Shao Miller [Tue, 9 Nov 2010 02:11:32 +0000 (02:11 +0000)]
[list] Fix typographical error from previous commit
Fix typographical error from commit ea631f6 ("[list] Add
list_first_entry()"). The symptom was PXELINUX 3.86 causing a stack
overflow under VMware.
Tested-by: Shao Miller <shao.miller@yrdsb.edu.on.ca> Signed-off-by: Shao Miller <shao.miller@yrdsb.edu.on.ca> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Fri, 5 Nov 2010 23:13:22 +0000 (23:13 +0000)]
[fc] Hold ULP's peer reference while ULP exists
Allow fc_ulp_decrement() to guarantee to fc_peer_decrement() that the
peer reference remains valid for the duration of the call, by ensuring
that ulp->peer remains valid while ulp is valid.
Michael Brown [Mon, 8 Nov 2010 03:06:14 +0000 (03:06 +0000)]
[retry] Process at most one timer's expiry in each call to retry_step()
Calling a timer's expiry method may cause arbitrary consequences,
including arbitrary modifications of the list of retry timers.
list_for_each_entry_safe() guards against only deletion of the current
list entry; it provides no protection against other list
modifications. In particular, if a timer's expiry method causes the
subsequent timer in the list to be deleted, then the next loop
iteration will access a timer that may no longer exist.
This is a particularly nasty bug, since absolutely none of the
list-manipulation or reference-counting assertion checks will be
triggered. (The first assertion failure happens on the next iteration
through list_for_each_entry(), showing that the list has become
corrupted but providing no clue as to when this happened.)
Fix by stopping traversal of the list of retry timers as soon as we
hit an expired timer.
Michael Brown [Fri, 5 Nov 2010 22:37:00 +0000 (22:37 +0000)]
[malloc] Avoid immediately clobbering reference count when freeing memory
Rearrange the fields in struct memory_block (without altering
MIN_MEMBLOCK_SIZE) so that the "count" field of a reference-counted
object is left intact when the memory containing the object is freed.
This allows for the possibility of detecting reference-counting errors
such as double-freeing.
Michael Brown [Sat, 6 Nov 2010 00:31:02 +0000 (00:31 +0000)]
[malloc] Use list_for_each_entry_safe() when we may delete a list entry
free_memblock() currently uses list_for_each_entry() to iterate over
the free list, and may delete an entry over which it iterates. While
there is no way that the deleted list entry could be overwritten
before we reference it, this does rely upon list_del() leaving the
"next" pointer intact, which is not guaranteed. Discovered while
tracking down a list-corruption bug (as a result of having modified
list_del() to sanitise the deleted list entry).
Michael Brown [Mon, 8 Nov 2010 02:51:18 +0000 (02:51 +0000)]
[list] Add list_first_entry()
There are several points in the iPXE codebase where
list_for_each_entry() is (ab)used to extract only the first entry from
a list. Add a macro list_first_entry() to make this code easier to
read.
Michael Brown [Wed, 3 Nov 2010 01:48:38 +0000 (01:48 +0000)]
[fcp] Fix potential memory leaks on error paths
Functions that instantiate objects generally own one reference to the
object being created. The error paths must therefore usually call
ref_put() to release this reference.
Dave Hansen [Thu, 21 Oct 2010 22:03:42 +0000 (15:03 -0700)]
[autoboot] Introduce "skip-san-boot" option
For some install-to-SAN scenarios, the OS needs to be able to reboot
to reread the partition table. On this second boot attempt, the SAN
disk will not be empty and so iPXE will attempt to boot from it,
rather than falling back to the OS' installation media.
Work around this problem by introducing the "skip-san-boot" option,
similar in spirit to "keep-san".
Modified-by: Michael Brown <mcb30@ipxe.org> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Sat, 16 Oct 2010 12:53:38 +0000 (13:53 +0100)]
[scsi] Wait for a successful TEST UNIT READY command
Some SCSI targets (observed with an EMC CLARiiON Fibre Channel target)
will not respond to commands correctly until a TEST UNIT READY has
been issued. In particular, a READ CAPACITY (10) command will return
with a success status, but no capacity data.
Fix by issuing a TEST UNIT READY command automatically, and delaying
further SCSI commands until the TEST UNIT READY has succeeded.
Reported-by: Hadar Hen Zion <hadarh@mellanox.co.il> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Tue, 19 Oct 2010 17:35:46 +0000 (18:35 +0100)]
[fc] Do not use the command reference number in FCP_CMND IUs
The FCP command reference number is intended to be used for
controlling precise delivery of FCP commands, rather than being an
essentially arbitrary tag field (as with iSCSI and SRP).
