[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Xen-ia64-devel] [PATCH 0/23] [RFC] VTi domain save/restore
I've been working on IA64 HVM domain save/restore.
Although it isn't completed yet, I'm posting for review/comment.
Please note that I tested the patches only with Linux which is idle
at this moment.
The main patches are 16, 21, 23.
The patch 17-19 should be sent to xen-devel. I'll send them unless there
The patch 14 is for consistency.
The patch 1-10 are random bug fixes/clean ups which aren't necessary
related to HVM domain save/restore.
Could you elaborate on struct vcpu_guest_context_regs rbs_nat, rbs_rnat?
The comment above them is somewhat vague.
I heavily rewrote the rbs related part of getting/setting vcpu context.
I'm not sure about them and I used only rbs_rnat.
I haven't forgotten hvm domain dump-core fix. I need to implement
HVM vcpu context save before fixing it.
- VTi break fault handler
The patch 14 is for consistent VTi cpu context save.
Although I don't see any reason why pUStk/pKStk are set
by VTi break fault handler, please review.
And people want to test it more before commit.
- RSE (both PV and HVM domain)
The number of physical stacked general register(RSE.N_STACKED_PHYS)
isn't virtualized. Guest OS utilizes it via PAL_RSE_INFO call.
If the number of cpu where domain is saved/restored aren't same,
what should be supposed to do?
The SDM says only that the number is cpu implementation specific.
So the easiest way is to refuse restore.
- Guest OS specific optimization. (both PV and HVM domain)
Xen/IA64 supports guest OS optimization and PV guest kernel or
GFW of HVM domain call opt_feature hypercall.
However the optimization information is lost between save/restore.
It should also be saved/restored.
struct opt_feature seems stable for save/restore ABI.
Or does someone else want to play with opt_feature hypercall?
- Timers of virtual devices
The current virtual device implementation in xen (vacpi, viosapic...)
uses the xen internal timer. However they aren't stopped/restarted when
guest domain pause/unpause. Something should be done here.
It isn't ia64 specific, x86 seems to have same issue.
- BUFFER PIO (HVM domain specific)
This isn't addressed yet. I'm going to attack it.
[PATCH 1/23] remove duplicate xc_get/set_hvm_param() definitions
[PATCH 2/23] minor clean up of sync_vcpu_execstate()
[PATCH 3/23] avoid set cr.ivt when context switch if possible
[PATCH 4/23] remove dead code
[PATCH 5/23] clean up: vmx_vcpu_set_rr()
[PATCH 6/23] fix vmx_emul_mov_from_dbr/ibr()
[PATCH 7/23] make vmx_setup_platfrom() return error.
[PATCH 8/23] clean up of ia64 hvm builder
[PATCH 9/23] minor clean up of hypervisor.h
[PATCH 10/23] xencomm: add resumedomain support.
[PATCH 11/23] vti domain save/restore: add show_stack() prototype
[PATCH 12/23] vti domain save/restore: fix stack unwinder
[PATCH 13/23] vti domain save/restore: add unwind directive to break fault
[PATCH 14/23] [RFC] vti domain save/restore: clean up vti break fault handler
[PATCH 15/23] vti domain save/restore: make vmx_vcpu_set_rr() accept non-current
[PATCH 16/23] [RFC] vti domain save/restore: implement
[PATCH 17/23] vti domain save/restore: split hvm/support.h into arch
[PATCH 18/23] vti domain save/restore: split save.h into arch generic/dependent
[PATCH 19/23] vti domain save/restore: split save.c into arch generic/dependent
[PATCH 20/23] vti domain save/restore: linux xencomm; add hvm_set/get_context
[PATCH 21/23] [RFC] vti domain save/restore: implement hvm_save/load. work in
[PATCH 22/23] vti domain save/restore: libxc: add support set/get_hvmcontext
[PATCH 23/23] [RFC] vti domain save/restore: libxc: implement vti domain
Xen-ia64-devel mailing list