Multicast MAC addresses will never have REMAC cache entries, and the
corresponding multicast IPoIB MAC address cannot be obtained simply by
issuing an ARP request.
For the trivial volume of multicast packets that we expect to send in
any realistic scenario, the simplest solution is to send them as
broadcasts instead.
Reported-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[tcp] Gracefully close connections during shutdown
We currently do not wait for a received FIN before exiting to boot a
loaded OS. In the common case of booting from an HTTP server, this
means that the TCP connection is left consuming resources on the
server side: the server will retransmit the FIN several times before
giving up.
Fix by initiating a graceful close of all TCP connections and waiting
(for up to one second) for all connections to finish closing
gracefully (i.e. for the outgoing FIN to have been sent and ACKed, and
for the incoming FIN to have been received and ACKed at least once).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[xen] Wait for and clear XenStore event before receiving data
Older, out-of-tree Xen kernel modules (such as those provided with
SuSE Linux Enterprise Server 11) do not clear the leftover "event
pending" bit when opening an event channel. Consequently, no event is
ever delivered to indicate that there is information in the XenStore
ring buffer, and the system hangs shortly after loading the
xen-platform-pci kernel module.
Work around this problem by always waiting for the XenStore event
channel to be signalled, and clearing the event before processing the
received data.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[ipoib] Attempt to generate ARPs as needed to repopulate REMAC cache
The only way to map an eIPoIB MAC address (REMAC) to an IPoIB MAC
address is to intercept an incoming ARP request or reply.
If we do not have an REMAC cache entry for a particular destination
MAC address, then we cannot transmit the packet. This can arise in at
least two situations:
- An external program (e.g. a PXE NBP using the UNDI API) may attempt
to transmit to a destination MAC address that has been obtained by
some method other than ARP.
- Memory pressure may have caused REMAC cache entries to be
discarded. This is fairly likely on a busy network, since REMAC
cache entries are created for all received (broadcast) ARP
requests. (We can't sensibly avoid creating these cache entries,
since they are required in order to send an ARP reply, and when we
are being used via the UNDI API we may have no knowledge of which
IP addresses are "ours".)
Attempt to ameliorate the situation by generating a semi-spurious ARP
request whenever we find a missing REMAC cache entry. This will
hopefully trigger an ARP reply, which would then provide us with the
information required to populate the REMAC cache.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
As with the neighbour cache, discarding an REMAC cache entry is
potentially very disruptive.
Originally-fixed-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[pxe] Always reconstruct packet for PXENV_GET_CACHED_INFO
Avoid accidentally returning stale packets (e.g. for a previously
attempted network device) by always constructing a fresh DHCP packet.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
If the link is blocked (e.g. due to a Spanning Tree Protocol port not
yet forwarding packets) then defer DHCP discovery until the link
becomes unblocked.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[stp] Add support for detecting Spanning Tree Protocol non-forwarding ports
A fairly common end-user problem is that the default configuration of
a switch may leave the port in a non-forwarding state for a
substantial length of time (tens of seconds) after link up. This can
cause iPXE to time out and give up attempting to boot.
We cannot force the switch to start forwarding packets sooner, since
any attempt to send a Spanning Tree Protocol bridge PDU may cause the
switch to disable our port (if the switch happens to have the Bridge
PDU Guard feature enabled for the port).
For non-ancient versions of the Spanning Tree Protocol, we can detect
whether or not the port is currently forwarding and use this to inform
the network device core that the link is currently blocked.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[netdevice] Add a generic concept of a "blocked link"
When Spanning Tree Protocol (STP) is used, there may be a substantial
delay (tens of seconds) from the time that the link goes up to the
time that the port starts forwarding packets.
Add a generic concept of a "blocked link" (i.e. a link which is up but
which is not expected to communicate successfully), and allow "ifstat"
to indicate when a link is blocked.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[ethernet] Add minimal support for receiving LLC frames
In some Ethernet framing variants the two-byte protocol field is used
as a length, with the Ethernet header being followed by an IEEE 802.2
LLC header. The first two bytes of the LLC header are the DSAP and
SSAP.
If the received Ethernet packet appears to use this framing, then
interpret the two-byte DSAP and SSAP as being the network-layer
protocol. This allows support for receiving Spanning Tree Protocol
frames (which use an LLC header with {DSAP,SSAP}=0x4242) to be added
without requiring a full LLC protocol layer.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[mromprefix] Report a dummy size at offset 0x02 of .mrom payload
The size of the .mrom payload (the second PCI ROM image) is defined in
its PCI header. The code type for the .mrom payload image is
deliberately set to an invalid value (0xff) to ensure that no BIOS
tries to parse anything in the image other than the PCI header.
Since the code type is not set to 0x00 ("Intel x86, PC-AT
compatible"), bytes 0x02-0x17 should not be interpreted by the BIOS as
being in the standard ISA expansion ROM format. In particular, the
byte at offset 0x02 does not represent the length of the ROM image (in
512-byte blocks).
