |
6 ++ X |
| |
4 ++ |
| X |
2 ++ X |
X + + + + + + + + + + +
0 ++--+---+---+---+---+----+---+---+---+---+---+
32 64 128 256 512 1 2 4 8 16 32 64
B B B B B KB KB KB KB KB KB KB
Buffer size
Fedora 22 guest, UDP, RX
3 ++------------------------------------------+
| X
| |
2.5 ++ |
| |
| |
2 ++ X |
G | |
b | |
/ 1.5 ++ |
s | X |
| |
1 ++ |
| |
| X |
0.5 ++ |
| X |
X + + + + +
0 ++-------+--------+-------+--------+--------+
32 64 128 256 512 1
B B B B B KB
Datagram size
Fedora 22 guest, UDP, TX
1 ++------------------------------------------+
| X
0.9 ++ |
| |
0.8 ++ |
0.7 ++ |
| |
G 0.6 ++ |
b | |
/ 0.5 ++ |
s | X |
0.4 ++ |
| |
0.3 ++ |
0.2 ++ X |
| |
0.1 ++ X |
X X + + + +
0 ++-------+--------+-------+--------+--------+
32 64 128 256 512 1
B B B B B KB
Datagram size
Windows 2012R2 guest, TCP, RX
3.2 ++------------------------------------------+
| X |
3 ++ |
| |
2.8 ++ |
| |
2.6 ++ X |
G | X X X X X
b 2.4 ++ X X |
/ | |
s 2.2 ++ |
| |
2 ++ |
| X X |
1.8 ++ |
| |
1.6 X+ |
+ + + + + + + + + + + +
1.4 ++--+---+---+---+---+---+---+---+---+---+---+
32 64 128 256 512 1 2 4 8 16 32 64
B B B B B KB KB KB KB KB KB KB
Buffer size
Windows 2012R2 guest, TCP, TX
14 ++-------------------------------------------+
| |
| X X
12 ++ |
| |
10 ++ |
| |
G | |
b 8 ++ |
/ | X |
s 6 ++ |
| |
| |
4 ++ X |
| |
2 ++ |
| X X X |
+ X X + + X X + + + + +
0 X+--+---+---+---+---+----+---+---+---+---+---+
32 64 128 256 512 1 2 4 8 16 32 64
B B B B B KB KB KB KB KB KB KB
Buffer size
Windows 2012R2 guest, UDP, RX
1.6 ++------------------------------------------X
| |
1.4 ++ |
| |
1.2 ++ |
| X |
| |
G 1 ++ |
b | |
/ 0.8 ++ |
s | |
0.6 ++ X |
| |
0.4 ++ |
| X |
| |
0.2 ++ X |
X + + + + +
0 ++-------+--------+-------+--------+--------+
32 64 128 256 512 1
B B B B B KB
Datagram size
Windows 2012R2 guest, UDP, TX
0.6 ++------------------------------------------+
| X
| |
0.5 ++ |
| |
| |
0.4 ++ |
G | |
b | |
/ 0.3 ++ X |
s | |
| |
0.2 ++ |
| |
| X |
0.1 ++ |
| X |
X X + + + +
0 ++-------+--------+-------+--------+--------+
32 64 128 256 512 1
B B B B B KB
Datagram size
Signed-off-by: Dmitry Fleytman <dmitry.fleytman@ravellosystems.com>
Signed-off-by: Leonid Bloch <leonid.bloch@ravellosystems.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Jan Kiszka
Ed Swierk via Qemu-devel
Kamil Rytarowski
Sameeh Jubran
Li Qiang
Paolo Bonzini
Kevin Wolf
Dmitry Fleytman
Gonglei
Changlong Xie
Peter Maydell
Dmitry Fleytman