ipxe.git
15 hours ago[efi] Use device path to locate filesystem from which we were loaded coverity_scan issue123 master github/coverity_scan github/issue123 github/master
Michael Brown [Mon, 3 Aug 2020 14:26:25 +0000 (15:26 +0100)] 
[efi] Use device path to locate filesystem from which we were loaded

The file:/ URI syntax may be used to refer to local files on the
filesystem from which the iPXE binary was loaded.  This is currently
implemented by directly using the DeviceHandle recorded in our
EFI_LOADED_IMAGE_PROTOCOL.

This mechanism will fail when a USB-enabled build of iPXE is loaded
from USB storage and subsequently installs its own USB host controller
drivers, since doing so will disconnect and reconnect the existing USB
storage drivers and thereby invalidate the original storage device
handle.

Fix by recording the device path for the loaded image's DeviceHandle
at initialisation time and later using the recorded device path to
locate the appropriate device handle.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
15 hours ago[efi] Match EDK2 numbering for USB ports
Michael Brown [Mon, 3 Aug 2020 13:42:18 +0000 (14:42 +0100)] 
[efi] Match EDK2 numbering for USB ports

The various USB specifications all use one-based numbering for ports.
This scheme is applied consistently across the various relevant
specifications, covering both port numbers that appear on the wire
(i.e. downstream hub port numbers) and port numbers that exist only
logically (i.e. root hub port numbers).

The UEFI specification is ambiguous about the port numbers as used for
the ParentPortNumber field within a USB_DEVICE_PATH structure.  As of
UEFI specification version 2.8 errata B:

- section 10.3.4.5 just states "USB Parent Port Number" with no
  indication of being zero-based or one-based

- section 17.1.1 notes that for the EFI_USB2_HC_PROTOCOL, references
  to PortNumber parameters are zero-based for root hub ports

- section 17.1.1 also mentions a TranslatorPortNumber used by
  EFI_USB2_HC_PROTOCOL, with no indication of being zero-based or
  one-based

- there are no other mentions of USB port numbering schemes.

Experimentation and inspection of the EDK2 codebase reveals that at
least the EDK2 reference implementation will use zero-based numbering
for both root and non-root hub ports when populating a USB_DEVICE_PATH
structure (though will inconsistently use one-based numbering for the
TranslatorPortNumber parameter).

Use zero-based numbering for both root and non-root hub ports when
constructing a USB_DEVICE_PATH in order to match the behaviour of the
EDK2 implementation.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 days ago[comboot] Fix stack pointer retrieval after COM32 binary returns
Dentcho Ludmilov Bankov [Fri, 1 Feb 2019 22:20:21 +0000 (00:20 +0200)] 
[comboot] Fix stack pointer retrieval after COM32 binary returns

This change fixes the offset used when retrieving the iPXE stack
pointer after a COM32 binary returns.  The iPXE stack pointer is saved
at the top of the available memory then the the top of the stack for
the COM32 binary is set just below it.  However seven more items are
pushed on the COM32 stack before the entry point is invoked so when
the COM32 binary returns the location of the iPXE stack pointer is 28
(and not 24) bytes above the current stack pointer.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 days ago[efi] Check the status code from AllocatePool()
Ignat Korchagin [Tue, 21 Jul 2020 21:27:08 +0000 (22:27 +0100)] 
[efi] Check the status code from AllocatePool()

According to the latest UEFI specification (Version 2.8 Errata B)
p. 7.2:

  "Buffer: A pointer to a pointer to the allocated buffer if the call
   succeeds; undefined otherwise."

So implementations are obliged neither to return NULL, if the
allocation fails, nor to preserve the contents of the pointer.

Make the logic more reliable by checking the status code from
AllocatePool() instead of checking the returned pointer for NULL

Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>
Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 days ago[cmdline] Add "--timeout" parameter to "ifconf" command
Joe Groocock [Fri, 20 Sep 2019 12:41:29 +0000 (13:41 +0100)] 
[cmdline] Add "--timeout" parameter to "ifconf" command

Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 days ago[intelx] Added PCI_ROM entry for Intel x553/x557-AT and x553 (SFP+) NICs
Rob Taglang [Tue, 2 Oct 2018 19:41:51 +0000 (15:41 -0400)] 
[intelx] Added PCI_ROM entry for Intel x553/x557-AT and x553 (SFP+) NICs

Signed-off-by: Rob Taglang <rob@taglang.io>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 days ago[intel] Set INTEL_NO_PHY_RST on i218v
Ben Bleything [Sun, 4 Nov 2018 06:53:27 +0000 (23:53 -0700)] 
[intel] Set INTEL_NO_PHY_RST on i218v

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 days ago[intel] Add INTEL_NO_PHY_RST for I218LM
Andrew Grasso [Thu, 13 Jul 2017 13:04:01 +0000 (09:04 -0400)] 
[intel] Add INTEL_NO_PHY_RST for I218LM

Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[deflate] Fix typo in comment describing length codes
Daniel Johnson [Tue, 21 Jul 2020 05:20:43 +0000 (22:20 -0700)] 
[deflate] Fix typo in comment describing length codes

Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[build] Fix a GNUism that FreeBSD's sed(1) cannot deal with
Tobias Kortkamp [Sun, 25 Nov 2018 16:57:13 +0000 (17:57 +0100)] 
[build] Fix a GNUism that FreeBSD's sed(1) cannot deal with

At the moment '\s*' is silently interpreted as just 's*', but in the
future it will be an error:

sed: 1: "s/\.o\s*:/_DEPS +=/": RE error: trailing backslash (\)

cf. https://bugs.freebsd.org/229925

Signed-off-by: Tobias Kortkamp <t@tobik.me>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[wpa] Fix erroneous debug message in wpa_derive_ptk
Michael Brown [Tue, 21 Jul 2020 14:34:39 +0000 (15:34 +0100)] 
[wpa] Fix erroneous debug message in wpa_derive_ptk

Split debug message since eth_ntoa() uses a static result buffer.

Originally-fixed-by: Michael Bazzinotti <bazz@bazz1.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[libc] Fix memcmp() to return proper values
Michael J. Bazzinotti [Tue, 14 Jan 2020 18:15:02 +0000 (13:15 -0500)] 
[libc] Fix memcmp() to return proper values

Fix memcmp() to return proper standard positive/negative values for
unequal comparisons.  Current implementation is backwards (i.e. the
functions are returning negative when should be positive and
vice-versa).

Currently most consumers of these functions only check the return value
for ==0 or !=0 and so we can safely change the implementation without
breaking things.

