[IA64] preparation for vga acceleration of VTI domain
This patch fixes several minor issues, as a prepare step to support VGA
acceleration for VTI domain:
- shared vram buffer needs to be mapped as WB in both sides, because
the shared vram is the true WB memory though VTI domain is told as UC
- Introduced a new pte_mem to indicate p2m entry containing valid mfn
when replace p2m entry. Pte_none doesn't work for VTI case, since IO
type is encoded into p2m entry which however doesn't contain valid mfn.
After above change, guest_physmap_add/remove_page is available to
be used by qemu to setup linear buffer for VTI domain.
Now xen drivers use is_running_on_xen() to check whether it runs on
xen or bare metal. There remains many if (running_on_xen) in ia64
specific codes. This patch replaces running_on_xen with is_running_on_xen()
[PAE] Allow pgdirs above 4GB for paravirt guests.
**NOTE**: This obviates the need for lowmem_emergency_pool.
Unpriv guests no longer need to be able to allocate memory
below 4GB for PAE PDPTs. Signed-off-by: Keir Fraser <keir@xensource.com>
Cast to a signed type before negating if we really want the result to be negative.
From: Jan Beulich <jbeulich@novell.com> Signed-off-by: Keir Fraser <keir@xensource.com>
Remove net_driver_util module. Inline the one mac-parsing function
into the netfront and netback drivers. Take the opportunity to fix
the mac-parsing logic. Signed-off-by: Keir Fraser <keir@xensource.com>
[XEND] Wait sufficient time for memory to balloon out before creating a new domain.
On the ES7000 when Dom0 boots up with all of system memory and you try
to bring up a DomU with more than 2GB of memory, xend times out before
the memory is freed causing the domain not to be created. This patch
increases the timeout depending on the amount of memory that needs to be
freed. It also places a cap on the sleep time so that it does not grow
without a limit. This fixes bug# 650
[HVM] Support multiple HVM time device models coming soon.
1: Adopt an abstract layer for periodic time on top of different
HV time device models like PIT, RTC and ACPI time.
2: Extract VMM knowledge from PIT DM code and keep close with
qemu for easy maintain in future.
3: Use guest time (TSC) to drive PIT CLK that helps us to remove
a lot of extra logic previously introduced to convert from host time to
guest time.
4: Some cleanup for SMP code like move cache_tsc_offset to per VP.
[NET] front: Use netif_carrier to model backend-connection state.
The tristate backend_state really only needs to be a boolean. In fact
other virtualisation network drivers such as the s390 already model this
using netif_carrier_*. So by converting backend_state over to the net
carrier flag we can tap into existing infrastructure rather than creating
new ones.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Separate xenbusstate enumeration from typedef. Avoid using the typedef
in Linux coe as it is consider bad coding style. Signed-off-by: Keir Fraser <keir@xensource.com>
Fix IDE CD-drive PIO mode.
CD drives in PIO mode don't work under Xen because of a change in Linux
between 2.6.12 and 2.6.16, as a result of the following thread:
http://lists.parisc-linux.org/pipermail/parisc-linux/2005-August/027197.html
The change breaks systems which have highmem and a swiotlb because the
ide-cd driver doesn't use the swiotlb, resulting in read/writes to/from
highmem pages in PIO mode not working any longer. Xen kernels usually have
both highmem and a swiotlb.
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
This patch adds support for ptc.l emulation for xenlinux.
In 2.6.16 kernel it flush_tlb_range will call global_tlb_purge
directly, so ptc.l shouldn't be used when CONFIG_SMP enable.
But in order to enhance performance (maybe), 2.6.17 kernel in
smp environment will do mm check first. If mm is current->active_mm
and the mm (corresponding process) just runs on the local processor,
kernel only needs to do ptc.l at local processor instead of global purge.
So ptc.l emulation is necessary for 2.6.17 kernel.
Signed-off-by: Kevin Tian <kevin.tian@intel.com> Signed-off-by: Zhang Xiantao <xiantao.zhang@intel.com>
Event-channel driver: Use for_each_possible_cpu instead
of for_each_cpu , or statically initialize data when possible. Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Remove broken inline-asm string functions from Xen/x86. The
generic implementations in string.c are quite sufficient. Signed-off-by: Keir Fraser <keir@xensource.com>
We should not panic the domain for an EINVAL return from the ioremap
hyerpcall. This was only meant for ENOSYS. Long term this is probably
not the best place for this kind of sanity checking.
