| Age | Commit message (Collapse) | Author |
|---|
|
Change the default purgatory sha256 code optimization from -O0 to -O2, and add a
new arch specific makefile variable $(ARCH)_PURGATORY_SHA256_CFLAGS which can
over ride this default. Set ia64_PURGATORY_SHA256_CFLAGS to -O0 to retain the
previous optimization level for ia64.
The purgatory sha256 code needs the be built with -O0 for the ia64
architecture. Currently this code is built with -O0 for all architectures,
which slows down the calculations for architectures which could otherwise
use -O2.
On arm64, it takes around 20 second to verify SHA in purgatory when
vmlinuz image is around 13MB and initramfs is around 30M with -O2
enabled. Otherwise, it takes more than 2 minutes.
Cc: Pratyush Anand <panand@redhat.com>
Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
The EFI Runtime Services Table contains pointers to ia64 function
descriptors. On existing, pre-Tiano, firmware, SetVirtualAddressMap()
converts *all* these pointers from physical to virtual. On Tiano-based
firmware, the pointer to the SetVirtualAddressMap() function descriptor
is not converted, so it remains a physical pointer.
The ia64 kexec purgatory patches the SetVirtualAddressMap() function
descriptor so that when the new kernel calls SetVirtualAddressMap(), it
never reaches firmware. Instead, it calls a dummy function that just
returns success.
Purgatory runs in physical mode, so it must convert the pointer from the
RuntimeServicesTable to a physical address. This patch makes that
conversion work both for old firmware (where the pointer is an identity-
mapped virtual address) and new Tiano firmware (where the pointer is a
physical address).
Without this patch, kexec on Tiano firmware causes an MCA because
ia64_env_setup() subtracts PAGE_OFFSET from a physical address and ends
up with an invalid physical address. Referencing that address causes
the MCA.
Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
From my observations the way that the EFI_LOAD_DATA is provided
on the inital boot works like this:
There is a large EFI_CONVENTIONAL_MEMORY region.
The portion begining at the first load segment of
the image to be loading and ending with the last segment,
aligned to 64K, is turned into a separate region
of type EFI_LOAD_DATA.
A truncated example of this:
...
mem04: type= 7, attr=0x0000000000000008, range=[0x0000000000100000-0x0000000004000000) ( 63MB)
mem05: type= 2, attr=0x0000000000000008, range=[0x0000000004000000-0x000000000481f000) ( 8MB)
mem06: type= 7, attr=0x0000000000000008, range=[0x000000000481f000-0x000000003e876000) ( 928MB)
...
Where type 7 is EFI_CONVENTIONAL_MEMORY and
type 3 is EFI_LOAD_DATA.
There is a patch to the user-space portion of kexec-tools that merges the
segments supplied to this code if they are adjacent. This seems to always
result in a single segment being passed to this code, that should start
at the same address as the existing EFI_LOAD_DATA segment.
So all that should be left to do is to merge the existing
EFI_LOAD_DATA region with the following EFI_CONVENTIONAL_MEMORY region,
and then split it up to accommodate the segment passed from user-space.
The new EFI_LOAD_DATA region created with this code will always
start at the same address as the old EFI_LOAD_DATA region. If
this proves to be overly simplistic it should be easy to update.
This code also allows merging of multiple regions to accommodate
the new EFI_LOAD_DATA region. I strongly doubt this will ever
be used, but it is in line with the way the existing code works.
If the same image is used after kexec, then the
EFI regions in question will turn out the same as the original regions.
This is important, otherwise kernel / hypervisor regions will not be able
to be inserted into /proc/iomem / /proc/iomem_machine.
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
With the recent build changes a number of unneded files
crept into tarballs, including .o and .d files.
This patch is farily verbose, but hopefully in the long
run this system will be obvious enough to be maintainable.
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
Since we use the implicit ruls for .c and .S, just colelct all sources
in the one variable.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
Purgatory seems to partially duplicate system headers.
It seems a log cleaner not to do so.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
This change makes kexec-tools work more like a standard configure-make-
make-install-type project:
* Remove $(OBJDIR) stuff. To do an out-of-tree build, just configure
from a different directory.