However, some Dell BIOSes seem to erroneously use the byte at offset
0x02 to determine the length of the .mrom payload when walking the
list of PCI ROM images. Since this byte is currently set to zero,
this can lead to the BIOS getting stuck in an infinite loop during
POST. (This problem may not arise if the .mrom payload is the final
image in the ROM, since the BIOS will then have no reason to attempt
to locate the next image.)
One possible workaround would be to put the real payload size in this
byte, but doing so would constrain the .mrom payload size to 128kB
(see commit 8049a52 ("[mromprefix] Allow for .mrom images larger than
128kB") for more details).
Another possible workaround would be to put the real payload size as a
word in bytes 0x02-0x03 (as is done for EFI ROMs). This would not
constrain the .mrom payload size, but a payload size which happened to
be exactly 128kB would result in a zero value in the byte at offset
0x02 and so could still result in infinite loops on BIOSes with this
bug.
We choose to place a fixed value of 0x01 in the byte at offset 0x02.
This should at least prevent the BIOS from getting stuck in an
infinite loop. (The BIOS may walk into the middle of the .mrom
payload, where it will almost certainly not find a valid {0x55,0xaa}
signature or a valid PCIR header, and will therefore hopefully abort
processing.)
Reported-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[tcp] Do not shrink window when discarding received packets
We currently shrink the TCP window permanently if we are ever forced
(by a low-memory condition) to discard a previously received TCP
packet. This behaviour was intended to reduce the number of
retransmissions in a lossy network, since lost packets might
potentially result in the entire window contents being retransmitted.
Since commit e0fc8fe ("[tcp] Implement support for TCP Selective
Acknowledgements (SACK)") the cost of lost packets has been reduced by
around one order of magnitude, and the reduction in the window size
(which affects the maximum throughput) is now the more significant
cost.
Remove the code which reduces the TCP maximum window size when a
received packet is discarded.
Reported-by: Wissam Shoukair <wissams@mellanox.com>
Tested-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Some HP BIOSes (observed with an HP ProLiant m710p Server Cartridge)
have a bug in the implementation of INT 1a,b101: they blithely assume
that real-mode code is able to read from anywhere in the 32-bit memory
space.
This problem affects the call to INT 1a,b101 made from within
pcibios_num_bus() (which uses REAL_CODE() and hence executes in
genuine real mode) but does not affect the call made from within
romprefix.S (since with a PMM BIOS, that call executes in flat real
mode anyway).
Work around the problem by explicitly calling flatten_real_mode()
before invoking INT 1a,b101. This is a rarely-used code path, and so
the extra overhead of emulating instructions in some VM configurations
(see commit 6d4deee ("[librm] Use genuine real mode to accelerate
operation in virtual machines") for more details) is negligible.
Reported-by: Wissam Shoukair <wissams@mellanox.com>
Debugged-by: Wissam Shoukair <wissams@mellanox.com>
Debugged-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[xhci] Ignore invalid protocol speed ID values on Intel Skylake platforms
Some Intel Skylake platforms (observed on a prototype Lenovo ThinkPad)
report the list of available USB3 protocol speed ID values as {1,2,3}
but then report a port's speed using ID value 4.
The value 4 happens to be the default value for SuperSpeed (when no
protocol speed ID value list is explicitly defined), and the hardware
seems to function correctly if we simply ignore its protocol speed ID
table and assume that it uses the default values.
Fix by adding a "broken PSI values" quirk for this controller.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[xhci] Fix comparison of signed and unsigned integers
gcc 4.8.2 fails to report this erroneous comparison unless assertions
are enabled.
Reported-by: Mary-Ann Johnson <MaryAnn.Johnson@displaylink.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[build] Fix .ids.o creation for drivers not in the all-drivers build
Commit dc19e63 ("[build] Construct all-drivers list based on driver
class") accidentally excluded the USB bus drivers from the list of
files parsed in order to create PCI 3.0 device ID lists.
Fix by returning $(DRIVERS) to its previous definition as a list of
all driver files, and use only $(DRIVERS_ipxe) to contain the
filtered list containing only those drivers which we want to include
in the "all-drivers" build.
Reported-by: Mary-Ann Johnson <MaryAnn.Johnson@displaylink.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The xHCI slot ID is one-based, not zero-based. Fix the length of the
xhci->slot[] array to account for this, and add assertions to check
that the hardware returns a valid slot ID in response to the Enable
Slot command.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[neighbour] Return success when deferring a packet
Deferral of a packet for neighbour discovery is not really an error.
If we fail to discover a neighbour then the failure will eventually be
reported by the call to neighbour_destroy() when any outstanding I/O
buffers are discarded.
The current behaviour breaks PXE booting on FreeBSD, which seems to
treat the error return from PXENV_UDP_WRITE as a fatal error and so
never proceeds to poll PXENV_UDP_READ (and hence never allows iPXE to
receive the ARP reply and send the deferred UDP packet).
Change neighbour_tx() to return success when deferring a packet. This
fixes interoperability with FreeBSD and removes transient neighbour
cache misses from the "ifstat" error output, while leaving genuine
neighbour discovery failures visible via "ifstat" (once neighbour
discovery times out, or the interface is closed).
