H. Peter Anvin [Tue, 8 Dec 2009 00:24:33 +0000 (16:24 -0800)]
NEWS: document MENU RESOLUTION
H. Peter Anvin [Tue, 8 Dec 2009 00:23:22 +0000 (16:23 -0800)]
Merge branch 'softres'
H. Peter Anvin [Tue, 8 Dec 2009 00:22:07 +0000 (16:22 -0800)]
doc: document MENU RESOLUTION
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
H. Peter Anvin [Tue, 8 Dec 2009 00:18:09 +0000 (16:18 -0800)]
vesamenu: add MENU RESOLUTION to be able to set nonstandard res
Add a MENU RESOLUTION command to make it possible to set a nonstandard
resolution. If the nonstandard resolution is unavailable, we revert
to the text mode screen.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
H. Peter Anvin [Tue, 8 Dec 2009 00:17:10 +0000 (16:17 -0800)]
vesacon: make it possible to request nonstandard resolutions
Make it possible to request nonstandard resolutions when enabling
vesacon.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
H. Peter Anvin [Mon, 7 Dec 2009 23:51:17 +0000 (15:51 -0800)]
vesacon: not all JPEG images are JFIF
There are a fair number of valid JPEG images which we can validly
decode that aren't JFIF-complient, so don't require the JFIF header.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
H. Peter Anvin [Mon, 7 Dec 2009 23:19:37 +0000 (15:19 -0800)]
NEWS: document mBFT
H. Peter Anvin [Mon, 7 Dec 2009 23:18:12 +0000 (15:18 -0800)]
NEWS: update for 3.84
H. Peter Anvin [Mon, 7 Dec 2009 23:02:14 +0000 (15:02 -0800)]
memdisk: slight cleanup of the "safe hook" structure
Make the "safe hook" structure code a bit more readable.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Shao Miller [Mon, 7 Dec 2009 04:03:43 +0000 (23:03 -0500)]
memdisk: "safe hook" and mBFT
Two additions to MEMDISK to support OS drivers.
The "safe hook" structure ("Safe Master Boot Record INT 13h Hook Routines")
is a means for an OS driver to follow a chain of INT 13h hooks, examining
the hooks' vendors and assuming responsibility for hook functionality along
the way. For MEMDISK, we guarantee an additional field which holds the
physical address for the mBFT.
The mBFT is an ACPI table which an OS driver can scan for. The mBFT
contains the official MEMDISK Info structure (MDI) which itself includes
parameters the OS will want to know about. The mBFT points back at the
"safe hook" structure's physical address so that an OS supporting both
"safe hook" chain-walking as well as mBFT-scanning can know that both
refer to the same MEMDISK instance.
Signed-off-by: Shao Miller <shao.miller@yrdsb.edu.on.ca>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Sebastian Herbszt [Sat, 28 Nov 2009 12:27:32 +0000 (13:27 +0100)]
doc: Fix some typos in the comboot documentation
Impact: Fixing typos
Fix some typos in the comboot documentation
Erwan Velu [Fri, 4 Dec 2009 16:40:06 +0000 (17:40 +0100)]
hdt: Fixing make release target
Impact: fixing com32 naming
Make release now generates the .c32 file accordingly to the scheme
Erwan Velu [Fri, 4 Dec 2009 16:29:55 +0000 (17:29 +0100)]
hdt: New release 0.3.6
Impact: new release
New release, thx to everyone for making that release possible
Erwan Velu [Fri, 4 Dec 2009 15:58:12 +0000 (16:58 +0100)]
hdt: Fixing about menu & show hdt
Impact: visual
Let's add website, rename contact to mailing list and put all people
together
Erwan Velu [Wed, 2 Dec 2009 21:31:18 +0000 (22:31 +0100)]
hdt: Fixing memory corruption in kernel>show
Impact: Fixing memory corruption
Fixing memory corruption which occurs during kernel>show
This patch makes the init cleaner.