However, there is one call that checks the polarity of this function,
and that is prf_sha1() for wireless WPA 4-way handshake.  Due to the
incorrect memcmp() polarity, the WPA handshake creates an incorrect
PTK, and the handshake would fail after step 2.  Undoubtedly, the AP
noticed the supplicant failed the mic check.  This commit fixes that
issue.

Similar to commit 3946aa9 ("[libc] Fix strcmp()/strncmp() to return
proper values").

Signed-off-by: Michael Bazzinotti <bazz@bazz1.com>
Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[pcbios] Take alignment into account when checking for available space
David Decotigny [Wed, 15 Jan 2020 07:42:05 +0000 (23:42 -0800)] 
[pcbios] Take alignment into account when checking for available space

Signed-off-by: David Decotigny <ddecotig@gmail.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[pcbios] Fix "out of memory" detection when expanding bottom area
David Decotigny [Wed, 15 Jan 2020 07:31:03 +0000 (23:31 -0800)] 
[pcbios] Fix "out of memory" detection when expanding bottom area

This caused iPXE to reject images even when enough memory was
available.

Signed-off-by: David Decotigny <ddecotig@gmail.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[libgcc] Change __divmoddi4 from int64 [unknown] to int64_t
Bobby Lockwood [Mon, 19 Dec 2016 12:01:15 +0000 (12:01 +0000)] 
[libgcc] Change __divmoddi4 from int64 [unknown] to int64_t

Matches the header file libgcc.h and solves

  __divmoddi4.c:3:56: error: unknown type name ‘int64’

Signed-off-by: Michael Brown <mcb30@ipxe.org>
13 days ago[build] Fix default target in sdsk image
bit [Sun, 19 Aug 2018 14:30:43 +0000 (16:30 +0200)] 
[build] Fix default target in sdsk image

gensdsk currently creates a syslinux.cfg file that is invalid if the
filename ends in lkrn.  Fix by setting the default target to label($b)
instead of filename($g).

Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
2 weeks ago[dns] Use all configured DNS servers
Michael Brown [Wed, 15 Jul 2020 18:10:30 +0000 (19:10 +0100)] 
[dns] Use all configured DNS servers

When no response is obtained from the first configured DNS server,
fall back to attempting the other configured servers.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
2 weeks ago[xfer] Remove address family from definition of a socket opener
Michael Brown [Wed, 15 Jul 2020 17:46:58 +0000 (18:46 +0100)] 
[xfer] Remove address family from definition of a socket opener

All implemented socket openers provide definitions for both IPv4 and
IPv6 using exactly the same opener method.  Simplify the logic by
omitting the address family from the definition.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
2 weeks ago[sfc] Add PCI ID for Solarflare SFC9120 and Solarflare SFC9140
LaDerrick H [Tue, 14 Jul 2020 15:59:59 +0000 (10:59 -0500)] 
[sfc] Add PCI ID for Solarflare SFC9120 and Solarflare SFC9140

This patch adds support for 2 Solarflare NICs.

Signed-off-by: LaDerrick H <ipxelist243@lacutt.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
2 weeks ago[efi] Claim SNP devices early in efi_download_start() tpl github/tpl
Michael Brown [Tue, 14 Jul 2020 17:24:02 +0000 (18:24 +0100)] 
[efi] Claim SNP devices early in efi_download_start()

Claiming the SNP devices has the side effect of raising the TPL to
iPXE's normal operating level of TPL_CALLBACK (see the commit message
for c89a446 ("[efi] Run at TPL_CALLBACK to protect against UEFI
timers") for details).  This must happen before executing any code
that relies upon the TPL having been raised to TPL_CALLBACK.

The call to efi_snp_claim() in efi_download_start() currently happens
only after the call to xfer_open().  Calling xfer_open() will
typically result in a retry timer being started, which will result in
a call to currticks() in order to initialise the timer.  The call to
currticks() will drop to TPL_APPLICATION and restore to TPL_CALLBACK
in order to allow a timer tick to occur.  Since this call happened
before the call to efi_snp_claim(), the restored TPL is incorrect.

This in turn results in efi_snp_claim() recording the incorrect
original TPL, causing efi_snp_release() to eventually restore the
incorrect TPL, causing the system to lock up when ExitBootServices()
is called at TPL_CALLBACK.

Fix by moving the call to efi_snp_claim() to the start of
efi_download_start().

Debugged-by: Jarrod Johnson <jjohnson2@lenovo.com>
Debugged-by: He He4 Huang <huanghe4@lenovo.com>
Debugged-by: James Wang <jameswang@ami.com.tw>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
3 weeks ago[efi] Ensure NUL byte is at lowest address within stack cookie cookie github/cookie
Michael Brown [Thu, 9 Jul 2020 13:20:53 +0000 (14:20 +0100)] 
[efi] Ensure NUL byte is at lowest address within stack cookie

The NUL byte included within the stack cookie to act as a string
terminator should be placed at the lowest byte address within the
stack cookie, in order to avoid potentially including the stack cookie
value within an accidentally unterminated string.

Suggested-by: Pete Beck <pete.beck@ioactive.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
3 weeks ago[efi] Distribute available entropy within stack cookie
Michael Brown [Thu, 9 Jul 2020 12:56:50 +0000 (13:56 +0100)] 
[efi] Distribute available entropy within stack cookie

Several of the values used to compute a stack cookie (in the absence
of a viable entropy source) will tend to have either all-zeroes or
all-ones in the higher order bits.  Rotate the values in order to
distribute the (minimal) available entropy more evenly.

Suggested-by: Pete Beck <pete.beck@ioactive.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
3 weeks ago[libc] Add bit-rotation functions for unsigned long values
Michael Brown [Thu, 9 Jul 2020 12:51:30 +0000 (13:51 +0100)] 
[libc] Add bit-rotation functions for unsigned long values

Generalise the bit rotation implementations to use a common macro, and
add roll() and rorl() to handle unsigned long values.

Each function will still compile down to a single instruction.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
3 weeks ago[efi] Avoid setting direction flag on EFI platforms
Michael Brown [Tue, 7 Jul 2020 12:49:17 +0000 (13:49 +0100)] 
[efi] Avoid setting direction flag on EFI platforms

The only remaining use case in iPXE for the CPU direction flag is in
__memcpy_reverse() where it is set to allow the use of "rep movsb" to
perform the memory copy.  This matches the equivalent functionality in
the EDK2 codebase, which has functions such as InternalMemCopyMem that
also temporarily set the direction flag in order to use "rep movsb".

