Michael Brown [Mon, 11 Mar 2013 00:50:52 +0000 (00:50 +0000)]
[pcbios] Inhibit all calls to INT 15,e820 and INT 15,e801 during POST
Many BIOSes do not construct the full system memory map until after
calling the option ROM initialisation entry points. For several
years, we have added sanity checks and workarounds to accommodate
charming quirks such as BIOSes which report the entire 32-bit address
space (including all memory-mapped PCI BARs) as being usable RAM.
The IBM x3650 takes quirky behaviour to a new extreme. Calling either
INT 15,e820 or INT 15,e801 during POST doesn't just get you invalid
data. We could cope with invalid data. Instead, these nominally
read-only API calls manage to trash some internal BIOS state, with the
result that the system memory map is _never_ constructed. This tends
to confuse subsequent bootloaders and operating systems.
[ GRUB 0.97 fails in a particularly amusing way. Someone thought it
would be a good idea for memcpy() to check that the destination memory
region is a valid part of the system memory map; if not, then memcpy()
will sulk, fail, and return NULL. This breaks pretty much every use
of memcpy() including, for example, those inserted implicitly by gcc
to copy non-const initialisers. Debugging is _fun_ when a simple call
to printf() manages to create an infinite recursion, exhaust the
available stack space, and shut down the CPU. ]
Fix by completely inhibiting calls to INT 15,e820 and INT 15,e801
during POST.
We do now allow relocation during POST up to the maximum address
returned by INT 15,88 (which seems so far to always be safe). This
allows us to continue to have a reasonable size of external heap, even
if the PMM allocation is close to the 1MB mark.
The downside of allowing relocation during POST is that we may
overwrite PMM-allocated memory in use by other option ROMs. However,
the downside of inhibiting relocation, when combined with also
inhibiting calls to INT 15,e820 and INT 15,e801, would be that we
might have no external heap available: this would make booting an OS
impossible and could prevent some devices from even completing
initialisation.
On balance, the lesser evil is probably to allow relocation during
POST (up to the limit provided by INT 15,88). Entering iPXE during
POST is a rare operation; on the even rarer systems where doing so
happens to overwrite a PMM-allocated region, then there exists a
fairly simple workaround: if the user enters iPXE during POST and
wishes to exit iPXE, then the user must reboot. This is an acceptable
cost, given the rarity of the situation and the simplicity of the
workaround.
Michael Brown [Fri, 8 Mar 2013 13:26:22 +0000 (13:26 +0000)]
[prefix] Use %cs as implicit parameter to uninstall()
romprefix.S currently calls uninstall() with an invalid value in %ax.
Consequently, base memory is not freed after a ROM boot attempt (or
after entering iPXE during POST).
The uninstall() function is physically present in .text16, and so can
use %cs to determine the .text16 segment address. The .data16 segment
address is not required, since uninstall() is called only by code
paths which set up .data16 to immediately follow .text16.
Michael Brown [Wed, 6 Mar 2013 17:35:30 +0000 (17:35 +0000)]
[tftp] Allow TFTP block size to be controlled via the PXE TFTP API
The PXE TFTP API allows the caller to request a particular TFTP block
size. Since mid-2008, iPXE has appended a "?blksize=xxx" parameter to
the TFTP URI constructed internally; nothing has ever parsed this
parameter. Nobody seems to have cared that this parameter has been
ignored for almost five years.
Fix by using xfer_window(), which provides a fairly natural way to
convey the block size information from the PXE TFTP API to the TFTP
protocol layer.
Michael Brown [Wed, 6 Mar 2013 15:05:30 +0000 (15:05 +0000)]
[menu] Prevent separators with shortcut keys from being selected
Nothing currently prevents a menu separator from being assigned a
shortcut key, and then from being selected using that shortcut key.
This produces an inconsistency in the user interface, since separators
cannot be selected by other means of menu navigation (arrow keys, page
up/down, etc).
It would be trivial to prevent separators from being assigned shortcut
keys, but this would eliminate one potentially useful use case: having
a large menu and using shortcut keys to jump to a section within the
menu.
Fix by treating a shortcut key on a separator as equivalent to "select
the separator, then press the down arrow key". This has the effect of
moving to the first non-separator menu item following the specified
separator, which is probably the most intuitive behaviour. (The
existing logic for moving the selection already handles the various
nasty corner cases such as a menu ending with one or more separators.)
Reported-by: Ján ONDREJ (SAL) <ondrejj@salstar.sk> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Wed, 6 Mar 2013 14:41:46 +0000 (14:41 +0000)]
[menu] Prevent character code zero from acting as a shortcut key
Unrecognised keys may be returned by getkey() as character code zero,
which currently matches against the first menu item with no shortcut
key defined.
Prevent this unintended behaviour by explicitly checking that the menu
item has a defined shortcut key.