Erwan Velu [Mon, 30 Nov 2009 23:11:25 +0000 (00:11 +0100)]
hdt: Fixing memtest entry
Impact: none
Fixing memtest entry
Erwan Velu [Mon, 30 Nov 2009 19:47:45 +0000 (20:47 +0100)]
hdt: Makefile isn't gzipping pci.ids
Impact: avoid missed gzipping
The iso wasn't gzipping pci.ids correctly
Erwan Velu [Mon, 30 Nov 2009 17:44:36 +0000 (18:44 +0100)]
dmi: dmitest have to use voltage_mv
Impact: fixing build error
Processor is now using voltage_mv
Erwan Velu [Mon, 30 Nov 2009 14:57:41 +0000 (15:57 +0100)]
hdt: summary mode only show memory size
Impact: visual
Summary mode only show the detected memory size.
Showing used banks is too much verbose for a summary
Erwan Velu [Mon, 30 Nov 2009 12:26:39 +0000 (13:26 +0100)]
hdt: Adding cpu voltage
Impact: Adding cpu voltage
Adding cpu voltage via DMI
Erwan Velu [Mon, 30 Nov 2009 12:25:00 +0000 (13:25 +0100)]
dmi: Switching processor voltage from V to mV
Impact: Changing units
Processor voltage is now millivolts
Erwan Velu [Mon, 30 Nov 2009 12:01:58 +0000 (13:01 +0100)]
dmi: more chassis types
Impact: more chassis types
More DMI chassis types
Erwan Velu [Mon, 30 Nov 2009 11:59:04 +0000 (12:59 +0100)]
dmi: returning out of spec instead of NULL
Impact: avoid null string
It's better returning out of spec instead of null
Erwan Velu [Mon, 30 Nov 2009 11:55:12 +0000 (12:55 +0100)]
dmi: More processors
Impact: More processors
More processors in the DMI table
Erwan Velu [Mon, 30 Nov 2009 11:39:08 +0000 (12:39 +0100)]
hdt: Fixing history cycling
Impact: prior to that commit, cycling the history failed
When we enter more than 32 commands, let's cycle the history
Erwan Velu [Mon, 30 Nov 2009 11:27:37 +0000 (12:27 +0100)]
hdt: Using del_multi_spaces for the cpu model
Impact: avoid multi spaces in the cpu model
Some cpu vendors like to put multi spaces in the model name.
We don't need it for a cleaner & shorter display
Erwan Velu [Mon, 30 Nov 2009 11:24:40 +0000 (12:24 +0100)]
dmi: More cpu upgrade options
Impact: many cpu upgrade were set out of spec
Fixing stupid typo which prevent cpu upgrade options to be shown
Erwan Velu [Mon, 30 Nov 2009 10:18:36 +0000 (11:18 +0100)]
hdt: removing vendor_id from cpu info
Impact: visual
No need to display the vendor_id as we already show the vendor name.
That confuse users, let's remove it
Erwan Velu [Mon, 30 Nov 2009 10:02:49 +0000 (11:02 +0100)]
pci: replace '-' by '_' in modules names
Impact: avoid kernel modules duplication
In modules.pcimap, kernel modules name are featuring '_' or '-' whereas modules.alias is only using '_'.
To avoid kernel modules duplication, let's rename all '-' by '_' to match what modules.alias provides
This avoid stupid duplications like "a-b" & "a_b" whereas they are in
fact the same kernel module
Erwan Velu [Mon, 30 Nov 2009 09:07:02 +0000 (10:07 +0100)]
hdt: Adding memtest entry in the menu
Imapct: Allow people starting memtest from the menu
Users can now starts memtest for our menu
Erwan Velu [Sun, 29 Nov 2009 20:40:29 +0000 (21:40 +0100)]
hdt: EDD uses hex numbers, not decimal
Impact: Fixing commit
4d2536e76d621bc7c5faa076a905051854f9c512
Commit
4d2536e76d621bc7c5faa076a905051854f9c512 was buggy as considering
EDD as decimal and not hex.