Use the Fibre Channel local exchange ID as the tag for FCP commands,
instead of the FCP command reference. The local exchange ID does not
appear within the FCP IU itself, but does appear within the FC frame
header; debug traces can therefore still be correlated with packet
captures.
Reported-by: Hadar Hen Zion <hadarh@mellanox.co.il> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Tue, 19 Oct 2010 05:14:22 +0000 (06:14 +0100)]
[hci] Use http://ipxe.org/<errno> instead of raw error numbers
Users tend to gloss over cryptic-looking error messages such as
"Boot failed: Exec format error (Error 0x2e852001)"
In particular, users tend not to report the error number, which is the
single most useful piece of diagnostic information in an iPXE error
message. Try replacing the "Error 0x2e852001" portion with a URL,
giving
"Boot failed: Exec format error (http://ipxe.org/2e852001)"
in the hope that users will, upon seeing something that is
recognisably a URL, try viewing it in a web browser. Such users will
be greeted by a web page containing a more detailed description of the
error (automatically generated from the einfo text), including links
to each line of code that might generate the error, and a section for
additional user-contributed notes. At the time of writing, a user who
visits http://ipxe.org/2e852001 would see a note saying
"This error usually indicates that the SAN disk is empty, and does
not yet contain a bootable operating system."
which may be more useful than "Exec format error (Error 0x2e852001)".
Michael Brown [Mon, 18 Oct 2010 13:37:11 +0000 (14:37 +0100)]
[iscsi] Ensure ISID is consistent within an iSCSI session
Commit 5f4ab0d ("[iscsi] Randomise a portion of the ISID to force new
session instantiation") introduced a regression by randomising the
ISID on each call to iscsi_start_login(), which may be called more
than once per connection, rather than on each call to
iscsi_open_connection(), which is guaranteed to be called only once
per connection. This is incorrect behaviour that causes our
connection to be rejected by some iSCSI targets (observed with a
COMSTAR target under OpenSolaris).
Fix by generating the ISID in iscsi_open_connection(), and storing the
randomised ISID as part of the session state.
Michael Brown [Sat, 16 Oct 2010 21:11:08 +0000 (22:11 +0100)]
[iscsi] Randomise a portion of the ISID to force new session instantiation
When a connection to an iSCSI target is broken without gracefully
closing the TCP socket, a subsequent connection attempt may fail
because the target believes that we are attempting session
reinstatement (see RFC3720 section 5.3.1). This has been observed
using the Microsoft iSCSI target.
Section 9.1.1 of RFC3720 states that initiators should use a stable
ISID, however section 5.3.1 shows that the only way to explicitly
request that a new session be created is to use a new ISID.
Fix by randomising the "qualifier" portion of the ISID.
Michael Brown [Thu, 14 Oct 2010 22:56:28 +0000 (23:56 +0100)]
[fcoe] Request SPMA iff FIP advertisement indicates support for SPMA
We currently set both the FP and SP bits in our FIP FLOGI, to allow
the FCF the choice of selecting either a fabric-provided or a server-
provided MAC address. This complies with the FCoE specification, but
has been observed to result in an FLOGI rejection from some FCFs.
Fix by recording whether or not the FCF supports SPMA, and requesting
only one of FPMA or SPMA in our FIP FLOGI. We choose to prefer SPMA
where available, because many iPXE drivers will not be able to receive
unicast packets sent to a non-default MAC address.
Reported-by: Hadar Hen Zion <hadarh@mellanox.co.il> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Sat, 9 Oct 2010 00:24:18 +0000 (01:24 +0100)]
[dhcp] Include session state metadata in packet traces
(Ab)use the "secs" field in transmitted DHCP packets to convey
metadata about the DHCP session state. In particular:
bit 0 represents the receipt of a ProxyDHCPOFFER
bit 1 represents the receipt of a DHCPOFFER
bits 2+ represent the transmitted packet sequence number
This allows some relevant information about the internal state of the
DHCP session to be read out from a packet trace from a non-debug build
of iPXE. It also potentially allows replies to be correlated to their
requests (for servers that copy the "secs" field from request to
reply).
Michael Brown [Fri, 8 Oct 2010 00:29:39 +0000 (01:29 +0100)]
[dhcp] Omit ProxyDHCPREQUEST if PXE options are present in ProxyDHCPOFFER
Some ProxyDHCP implementations seem to violate the PXE specification
by expecting the client to retain options from the ProxyDHCPOFFER
rather than issuing a separate ProxyDHCPREQUEST.
Work around such broken clients by retaining the ProxyDHCPOFFER
packet, and proceeding to a ProxyDHCPREQUEST only if the
ProxyDHCPOFFER does not already contain PXE options.