* Use the implicit Makefile rules more, and just edit the compiler
flags for specific targets.
* Simplify compiler/linker flags - no need for EXTRA_*
* Add TARGET_CC, and improve checks for BUILD_CC too.
* Set arch-specific flags in arch-specific makefiles, not conditional
on $(ARCH).
* Generate dependency files in the main compile, rather than as a
separate step.
* Don't #include sha256.c, but re-build it into the purgatory.
Still a work-in-progress.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
The address where the ELF core header is stored is passed to the secondary
kernel as a kernel command line option. The memory area for this header is
also marked as a separate EFI memory descriptor on ia64.
The separate EFI memory descriptor is at the moment of the type
EFI_UNUSABLE_MEMORY. With such a type the secondary kernel skips over the
entire memory granule (config option, 16M or 64M) when detecting memory.
If we are lucky we will just lose some memory, but if we happen to have data
in the same granule (such as an initramfs image), then this data will never
get mapped and the kernel bombs out when trying to access it.
So this is an attempt to fix this by changing the EFI memory descriptor
type into EFI_LOADER_DATA. This type is the same type used for the kernel
data and for initramfs. In the secondary kernel we then handle the ELF core
header data the same way as we handle the initramfs image.
This strategy requires changes in the secondary kernel as well, I'll
post the kernel patches in a little while.
Signed-off-by: Magnus Damm <magnus@valinux.co.jp>
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
Hi,
I've run into problems testing kexec/kdump on a Montecito revision C
processor. In purgatory, __dummy_efi_function is copied onto the end of
the command line boot parameter (command_line + command_line_len) and this
address is used to replace the EFI call to set_virtual_address_map(). The
copied range is then icache flushed.
The destination address is aligned to 16-bytes (in kexec-elf-ia64.c), but
the fc.i instruction flushes a 32-byte range "associated" with that
address. When my command line length is 16-byte aligned but not 32-byte
aligned, this results in the first 16-bytes of __dummy_efi_function getting
flushed (and the 16 bytes prior to that), but the second half of the
function (the part with the br.ret) does not get flushed. kdump then hangs
in purgatory. By adding a few spaces to my command line, it becomes both
16 and 32-byte aligned, and kdump works.
This patch makes icache_flush_range() align the start address to 32-bytes
and account for the difference. The patch is against Horms
kexec-tools-testing tree. As a side note, you could also fix this by just
adding 32 to the length passed to flush_icache_range() but that hides the
dependent behavior.
Thanks,
-T
It seems I was always testing with command line more than 16 bytes
length.....
Thanks.
Acked-by: Zou Nan hai <nanhai.zou@intel.com>
Manually applied - not sure why that was neccessary.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
This adds vmm support to kexec-tool for ia64. This is annalogous
to the same feature that is present in the latest version of elilo.
It is a method of booting a vmm (hypervisor) such as Xen.
Essentially it works as follows.
* If the --vmm argument is not provided, then the kernel is booted as normal,
no changes
* Else, the image specified by --vmm is placed loaded into the elf
segments, where the Linux kernel image would otherwise have gone.
And the Linux kernel image, allong with its length is loaded into a
seprate segment, and passed as new entry at the end of the boot parameters.
This is somewhat similar to how initramfs/initramd images are passed
to a booting kernel, and can work in conjunction with that feature.
On boot (or in this case on kexec) the vmm (hypervisor) will be
loaded instead of a Linux kernel, and the hypervisor will then load up
the Linux kernel as it sees fit.
This is needed in order for kexec from Xen to Xen, using the
port of kexec to Xen that I am working on, to work.
I am not entirely fond of this design, and i think that developing
an ia64 variant of multiboot would be much nicer. However it is
an existing method that is currently in widespread use through
its incarntation in elilo. And if multiboot is added in future,
it can be done as a separate boot method, and thus orthogonal to this
patch.
In order to use this code a number of other changes are needed,
in particular:
1. Xen and the corresponding Linux Kernel needs to be patched with
the port of kexec to ia64-xen that I have been working on.
I will post the latest version of these patches to xen-devel
shortly.