Erwan Velu [Sun, 29 Nov 2009 20:13:00 +0000 (21:13 +0100)]
hdt: Fixing cpu flags splitting
Impact: improve rendering
Under the cli, the cpu flags were sometimes leaking to another line
Erwan Velu [Sun, 29 Nov 2009 19:48:41 +0000 (20:48 +0100)]
Adding ifcpu.c32, removing ifcpuhvm
Impact: ifcpu holds more cases
ifcpu allow users defining a set of required cpu features for booting an
entry
Erwan Velu [Sun, 29 Nov 2009 14:35:13 +0000 (15:35 +0100)]
hdt: Adding gPXE detection
Impact: gPXE is now detected
HDT is now able to report is gPXE is loaded on this given system
Erwan Velu [Sun, 29 Nov 2009 14:20:33 +0000 (15:20 +0100)]
com32: Avoid gpxe detection code duplication
Impact: avoid code duplication
This will make code more generic, hdt will need it ;)
Erwan Velu [Sun, 29 Nov 2009 13:38:43 +0000 (14:38 +0100)]
hdt: Do not display hostbus & interface for EDD<3
Impact: Avoid corruption while displaying host bus interface when
EDD<3.0
If EDD < 3.0 displaying host bus & interface for disk devices return
unexpected content. We have to avoid displaying that informations in
such case
Erwan Velu [Sun, 29 Nov 2009 13:14:19 +0000 (14:14 +0100)]
hdt: Fixing wrong Part Number display
Impact: Visual
In the memory bank display, the part number wasn't correctly padded.
Erwan Velu [Sun, 29 Nov 2009 13:11:09 +0000 (14:11 +0100)]
hdt: Keeping symlinks on generated binaries
Impact: none
Building binary images let some symlinks useful for testing
Erwan Velu [Sun, 29 Nov 2009 13:07:52 +0000 (14:07 +0100)]
hdt: Preventing agressive disk's size rounding
Impact: Visual
Prevent reported case where 1997MB could be displayed as 1GB.
Erwan Velu [Sat, 28 Nov 2009 12:22:03 +0000 (13:22 +0100)]
hdt: Adding memtest is our binary images
Impact: memtest is now included
Downloading memtest while generating floppy/iso images
Erwan Velu [Fri, 27 Nov 2009 21:22:18 +0000 (22:22 +0100)]
Fixing comments on ifcpu64/ifcpuhvm
Impact: Fixing typo
Fixing typo
Erwan Velu [Fri, 27 Nov 2009 21:12:40 +0000 (22:12 +0100)]
Adding ifcpuhvm.c32
Impact: new module to boot hvm systems
This module allow users to define a boot entry regarding if the cpu is
supporting hvm (vmx|svm). This could used to boot xen, or any hypervisor
Erwan Velu [Fri, 27 Nov 2009 19:41:16 +0000 (20:41 +0100)]
hdt: Fixing cpu cores naming
Impact: Visual
Fixing cpu cores naming
Erwan Velu [Fri, 27 Nov 2009 19:30:51 +0000 (20:30 +0100)]
hdt: Adding 'release' target to Makefil
Impact: Automate releasing files
'make release' will generate the proper binaries
Erwan Velu [Fri, 27 Nov 2009 19:21:05 +0000 (20:21 +0100)]
hdt: 0.3.6-pre2
Impact: new subrelease
New subrelease
Erwan Velu [Fri, 27 Nov 2009 19:11:34 +0000 (20:11 +0100)]
hdt: Adding cpu cache management
Impact: Adding cpu cache
CPU cache is now visible from HDT
Erwan Velu [Fri, 27 Nov 2009 18:47:46 +0000 (19:47 +0100)]
cpuid: Adding L1 & L2 cache detection
Impact: Adding L1 & L2 cache detection
Adding L1 & L2 cache detection
Erwan Velu [Fri, 27 Nov 2009 15:09:01 +0000 (16:09 +0100)]
hdt: Fixing memory detection
Impact: Fix memory detection under cli
Under the CLI, the memory detection wasn't started...
This commit fixes the memory initialisation path
Erwan Velu [Fri, 27 Nov 2009 14:53:07 +0000 (15:53 +0100)]
hdt: Adding cpu core management
Impact: we now display the number of cores
CPU cores are now displayed
Erwan Velu [Fri, 27 Nov 2009 14:44:33 +0000 (15:44 +0100)]
cpuid: Adding cpu core detection
Impact: we are now able to detect the number of cpu cores
We can detect the number of cores using cpuid(4) or cpuid_ecx(0x80000008)
Erwan Velu [Fri, 27 Nov 2009 10:45:46 +0000 (11:45 +0100)]
hdt: Improving cpu flags report
Impact: more cpu flags detected
This commit allow hdt reporting more cpu flags but also simplify from
the developer point of view the cpu flags reporting.