Reported-by: Ján ONDREJ (SAL) <ondrejj@salstar.sk> Signed-off-by: Michael Brown <mcb30@ipxe.org>
[zbin] Fix size used for memset in alloc_output_file
The output->buf field is a pointer, not an array, so sizeof() is not
applicable. We must use the allocated string length instead.
Identified by gcc:
util/zbin.c: In function ‘alloc_output_file’:
util/zbin.c:146:37: warning: argument to ‘sizeof’ in ‘memset’ call
is the same expression as the destination; did you mean to
dereference it? [-Wsizeof-pointer-memaccess]
memset ( output->buf, 0xff, sizeof ( output->buf ) );
Signed-off-by: Daniel P. Berrange <berrange@redhat.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Tue, 5 Mar 2013 14:22:53 +0000 (14:22 +0000)]
[undi] Work around specific devices with known broken interrupt behaviour
Some PXE stacks are known to claim that IRQs are supported, but then
never generate interrupts. No satisfactory solution has been found to
this problem; the workaround is to add the PCI vendor and device IDs
to a list of devices which will be treated as simply not supporting
interrupts.
This is something of a hack, since it will generate false positives
for identical devices with a working PXE stack (e.g. those that have
been reflashed with iPXE), but it's an improvement on the current
situation.
Reported-by: Richard Moore <rich@richud.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Fri, 1 Mar 2013 16:32:29 +0000 (16:32 +0000)]
[netdevice] Separate VLAN support from presence of VLAN-supporting drivers
Some NICs (e.g. Hermon) provide hardware support for stripping the
VLAN tag, but do not provide any way for this support to be disabled.
Drivers for this hardware must therefore call vlan_find() to identify
a suitable receiving network device.
Provide a weak version of vlan_find() which will always return NULL if
VLAN support has not been enabled (either directly, or by enabling
a feature such as FCoE which requires VLAN support). This allows the
VLAN code to be omitted from builds where the user has not requested
support for VLANs.
Inspired-by: Stefan Hajnoczi <stefanha@redhat.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Stefan Hajnoczi [Fri, 1 Mar 2013 15:32:16 +0000 (16:32 +0100)]
[iscsi] Include 802.1Q VLAN identifier in iBFT
The iBFT NIC section has a VLAN field which must be filled in so that
iSCSI booting works over VLANs.
Unfortunately it is unclear from the IBM specification linked in
ibft.c whether the VLAN field is just the 802.1Q VLAN Identifier or
the full 802.1Q TCI. For now just fill in the VID, the Priority Code
Point and Drop Eligible Indicator could be set in the future if it
turns out they should be present too.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> Modified-by: Michael Brown <mcb30@ipxe.org> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Mon, 25 Feb 2013 17:22:01 +0000 (17:22 +0000)]
[efi] Ensure EFI binaries comply with Authenticode requirements
Authenticode requires that the size of the raw file must equal the
size of the OptionalHeader.SizeOfHeaders plus the sum of all sections'
SizeOfRawData.
Michael Brown [Wed, 21 Nov 2012 02:37:55 +0000 (02:37 +0000)]
[efi] Delegate to child device's EFI_COMPONENT_NAME2_PROTOCOL, if present
EFI's device naming model requires drivers to provide names for child
devices. Allow the driver's GetControllerName() method to delegate to
an instance of EFI_COMPONENT_NAME2_PROTOCOL installed on the child
device itself (if present); this allows the SNP device to expose its
own device name via the PCI driver's GetControllerName() method.
Michael Brown [Mon, 12 Nov 2012 16:16:31 +0000 (16:16 +0000)]
[bzimage] Allow initrds to be rearranged in place
At present, loading a bzImage via iPXE requires enough RAM to hold two
copies of each initrd file. Remove this constraint by rearranging the
initrds in place.
Thomas Miletich [Fri, 2 Nov 2012 15:00:09 +0000 (15:00 +0000)]
[build] Make version.o depend on the git index
The version number string is currently updated only if version.o
happens to be rebuilt due to changes in its dependencies. Add a
dependency upon the git index, so that the version number is updated
after any checkout.
Signed-off-by: Thomas Miletich <thomas.miletich@gmail.com> Modified-by: Michael Brown <mcb30@ipxe.org> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Fri, 2 Nov 2012 12:15:36 +0000 (12:15 +0000)]
[build] Inhibit .eh_frame on newer gcc versions
Using -fno-dwarf2-cfi-asm is not sufficient to prevent the .eh_frame
section from being generated on newer versions of gcc. Add
-fno-exceptions -fno-unwind-tables -fno-asynchronous-unwind-tables;
this is sufficient to inhibit the .eh_frame section on gcc 4.7.1.
This does not affect the overall binary size, but does fix the numbers
reported by "size" for individual object files.
Michael Brown [Fri, 2 Nov 2012 13:21:52 +0000 (13:21 +0000)]
[build] Use -maccumulate-outgoing-args if required by gcc
Current versions of gcc require -maccumulate-outgoing-args if any
sysv_abi functions call ms_abi functions. This requirement is likely
to be lifted in future gcc versions, so test explicitly to see if the
current version of gcc requires -maccumulate-outgoing-args.
This problem is currently masked since the implied
-fasynchronous-unwind-tables (which is the default in current gcc
versions) implies -maccumulate-outgoing-args.
Michael Brown [Mon, 29 Oct 2012 10:14:00 +0000 (10:14 +0000)]
[forcedeth] Increase TX ring size to prevent dropped packets
Commit 947976d ("[netdevice] Do not force a poll on net_tx()")
requires network devices to have TX rings that are sufficiently large
to allow a transmitted response to all packets received during a
single poll.
Reported-by: Robin Smidsrød <robin@smidsrod.no> Tested-by: Robin Smidsrød <robin@smidsrod.no> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Thu, 25 Oct 2012 23:09:09 +0000 (16:09 -0700)]
[intel] Poll RX queue if hardware reports RX overflow
The Intel NIC emulation in some versions of VMware seems to suffer
from a flaw whereby the Interrupt Cause Register (ICR) fails to assert
the usual "packet received" bit (ICR.RXT0) if a receive overflow
(ICR.RXO) has also occurred.
Work around this flaw by polling for completed descriptors whenever
either ICR.RXT0 or ICR.RXO is asserted.
Reported-by: Miroslav Halas <miroslav.halas@bankofamerica.com> Debugged-by: Miroslav Halas <miroslav.halas@bankofamerica.com> Tested-by: Miroslav Halas <miroslav.halas@bankofamerica.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Wed, 24 Oct 2012 20:26:51 +0000 (13:26 -0700)]
[netdevice] Do not force a poll on net_tx()
Including a netdev_poll() within net_tx() can cause the net_step()
loop to end up processing hundreds or thousands of packets within a
single step, since each received packet being processed may trigger a
response which, in turn causes a poll for further received packets.
Network devices must now ensure that the TX ring is at least as large
as the RX ring, in order to avoid running out of TX descriptors. This
should not cause any problems; unlike the RX ring, there is no
substantial memory cost incurred by increasing the TX ring size.
Michael Brown [Sat, 20 Oct 2012 21:44:34 +0000 (14:44 -0700)]
[test] Remove self-test for empty formatted setting value
Commit 5ad445f ("[settings] Treat an empty formatted value as meaning
"delete setting"") (re)defined the semantics of storing an empty
formatted setting as meaning "delete setting".
Remove the existing self-test using an empty formatted hex setting
value, since it no longer conforms to the defined semantics.
Shao Miller [Fri, 5 Oct 2012 14:14:12 +0000 (15:14 +0100)]
[serial] Don't enable serial console without serial support
serial_console_init() would enable serial console support without
knowing if the serial driver succeeded or not. As a result, the
serial console would interfere with a normal keyboard on a system
lacking serial support.
Reported-by: Jan ONDREJ (SAL) <ondrejj(at)salstar.sk> Signed-off-by: Shao Miller <sha0.miller@gmail.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Fri, 5 Oct 2012 14:02:30 +0000 (15:02 +0100)]
[int13] Do not zero %edx when jumping to a boot sector
Commit 73eb3f1 ("[int13] Zero all possible registers when jumping to a
boot sector") introduced a regression preventing the SAN-booting of
boot sectors which rely upon %dl containing the correct drive number
(such as most CD-ROM boot sectors).
Michael Brown [Wed, 26 Sep 2012 20:42:23 +0000 (21:42 +0100)]
[tls] Split received records over multiple I/O buffers
TLS servers are not obliged to implement the RFC3546 maximum fragment
length extension, and many common servers (including OpenSSL, as used
in Apache's mod_ssl) do not do so. iPXE may therefore have to cope
with TLS records of up to 16kB. Allocations for 16kB have a
non-negligible chance of failing, causing the TLS connection to abort.
Fix by maintaining the received record as a linked list of I/O
buffers, rather than a single contiguous buffer. To reduce memory
pressure, we also decrypt in situ, and deliver the decrypted data via
xfer_deliver_iob() rather than xfer_deliver_raw().
Michael Brown [Fri, 14 Sep 2012 16:49:53 +0000 (17:49 +0100)]
[efi] Provide guaranteed space in transmitted packets
eIPoIB requires space to expand a transmitted ARP packet. This
guarantee is met by ensuring that a transmitted packet consists of at
least MAX_LL_HEADER_LEN bytes from the start of the I/O buffer up to
the end of the link-layer header, and at least IOB_ZLEN bytes
thereafter.