2. The currently hardcoded PAGE_OFFSET value in purgatory needs
to be changed from the Linux value to the Xen value. I will
post a very hackish, definately not to be released, patch
after this patch which includes a comment that explains this problem
more clearly.
Also, xen->linux and linux->xen is still very much work in progress due
to the problem described at the following link
http://permalink.gmane.org/gmane.linux.ports.ia64/14995
However, from an infastructure point of view I think it would be good to
apply this code, so that kexec-tool is one step closer to being able to
support vmms (hypervisors). The patch does not alter any existing
behaviour, it just adds a new feature. Bugs asside, the only real danger
seems to be confusion for end-users, perhaps we could comment out the help
text to hide the feature from the lay user, or attach a big fat warning to it.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
Add a more verbose comment to explain why set_virtual_address_map is
replaced why a dummy function
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
This unifies the comments in purgatory-ia64.c to always use C style
comments. Previously some comments where C style, while others were
C++ style.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
This makes the tail of patch_efi_memmap() slightly less nested,
and thus a little easier to read.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
Just use boot_param->efi_memdesc_size directly instead, it seems
at least as clean, and possibly easier to follow.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
len is assigned once and used once, just use boot_param->efi_memmap_size
directly instead.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
* Rename md1 and md2 to md_src and md_dest respectively to make things
a little easier to follow.
* Remove p1 and p2, use src and dest instead
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
There is a farily complex if, for construct in patch_efi_memmap(),
that seems to be simplifyable to a somewhat simpler while statement.
Note that this does change the logic statement. In particular
the original code has if (seg->end < mend) towards the end,
and the new code effectively replaces this with if (seg->end <= mend).
However, in the original code this is copled with a separate
if (seg->end > mend) check at the begining, so I believe that
this is actually a minor (possibly never seen) logic error in
the original code. The node code just always checks (seg->end > mend).
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
This patch removes a duplicated assignment of *md2.
It also replaces a switch statement with an if statement
which is much more compact in this instance.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
This patch reduces the nesting in patch_efi_memmap() by
jumping to the next interation of the inner for loop
if the following condition is true.
if (seg->start < mstart || seg->start >= mend)
This is instead of a reasonably large ammount of code inside the if
conditional if the converse is true.
This makes things somewhat easier to read as the nesting is already
quite deep, and many lines do not fit easily within 80 columns.
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
Make purgatory code 80 columns wide
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
Remove kexec/arch/ia64/kexec-ia64.h.orig and purgatory/arch/ia64/Makefile.orig
which were (presumably accidently) introduced in changest
9241000f28eb6b86a06c0be2d6cf31498373bc1c, "kdump ia64".
Signed-Off-By: Simon Horman <horms@verge.net.au>
|
|
SN platform support PIO in a different way to generic IA64 platform. It
does not support most of the legacy I/O ports.
Give an --noio option to kexec-tools to disable I/O in purgatory code.
This patch also removed an unused io.h in kexec-tools.
Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>
Edited to consistently use tabs instead of spaces for intentation,
remove one instance of trailing whitespace, and fix indentation
of noio line in options[].
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
without this patch, crash tool will not able to analyze efi memmap of
first kernel from vmcore file.
This patch is against kexec-tools-1.101 with kdump10 patch.
Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>
Removed bogus fragments caused by whitespace addition
Signed-off-by: Simon Horman <horms@verge.net.au>
|
|
On Fri, 2006-06-09 at 19:50, Welterlen Benoit wrote:
> Zou Nan hai wrote:
> > The ia64 kdump patch is in 2 parts.
> >
> > the kexec-kdump-ia64-2.6.16.patch should apply on top of the previous
> > kexec patch by Khalid in Tony's test tree.
> >
> > the kexec-tools-kdump-ia64.patch should apply to kexec-tools-1.101
> > with kexec-tools-1.101-kdump.patch
> >
> >
> > To test it.
> > Build first SMP kernel with KEXEC and KDUMP enabled.
> >
> > Boot it with kernel parameter "crashkernel=XXX@YYY"
> > means reserver XXX from YYY for crashdumping.
> > Build an UP kernel with KEXEC KDUMP VMCORE enabled.
> > load this kernel as a crashdumping kernel
> > kexec -p vmlinux.gz --initrd=initrd --append="...."
> >
> > trigger a crash,
> > maybe "echo c > /proc/sysrq-trigger"
> > after the crash kernel boots,
> > cp /proc/vmcore core
> >
> > gdb first_kernel_vmlinux core
> >
> > please test and review.
> >
> > Signed-off-by: Khalid Aziz <khalid_aziz@hp.com>
> > Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>
> >
> >
> > https://lists.osdl.org/mailman/listinfo/fastboot
> >
>
> Hello Nan hai,
>
> I tried your patches. It seems that the kexec-tools-kdump-ia64.patch
> file can not be applied after the latest release of kexec-tools
> http://lse.sourceforge.net/kdump/patches/1.101-kdump9/kexec-tools-1.101-kdump9.patch
>
> I modified it for that. (attached file).
>
> I have a question about kdump :
>
> When the second kernel is loaded, kexec checks if the segments of the
> new kernel are in the reserved memory
>
> valid_memory_range in kexec/kexec.c :
> if ((send > mem_max) || (sstart < mem_min)) return 0;
>
> but mem_min and mem_max are defined by the XXX@YYY argument of the
> first kernel.
> For me, with 512@512 :
> more /proc/iomem
> ...
> 049cc000-77ffffff : System RAM
> 20000000-3fffffff : Crash kernel
> ...
> So, I can not load the second kernel : Invalid memory segment
> 0x4000000 - 0x469ffff
>
> When I set 64@64 argument for the first kernel, the checking is ok,
> but I have another issue :
> kexec_load failed: Cannot assign requested address
> entry = 0x80020 flags = 320001
> nr_segments = 6
> segment[0].buf = 0x6000000000021b90
> segment[0].bufsz = 20
> segment[0].mem = (nil)
> segment[0].memsz = 10000
> segment[1].buf = 0x60000000000222d0
> segment[1].bufsz = 10638
> segment[1].mem = 0x80000
> segment[1].memsz = 20000
> segment[2].buf = 0x2000000003b50010
> segment[2].bufsz = 23473c
> segment[2].mem = 0x100000
> segment[2].memsz = 240000
> segment[3].buf = 0x20000000002f0010
> segment[3].bufsz = 692dd8
> segment[3].mem = 0x4000000
> segment[3].memsz = 6a0000
> segment[4].buf = 0x2000000000990010
> segment[4].bufsz = 42c8
> segment[4].mem = 0x46a0000
> segment[4].memsz = 10000
> segment[5].buf = 0x20000000009a0010
> segment[5].bufsz = 17c3ec
> segment[5].mem = 0x46b0000
> segment[5].memsz = 2d0000
>
>
> Segments of the second kernel are the same than the first one
> (0x0000000004000000, 0x00000000046a0000 ...)
> We can not change the PHYSICAL_START as in other architectures (x86,
> x86_64, powerpc).
>
> So, I don't understand how it should work. Can you please have some
> explanation on this ?
>
> Thank you very much !
>
> Best regards,
>
> Benoit Welterlen
>
>
> ______________________________________________________________________
I modify the patch based on this one, fixed some bugs in it.
please test.
Thanks
Zou Nan hai
Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>
Signed-off-by: Maneesh Soni <maneesh@in.ibm.com>
|
|
This patch adds support for kexec-tools on ia64. This patch applies on
top of -kdump7 patch from <http://lse.sourceforge.net/kdump/>.
Signed-off-by: Khalid Aziz <khalid.aziz@hp.com>
Signed-off-by: Maneesh Soni <maneesh@in.ibm.com>
|
|
o This patch adds support for reserving space for backup region. Also adds code
in purgatory to copy the first 640K to backup region.
o Moved kexec_flags inside kexec_info structure.
Signed-off-by: Vivek Goyal <vgoyal@in.ibm.com>
Signed-off-by: Maneesh Soni <maneesh@in.ibm.com>
|
|
- Initial import into git
- initial nbi image formage support
- ppc32 initial register setting fixes.
- gzipped multiboot file support
|