Signed-off-by: Alex Williamson <alex.williamson@hp.com>
[IA64] make xenLinux/ia64 privcmd mmap not to use dom0 memory
xenLinux/ia64 privcmd mmap uses pseudo physical address space.
it used alloc_pages() to allocate the space.
It wastes dom0 memory and sometimes several hundreds megabytes is
allocated depending on domU memory size.
With this patch xenLinux/ia64 trys to find the region which can be
used safely and uses the reasion.
[IA64] Final patch to make evtchn fully working for xen/ia64
OK, this is the last one of patch sets, which enables xen/ia64 to work
on event channel mechanism fully. Now user may observe /proc/interrupts
of dom0 as:
Based on this patch, dom0's performance is not affected, while domU's
network output is seen with several percents increase. More important,
now xen/ia64 can progress a base step towards driver domain support!
New memory_op subops which return the apparent or actual physical address map.
The new subops return a memory map in e820 format. This will allow the
removal of some Xen special casing in the Linux port by using the same
code as native.
Signed-off-by: Ian Campbell <ian.campbell@xensource.com>
[LOADER] More sanity checks when parsing Elf images to avoid
out-of-bounds array accesses when loading the image. Signed-off-by: Keir Fraser <keir@xensource.com>
[ACM] Replace the union structure parameter of the ACM hypercalls
with command-specific request structures. It aligns the ACM
hypercalls with the way parameters are passed in the event channel
hypercalls.
Advantages include backward-compatibility regarding old guests when
new calls are added and clarity of the code.
[LOADER] Another change to the elf_paddr_offset/virt_base logic.
This avoids using zero to indicate 'undefined'. Instead we use
an explicit boolean to indicate whether suitable defaults should
be used. As well as making the logic clearer (I believe) it also
avoids problems if anyone ever explicitly specifies virt_base
(and also elf_paddr_offset) as zero.
[IA64] sparse-merge: check SPARSEDIR for modifications
Check SPARSEDIR for modifications before allowing the merge.
This removes the need for the FIXME. Additionally clean up
empty diffs at the end of the run.
Here's another sparse-merge patch to allow ARCH to be set on the
command-line, and to add -C to the final hg update since it's probably
switching between branches.
[LOADER] Remove check for VIRT_BASE existence: it's not needed
to prevent xend crashign (when Aravindh's latest patch is
applied) and breaks ia64. Also, ensure we set a reasonable
default for elf_paddr_offset.
[BUILDER] Always build initial PAE pagetables with a fourth mid-level
pgdir. This is required for Xen to be able to insert its own protected
mappings. At the same time, clean up 32b and pae pagetable-size
calculations to use the helper macro already used for x86/64.
Based on an original patch from Bruce Rogers at Novell.
The enforce_dom0_cpus test sets the number of vcpus for the Dom0 and
checks to see if they have been set correctly. We found that on
multi-proc systems, it takes a certain amount of time for the number of
vcpus to change.
This patch checks the number of vcpus inside a loop until it is correct
or 20 seconds elapse, whichever happens first. If after 20 seconds the
number of vcpus is not changed correctly, an error is raised.
Several page allocations, done in the hypervisor when starting an HVM
domain, are not checked. This can cause the physical machine to crash
when starting the HVM domain during low-memory conditions.
Kudos to Charles Arnold for catching the problem with
shadow_direct_map_init.
Signed-off-by: Charles Coffing <ccoffing@novell.com>
[XEND] An empirical and more conservative memory-overhead estimate for PV and HVM guests.
This patch calculates the overhead needed for HVM domains. If HVM is
supported by the hardware, I add a little ballooning overhead to
paravirtualized VMs also, to avoid low-memory situations. (There are
various unchecked alloc_domheap_pages calls in shadow*.c that I am
trying to avoid tripping over for now...) The values in this patch work
fine on 32 bit; I may update them later based on feedback and/or testing
on 64 bit.
Signed-off-by: Charles Coffing <ccoffing@novell.com>
[XEND] separate concept of initial memory size and overhead memory size.
When a domain (whether para- or fully-virtualized) reports how much
overhead memory it requires (via getDomainMemory in image.py), all such
memory was immediately allocated to the domain itself. This is
certainly incorrect for HVM domains, since additional
increase_reservation calls are made later in qemu. Since all ballooned
memory is already taken, qemu will fail. The fix is to treat the
requested memory size and the overhead size as separate values. The
requested memory size is immediately allocated to the new domain; the
overhead is left unallocated for whatever else might need it later.
Signed-off-by: Charles Coffing <ccoffing@novell.com>
projects / xen.git / log