As noted in commit d2fb317 ("[crypto] Avoid temporarily setting
direction flag in bigint_is_geq()"), some UEFI implementations are
known to have buggy interrupt handlers that may reboot the machine if
a timer interrupt happens to occur while the direction flag is set.

Work around these buggy UEFI implementations by using the
(unoptimised) generic_memcpy_reverse() on i386 or x86_64 UEFI
platforms.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
3 weeks ago[libc] Provide an unoptimised generic_memcpy_reverse()
Michael Brown [Tue, 7 Jul 2020 12:13:28 +0000 (13:13 +0100)] 
[libc] Provide an unoptimised generic_memcpy_reverse()

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[crypto] Avoid temporarily setting direction flag in bigint_is_geq()
Michael Brown [Mon, 6 Jul 2020 23:09:40 +0000 (00:09 +0100)] 
[crypto] Avoid temporarily setting direction flag in bigint_is_geq()

The UEFI specification states that the calling convention for IA-32
and x64 includes "Direction flag in EFLAGS is clear".  This
specification covers only the calling convention used at the point of
calling functions annotated with EFIAPI.  The specification explicitly
states that other functions (such as private functions or static
library calls) are not required to follow the UEFI calling
conventions.

The reference EDK2 implementation follows this specification.  In
particular, the EDK2 interrupt handlers will clear the direction flag
before calling any EFIAPI functions, and will restore the direction
flag when returning from the interrupt handler.  Some EDK2 private
library functions (most notably InternalMemCopyMem) may set the
direction flag temporarily in order to make efficient use of CPU
string operations.

The current implementation of iPXE's bigint_is_geq() for i386 and
x86_64 will similarly set the direction flag temporarily in order to
make efficient use of CPU string operations.

On some UEFI implementations (observed with a Getac RX10 tablet), a
timer interrupt that happens to occur while the direction flag is set
will reboot the machine.  This very strongly indicates that the UEFI
timer interrupt handler is failing to clear the direction flag before
performing an affected operation (such as copying a block of memory).

Work around such buggy UEFI implementations by rewriting
bigint_is_geq() to avoid the use of string operations and so obviate
the requirement to temporarily set the direction flag.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[usb] Leave port enabled after a failed device registration
Michael Brown [Sat, 4 Jul 2020 10:52:26 +0000 (11:52 +0100)] 
[usb] Leave port enabled after a failed device registration

A failure in device registration (e.g. due to a device with malformed
descriptors) will currently result in the port being disabled as part
of the error path.  This in turn causes the hardware to detect the
device as newly connected, leading to an endless loop of failed device
registrations.

Fix by leaving the port enabled in the case of a registration failure.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[axge] Reapply USB device configuration when opening network device
Michael Brown [Fri, 3 Jul 2020 19:17:25 +0000 (20:17 +0100)] 
[axge] Reapply USB device configuration when opening network device

When connected to a USB3 port, the AX88179 seems to have an
approximately 50% chance of producing a USB transaction error on each
of its three endpoints after being closed and reopened.  The root
cause is unclear, but rewriting the USB device configuration value
seems to clear whatever internal error state has accumulated.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[xhci] Increase link state settling delay to 100ms
Michael Brown [Fri, 3 Jul 2020 11:52:05 +0000 (12:52 +0100)] 
[xhci] Increase link state settling delay to 100ms

Experimentation shows that the existing 20ms delay is insufficient,
and often results in device detection being deferred until after iPXE
has completed startup.

Fix by increasing the delay to 100ms.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[usb] Avoid unnecessary calls to usb_hub_set_drvdata()
Michael Brown [Fri, 3 Jul 2020 10:29:25 +0000 (11:29 +0100)] 
[usb] Avoid unnecessary calls to usb_hub_set_drvdata()

The driver-private data for root hubs is already set immediately after
allocating the USB bus.  There seems to be no reason to set it again
when opening the root hub.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[xhci] Set link state to RxDetect after disabling USB3 root hub port
Michael Brown [Thu, 2 Jul 2020 21:53:11 +0000 (22:53 +0100)] 
[xhci] Set link state to RxDetect after disabling USB3 root hub port

The "disabled" port states for USB2 and USB3 are not directly
equivalent.  In particular, a disabled USB3 port will not detect new
device connections.  The result is that a USB3 device disconnected
from and reconnected to an xHCI root hub port will end up reconnecting
as a USB2 device.

Fix by setting the link state to RxDetect after disabling the port, as
is already done during initialisation.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[usb] Do not attempt to disable USB3 hub ports
Michael Brown [Thu, 2 Jul 2020 15:48:17 +0000 (16:48 +0100)] 
[usb] Do not attempt to disable USB3 hub ports

The USB3 specification removes PORT_ENABLE from the list of features
that may be cleared via a CLEAR_FEATURE request.  Experimentation
shows that omitting the attempt to clear PORT_ENABLE seems to result
in the correct hotplug behaviour.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[usb] Add missing usb_recycle() for completed hub interrupt transfers
Michael Brown [Thu, 2 Jul 2020 13:19:02 +0000 (14:19 +0100)] 
[usb] Add missing usb_recycle() for completed hub interrupt transfers

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[usb] Clear device endpoint halt before resetting host endpoint
Michael Brown [Thu, 2 Jul 2020 01:51:58 +0000 (02:51 +0100)] 
[usb] Clear device endpoint halt before resetting host endpoint

Resetting the host endpoint may immediately restart any pending
transfers for that endpoint.  If the device endpoint halt has not yet
been cleared, then this will probably result in a second failed
transfer.

This second failure may be detected within usb_endpoint_reset() while
waiting for usb_clear_feature() to complete.  The endpoint will
subsequently be removed from the list of halted endpoints, causing the
second failure to be effectively ignored and leaving the host endpoint
in a permanently halted state.

Fix by deferring the host endpoint reset until after the device
endpoint is ready to accept new transfers.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[axge] Handle non-gigabit link speeds
Michael Brown [Wed, 1 Jul 2020 19:40:09 +0000 (20:40 +0100)] 
[axge] Handle non-gigabit link speeds

The ASIX USB NICs are capable of autodetecting the Ethernet link speed
and reporting it via PLSR but will not automatically update the
relevant GM and PS bits in MSR.  The result is that a non-gigabit link
will fail to send or receive any packets.

The interrupt endpoint used to report link state includes the values
of the PHY BMSR and LPA registers.  These are not sufficient to
differentiate between 100Mbps and 1000Mbps, since the LPA_NPAGE bit
does not necessarily indicate that the link partner is advertising
1000Mbps.

Extend axge_check_link() to write the MSR value based on the link
speed read from PLSR, and simplify the interrupt endpoint handler to
merely trigger a call to axge_check_link().