Adjust the I/O buffer allocation for SNP transmitted packets to ensure
that this guarantee is met.
Michael Brown [Fri, 14 Sep 2012 14:42:54 +0000 (15:42 +0100)]
[int13] Zero all possible registers when jumping to a boot sector
At least one boot sector (the DUET boot sector used for bootstrapping
EFI from a non-EFI system) fails to initialise the high words of
registers before using them in calculations, leading to undefined
behaviour.
Work around such broken boot sectors by explicitly zeroing the
contents of all registers apart from %cs:%ip and %ss:%sp.
Michael Brown [Fri, 31 Aug 2012 02:52:06 +0000 (03:52 +0100)]
[settings] Use a generic setting's own type as its default type
When fetching a named setting using a name that does not explicitly
specify a type, default to using the type stored when the setting was
created, rather than always defaulting to "string". This allows the
behaviour of user-defined settings to match the behaviour of
predefined settings (which have a sensible default type).
For example:
set server:ipv4 192.168.0.1
echo ${server}
will now print "192.168.0.1", rather than trying to print out the raw
IPv4 address bytes as a string.
The downside of this change is that existing tricks for printing
special characters within scripts may require (backwards-compatible)
modification. For example, the "clear screen" sequence:
set esc:hex 1b
set cls ${esc}[2J
echo ${cls}
will now have to become
set esc:hex 1b
set cls ${esc:string}[2J # Must now explicitly specify ":string"
echo ${cls}
Michael Brown [Wed, 29 Aug 2012 22:34:14 +0000 (23:34 +0100)]
[ipoib] Expose Ethernet-compatible eIPoIB link-layer addresses and headers
Almost all clients of the raw-packet interfaces (UNDI and SNP) can
handle only Ethernet link layers. Expose an Ethernet-compatible link
layer to local clients, while remaining compatible with IPoIB on the
wire. This requires manipulation of ARP (but not DHCP) packets within
the IPoIB driver.
This is ugly, but it's the only viable way to allow IPoIB devices to
be driven via the raw-packet interfaces.
Michael Brown [Fri, 24 Aug 2012 15:44:29 +0000 (16:44 +0100)]
[realtek] Force EEPROM CS low before disabling EEPROM access mode
Some RTL8169 cards seem to drive the EEPROM CS line high (i.e. active)
when 9346CR.EEM is set to "normal operating mode", with the result
that the CS line is never deasserted. The symptom of this is that the
first read from the EEPROM will work, while all subsequent reads will
return garbage data.
Reported-by: Thomas Miletich <thomas.miletich@gmail.com> Debugged-by: Thomas Miletich <thomas.miletich@gmail.com> Tested-by: Thomas Miletich <thomas.miletich@gmail.com> Signed-off-by: Michael Brown <mcb30@ipxe.org>
Michael Brown [Thu, 23 Aug 2012 21:44:20 +0000 (22:44 +0100)]
[realtek] Forcibly enable advertisement of 1000Mbps speeds
Some RTL8169 cards (observed with an RTL8169SC) power up advertising
only 100Mbps, despite being capable of 1000Mbps. Forcibly enable
advertisement of 1000Mbps on any RTL8169-like card.
This change relies on the assumption that the CTRL1000 register will
not exist on 100Mbps-only RTL8169 cards such as the RTL8101.
Michael Brown [Thu, 23 Aug 2012 12:35:19 +0000 (13:35 +0100)]
[realtek] Enable DAC only when built as a 64-bit binary
Some RTL8169 cards (observed with an RTL8169SC) crash and burn if DAC
is enabled, even if only 32-bit addresses are used. Observed
behaviour includes system lockups and repeated transmission of garbage
data onto the wire.
This seems to be a known problem. The Linux r8169 driver disables DAC
by default and provides a "use_dac" module parameter.
There appears to be no known test for determining whether or not DAC
will work. As a workaround, enable DAC only if we are built as as
64-bit binary. This at least eliminates the problem in the common
case of a 32-bit build, which will never use 64-bit addresses anyway.
Michael Brown [Thu, 23 Aug 2012 12:27:31 +0000 (13:27 +0100)]
[realtek] Use explicit value for TCR.MXDMA
Some RTL8169 cards (observed with an RTL8169SC) power up with
TCR.MXDMA set to 16 bytes. While this does not prevent proper
operation, it almost certainly degrades performance.
Fix by explicitly setting TCR.MXDMA to "unlimited".
Michael Brown [Thu, 23 Aug 2012 11:25:08 +0000 (12:25 +0100)]
[realtek] Use explicit values for RCR.RXFTH and RCR.MXDMA
Some RTL8169 cards (observed with an RTL8169SC) power up with invalid
values in RCR.RXFTH and RCR.MXDMA, causing receive DMA to fail. Fix
by setting explicit values for both fields.
projects / thirdparty / ipxe.git / log
