If there is more than one loaded image, refuse to automatically select
the image to execute.  There are at least two possible cases, with
different "correct" answers:

1. User loads image A by mistake, then loads image B and types "boot".
   User wants to execute image B.

2. User loads image A, then loads image B (which patches image A), then
   types "boot".  User wants to execute image A.

If a user actually wants to load multiple images, they must explicitly
specify which image is to be executed.

Clearing the LOADED flag actually prevents users from doing clever things
such as loading an image, then loading a patch image, then executing the
first image.  (image_exec() checks for IMAGE_LOADED, so this sequence of
operations will fail if the LOADED flag gets cleared.)

This reverts commit 14c080020f.

Loading an image may overwrite part or all of any previously-loaded
images, so we should clear the LOADED flag for all images prior to
attempting to load a new image.

We can just treat all non-kernel images as initrds, which matches our
behaviour for multiboot kernels.  This allows us to eliminate initrd as
an image type, and treat the "initrd" command as just another synonym for
"imgfetch".

dhcppkt_store() is supposed to clear the setting if passed NULL for the
setting data.  In the case of fixed-location fields (e.g. client IP
address), this requires setting the content of the field to all-zeros.

__from_data16 and __from_text16 now take a pointer to a
.data16/.text16 variable, and return the real-mode offset within the
appropriate segment.  This matches the use case for every occurrence
of these macros, and prevents potential future bugs such as that fixed
in commit d51d80f.  (The bug arose essentially because "&pointer" is
still syntactically valid.)

__from_data16 takes the value pointed to, rather than the pointer
itself.  This was silently causing gPXE to return a dud buffer pointer
when the caller did not supply a buffer for PXENV_GET_CACHED_INFO.

Perform the same test for a matching DHCP_SERVER_IDENTIFIER on
ProxyDHCPACKs as we do for DHCPACKs.  Otherwise, a retransmitted
DHCPACK can end up being treated as the ProxyDHCPACK.

I have a vague and unsettling memory that this test was deliberately
omitted, but I can't remember why, and can't find anything in the VC
logs.

Remove src/Config as it has no more users, and conflicts with
src/config on case-deficient filesystems.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

ns8390.c can produce four different drivers (one PCI, three ISA.)  The
ISA driver requires setting a few macros; do that by setting defines
in stub files instead of using src/Config.

Currently, all the ISA drivers are broken (they were not enabled by
default), so #if 0 them out.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

zImage kernels require the real-mode portion to be loaded at 9000:0000
rather than 1000:0000.

When the 16-bit segment registers are accessed using 32-bit instructions
the high order bytes are undefined on older CPUs.  We now explicitly
zero the high order bytes when snapshotting the CPU state.  This ensures
that the GDB stub reports consistent values for the segment registers.

This commit implements GDB over UDP.  Using UDP is more complex than
serial and has required some restructuring.

The GDB stub is now built using one or both of GDBSERIAL and GDBUDP
config.h options.

To enter the debugger, execute the gPXE shell command:
gdbstub <transport> [<options>...]

Where <transport> is "serial" or "udp".  For "udp", the name of a
configured network device is required:
gdbstub udp net0

The GDB stub listens on UDP port 43770 by default.

This commit also includes a test to ensure that single stepping works,
since continue, kill, detach, and single step all share code.

Commit fd0aef9 introduced a typo that caused PMM detection to start at
paragraph 0xe00 rather than 0xe000.  (Detection would still work, since it
would scan until it ran out of base memory, but it would end up scanning
an unnecessarily large portion of base memory.)

Spotted by Sebastian Herbszt <herbszt@gmx.de>.

OpenBSD throws compiler warnings that we can't reproduce on Linux, for
some reason.

Original patch from Dewey Hylton <dewey@hyltown.com>.