Michael Brown [Thu, 7 Oct 2010 15:19:00 +0000 (16:19 +0100)]
[fc] Update ELS port IDs when receiving an ELS frame
The port ID assigned by the FLOGI response is implicit in the
destination ID used for the response (which will differ from the
source ID used for the corresponding request).
Michael Brown [Thu, 7 Oct 2010 15:03:16 +0000 (16:03 +0100)]
[netdevice] Pass both link-layer addresses in net_tx() and net_rx()
FCoE requires the use of fabric-provided MAC addresses, which breaks
the assumption that the net device's MAC address is implicitly the
source address for net_tx() and the (unicast) destination address for
net_rx().
Michael Brown [Wed, 22 Sep 2010 18:59:37 +0000 (19:59 +0100)]
[libc] Ensure that error numbers from EUNIQ() have the correct type
Error numbers are signed ints. EUNIQ() should not allow implicit type
promotion based on the supplied error diambiguator, because this
causes problems with statements such as
Michael Brown [Tue, 21 Sep 2010 23:12:23 +0000 (00:12 +0100)]
[int13] Add infrastructure to support EDD version 4.0
Support the extensions mandated by EDD 4.0, including:
o the ability to specify a flat physical address in a disk address
packet,
o the ability to specify a sector count greater than 127 in a disk
address packet,
o support for all functions within the Fixed Disk Access and EDD
Support subsets,
o the ability to describe a device using EDD Device Path Information.
This implementation is based on draft revision 3 of the EDD 4.0
specification, with reference to the EDD 3.0 specification. It is
possible that this implementation may need to change in order to
conform to the final published EDD 4.0 specification.
Michael Brown [Sat, 18 Sep 2010 21:37:54 +0000 (22:37 +0100)]
[hermon] Minimise the number of VPM mapping operations
Mapping a single page at a time causes a several-second delay at
device initialisation time. Reduce this by mapping multiple pages at
a time, using the largest block sizes possible given the alignment
constraints.
Michael Brown [Sat, 18 Sep 2010 21:37:54 +0000 (22:37 +0100)]
[arbel] Minimise the number of VPM mapping operations
Mapping a single page at a time causes a several-second delay at
device initialisation time. Reduce this by mapping multiple pages at
a time, using the largest block sizes possible given the alignment
constraints.
Michael Brown [Fri, 17 Sep 2010 23:57:09 +0000 (00:57 +0100)]
[arbel] Allocate only as much memory as is needed for firmware and ICM
Use individual page mappings rather than a single whole-region
mapping, to avoid the waste of memory that occurs due to the
constraint that each mapped block must be aligned on its own size.
Michael Brown [Fri, 17 Sep 2010 03:56:23 +0000 (04:56 +0100)]
[arbel] Perform ICM allocations according to the specification
The current method for ICM allocation exactly matches the addresses
chosen by the old Etherboot driver, but does not match the
specification. Some ICM tables (notably the queue pair context table)
therefore end up incorrectly aligned.
Fix by performing allocations as per the specification.
Michael Brown [Fri, 17 Sep 2010 02:30:28 +0000 (03:30 +0100)]
[arbel] Improve debugging output and facilities
Improve the utility of debugging messages by including the relevant
port number, queue number (QPN, CQN, EQN), work queue entry (WQE)
number, and physical addresses wherever applicable.
Add arbel_dump_cqctx() for dumping a completion queue context and
arbel_dump_qpctx() for dumping a queue pair context.
Michael Brown [Thu, 16 Sep 2010 21:06:20 +0000 (22:06 +0100)]
[arbel] Map all event types to our event queue
Only port state change events are currently mapped to our event queue,
since those are the only events we are prepared to handle. This
ignores a potentially useful source of diagnostic information in the
case of unexpected failures.
Fix by mapping all events to the event queue; a build with debugging
enabled will therefore at least dump the raw content of the unexpected
events.
Michael Brown [Thu, 16 Sep 2010 21:06:20 +0000 (22:06 +0100)]
[hermon] Map all event types to our event queue
Only port state change events are currently mapped to our event queue,
since those are the only events we are prepared to handle. This
ignores a potentially useful source of diagnostic information in the
case of unexpected failures.
Fix by mapping all events to the event queue; a build with debugging
enabled will therefore at least dump the raw content of the unexpected
events.
Michael Brown [Wed, 15 Sep 2010 22:10:15 +0000 (23:10 +0100)]
[infiniband] Add node GUID as distinct from the first port GUID
iPXE currently uses the first port's port GUID as the node GUID,
rather than using the (possibly distinct) real node GUID. This can
confuse opensm during the handover to a loaded OS: it thinks the port
already belongs to a different node and so discards our port
information with a warning message about duplicate ports. Everything
is picked up correctly on the second subnet sweep, after opensm has
established that the "old" node no longer exists, but this can delay
link-up unnecessarily by several seconds.