Signed-off-by: Michael Brown <mcb30@ipxe.org>
4 weeks ago[efi] Raise TPL during driver entry point
Michael Brown [Tue, 30 Jun 2020 15:32:59 +0000 (16:32 +0100)] 
[efi] Raise TPL during driver entry point

As per commit c89a446 ("[efi] Run at TPL_CALLBACK to protect against
UEFI timers") we expect to run at TPL_CALLBACK almost all of the time.
Various code paths rely on this assumption.  Code paths that need to
temporarily lower the TPL (e.g. for entropy gathering) will restore it
to TPL_CALLBACK.

The entropy gathering code will be run during DRBG initialisation,
which happens during the call to startup().  In the case of iPXE
compiled as an EFI application this code will run within the scope of
efi_snp_claim() and so will execute at TPL_CALLBACK as expected.

In the case of iPXE compiled as an EFI driver the code will
incorrectly run at TPL_APPLICATION since there is nothing within the
EFI driver entry point that raises (and restores) the TPL.  The net
effect is that a build that includes the entropy-gathering code
(e.g. a build with HTTPS enabled) will return from the driver entry
point at TPL_CALLBACK, which causes a system lockup.

Fix by raising and restoring the TPL within the EFI driver entry
point.

Debugged-by: Ignat Korchagin <ignat@cloudflare.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[efi] Detect and disable seriously broken EFI_RNG_PROTOCOL implementations
Michael Brown [Sun, 28 Jun 2020 18:24:30 +0000 (19:24 +0100)] 
[efi] Detect and disable seriously broken EFI_RNG_PROTOCOL implementations

The EFI_RNG_PROTOCOL on the Microsoft Surface Go does not generate
random numbers.  Successive calls to GetRNG() without any intervening
I/O operations (such as writing to the console) will produce identical
results.  Successive reboots will produce identical results.

It is unclear what the Microsoft Surface Go is attempting to use as an
entropy source, but it is demonstrably producing zero bits of entropy.

The failure is already detected by the ANS-mandated Repetition Count
Test performed as part of our GetEntropy implementation.  This
currently results in the entropy source being marked as broken, with
the result that iPXE refuses to perform any operations that require a
working entropy source.

We cannot use the existing EFI driver blacklisting mechanism to unload
the broken driver, since the RngDxe driver is integrated into the
DxeCore image.

Work around the broken driver by checking for consecutive identical
results returned by EFI_RNG_PROTOCOL and falling back to the original
timer-based entropy source.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[build] Disable position-independent code for ARM64 EFI builds
Michael Brown [Fri, 26 Jun 2020 20:21:31 +0000 (21:21 +0100)] 
[build] Disable position-independent code for ARM64 EFI builds

Some versions of gcc (observed with the cross-compiling gcc 9.3.0 in
Ubuntu 20.04) default to enabling -fPIE.  Experimentation shows that
this results in the emission of R_AARCH64_ADR_GOT_PAGE relocation
records for __stack_chk_guard.  These relocation types are not
supported by elf2efi.c.

Fix by explicitly disabling position-independent code for ARM64 EFI
builds.

Debugged-by: Antony Messerli <antony@mes.ser.li>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[golan] Add explicit type casts for nodnic_queue_pair_type
Michael Brown [Sat, 27 Jun 2020 19:43:32 +0000 (20:43 +0100)] 
[golan] Add explicit type casts for nodnic_queue_pair_type

GCC 10 emits warnings for implicit conversions of enumerated types.

The flexboot_nodnic code defines nodnic_queue_pair_type with values
identical to those of ib_queue_pair_type, and implicitly casts between
them.  Add an explicit cast to fix the warning.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[intel] Avoid spurious compiler warning on GCC 10
Michael Brown [Sat, 27 Jun 2020 19:21:11 +0000 (20:21 +0100)] 
[intel] Avoid spurious compiler warning on GCC 10

GCC 10 produces a spurious warning about an out-of-bounds array access
for the unsized raw dword array in union intelvf_msg.

Avoid the warning by embedding the zero-length array within a struct.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[build] Be explicit about -fcommon compiler directive temp github/temp
Bruce Rogers [Wed, 6 May 2020 21:03:02 +0000 (15:03 -0600)] 
[build] Be explicit about -fcommon compiler directive

gcc10 switched default behavior from -fcommon to -fno-common.  Since
"__shared" relies on the legacy behavior, explicitly specify it.

Signed-off-by: Bruce Rogers <brogers@suse.com>
Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[ocsp] Accept SHA1 certID responses even if SHA1 is not enabled
Michael Brown [Thu, 25 Jun 2020 12:04:02 +0000 (13:04 +0100)] 
[ocsp] Accept SHA1 certID responses even if SHA1 is not enabled

Various implementation quirks in OCSP servers make it impractical to
use anything other than SHA1 to construct the issuerNameHash and
issuerKeyHash identifiers in the request certID.  For example: both
the OpenCA OCSP responder used by ipxe.org and the Boulder OCSP
responder used by LetsEncrypt will fail if SHA256 is used in the
request certID.

As of commit 6ffe28a ("[ocsp] Accept response certID with missing
hashAlgorithm parameters") we rely on asn1_digest_algorithm() to parse
the algorithm identifier in the response certID.  This will fail if
SHA1 is disabled via config/crypto.h.

Fix by using a direct ASN.1 object comparison on the OID within the
algorithm identifier.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 weeks ago[efi] Enable stack protection where possible
Michael Brown [Tue, 23 Jun 2020 22:08:49 +0000 (23:08 +0100)] 
[efi] Enable stack protection where possible

Enable -fstack-protector for EFI builds, where binary size is less
critical than for BIOS builds.

The stack cookie must be constructed immediately on entry, which
prohibits the use of any viable entropy source.  Construct a cookie by
XORing together various mildly random quantities to produce a value
that will at least not be identical on each run.

On detecting a stack corruption, attempt to call Exit() with an
appropriate error.  If that fails, then lock up the machine since
there is no other safe action that can be taken.

The old conditional check for support of -fno-stack-protector is
omitted since this flag dates back to GCC 4.1.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[parseopt] Treat empty integer strings in user input as invalid
Michael Brown [Fri, 19 Jun 2020 16:29:46 +0000 (17:29 +0100)] 
[parseopt] Treat empty integer strings in user input as invalid

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[util] Treat empty integer strings as invalid
Michael Brown [Fri, 19 Jun 2020 15:56:02 +0000 (16:56 +0100)] 
[util] Treat empty integer strings as invalid

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[snp] Retry initialisation if link is reported as down
Michael Brown [Thu, 18 Jun 2020 23:18:22 +0000 (00:18 +0100)] 
[snp] Retry initialisation if link is reported as down

Some devices (observed with a Getac RX10 tablet and docking station
containing an embedded AX88179 USB NIC) seem to be capable of
detecting link state only during the call to Initialize(), and will
occasionally erroneously report that the link is down for the first
few such calls.