Debugged-by: Wissam Shoukair <wissams@mellanox.com>
Tested-by: Wissam Shoukair <wissams@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[intel] Fix operation when physical function has jumbo frames enabled
When jumbo frames are enabled, the Linux ixgbe physical function
driver will disable the virtual function's receive datapath by
default, and will enable it only if the virtual function negotiates
API version 1.1 (or higher) and explicitly selects an MTU.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[int13con] Add basic ability to log to a local disk via INT 13
Several popular public cloud providers do not provide any sensible
mechanism for obtaining debug output from an OS which is failing to
boot. For example, Amazon EC2 provides the "Get System Log" facility,
which occasionally deigns to report a random subset of the characters
emitted via the VM's serial port, but usually returns only a blank
screen. (Amazingly, this is still superior to the debugging
facilities provided by Azure.)
Work around these shortcomings by adding a console type which sends
output to a magically detected raw disk partition, and including such
a partition within any iPXE .usb-format image.
To use this facility:
- build an iPXE .usb image with CONSOLE_INT13 enabled
- boot the cloud VM from this image
- after the boot fails, attach the VM's boot disk to a second VM
- from this second VM, use "less -f -R /dev/sdb3" (or similar) to
view the iPXE output.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[intel] Add support for mailbox used by virtual functions
Virtual functions use a mailbox to communicate with the physical
function driver: this covers functionality such as obtaining the MAC
address.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[intel] Allow for the use of advanced TX descriptors
Intel virtual function NICs almost work with the use of "legacy"
transmit and receive descriptors (which are backwards compatible right
back to the original Intel Gigabit NICs).
Unfortunately the "TX switching" feature (which allows for VM<->VM
traffic to be looped back within the NIC itself) does not work when a
legacy TX descriptor is used: the packet is instead sent onto the
wire.
Fix by allowing for the use of an "advanced" TX descriptor (containing
exactly the same information as is found in the "legacy" descriptor).
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[usb] Always clear recorded disconnections after performing hotplug actions
The recorded disconnections (in port->disconnected) will currently be
left uncleared if usb_attached() returns an error (e.g. because there
are no drivers for a particular USB device). This is incorrect
behaviour: the disconnection has been handled and the record should be
cleared until the next physical disconnection is detected (via the CSC
bit).
The problem is masked for EHCI, UHCI, and USB hubs, since these will
report a changed port (via usb_port_changed()) only when the
underlying hardware reports a change. xHCI will call
usb_port_changed() in response to any port status event, at which
point the stale value of port->disconnected will be erroneously acted
upon. This can lead to an endless loop of repeatedly enumerating the
same device when a driverless device is attached to an xHCI root hub
port.
Fix by unconditionally clearing port->disconnected in usb_hotplugged().
Reported-by: Robin Smidsrød <robin@smidsrod.no>
Tested-by: Robin Smidsrød <robin@smidsrod.no>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[usb] Do not call usb_hotplug() when registering a new hub
The action of registering a new hub can itself happen in only two
ways: either a new USB hub has been created (in which case we are
already inside a call to usb_hotplug()), or a new root hub has been
created.
In the former case, we do not need to issue a further call to
usb_hotplug(), since the hub's ports will all be marked as changed and
so will be handled after the return from register_usb_hub() anyway.
Calling usb_hotplug() within register_usb_hub() leads to a confusing
order of events, such as:
- root hub port 1 detects a change
- root hub port 2 detects a change
- usb_hotplug() is called
- root hub port 1 finds a USB hub
- usb_hotplug() is called
- this inner call to usb_hotplug() handles root hub port 2
Fix by calling usb_hotplug() only from usb_step() and from
register_usb_bus(). This avoids recursive calls to usb_hotplug() and
ensures that devices are enumerated in the order of detection.
Tested-by: Robin Smidsrød <robin@smidsrod.no>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
When USB network card drivers are used, the BIOS' legacy USB
capability is necessarily disabled since there is no way to share the
host controller between the BIOS and iPXE. This currently results in
USB keyboards becoming non-functional in USB-enabled builds of iPXE.
Fix by adding basic support for USB keyboards, enabled by default in
iPXE builds which include USB support.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
[usb] Maintain single lists of halted endpoints and changed ports
When an EHCI hotplug action results in the controller disowning the
port, it will result in a hotplug action on the corresponding UHCI or
OHCI controller. Allow such hotplug actions to be carried out as part
of the same call to usb_step() or usb_register_bus(), by maintaining a
single central list of changed ports.
Signed-off-by: Michael Brown <mcb30@ipxe.org>
The USB core will currently fail to detect disconnections if a new
device has attached by the time the port is examined in
usb_hotplug().
Fix by recording the fact that a disconnection has taken place
whenever the "connection status changed" (CSC) bit is observed to be
set. (Whether the change represents a disconnection or a
reconnection, it indicates that the port has experienced some time of
being disconnected.)
Note that the time at which a disconnection can be detected varies by
hub type. In particular: root hubs can observe the CSC bit when
polling, and so will record the disconnection before calling
usb_port_changed(), but USB hubs read the port status (and hence the
CSC bit) only during the call to hub_speed(), long after the call to
usb_port_changed().
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown