Originally written: 3/14/2012; last Web page update:
-11/6/2012, referencing rEFInd 0.4.7
+4/24/2016, referencing rEFInd 0.10.3
-
I'm a technical writer and consultant specializing in Linux technologies. This Web page is provided free of charge and with no annoying outside ads; however, I did take time to prepare it, and Web hosting does cost money. If you find this Web page useful, please consider making a small donation to help keep this site up and running. Thanks!
+
This Web page is provided free of charge and with no annoying outside ads; however, I did take time to prepare it, and Web hosting does cost money. If you find this Web page useful, please consider making a small donation to help keep this site up and running. Thanks!
+
@@ -110,12 +144,6 @@ href="mailto:rodsmith@rodsbooks.com">rodsmith@rodsbooks.com
systems with a limited number of configurations. Therefore, if you
try rEFInd and run into bugs, please report them to me!
-
I have little talent with graphics manipulation programs, so
- rEFInd's boot logo, such as it is, is pretty weak. If you have
- artistic talent and would like to create a rEFInd logo, please feel
- free to send it to me. I won't make any final decision about
- changes until at least June 30 of 2012.
-
rEFIt's original design, and hence rEFInd's design, enables easy
theming by replacing icon files. If you'd like to design a new
theme for rEFInd, feel free to submit it. I might or might not
@@ -131,6 +159,12 @@ href="mailto:rodsmith@rodsbooks.com">rodsmith@rodsbooks.com
+
As described in reference to version 0.9.2 on the Revisions page, rEFInd includes a
+ delicate and hackish workaround to a problem introduced by Shim
+ 0.8. Developing a better solution to that problem is a high
+ priority.
+
The support for booting legacy (BIOS) OSes on UEFI-based PCs
currently has a number of limitations. Most importantly, it works
off of the list of boot devices stored in the computer's NVRAM. I'd
@@ -139,64 +173,85 @@ href="mailto:rodsmith@rodsbooks.com">rodsmith@rodsbooks.com
optical drives and uses generic icons rather than OS-specific
icons.
-
Currently, rEFInd can detect whether it's compiled for x86
- or x86-64 systems and displays this information in its
- "About" screen (AboutrEFInd() in main.c). I'd
- like to add detection for Itanium and ARM systems, but I have no
- way to test such changes.
-
-
The code could be more flexible in its handling of the sizes of
- various graphical elements, and particularly drawn text. Prior to
- version 0.2.2, submenu text was invisible on UEFI-based PCs with
- 800x600 and smaller displays because of an inability to properly
- crop the graphics fields that hold the text. With version 0.2.2,
- I've put a band-aid on this problem by reducing the field size so
- that it now works on 800x600 displays, but smaller displays still
- suffer from this problem. This is just an example of the
- inflexibility of certain layout issues within rEFInd.
-
-
Although the ICNS file format used by rEFInd supports multiple
- image sizes, if a size that rEFInd needs isn't present in the file,
- rEFInd can't use the icon. The ability to scale images to the
- desired size would be useful.
-
-
I would like to be able to specify the volume on which a boot
- loader resides using a partition GUID value, but extracting a GUID
- from the partition data is harder than extracting the volume's
- label or counting up the filesystem numbers.
-
-
The default_selection option in refind.conf could be improved by
- supporting a list of default options, so that if the first item
- isn't found, rEFInd will try to boot the second one in the list,
- and so on. This could be handy in case a driver fails to load, or
- to provide an override in case the user inserts a specific
- removable disk—by placing the removable disk's name first in
- the list, it will take precedence over the normal hard disk
- default.
-
-
Along the lines of the previous item, the default_selection might
- be expanded to support some form of specification of disk types, as
- in a special entry for any optical disk or any external disk, no
- matter what its name is.
+
rEFInd's Makefiles and, to a lesser extent, C code,
+ support x86, x86-64, and ARM64 CPUs. EFI is also
+ available for Itanium (IA-64) and ARM32 CPUs, so I'd like to add
+ this support.
+
+
Currently, rEFInd can detect whether it's compiled for x86,
+ x86-64, or ARM64 systems and displays this information in
+ its "About" screen (AboutrEFInd() in main.c). I'd
+ like to add detection for Itanium and 32-bit ARM systems, but I
+ have no way to test such changes.
+
+
Further to the preceding, rEFInd's GPT-scanning code (used to
+ extract partition names) includes assumptions about byte order, and
+ so will work only on little-endian CPUs such as the x86 and
+ x86-64.
+
+
A way to set the color of the font would be useful for theming
+ purposes.
+
+
The program's font features could be greatly improved by enabling
+ use of a standard font format, by enabling use of non-ASCII
+ characters, and by enabling use of variable-width as well as
+ monospace fonts.
+
+
The default_selection might be expanded to support some
+ form of specification of disk types, as in a special entry for any
+ optical disk or any external disk, no matter what its name is.
It would be useful to be able to specify paths to boot loaders
and/or initial RAM disks relative to the rEFInd directory (or the
boot loader's directory, in the case of initrds).
-
Various options (dont_scan_dirs, also_scan_dirs,
- scan_driver_dirs, etc.) refer to directories or files,
- either on the ESP or on all partitions. A way to identify specific
- partitions for these options would be useful in some
- situations.
+
Currently the background for certain subscreens (such as the
+ information page or submenu listings) is a solid color based on the
+ upper-left corner of the screen. Having an option to support a
+ transparent background is desirable to some users.
+
+
When delivering rEFInd as a boot loader from a network server,
+ rEFInd is limited to its default options and can boot only local
+ OSes, not network OSes. The cause is that the server delivers a
+ single file, so rEFInd is divorced from its configuration and
+ support files.
-
+
A way to identify specific Windows versions and present unique
+ icons or change the text is desirable. Currently, a crude
+ distinction of XP and earlier vs. Vista and later is possible for
+ BIOS-booting on Macs, but no such distinction is made for EFI-mode
+ booting, and nothing finer-grained is attempted. Improvements will
+ probably require identifying unique features of each version's boot
+ loader files or boot sector code.
I've been receiving reports of blank screens when using rEFInd on
+ some recent Mac models. I've investigated this with the help of one
+ user, and I suspect that Apple has made changes to its firmware
+ that are likely to affect just about any EFI program. I don't have
+ a definitive solution, but at least one user has reported that
+ removing rEFInd's drivers has caused the problem to go into
+ remission.
+
+
Another Mac-specific display problem relates to "retina" displays:
+ Some users report that rEFInd comes up in a lower resolution than
+ the screen supports, and that this setting persists into the running
+ OS X instance, and can't be adjusted using the usual OS X means.
+ Unfortunately, I lack the hardware necessary to experiment and find
+ a solution to this problem within rEFInd. Thus, a fix will have to
+ wait for me to get my hands on such hardware (which is not on my
+ immediate purchase list) or until somebody with such hardware and
+ the necessary skills submits a fix. (Note that a fix could
+ conceivably involve passing options to the OS X boot loader or
+ something else that would require trivial or no changes to
+ rEFInd.)
+
+
Some EFIs have bugs that cause the allegedly case-insensitive
StriCmp() function to perform a case-sensitive comparison.
This causes any number of bugs in file matching. For instance:
Changing the case of icon filename extensions (or various other
@@ -204,16 +259,12 @@ href="mailto:rodsmith@rodsbooks.com">rodsmith@rodsbooks.com
"generic" ones; and rEFInd sometimes appears in its own menu (the
firmware sometimes returns an all-caps version of the filename, but
other times returns the filename with the correct case, causing a
- mismatch if the path includes lowercase elements). Some of these
- problems can be overcome by converting both strings to be compared
- to one case before doing the comparison, but others aren't so easy,
- since I think StriCmp() is being called internally to the
- EFI. In any event, it'd be nice to fix some of these problems.
- OTOH, this is a workaround for a bug on just one EFI
- implementation, and a dismal one at that, so I'm inclined to just
- let it go.
-
-
The Shutdown option works correctly on Macs, but not on UEFI-based
+ mismatch if the path includes lowercase elements). This problem is
+ worse when compiling rEFInd with GNU-EFI than with Tianocore.
+ Version 0.9.1 has made improvements on this score, but some issues
+ may continue to lurk.
+
+
The Shutdown option works correctly on Macs, but not on many UEFI-based
PCs. On such systems, Shutdown reboots the computer. This should be
fixed.
@@ -222,6 +273,23 @@ href="mailto:rodsmith@rodsbooks.com">rodsmith@rodsbooks.com
Apple-specific EFI extension, and I know of no standard EFI way to
do it.