Work around these devices by retrying the Initialize() call multiple
times, terminating early if a link is detected.  The eventual absence
of a link is treated as a non-fatal error, since it is entirely
possible that the link really is down.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[crypto] Disable MD5 as an OID-identifiable algorithm by default
Michael Brown [Tue, 16 Jun 2020 22:17:21 +0000 (23:17 +0100)] 
[crypto] Disable MD5 as an OID-identifiable algorithm by default

Disable the use of MD5 as an OID-identifiable algorithm.  Note that
the MD5 algorithm implementation will still be present in the build,
since it is used implicitly by various cryptographic components such
as HTTP digest authentication; this commit removes it only from the
list of OID-identifiable algorithms.

It would be appropriate to similarly disable the use of SHA-1 by
default, but doing so would break the use of OCSP since several OCSP
responders (including the current version of openca-ocspd) are not
capable of interpreting the hashAlgorithm field and so will fail if
the client uses any algorithm other than the configured default.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[crypto] Ensure that test code drags in required ASN.1 object identifiers
Michael Brown [Tue, 16 Jun 2020 22:40:58 +0000 (23:40 +0100)] 
[crypto] Ensure that test code drags in required ASN.1 object identifiers

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[crypto] Allow algorithms to be included without being OID-identifiable
Michael Brown [Tue, 16 Jun 2020 16:14:54 +0000 (17:14 +0100)] 
[crypto] Allow algorithms to be included without being OID-identifiable

There are many ways in which the object for a cryptographic algorithm
may be included, even if not explicitly enabled in config/crypto.h.
For example: the MD5 algorithm is required by TLSv1.1 or earlier, by
iSCSI CHAP authentication, by HTTP digest authentication, and by NTLM
authentication.

In the current implementation, inclusion of an algorithm for any
reason will result in the algorithm's ASN.1 object identifier being
included in the "asn1_algorithms" table, which consequently allows the
algorithm to be used for any ASN1-identified purpose.  For example: if
the MD5 algorithm is included in order to support HTTP digest
authentication, then iPXE would accept a (validly signed) TLS
certificate using an MD5 digest.

Split the ASN.1 object identifiers into separate files that are
required only if explicitly enabled in config/crypto.h.  This allows
an algorithm to be omitted from the "asn1_algorithms" table even if
the algorithm implementation is dragged in for some other purpose.

The end result is that only the algorithms that are explicitly enabled
in config/crypto.h can be used for ASN1-identified purposes such as
signature verification.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
6 weeks ago[tls] Default to supporting only TLSv1.1 or above
Michael Brown [Tue, 16 Jun 2020 12:14:12 +0000 (13:14 +0100)] 
[tls] Default to supporting only TLSv1.1 or above

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 weeks ago[tls] Allow a minimum TLS protocol version to be specified
Michael Brown [Fri, 12 Jun 2020 20:40:33 +0000 (21:40 +0100)] 
[tls] Allow a minimum TLS protocol version to be specified

The supported ciphers and digest algorithms may already be specified
via config/crypto.h.  Extend this to allow a minimum TLS protocol
version to be specified.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 weeks ago[efi] Attempt to connect our driver directly if ConnectController fails
Michael Brown [Wed, 10 Jun 2020 21:52:11 +0000 (22:52 +0100)] 
[efi] Attempt to connect our driver directly if ConnectController fails

Some platforms (observed with an AMI BIOS on an Apollo Lake system)
will spuriously fail the call to ConnectController() when the UEFI
network stack is disabled.  This appears to be a BIOS bug that also
affects attempts to connect any non-iPXE driver to the NIC controller
handle via the UEFI shell "connect" utility.

Work around this BIOS bug by falling back to calling our
efi_driver_start() directly if the call to ConnectController() fails.
This bypasses any BIOS policy in terms of deciding which driver to
connect but still cooperates with the UEFI driver model in terms of
handle ownership, since the use of EFI_OPEN_PROTOCOL_BY_DRIVER ensures
that the BIOS is aware of our ownership claim.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
8 weeks ago[uri] Avoid appearing to access final byte of a potentially empty string
Michael Brown [Fri, 5 Jun 2020 09:01:19 +0000 (10:01 +0100)] 
[uri] Avoid appearing to access final byte of a potentially empty string

The URI parsing code for "host[:port]" checks that the final character
is not ']' in order to allow for IPv6 literals.  If the entire
"host[:port]" portion of the URL is an empty string, then this will
access the preceding character.  This does not result in accessing
invalid memory (since the string is guaranteed by construction to
always have a preceding character) and does not result in incorrect
behaviour (since if the string is empty then strrchr() is guaranteed
to return NULL), but it does make the code confusing to read.

Fix by inverting the order of the two tests.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
8 weeks ago[efi] Work around UEFI specification bug in LoadImage for SAN boot
Michael Brown [Fri, 5 Jun 2020 08:40:36 +0000 (09:40 +0100)] 
[efi] Work around UEFI specification bug in LoadImage for SAN boot

As described in the previous commit, work around a UEFI specification
bug that necessitates calling UnloadImage if the return value from
LoadImage is EFI_SECURITY_VIOLATION.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
8 weeks ago[efi] Work around UEFI specification bug in LoadImage
Michael Brown [Thu, 4 Jun 2020 21:24:21 +0000 (22:24 +0100)] 
[efi] Work around UEFI specification bug in LoadImage

iPXE currently assumes that any error returned from LoadImage()
indicates that the image was not loaded.  This assumption was correct
at the time the code was written and remained correct for UEFI
specifications up to and including version 2.1.

In version 2.3, the UEFI specification broke API and ABI compatibility
by defining that a return value of EFI_SECURITY_VIOLATION would now
indicate that the image had been loaded and a valid image handle had
been created, but that the image should not be started.

The wording in version 2.2 is ambiguous, and does not define whether
or not a return value of EFI_SECURITY_VIOLATION indicates that a valid
image handle has been created.