Michael Brown [Thu, 16 Sep 2010 02:23:45 +0000 (03:23 +0100)]
[infiniband] Always call ib_link_state_changed() in ib_smc_update()
ib_smc_update() potentially updates the Infiniband port state, and so
should almost always be followed by a call to ib_link_state_changed().
The one exception is the call made to ib_smc_update() before the
device is registered.
Fix by removing explicit calls to ib_link_state_changed() from drivers
using ib_smc_update(), including a call to ib_link_state_changed()
within ib_smc_update(), and creating a separate ib_smc_init() for use
prior to device registration.
Michael Brown [Wed, 15 Sep 2010 04:00:44 +0000 (05:00 +0100)]
[fcoe] Disambiguate the various error cases and add a CRC failure message
It seems as though several drivers neglect to strip the Ethernet CRC,
which will cause the FCoE footer to be misplaced and result
(coincidentally) in an "invalid CRC" error from FCoE.
Add a human-visible message indicating this, to aid in diagnosis.
Michael Brown [Wed, 15 Sep 2010 03:56:28 +0000 (04:56 +0100)]
[netdevice] Report network-layer errors via network device statistics
Errors generated by the network layer in response to received packets
are liable to be lost, since nothing systematically records these
errors and often the packets do not propagate far enough through the
stack to impact upon user-visible processes.
Improve this situation by recording network-layer errors in the
network device statistics.
Michael Brown [Fri, 3 Sep 2010 15:11:51 +0000 (16:11 +0100)]
[block] Replace gPXE block-device API with an iPXE asynchronous interface
The block device interface used in gPXE predates the invention of even
the old gPXE data-transfer interface, let alone the current iPXE
generic asynchronous interface mechanism. Bring this old code up to
date, with the following benefits:
o Block device commands can be cancelled by the requestor. The INT 13
layer uses this to provide a global timeout on all INT 13 calls,
with the result that an unexpected passive failure mode (such as
an iSCSI target ACKing the request but never sending a response)
will lead to a timeout that gets reported back to the INT 13 user,
rather than simply freezing the system.
o INT 13,00 (reset drive) is now able to reset the underlying block
device. INT 13 users, such as DOS, that use INT 13,00 as a method
for error recovery now have a chance of recovering.
o All block device commands are tagged, with a numerical tag that
will show up in debugging output and in packet captures; this will
allow easier interpretation of bug reports that include both
sources of information.
o The extremely ugly hacks used to generate the boot firmware tables
have been eradicated and replaced with a generic acpi_describe()
method (exploiting the ability of iPXE interfaces to pass through
methods to an underlying interface). The ACPI tables are now
built in a shared data block within .bss16, rather than each
requiring dedicated space in .data16.
o The architecture-independent concept of a SAN device has been
exposed to the iPXE core through the sanboot API, which provides
calls to hook, unhook, boot, and describe SAN devices. This
allows for much more flexible usage patterns (such as hooking an
empty SAN device and then running an OS installer via TFTP).
Michael Brown [Sun, 12 Sep 2010 18:11:59 +0000 (19:11 +0100)]
[hermon] Use correct alignment for doorbell records
Doorbell records are currently embedded within the completion queue
and receive work queue strucures, which are allocated using zalloc()
and so have an alignment guarantee of only sizeof(void*), i.e. four
bytes. This is sufficient for the receive work queue, but not for the
completion queue, which requires an alignment guarantee of eight
bytes.
Though not guaranteed, it so happens that zalloc() will always return
a pointer that is exactly four bytes above a sixteen-byte boundary.
The completion queue doorbell record is therefore always misaligned,
and the value passed to the hardware via SW2HW_CQ is actually always
pointing to the page_offset value within the MTT descriptor (which
directly precedes the inline doorbell record). Provided that the page
offset is greater than 0x100, this looks to the hardware like an
update_ci value of greater than 0x010000 (taking into account
endianness differences), and so the hardware will happily deliver more
than 0x010000 completions before stopping. Hence this problem is
rarely observable.
Fix by allocating the doorbell records separately and using the
correct alignment constraints.
Michael Brown [Sun, 12 Sep 2010 16:35:35 +0000 (17:35 +0100)]
[hermon] Set event queue number for completion queues
Give completion queues a chance to deliver exception events by
programming in the number of our event queue (currently used only for
port state changes).
projects / thirdparty / ipxe.git / log
