Pass in 'true' if you'd like to search this device's partitions or
'false' if you should just search the device. EFI and (in the future)
kboot have discrete partitions that aren't accessed via the full disk
device. Weird things happen if you try to search in these cases.
Sponsored by: Netflix
To support more flexible device matching, we now pass in the full
devspec to the parsedev routines. For everything execpt uboot, this is
just a drop in (since everything except uboot and openfirmware always
uses disk...: and/or zfs:, but openfirmware isn't really affected).
uboot we kludge around it by subtracting 4 from where the rest of the
device name starts. This is unforunate, and can compute the address one
before the string. But we never dereference that address. uboot needs
more work, and this is an acceptable UB until that other work happens.
OFW doesn't really use the parsedev routines these days (since none of
the supported device uses this... yet). It too needs more work, but it
needs device matching support first.
Sponsored by: Netflix
Reviewed by: delphij
Differential Revision: https://reviews.freebsd.org/D37553
devinit() marches through all the devices, calling the inint routines if
any exist. Replace all the identical copies of this code.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D37349
We no longer need the zfs stubs since we're no longer referencing these
functions outside of zfs.c.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D37345
We no longer need to have to hand-code this for each boot loader since
devparse() handles them all with dv_parsedev().
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D37341
Change the first argument to zfs_parsedev() to be a pointer to a struct
devdesc *. This now gets filled in with a malloc'd structure that's
returned to the caller that the caller is repsonsible for freeing. Most
nplaces in the tree passed in a malloc'd pointer anyway, and this moves
knowledge of zfs_devdesc more firmly into the zfs.c code.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D37336
Change the first argument to disk_parsedev() to be a pointer to a struct
devdesc *. This now gets filled in with a malloc'd structure that's
returned to the caller that the caller is repsonsible for freeing. Most
places in the tree passed in a malloc'd pointer anyway, and this moves
knowledge of disk_devdesc more firmly into the disk.[ch] code.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D37335
Remove support for booting off of firewire, and for having dcons via
firewire in the loader. Kernel support for these things is unchanged.
Discussed on arch@ and the current state is not working (and the build
was wrong to boot).
Sponsored by: Netflix
Discussed: https://lists.freebsd.org/archives/freebsd-arch/2022-November/000267.html
Reviewed by: kevans, melifaro, emaste
Differential Revision: https://reviews.freebsd.org/D37334
This saves 80 bytes (the new bootinfo structure was 84 bytes, and a
pointer is 4 bytes). The bi_load32 code is the same size.
Sponsored by: Netflix
Reviewed by: tsoome
Differential Revision: https://reviews.freebsd.org/D36575
FreeBSD 4.x and earlier used the bi_bios_geom to get the geometry of the
device. Starting in 5.x, with the wdc -> ata rewrite, it was used only
in pc98 kernels to report geometry of the drives. It can be safely
removed as booting kernels this old is no longer supported. This saves
176 bytes in the BIOS loader.
Sponsored by: Netflix
Reviewed by: adrian, emaste
Differential Revision: https://reviews.freebsd.org/D36543
We set this value twice: once to 0 and once to the VA that has the name
of the kernel. The first store is redundant. In addition, these two
stores of 0 are also redundant. Since we never set them, they will
always be zero, even if we're called multiple times. This saves 21
bytes on BIOS loader.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D36541
Use the efi's bi_copyenv to md_copyenv and place it in modinfo.c. Remove
all other nearly identical and efi's has the best error handling.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D36574
md_copymodules, bi_copymdoules, bi_copymodules32 (x2) and
bi_copymodules64 (x2) are all the same routine... Replace them all with
md_copymodules. This saves about 800 bytes on i386 BIOS loader, which is
a nice bonus.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D36572
Move all the MOD_xxx macros to this header. Each user of this interface
is currently required to define MOD_ALIGNMENT(l). modinfo was selected
because it sits inbetween modules and metadata and will make it easier
to migrate to new, shared intefaces.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D36571
To further reduce the differences between the different MOD_xxx macros,
use MOD_ALIGN to do the proper alignment for the given use.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D36570
Since archsw.arch_copyin is always i386_copyin, this will be a nop in
terms of functionality. This is a diff reduction against other copies of
the code that differ only by what copyin routine they call.
Sponsored by: Netflix
Reviewed by: tsoome
Differential Revision: https://reviews.freebsd.org/D36568
disk_blocks assumes BIOSDISK_SECSIZE, but the media may not be using
it. In particular, bioscd on Parallels presents a 2K sector size, so
we end up with a short disk_blocks and subsequent validation fails when
trying to read /boot/lua.