Attempt to work around all of these incompatible and partially
undefined APIs by calling UnloadImage if we get a return value of
EFI_SECURITY_VIOLATION.  Minimise the risk of passing an uninitialised
pointer to UnloadImage by setting ImageHandle to NULL prior to calling
LoadImage.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
8 weeks ago[png] Fix potential integer overflow
Michael Brown [Thu, 4 Jun 2020 21:09:11 +0000 (22:09 +0100)] 
[png] Fix potential integer overflow

Signed-off-by: Michael Brown <mcb30@ipxe.org>
2 months ago[bnxt] Add driver support for Broadcom NetXtreme-E Adapters
Joseph Wong [Tue, 30 Apr 2019 21:17:04 +0000 (14:17 -0700)] 
[bnxt] Add driver support for Broadcom NetXtreme-E Adapters

Signed-off-by: Joseph Wong <joseph.wong@broadcom.com>
4 months ago[efi] Disambiguate errors returned by ConnectController
Michael Brown [Sat, 14 Mar 2020 09:49:49 +0000 (09:49 +0000)] 
[efi] Disambiguate errors returned by ConnectController

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[int13con] Create log partition only when CONSOLE_INT13 is enabled
Michael Brown [Sun, 1 Mar 2020 12:56:28 +0000 (12:56 +0000)] 
[int13con] Create log partition only when CONSOLE_INT13 is enabled

Reduce the size of the USB disk image in the common case that
CONSOLE_INT13 is not enabled.

Originally-implemented-by: Romain Guyard <romain.guyard@mujin.co.jp>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[bios] Define macros for constructing partition table entries
Michael Brown [Sun, 1 Mar 2020 11:58:47 +0000 (11:58 +0000)] 
[bios] Define macros for constructing partition table entries

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[iscsi] Eliminate variable-length stack allocation in URI parsing
Michael Brown [Sun, 16 Feb 2020 23:46:33 +0000 (23:46 +0000)] 
[iscsi] Eliminate variable-length stack allocation in URI parsing

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[iscsi] Eliminate variable-length stack allocations in CHAP handlers
Michael Brown [Sun, 16 Feb 2020 23:19:03 +0000 (23:19 +0000)] 
[iscsi] Eliminate variable-length stack allocations in CHAP handlers

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[settings] Eliminate variable-length stack allocation
Michael Brown [Sun, 16 Feb 2020 22:30:38 +0000 (22:30 +0000)] 
[settings] Eliminate variable-length stack allocation

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[slam] Allow for the possibility of IPv6 multicast addresses
Michael Brown [Sun, 16 Feb 2020 22:02:25 +0000 (22:02 +0000)] 
[slam] Allow for the possibility of IPv6 multicast addresses

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[slam] Eliminate variable-length stack allocation
Michael Brown [Sun, 16 Feb 2020 21:55:59 +0000 (21:55 +0000)] 
[slam] Eliminate variable-length stack allocation

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[infiniband] Eliminate variable-length stack allocation
Michael Brown [Sun, 16 Feb 2020 21:25:27 +0000 (21:25 +0000)] 
[infiniband] Eliminate variable-length stack allocation

Signed-off-by: Michael Brown <mcb30@ipxe.org>
5 months ago[tftp] Eliminate unnecessary variable-length stack allocation
Michael Brown [Sun, 16 Feb 2020 20:08:20 +0000 (20:08 +0000)] 
[tftp] Eliminate unnecessary variable-length stack allocation

Eliminate an unnecessary variable-length stack allocation and memory
copy by allowing TFTP option processors to modify the option string
in-place.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 months ago[travis] Ensure that most recent tag is always available
Michael Brown [Thu, 2 Jan 2020 23:14:03 +0000 (00:14 +0100)] 
[travis] Ensure that most recent tag is always available

Remove clone depth limit, to ensure that the most recent tag (from
which the version should be constructed) is always present.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 months ago[build] Construct full version number automatically from git revision v1.20.1
Michael Brown [Thu, 2 Jan 2020 22:43:15 +0000 (23:43 +0100)] 
[build] Construct full version number automatically from git revision

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 months ago[snp] Set EFI_SIMPLE_NETWORK_RECEIVE_MULTICAST bit as per UEFI spec v20.01
Ignat Korchagin [Fri, 13 Dec 2019 16:17:58 +0000 (16:17 +0000)] 
[snp] Set EFI_SIMPLE_NETWORK_RECEIVE_MULTICAST bit as per UEFI spec

According to UEFI specification 2.8 p 24.1 we must set the
EFI_SIMPLE_NETWORK_RECEIVE_MULTICAST bit in the "Disable" mask, when
"ResetMCastFilter" is TRUE.

Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>
Split-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 months ago[snp] Try promiscuous multicast receive filter if the regular one fails
Ignat Korchagin [Fri, 13 Dec 2019 16:17:58 +0000 (16:17 +0000)] 
[snp] Try promiscuous multicast receive filter if the regular one fails

Currently, if the SNP driver for whatever reason fails to enable
receive filters for multicast frames, it falls back to enabling just
unicast and broadcast filters.  This breaks some IPv6 functionality as
the network card does not respond to neighbour solicitation requests.

Some cards refuse to enable EFI_SIMPLE_NETWORK_RECEIVE_MULTICAST, but
do support enabling EFI_SIMPLE_NETWORK_RECEIVE_PROMISCUOUS_MULTICAST,
so try it before falling back to just unicast+broadcast.

Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>
Split-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 months ago[peerdist] Allow for the use of a hosted cache server
Michael Brown [Sun, 15 Dec 2019 23:26:02 +0000 (23:26 +0000)] 
[peerdist] Allow for the use of a hosted cache server

Allow a PeerDist hosted cache server to be specified via the
${peerhost} setting, e.g.:

  # Use 192.168.0.1 as hosted cache server
  set peerhost 192.168.0.1

Note that this simply treats the hosted cache server as a permanently
discovered peer for all segments.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
7 months ago[peerdist] Allow PeerDist to be globally enabled or disabled
Michael Brown [Fri, 13 Dec 2019 14:44:22 +0000 (14:44 +0000)] 
[peerdist] Allow PeerDist to be globally enabled or disabled

Allow the use of PeerDist content encoding to be enabled or disabled
via the ${peerdist} setting, e.g.:

  # Disable PeerDist
  set peerdist 0

Signed-off-by: Michael Brown <mcb30@ipxe.org>
10 months ago[lan78xx] Always enable automatic speed and duplex detection
Michael Brown [Sun, 29 Sep 2019 19:59:22 +0000 (20:59 +0100)] 
[lan78xx] Always enable automatic speed and duplex detection

On devices with no EEPROM or OTP, the MAC_CR register defaults to not
using automatic link speed detection, with the result that no packets
are successfully sent or received.

Fix by always enabling automatic speed and duplex detection, since
iPXE provides no mechanism for manual configuration of either link
speed or duplex.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
10 months ago[efi] Do not attempt EFI_USB_IO_PROTOCOL transfers during shutdown
Michael Brown [Sun, 15 Sep 2019 09:40:23 +0000 (10:40 +0100)] 
[efi] Do not attempt EFI_USB_IO_PROTOCOL transfers during shutdown

On at least some platforms (observed with a Raspberry Pi), any attempt
to perform USB transfers via EFI_USB_IO_PROTOCOL during EFI shutdown
will lock up the system.  This is quite probably due to the already
documented failure of all EFI timers when ExitBootServices() is
called: see e.g. commit 5cf5ffea2 "[efi] Work around temporal anomaly
encountered during ExitBootServices()".

Work around this problem by refusing to poll endpoints if shutdown is
in progress, and by immediately failing any attempts to enqueue new
transfers.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
10 months ago[efi] Report failed control transfers as expected by the USB core
Michael Brown [Sun, 15 Sep 2019 09:25:46 +0000 (10:25 +0100)] 
[efi] Report failed control transfers as expected by the USB core

The USB core reuses the I/O buffer space occupied by the USB setup
packet to hold the completion status for message transfers, assuming
that the message() method will always strip the setup packet before
returning.  This assumption is correct for all of the hardware
controller drivers (XHCI, EHCI, and UHCI), since these drivers are
able to enqueue the transfer as a separate action from waiting for the
transfer to complete.

The EFI_USB_IO_PROTOCOL does not allow us to separate actions in this
way: there is only a single blocking method that both enqueues and
waits for completion.  Our usbio driver therefore currently defers
stripping the setup packet until the control endpoint is polled.

This causes a bug if a message transfer is enqueued but never polled
and is subsequently cancelled, since the cancellation will be reported
with the I/O buffer still containing the setup packet.  This breaks
the assumption that the setup packet has been stripped, and triggers
an assertion failure in usb_control_complete().

Fix by always stripping the setup packet in usbio_endpoint_message(),
and adjusting usbio_control_poll() to match.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[golan] Fix address-of-pointer bug for multicast attach/detach
Michael Brown [Sat, 17 Aug 2019 16:51:18 +0000 (17:51 +0100)] 
[golan] Fix address-of-pointer bug for multicast attach/detach

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[ethernet] Avoid false positive Coverity warning
Michael Brown [Sat, 17 Aug 2019 16:30:09 +0000 (17:30 +0100)] 
[ethernet] Avoid false positive Coverity warning

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[coverity] Override assumptions about wcrtomb() and hmac_init()
Michael Brown [Sat, 17 Aug 2019 16:18:54 +0000 (17:18 +0100)] 
[coverity] Override assumptions about wcrtomb() and hmac_init()

Newer versions of Coverity use built-in models for wcrtomb() and
hmac_init() that are capable of returning errors, and reports defects
due to code failing to check for these errors.  The actual iPXE
implementations are simpler than Coverity's models and can never
return errors, so these defects are false positives.

Fix by overriding Coverity's built-in models for these functions.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[crypto] Profile the various stages of modular multiplication
Michael Brown [Sat, 17 Aug 2019 00:24:04 +0000 (01:24 +0100)] 
[crypto] Profile the various stages of modular multiplication

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[crypto] Drag in configured digestInfo prefixes for any use of RSA
Michael Brown [Sat, 17 Aug 2019 00:18:34 +0000 (01:18 +0100)] 
[crypto] Drag in configured digestInfo prefixes for any use of RSA

Ensure that the configured RSA digestInfo prefixes are included in any
build that includes rsa.o (rather than relying on x509.o or tls.o also
being present in the final binary).

This allows the RSA self-tests to be run in isolation.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[tls] Add missing call to tls_tx_resume() when restarting negotiation
Michael Brown [Fri, 16 Aug 2019 21:40:19 +0000 (22:40 +0100)] 
[tls] Add missing call to tls_tx_resume() when restarting negotiation

The restart of negotiation triggered by a HelloRequest currently does
not call tls_tx_resume() and so may end up leaving the connection in
an idle state in which the pending ClientHello is never sent.

Fix by calling tls_tx_resume() as part of tls_restart(), since the
call to tls_tx_resume() logically belongs alongside the code that sets
bits in tls->tx_pending.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[peerdist] Limit number of concurrent raw block downloads
Michael Brown [Fri, 16 Aug 2019 20:42:49 +0000 (21:42 +0100)] 
[peerdist] Limit number of concurrent raw block downloads

Raw block downloads are expensive if the origin server uses HTTPS,
since each concurrent download will require local TLS resources
(including potentially large received encrypted data buffers).

Raw block downloads may also be prohibitively slow to initiate when
the origin server is using HTTPS and client certificates.  Origin
servers for PeerDist downloads are likely to be running IIS, which has
a bug that breaks session resumption and requires each connection to
go through the full client certificate verification.

Limit the total number of concurrent raw block downloads to ameliorate
these problems.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[peerdist] Start block download timers from within opener methods
Michael Brown [Fri, 16 Aug 2019 20:23:55 +0000 (21:23 +0100)] 
[peerdist] Start block download timers from within opener methods

Move the responsibility for starting the block download timers from
peerblk_expired() to peerblk_raw_open() and peerblk_retrieval_open(),
in preparation for adding the ability to defer calls to
peerblk_raw_open() via a block download queue.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
11 months ago[process] Add PROC_INIT() for initialising static processes
Michael Brown [Fri, 16 Aug 2019 20:20:05 +0000 (21:20 +0100)] 
[process] Add PROC_INIT() for initialising static processes

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Add predefined shortcut for Raspberry Pi builds
Michael Brown [Fri, 2 Aug 2019 10:57:35 +0000 (11:57 +0100)] 
[build] Add predefined shortcut for Raspberry Pi builds

Add a build shortcut "rpi", allowing for e.g.

  make CONFIG=rpi CROSS=aarch64-linux-gnu- bin-arm64-efi/rpi.efi

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Move predefined all-drivers build shortcut to Makefile
Michael Brown [Fri, 2 Aug 2019 10:00:43 +0000 (11:00 +0100)] 
[build] Move predefined all-drivers build shortcut to Makefile

The (very approximate) split between Makefile.housekeeping and
Makefile is that the former provides mechanism and the latter provides
policy.

Provide a section within Makefile as a home for predefined build
shortcuts such as the existing all-drivers build.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Do not apply WORKAROUND_CFLAGS for host compiler
Michael Brown [Mon, 22 Jul 2019 13:51:28 +0000 (14:51 +0100)] 
[build] Do not apply WORKAROUND_CFLAGS for host compiler

The WORKAROUND_CFLAGS list is constructed based on running tests on
the target compiler, and the results may not be valid for the host
compiler.