A visual improvement is also done for showing lists of cpu flags.
Erwan Velu [Fri, 27 Nov 2009 09:10:52 +0000 (10:10 +0100)]
cpuid: Adding more flags
Impact: sync with linux kernel
cpu flags are now in sync with the linux kernel. Too many flags were
missing.
Erwan Velu [Thu, 26 Nov 2009 22:44:22 +0000 (23:44 +0100)]
hdt: show error message when no disk is found
Impact: Visual
Display an error message when no disk is found on the system
Erwan Velu [Thu, 26 Nov 2009 22:14:51 +0000 (23:14 +0100)]
hdt: Adding CODENAME in title (menu.c32)
Impact: visual
Let's display the CODENAME in the menu title
Erwan Velu [Thu, 26 Nov 2009 22:06:35 +0000 (23:06 +0100)]
hdt: clear_screen have to reset the xy position
Impact: visual
This commit avoid clear_screen letting the cursor at the same place
after the clearing. It have to be reset to
Erwan Velu [Thu, 26 Nov 2009 21:25:34 +0000 (22:25 +0100)]
hdt: Auto generating versions for menu mode
Impact: visual
The menu mode now display the current version
Erwan Velu [Thu, 26 Nov 2009 20:41:26 +0000 (21:41 +0100)]
hdt: Adding vesa mode switching under CLI
Impact: its now possible to switch the vesa mode
Users can use vesa>enable to switch to vesa mode or
vesa>disable to switch to console mode
Erwan Velu [Thu, 26 Nov 2009 20:27:40 +0000 (21:27 +0100)]
hdt: Improving help messages' display in menu
Impact: Visual
Help messages need a space for increasing readbility
Erwan Velu [Thu, 26 Nov 2009 20:23:07 +0000 (21:23 +0100)]
hdt: Improving menu layout
Impact: visual
Improving layout. Thx gert.
Erwan Velu [Thu, 26 Nov 2009 20:07:37 +0000 (21:07 +0100)]
hdt: Adding a menu for floppy & iso images
Impact: none
Floppy & ISO images are now featuring a menu mode.
Thx gert for this idea.
Erwan Velu [Thu, 26 Nov 2009 19:14:18 +0000 (20:14 +0100)]
hdt: Fixing typo in variable's name
Impact: Building is working again
Ooops ....
Erwan Velu [Thu, 26 Nov 2009 15:48:35 +0000 (16:48 +0100)]
hdt: Adding novesa for future use
Impact: none
In a close future, as vesa will be enabled by default, let's anticipate
the novesa command
Erwan Velu [Wed, 25 Nov 2009 23:01:12 +0000 (00:01 +0100)]
hdt: print_history have to use more_printf
Impact: managing scrolling in history()
History can generates some scrolling, so let's use more_printf
Erwan Velu [Wed, 25 Nov 2009 22:50:03 +0000 (23:50 +0100)]
hdt: Fixing history behavior to match bash's one
Impact: history usage is more obvious
History (up/down) is more intiutive and copy bash's behavior
Erwan Velu [Wed, 25 Nov 2009 22:10:34 +0000 (23:10 +0100)]
hdt: Adding history command
Impact: new command 'history'
This command allow to see the content of the history
Erwan Velu [Wed, 25 Nov 2009 21:35:04 +0000 (22:35 +0100)]
hdt: reworking history to prevent corruption
Impact: avoid history corruption
No need to copy history to input. We can directly use the history stack.
Steffen Winterfeldt [Thu, 26 Nov 2009 20:00:17 +0000 (21:00 +0100)]
gfxboot.c32
Ok, here is the first try to turn the gfxboot wrapper into a com32 module. I
had to extend the interface to the gfxboot core a bit to get it working. So
it works only with latest gfxboot from
git://gitorious.org/gfxboot/gfxboot.git
Steffen
Signed-off-by: Sebastian Herbszt <herbszt@gmx.de>
Sebastian Herbszt [Fri, 20 Nov 2009 14:42:19 +0000 (15:42 +0100)]
pcitest: fix wrong statements about file location
Impact: Visual
The module tries to access the files in the current directory.
So this patch fixes the error message accordingly.
Erwan Velu [Tue, 24 Nov 2009 12:42:41 +0000 (13:42 +0100)]
gpllibs: Running Nindent
Impact: visual
Running Nindent on gpllib / gplinclude
Erwan Velu [Tue, 24 Nov 2009 09:26:28 +0000 (10:26 +0100)]
Merge branch 'master' of ssh://terminus.zytor.com//pub/git/users/erwan/hdt
Erwan Velu [Tue, 24 Nov 2009 09:24:49 +0000 (10:24 +0100)]
hdt: Running Nindent
Impact: Visual cleanup
HDT now uses the same indent as syslinux
Erwan Velu [Tue, 24 Nov 2009 09:05:48 +0000 (10:05 +0100)]
Merge remote branch 'pam/hdt-0.3.6'
Pierre-Alexandre Meyer [Tue, 24 Nov 2009 02:40:01 +0000 (18:40 -0800)]
hdt: fix memory corruption in the CLI
More memory corruption with overlooked. parse_command_line and
expand_aliases both allocate memory (for command, module, argc and argv).
We did fix the memory corruption for the autocompletion feature but
but didn't do it for the actual execution of the commands.
Signed-off-by: Pierre-Alexandre Meyer <pierre@ning.com>
root [Mon, 23 Nov 2009 21:04:05 +0000 (22:04 +0100)]
memory: Commenting workaround
Impact: none
HDT is using a memory detection code which duplicates some syslinux
code. Once addr_t will be moved into 64bit, this can be removed by using
syslinux_scan_memory().
Erwan Velu [Mon, 23 Nov 2009 16:16:41 +0000 (17:16 +0100)]
hdt: replacing e88 by 88
Impact: e88 doesn't have any meaning
Fixing e88 by 88
Erwan Velu [Mon, 23 Nov 2009 14:16:32 +0000 (15:16 +0100)]
hdt: Rounding detected memory size
Impact: Visual
Rounding values to have more realistic values
Erwan Velu [Mon, 23 Nov 2009 12:44:51 +0000 (13:44 +0100)]
hdt: Improving memory detection
Impact: Detect memory size even if dmi is broken
This commit uses e820/e801/e88 memory detection to find the memory size.
That allow to display the memory size even if the dmi table is broken
Erwan Velu [Mon, 23 Nov 2009 12:39:55 +0000 (13:39 +0100)]
memory: Adding memory size detection
Impact: allow com32 module to detect the amount of memory installed
This commit adds several function:
sanitize_e820_map() to generate a new e820 map that doesn't have
overlaping
memsize_e820() to detect the available memory (in KB)
detect_memsize() to detect the amount of memory by using e820/e801/e88.
It first try to detect the memory via e820. If it fails, it uses e801 as
fallback. If this e801 fails, e88 is used.
Erwan Velu [Sun, 22 Nov 2009 21:48:49 +0000 (22:48 +0100)]
hdt: Cleaning mbr_ids
Impact: none
Let's just clean mbr_ids as we do for the orther structures
Erwan Velu [Sun, 22 Nov 2009 21:30:57 +0000 (22:30 +0100)]
hdt: More coherency in messages
Impact: more coherency
More coherency in messages
Erwan Velu [Sun, 22 Nov 2009 21:26:10 +0000 (22:26 +0100)]
hdt: Adding more coherency in memory messages
Impact: more coherency
More coherency in memory messages
Erwan Velu [Sun, 22 Nov 2009 20:57:47 +0000 (21:57 +0100)]
hdt: memory menu have to be shown in every case
Impact: memory menu have to be always visible
In virtualbox, dmi table isn't correctly feed. In that situation, the
memory menu isn't displayed whereas e88, e820 & e801 are present.
So let's remove the condition where the memory menu have to be shown
only when memory modules are detected.
Tested by Gert.
Erwan Velu [Sun, 22 Nov 2009 20:50:46 +0000 (21:50 +0100)]
hdt: Renaming "show 88" to "show e88"
Impact: increasing coherency
Let's rename that function to keep more coherency with e820 & e801
Erwan Velu [Sun, 22 Nov 2009 18:27:21 +0000 (19:27 +0100)]
Merge remote branch 'pam/hdt-0.3.6'
Pierre-Alexandre Meyer [Sun, 22 Nov 2009 00:49:37 +0000 (16:49 -0800)]
gpllib: mbr: fix grub2 detection
gerth reports that hdt-0.3.6 doesn't detect grub2 anymore.
It looks like some break were missing in the case statement.
Signed-off-by: Pierre-Alexandre Meyer <pierre@mouraf.org>
Erwan Velu [Sat, 21 Nov 2009 11:37:08 +0000 (12:37 +0100)]
hdt: Only copy needed art files
Impact: Reduces binary image size
This commit only copy the needed art files to put in the floppy/iso
images.
Erwan Velu [Sat, 21 Nov 2009 11:34:53 +0000 (12:34 +0100)]
hdt: Allow user to override the background image
Impact: none
This commit allow user using vesa=<filename> to override the background
image in vesa mode
Erwan Velu [Fri, 20 Nov 2009 22:14:00 +0000 (23:14 +0100)]
Merge remote branch 'hpa/master'
Erwan Velu [Fri, 20 Nov 2009 21:54:16 +0000 (22:54 +0100)]
hdt: Adding another logo
Impact: none
Adding another logo (black)
Erwan Velu [Fri, 20 Nov 2009 21:50:50 +0000 (22:50 +0100)]
hdt: Adding logo
Impact: none
Adding transparent logo
Erwan Velu [Fri, 20 Nov 2009 21:35:41 +0000 (22:35 +0100)]
hdt: Updating contributors
Impact: Visual
Adding Alexander Andino for design. Adding other contributors's role
Erwan Velu [Fri, 20 Nov 2009 21:33:55 +0000 (22:33 +0100)]
hdt: fixing wrong type
Impact: Avoid compilation error
Fixing return type
Erwan Velu [Fri, 20 Nov 2009 21:13:28 +0000 (22:13 +0100)]
hdt: 0.3.6-pre1
Impact: Pre-release
Releasing 0.3.6-pre1
Erwan Velu [Fri, 20 Nov 2009 21:00:50 +0000 (22:00 +0100)]
hdt: Adding background image for the CLI
Impact: Visual
The cli now features a background image when started in vesa mode
Erwan Velu [Fri, 20 Nov 2009 15:49:42 +0000 (16:49 +0100)]
hdt: Adding vesa mode support
Impact: cli can now be turned into vesa
Using the "vesa" keyword at boot time, you can enable the vesa mode for
the cli.
Erwan Velu [Fri, 20 Nov 2009 15:37:36 +0000 (16:37 +0100)]
hdt: hdt.img must use 8.3 naming
Impact: Loading files is now working again
Prior to that commit, the hdt.img was failing loading files
H. Peter Anvin [Fri, 20 Nov 2009 04:17:43 +0000 (20:17 -0800)]
memdisk: pptr->cd_pkt.sect_count is in virtual sectors
Per the El Torito spec, pptr->cd_pkt.sect_count are in virtual
sectors, i.e. 512-byte sectors.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Shao Miller [Fri, 20 Nov 2009 01:49:34 +0000 (20:49 -0500)]
memdisk: Use boot_lba logic for booting an offset within the disk image
We have two fundamental quantities: the offset of the disk image
itself and the LBA from which we boot. Keep them straight. This
should make at least some no-emulation isos work correctly.
Signed-off-by: Shao Miller <shao.miller@yrdsb.edu.on.ca>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Shao Miller [Fri, 20 Nov 2009 01:20:33 +0000 (20:20 -0500)]
memdisk: Restore DL for INT 13h, AH=15h, DL=floppy
At least PC-DOS requires this to avoid an infinite loop; the DL we
restore is not the same DL that we call the underlying BIOS with, but
rather the original one (before the decrement.) PC-DOS loops over all
the floppies and relies on the call preserving DL in order to advance
the loop.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
H. Peter Anvin [Fri, 20 Nov 2009 01:17:36 +0000 (17:17 -0800)]
open: close the file descriptor if the file fails to open
Close the file descriptor (freeing the file structure) when opening a
file fails.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>