+
A couple of Mac users have reported that the brightness-adjustment
+ features in Windows don't work when Windows is booted via rEFInd,
+ but that these features do work when Windows is booted via the
+ Mac's built-in boot manager. Unfortunately, I have no idea what
+ causes this problem, I have no Windows installation on my one
+ (elderly) Mac, and I have no way to debug it. Therefore, it's
+ unlikely that I'll be able to fix this problem myself; but if you
+ have the equipment and skill to do so, I'd be interested in
+ receiving a patch.
+
+
If you use a true MBR disk on a Mac to boot Windows or some other
+ BIOS-only OS, and if that disk has an extended partition, bogus
+ additional BIOS/legacy-bootable options may appear in the rEFInd
+ menu. The reason appears to be a bug in the handling of
+ extended/logical partitions in the refind/lib.c file, but
+ I haven't fully tracked it down.
+
The re-scan feature occasionally produces odd results, such as
ignoring new media or keeping old media that have been ejected.
This should be investigated and fixed.
@@ -237,111 +305,155 @@ href="mailto:rodsmith@rodsbooks.com">rodsmith@rodsbooks.com
caused by a truncated DevicePath to the shell, which includes the
shell's pathname but not the device identifier.
+
When specifying a volume by name in dont_scan_dirs,
+ slashes are converted to backslashes in the specification but not
+ in the actual volume name read from disk. Thus, you can't specify a
+ volume by name if it includes a slash (as in Fedora
+ /boot). Workarounds are to rename the volume to omit the slash
+ and to use a filesystem number rather than a volume label.
+
The code is in need of review to search for memory leaks and
similar problems.
+
If the user has a Linux software RAID 1 array with Btrfs, HFS+,
+ or FAT filesystem, rEFInd will detect kernels or boot loaders in
+ RAID 1 twice. Checks to prevent this with ext2/3/4fs and ReiserFS
+ already exist; these checks could be expanded to block such
+ duplication with more filesystems.
+
+
Some Macs experience problems with waking up from suspend states
+ when rEFInd is installed. Unfortunately, I lack the hardware to
+ test and experiment with this (my only Intel-based Mac doesn't
+ exhibit this problem), so I can't fix this myself. Using
+ pmset to disable the autopoweroff option is
+ claimed
+ by some to at least partially fix the problem, though. Using
+ the --ownhfs installation option may also help in some
+ cases.
+
+
If you activate BIOS-mode support on UEFI-based PCs, you may find
+ multiple copies of the BIOS-mode loaders added to your firmware's
+ boot manager. Only one copy shows up in rEFInd, though.
+
New features I'd like to add:
-
With the arrival of PCs preloaded with Windows 8 and with Secure
- Boot enabled, some way to cope is in order. I'm thinking of adding
- code to limit or prohibit booting of unsigned boot loaders if
- rEFInd detects that Secure Boot is active, and link with the Shim
- pre-bootloader to help handle signing and authentication. I need to
- research the technical details more, though.
-
-
EFI supports network boots. rEFInd doesn't, but it would be nice if
- it would.
+
Currently, debugging rEFInd requires adding Print()
+ statements to the code. Adding a logging facility that supports
+ multiple log levels and writes the output to a file would help with
+ debugging, especially when dealing with problem reports from
+ users.
There's currently no way to create a manual boot stanza for a
BIOS-booted OS. This isn't a big priority for me personally, but I
can see how it could be for some people.
-
I've received queries about rEFInd's ability to work with Apple's
- whole-disk encryption scheme that's new with OS X 10.7.
- Unfortunately, I lack the hardware to test this, but my
- understanding is that it will work correctly if rEFInd is
- installed in the ESP rather than on the Mac OS X root partition.
- See this
- forum thread for more information.
-
I'd like to find a way to enable users to enter customizations for
boot options and then save them to the refind.conf file.
One possible way to implement this would be to have manual boot
stanzas override auto-detected boot loader definitions for the same
boot loader file.
+
Along similar lines, some users have asked for a way to take
+ detected boot programs and create a set of manual boot stanzas for
+ them, so that they can be modified manually.
+
+
Support for touchscreens and/or configurable buttons for rEFInd's
+ actions would enable use of rEFInd on tablet computers that lack
+ complete keyboards.
+
+
The ability to rotate the display for users who rotate their
+ monitors or who use tablets would be helpful.
+
+
GRUB provides a configuration-file command called outb
+ that enables manipulating hardware registers. Something similar,
+ via the mm command, can be done in the EFI shell. I'd like
+ to add such a feature to rEFInd, since it enables doing things like
+ disabling one or another video output on Macs with two video
+ cards.
+
+
I have thoughts about creating an EFI configuration tool and
+ information utility—something to tell you about your hard
+ disks, enable you to manage MOKs, adjust boot loader priority in
+ the NVRAM, and so on. This would be useful in system maintenance
+ and in recovering from boot problems.
+
+
An installation tool for the EFI environment would be useful.
+ A simple EFI shell script might work, but because this function
+ requires access to the bcfg command, this would work
+ only from a version 2 shell or if bcfg were implemented
+ as a standalone program. Another alternative would be a program
+ written in C.
+
It should be possible to override specific auto-detected boot
loader settings—say, to disable one specific boot loader or
change its icon.
-
A way to read boot options set via efibootmgr,
- bless, or similar options from NVRAM to add to the boot
- set would be useful.
+
A GUI configuration tool for host OSes (Linux, OS X, Windows, etc.)
+ would be nice, but it's low on my personal priority list. If you'd
+ like to contribute, I prefer something written in a cross-platform
+ GUI toolkit, so that a single code base can be used on any of the
+ major OSes.
-
A way to examine and change the NVRAM settings could be useful.
- This would enable a CD-based boot of rEFInd to fix a broken disk
- boot. Perhaps this could be done via a separate tool that could be
- launched much like the shell or gptsync.
+
The ability to produce audio output (at least a simple "beep") to signal
+ to visually impaired users when rEFInd is ready to accept input would be
+ helpful.
-
I'd like to give the user the ability to set custom options on a
- single-boot basis, similar to what's possible in GRUB.
+
-
A way to set the color of the font would be useful for theming
- purposes.
+
Improvements to the EFI drivers:
-
Going further, the ability to load arbitrary other fonts, ideally
- in a standard format, would be desirable for theming purposes.
+
-
A GUI configuration tool would be nice, but it's low on my personal
- priority list. If you'd like to contribute, I prefer something
- written in a cross-platform GUI toolkit, so that a single code base
- can be used on any of the major OSes.
+
Drivers for additional filesystems are desirable. Only XFS and JFS
+ are missing from the major Linux filesystems. UDF would also be a
+ welcome addition, as might drivers for other OSes (say, for the
+ BSDs, especially if BSD developers create a boot loader similar to
+ Linux's EFI stub loader). Also along these lines, adding drivers
+ for Linux LVM and RAID setups would be useful.
-
A way to "source" one configuration file from another one would be
- helpful for some types of configuration scripts. (This would enable
- overriding options in a secondary file without modifying the
- default original file, for instance.)
+
This may not be possible, or it may require a new driver, but a way
+ to have the drivers access files (like a Linux loopback mount) is
+ desirable.
-
+
When built with the GNU-EFI package, an attempt to load more than
+ one driver on my 32-bit Mac Mini causes the computer to hang. I do
+ not have this problem with 64-bit drivers on my UEFI-based
+ computers. I don't know if this is a 32-bit issue or a Mac issue.
+ This is not relevant if you're using my binary package,
+ since I build it with the TianoCore EDK2, and the drivers built in
+ that way don't exhibit this bug.
-
Improvements to the EFI drivers:
+
-
+
Improvements to gptsync, refind-install, or other
+ support tools:
-
Drivers for additional filesystems are required. Given the recent
- shift to ext4fs, that should be the priority; however, other Linux
- filesystems, UDF, and perhaps others would all be welcome
- additions. Also along these lines, adding drivers for Linux LVM and
- RAID setups would be useful, too.
+
-
As detailed on the drivers page, there
- are performance issues with the drivers on some systems. I suspect
- that most "real" computers aren't greatly affected (in my tests,
- the problem is worst with VirtualBox, and the next worst is a
- system that uses DUET). Nonetheless,
- I'd like to track down the cause and fix it.
+
The gptsync program can return misleading error codes
+ under some circumstances, such as when it makes no changes to the
+ partition table. Fix this.
-
The HFS+ driver returns a volume label of "HFS+ volume", no matter
- what the volume's real label is.
+
rEFInd's support for network booting is primitive and relies on the
+ external iPXE package. In my own testing, iPXE retrieves the
+ BIOS-mode boot loader from some servers that offer both, which
+ makes it useless on those networks.
-
This may not be possible, or it may require a new driver, but a way
- to have the drivers access files (like a Linux loopback mount) is
- desirable.