The only relevant workaround required for the host compiler is
-Wno-stringop-truncation, which is needed to avoid a spurious compiler
warning for a totally correct usage of strncpy() in util/elf2efi.c.

Duplicating the workaround tests for the host compiler is messy, as is
conditionally applying __attribute__((nonstring)).  Fix instead by
disapplying WORKAROUND_CFLAGS for the host compiler, and using
memcpy() with an explicitly calculated length instead of strncpy() in
util/elf2efi.c.

Reported-by: Ignat Korchagin <ignat@cloudflare.com>
Reported-by: Christopher Clark <christopher.w.clark@gmail.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Workaround compilation error with gcc 9.1
Valentine Barshak [Mon, 22 Jul 2019 09:47:50 +0000 (10:47 +0100)] 
[build] Workaround compilation error with gcc 9.1

Compiling with gcc 9.1 generates lots of "taking address of packed
member of ... may result in an unaligned pointer value" warnings.

Some of these warnings are genuine, and indicate correctly that parts
of iPXE currently require the CPU (or runtime environment) to support
unaligned accesses.  For example: the TCP/IP receive data path will
attempt to access 32-bit fields that may not be aligned to a 32-bit
boundary.

Other warnings are either spurious (such as when the pointer is to a
variable-length byte array, which can have no alignment requirement
anyway) or unhelpful (such as when the pointer is used solely to
provide a debug colour value for the DBGC() macro).

There appears to be no easy way to silence the spurious warnings.
Since the ability to perform unaligned accesses is already a
requirement for iPXE, work around the problem by silencing this class
of warnings.

Signed-off-by: Valentine Barshak <gvaxon@gmail.com>
Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Fix "'%s' directive argument is null" error
Valentine Barshak [Sun, 9 Jun 2019 10:30:11 +0000 (13:30 +0300)] 
[build] Fix "'%s' directive argument is null" error

Use '%p' directive, and print handle's address if the address is null
and the handle doesn't have a name.  This fixes the following
compilation error:

  interface/efi/efi_debug.c:334:3: error: '%s' directive
  argument is null [-Werror=format-overflow=]

Signed-off-by: Valentine Barshak <gvaxon@gmail.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[smscusb] Fetch MAC from device tree for Raspberry Pi Model B+
Michael Brown [Fri, 19 Jul 2019 18:15:33 +0000 (19:15 +0100)] 
[smscusb] Fetch MAC from device tree for Raspberry Pi Model B+

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Add named configuration for Raspberry Pi
Michael Brown [Fri, 19 Jul 2019 16:45:22 +0000 (17:45 +0100)] 
[build] Add named configuration for Raspberry Pi

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[smsc95xx] Fetch MAC from device tree for Raspberry Pi
Michael Brown [Fri, 19 Jul 2019 16:43:39 +0000 (17:43 +0100)] 
[smsc95xx] Fetch MAC from device tree for Raspberry Pi

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[efi] Register a device tree if provided by the platform firmware
Michael Brown [Fri, 19 Jul 2019 16:42:12 +0000 (17:42 +0100)] 
[efi] Register a device tree if provided by the platform firmware

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[fdt] Add ability to parse a MAC address from a flattened device tree
Michael Brown [Fri, 19 Jul 2019 16:35:39 +0000 (17:35 +0100)] 
[fdt] Add ability to parse a MAC address from a flattened device tree

The Raspberry Pi NIC has no EEPROM to hold the MAC address.  The
platform firmware (e.g. UEFI or U-Boot) will typically obtain the MAC
address from the VideoCore firmware and add it to the device tree,
which is then made available to subsequent programs such as iPXE or
the Linux kernel.

Add the ability to parse a flattened device tree and to extract the
MAC address.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[efi] Return only registered EFI devices from efidev_parent()
Michael Brown [Mon, 15 Jul 2019 11:49:47 +0000 (12:49 +0100)] 
[efi] Return only registered EFI devices from efidev_parent()

efidev_parent() currently assumes that any device with BUS_TYPE_EFI is
part of a struct efi_device.  This assumption is not valid, since the
code in efi_device_info() may also create a device with BUS_TYPE_EFI.

Fix by searching through the list of registered EFI devices when
looking for a match, instead of relying on the bus type value.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[arm] Provide dummy implementations for {in,out}[s]{b,w,l}
Michael Brown [Sun, 14 Jul 2019 14:27:01 +0000 (15:27 +0100)] 
[arm] Provide dummy implementations for {in,out}[s]{b,w,l}

It is currently not possible to build the all-drivers iPXE binaries
for ARM, since there is no implementation for inb(), outb(), etc.

There is no common standard for accessing I/O space on ARM platforms,
and there are almost no ARM-compatible peripherals that actually
require I/O space accesses.

Provide dummy implementations that behave as though no device is
present (i.e. ignore writes, return all bits high for reads).  This is
sufficient to allow the all-drivers binaries to link, and should cause
drivers to behave as though no I/O space peripherals are present in
the system.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
12 months ago[build] Fix use of inline assembly on GCC 8 ARM64 builds
Michael Brown [Sun, 14 Jul 2019 13:05:48 +0000 (14:05 +0100)] 
[build] Fix use of inline assembly on GCC 8 ARM64 builds

Commit 1a7746603 ("[build] Fix use of inline assembly on GCC 4.8 ARM64
builds") switched from using "%c0" to "%a0" in order to avoid an
"invalid operand prefix" error on the ARM64 version of GCC 4.8.

It appears that the ARM64 version of GCC 8 now produces an "invalid
address mode" error for the "%a0" form, but is happy with the original
"%c0" form.

Switch back to using the "%c0" form, on the assumption that the
requirement for "%a0" was a temporary aberration.

Originally-fixed-by: John L. Jolly <jjolly@suse.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
15 months ago[golan] Add various new PCI device IDs
Mohammed [Thu, 2 May 2019 10:00:18 +0000 (11:00 +0100)] 
[golan] Add various new PCI device IDs

Signed-off-by: Mohammed <mohammedt@mellanox.com>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
15 months ago[intelxl] Add driver for Intel 40 Gigabit Ethernet NIC virtual functions
Michael Brown [Wed, 24 Apr 2019 16:15:49 +0000 (17:15 +0100)] 
[intelxl] Add driver for Intel 40 Gigabit Ethernet NIC virtual functions

Signed-off-by: Michael Brown <mcb30@ipxe.org>