PR: 233098
Reviewed by: imp, tsoome
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D36490
And thus has a limited range of supported baud rates. Also add that
setting BOOT_BOOT0_COMCONSOLE_SPEED=0 will leave it unchanged which
sometimes can give you 115200 if the BIOS initialized things outside of
the normal BIOS baud rates (which many x86 enbedded-targetted boards
do).
Sponsored by: Netflix
Reviewed by: emaste, manu (earlier versions)
Suggestions by: jhb
Differential Revision: https://reviews.freebsd.org/D36300
The commit message documented it as /etc/src.conf but the comment in the
source mentioned the non-existent /etc/loader.conf.
Fixes: f8a199f28f ("stand: Raise limit to 550,000 bytes for loader")
Raise the limit for /boot/loader to be 550k. The IBM PC imposes a limit
of 640k of RAM below 1MB, which is needed for real mode calls. BTX takes
40k of that. The BIOS takes some amount (25k seems a good "99% take less
than or equal to this" estimate for that, though some systems consume
more). Most typical setups need 25k of stack. This leaves 550k for
code. We set the limit to 550,000 which gives about an extra 13,000
bytes of buffer for machines that whose setups use a little more stack
or whose BIOS reserves a bit more...
Add this derivation in the Makefile. Also recommend setting LOADERSIZE
lower in /etc/src.conf when the loader has to run on a system whose BIOS
takes up more space, or for a complex setup. Add a recipe for how to
find how much RAM your BIOS uses as well (thanks to jhb@ for the
trick). Network cards that boot via PXE and HBAs with their BIOS enabled
are known to be large consumers of lomem space.
Sponsored by: Netflix
Reviewed by: jhb
Differential Revision: https://reviews.freebsd.org/D36152
Turns out there's two hidden a.out dependencies. pxeldr.S assumes it has
access to the a.out header from /boot/loader and cdboot.S assumes that
/boot/loader is also a.out and doesn't use boot2.
So, go back to making a.out files for these and adjust the size checks
to use ls, but we only need to check loader.bin. Trim the size we check
against by 2,000. The difference in size between loader and loader.bin
is about 3000 bytes, but clang15 produces binaries that are a smidge
bigger so we need to relax the check just a little and accept some
additional risk for the moment.
Add some comments to loader's Makefile about this.
Sponsored by: Netflix
Reviewed by: emaste
Differential Revision: https://reviews.freebsd.org/D36142
It's sometimes desirable to override the size limit: It's a soft limit
and there are times we exceed the limit by just a little bit and don't
want the build to fail (or we are hitting runtime failures below the
510,000 byte limit).
Sponsored by: Netflix
devformat produces the same output as i386_fmtdev, so just use it to
reduce on the dependencies.
Sponsored by: Netflix
Reviewed by: tsoome
Differential Revision: https://reviews.freebsd.org/D35927
All of the archsw fmtdev functions treat DEVT_DISK as a call to
disk_fmtdev. Set all disks' dv_fmtdev to disk_fmtdev so devformat
will return the same thing.
Sponsored by: Netflix
Reviewed by: tsoome (prior version)
Differential Revision: https://reviews.freebsd.org/D35917
Rather than have the magic, hand-crafted fields that have to align with
fields in other structures at the end of i386_devdesc, make it into
anonymous union and adjust the code accordingly. This is safer and
similar to what CAM does.
Sponsored by: Netflix
Reviewed by: kevans, tsoome (prior version)
Differential Revision: https://reviews.freebsd.org/D35965
The BIOS method of booting imposes an absolute limit of 640k for the
size of the program being run due to btx. In practice, this means that
programs larger than about 500kiB will fail in odd ways as the stack /
heap will overflow.
Pick 510,000 as the cutoff line semi-arbitrarily. loader_lua is now
almost too big and we want to break the build when it crosses this
threshold. In my experience, below 500,000 always works, above 520,000
always seems to fail with things getting bad somewhere between 512,000
to 515,000. 510,000 is as close to the line as I think we can go, though
experience may dictate we need to lower this in the future.
This is at-best a stop-breakage until we have a better way to subset the
boot loader for BIOS booting to allow better, more fined-tuned
/boot/loaders for the many different environments they have to run
in. This likely means we'll have a graphical loader than understands a
few filesystmes for installation, and a non-graphical loader that
understands the most filesystems possible for everything else in the
future. Our build infrastructure needs some work before we can do that,
however.