Send a null command, specifically "pulse outputs" with no outputs
selected, to the KBC after changing A20.  This was apparently done by DOS,
presumably as a synchronization hack, and the authors of the UHCI spec
thought it was inherent.  Therefore, there are systems out there (e.g. HP
DL360 G5) which will stop responsing to "legacy USB" unless they see the
null command, 0xFF, written to port 0x64 at the end of the A20 toggling
sequence.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>

If we ever need these macros, they should be defined in
include/{big,little}_bswap.h, and the whole family should be defined.

Add an entry to tg3_nics for the BCM5789

Signed-off-by: Steven Walter <stevenrwalter@gmail.com>

This fixes a bug introduced in commit 4c85017.

Some drivers that still use the legacy-driver wrapper (tg3 in particular)
apparently do not specify their alignment constraints properly.  This
hack forces any __shared data to be maximally aligned.

Note that this provides only 16-byte alignment; it is not possible to
request alignment to any greater than 16 bytes using
__attribute__((aligned)), since the relocation code will preserve only 16
byte alignment (and operation under -DKEEP_IT_REAL cannot preserve more
that 16 byte alignment).

Idea proposed by Tim Hockin <thockin@google.com>

When the BIOS doesn't support BBS, hooking INT 19 is the only way to add
ourselves as a boot device.  If we have to do this, we should at least
try to chain to the original INT 19 vector if our boot fails.

Idea suggested by Andrew Schran <aschran@google.com>

2.6.22+ kernels have an extra field in the bzimage_header structure to
indicate the maximum permitted command-line length.  Use this if it is
available.

A bug in read_smbios_string() was causing the starting offset of the
SMBIOS structure to be added twice, resulting in completely the wrong
strings being returned.

Bug identified by Martin Herweg <m.herweg@gmx.de>

Add the definition of SLAM_MAX_BLOCKS_PER_NACK, which is roughly
equivalent to a TCP window size; it represents the maximum number of
packets that will be requested in a single NACK.

Note that, to keep the code size down, we still limit ourselves to
requesting only a single range per NACK; if the missing-block list is
discontiguous then we may request fewer than SLAM_MAX_BLOCKS_PER_NACK
blocks.

Avoid calling cpu_nap() until after we have determined that there is
no input ready to read.  This avoids delaying for one timer interrupt
(~50ms) in the case of

  if ( iskey() )
     char = getkey()

which happens to be present in monojob.c, which is where we spend most
of our time looping (e.g. during any download).

This should eliminate the irritating tendency of gPXE to lose
keypresses.

Discovered on a Dell system where the serial port seems to send in a
constant stream of 0xff characters; this wouldn't be a problem in
itself except that each one triggers the 50ms delay (as mentioned
above), which really kills performance.

On any fast network, or with any driver that may drop packets
(e.g. Infiniband, which has very small RX rings), the traditional
usage of the SLAM protocol will result in enormous numbers of packet
drops and a consequent large number of retransmissions.

By adapting the client behaviour, we can force the server to act more
like a multicast TFTP server, with flow control provided by a single
master client.

This behaviour should interoperate with any traditional SLAM client
(e.g. Etherboot 5.4) on the network.  The SLAM protocol isn't actually
documented anywhere, so it's hard to define either behaviour as
compliant or otherwise.

Allow for timers to be started with fixed timeouts.

A missing test for dhcp->dhcpoffer in dhcp_timer_expired() was causing
the client to transition to DHCPREQUEST after timing out on waiting
for ProxyDHCP even if no DHCPOFFERs had been received.

In a SLAM NACK packet, if we run out of space to represent the
missing-block list, then indicate all remaining blocks as missing.
This avoids the need to wait for the one-second timeout before
receiving the blocks that otherwise wouldn't have been requested due
to running out of space.

Shorter NACK packets take less time to construct and spew out less
debug output, and there's a limit to how useful it is to send a
complete missing-block list anyway; if the loss rate is high then
we're going to have to retransmit an updated missing-block list
anyway.

Also add pretty debugging output to show the list of requested blocks.