At this late date, it likely isn't worth the efforts to move parts of
the loader into high memory. There's a number of assumptions about where
the stack is, where buffers reside, etc that are fulfilled when it lives
in the first 640k that would need bounce buffers and/or other counter
measures if we were to split it up. All BIOS calls are done in 16-bit
mode with SEG:OFF addresses, requiring them to be in the first 640k of
RAM. And nearly all machines in the last decade can boot with UEFI
(though there's some exceptions, so it isn't worth killing outright
yet).
Sponsored by: Netflix
Reviewed by: kevans
Differential Revision: https://reviews.freebsd.org/D36129
The first level boot blocks have understood how to load ELF code since
1999. Switch /boot/loader and /boot/pxeldr over to being ELF format so
that in-tree tools can examine them more closely. In addition, one
could, in theory, now have a 'lo-mem' and a 'hi-mem' segment (though a
lot of work would need to be done with bounce buffers, btx, code segment
marking, etc for an arrangement like that to work).
As far as I can tell, this is the last a.out binary in the tree. There
are several raw binaries left, but everything else is ELF.
Reviewed by: emaste, kevans
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D36130
dv_cleanup is specified almost everywhere. Use nullsys instead of NULL
to indicate 'do nothing'. Also, be consistent in trailing commas that
were missing before.
Sponsored by: Netflix
Reviewed by: tsoome
Differential Revision: https://reviews.freebsd.org/D35913
With clang 15, the following -Werror warning is produced:
stand/i386/boot2/boot2.c:358:6: error: a function declaration without a prototype is deprecated in all versions of C [-Werror,-Wstrict-prototypes]
This is because parse() is declared with a (void) argument list, and
defined with an empty argument list. Make the definition match the
declaration.
MFC after: 3 days
The console escape code standard (ECMA-48) specifies color #3 (escape
code 33) as yellow. A brown console color is an artifact of the VGA
palette, which replaces dim (but not bright) yellow with brown.
Reviewed by: adrian, imp
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D34531
For historical reasons console color number 3 may be either yellow (most
consoles) or brown (VGA palette). The console escape code standard
uses "yellow", but teken color name constants appear to be based on the
VGA scheme and use TC_BROWN for color 3. Even so, the palette table
used 50,50,0 as the RGB percentage tuple, resulting in a dim yellow for
framebuffer consoles at the time teken was introduced.
Amusingly, in 19e2ce2d83 the comment on the palette entry was changed
from "brown" to "dark yellow" but the colour itself was changed from
a pure yellow to being somewhat brown.
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Accept "bright" or "light" prefix for named colors.
For numeric colors, update error message to specify that values 0 to 15
are allowed, and verify that values are in that range.
Reviewed by: imp, tsoome (both earlier version)
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D34512
The text after .error et al is emitted verbatim.
Reviewed by: sjg
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D33904
Store whether or not we found a vbefb module (eg, a tg supported kernel)
in the preloaded_file structure. This automatically resets on reload and
eliminates load_elf knowing about any gfx_* interface. Restrict this to
i386, which is the only place it's used. Update libi386 to check in the
preloaded_file struct. Eliminate this from the teken_gfx
structure. Rewrite the parsing code to be more inline. Check this from
the same place we check for a relocatable amd64 kernel.
Sponsored by: Netflix
Reviewed by: manu, tsoome
Differential Revision: https://reviews.freebsd.org/D33427
Reading from Write Combining memory can be very-very slow. Try to use
shadow buffer to avoid such reads.
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D33305
We want to keep our root file system open to preserve bcache segment
between file accesses, thus reducing physical disk IO.
Reviewed by: imp, allanjude, kevans (previous version)
Differential Revision: https://reviews.freebsd.org/D30848
MFC after: 1 month
Because lld 13 and higher default to garbage collecting start/stop
symbols when using --gc-sections, the linker sets used in the i386 boot
loaders will disappear. This leads to the loaders not recognizing any
commands, and failure to boot.
Until we have a good set of linker scripts for the loaders, work around
it by disabling the start-stop-gc feature.
MFC after: 1 week
This pushes the bulk of the rx servicing into a single loop that's only
slightly convoluted, and it addresses a problem with rx handling in the
process. If we hit a tx interrupt while we're processing, we'd
previously drop the frame on the floor completely and ultimately
timeout, increasing boot time on particularly busy hosts as we keep
having to backoff and resend.
After this patch, we don't seem to hit timeouts at all on zoo anymore
though loading a 27M kernel is still relatively slow (~1m20s).
Reviewed by: tsoome
Triage by: Ash Gokhale <ashfixit gmail com>
Sponsored By: National Bureau of Economic Research
Sponsored by: Klara, Inc.
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D31512
