tools/libs: add missing support of linear p2m_list, cleanup
v2 cover tools/libs: add missing support of linear p2m_list, cleanup
There are some corners left which don't support the not so very new linear p2m list of pv guests, which has been introduced in Linux kernel 3.19 and which is mandatory for non-legacy versions of Xen since kernel 4.14.
This series adds support for the linear p2m list where it is missing (colo support and "xl dump-core").
In theory it should be possible to merge the p2m list mapping code from migration handling and core dump handling, but this needs quite some cleanup before this is possible.
The first three patches of this series are fixing real problems, so I've put them at the start of this series, especially in order to make backports easier.
The other three patches are only the first steps of cleanup. The main work done here is to concentrate all p2m mapping in libxenguest instead of having one implementation in each of libxenguest and libxenctrl.
Merging the two implementations should be rather easy, but this will require to touch many lines of code, as the migration handling variant seems to be more mature, but it is using the migration stream specific structures heavily. So I'd like to have some confirmation that my way to clean this up is the right one.
My idea would be to add the data needed for p2m mapping to struct domain_info_context and replace the related fields in struct xc_sr_context with a struct domain_info_context. Modifying the interface of xc_core_arch_map_p2m() to take most current parameters via struct domain_info_context would then enable migration coding to use xc_core_arch_map_p2m() for mapping the p2m. xc_core_arch_map_p2m() should look basically like the current migration p2m mapping code afterwards.
Any comments to that plan?
Changes in V2:
- added missing #include in ocaml stub
Juergen Gross (6):
tools/libs/guest: fix max_pfn setting in map_p2m() tools/libs/ctrl: fix xc_core_arch_map_p2m() to support linear p2m
table
tools/libs/ctrl: use common p2m mapping code in xc_domain_resume_any() tools/libs: move xc_resume.c to libxenguest
tools/libs: move xc_core* from libxenctrl to libxenguest tools/libs/guest: make some definitions private to libxenguest
tools/include/xenctrl.h | 63 ---
tools/include/xenguest.h | 63 +++
tools/libs/ctrl/Makefile | 4 -
tools/libs/ctrl/xc_core_x86.c | 223 ----------
tools/libs/ctrl/xc_domain.c | 2 -
tools/libs/ctrl/xc_private.h | 43 +-
tools/libs/guest/Makefile | 4 +
.../libs/{ctrl/xc_core.c => guest/xg_core.c} | 7 +-
.../libs/{ctrl/xc_core.h => guest/xg_core.h} | 15 +-
.../xc_core_arm.c => guest/xg_core_arm.c} | 31 +-
.../xc_core_arm.h => guest/xg_core_arm.h} | 0
tools/libs/guest/xg_core_x86.c | 399 ++++++++++++++++++
.../xc_core_x86.h => guest/xg_core_x86.h} | 0
tools/libs/guest/xg_dom_boot.c | 2 +-
tools/libs/guest/xg_domain.c | 19 +-
tools/libs/guest/xg_offline_page.c | 2 +-
tools/libs/guest/xg_private.h | 16 +-
.../{ctrl/xc_resume.c => guest/xg_resume.c} | 69 +--
tools/libs/guest/xg_sr_save_x86_pv.c | 2 +-
tools/ocaml/libs/xc/xenctrl_stubs.c | 1 +
20 files changed, 545 insertions(+), 420 deletions(-)
delete mode 100644 tools/libs/ctrl/xc_core_x86.c
rename tools/libs/{ctrl/xc_core.c => guest/xg_core.c} (99%)
rename tools/libs/{ctrl/xc_core.h => guest/xg_core.h} (92%)
rename tools/libs/{ctrl/xc_core_arm.c => guest/xg_core_arm.c} (72%)
rename tools/libs/{ctrl/xc_core_arm.h => guest/xg_core_arm.h} (100%)
create mode 100644 tools/libs/guest/xg_core_x86.c
rename tools/libs/{ctrl/xc_core_x86.h => guest/xg_core_x86.h} (100%)
rename tools/libs/{ctrl/xc_resume.c => guest/xg_resume.c} (80%)Signature · 1 line
2.26.2
PATCH v2 1/6 tools/libs/guest: fix max_pfn setting in map_p2m()
When setting the highest pfn used in the guest, don't subtract 1 from the value read from the shared_info data. The value read already is the correct pfn.
---
This is a backport candidate
--- tools/libs/guest/xg_sr_save_x86_pv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/libs/guest/xg_sr_save_x86_pv.c +1 -1
@@ -468,7 +468,7 @@ static int map_p2m(struct xc_sr_context *ctx) ctx->x86.pv.p2m_generation = ~0ULL ; ctx->x86.pv.max_pfn =GET_FIELD (ctx->x86.pv.shinfo, arch.max_pfn, - ctx->x86.pv.width) -1 ; + ctx->x86.pv.width); p2m_cr3 =GET_FIELD (ctx->x86.pv.shinfo, arch.p2m_cr3, ctx->x86.pv.width);return p2m_cr3 ?map_p2m_list (ctx, p2m_cr3) :map_p2m_tree (ctx);
Signature · 1 line
2.26.2
Re: [PATCH v2 1/6] tools/libs/guest: fix max_pfn setting in map_p2m()
Quoted 6 lines · Juergen Gross
When setting the highest pfn used in the guest, don't subtract 1 from the value read from the
shared_infodata. The value read already is the correct pfn.Fixes 91e204d37f449 ("libxc: try to find last used pfn when migrating")Signed-off-by Juergen Gross <[email protected]>
PATCH v2 2/6 tools/libs/ctrl: fix xc_core_arch_map_p2m() to support linear p2m table
In reply to Juergen GrossThe core of a pv linux guest produced via "xl dump-core" is not usable as since kernel 4.14 only the linear p2m table is kept if Xen indicates it is supporting that. Unfortunately xc_core_arch_map_p2m() is still supporting the 3-level p2m tree only.
Fix that by copying the functionality of map_p2m() from libxenguest to libxenctrl.
Additionally the mapped p2m isn't of a fixed length now, so the interface to the mapping functions needs to be adapted. In order not to add even more parameters, expand struct domain_info_context and use a pointer to that as a parameter.
---
This is a backport candidate.
--- tools/include/xenguest.h | 1 + tools/libs/ctrl/xc_core.c | 5 +- tools/libs/ctrl/xc_core.h | 8 +- tools/libs/ctrl/xc_core_arm.c | 23 +-- tools/libs/ctrl/xc_core_x86.c | 256 ++++++++++++++++++++++++++++------ tools/libs/ctrl/xc_private.h | 1 + tools/libs/guest/xg_domain.c | 17 +-- 7 files changed, 233 insertions(+), 78 deletions(-)
tools/include/xenguest.h +1 -0
@@ -700,6 +700,7 @@ struct xc_domain_meminfo {xen_pfn_t *pfn_type;xen_pfn_t *p2m_table;unsigned long p2m_size; +unsigned int p2m_frames; }; int xc_map_domain_meminfo(xc_interface *xch, uint32_t domid,
tools/libs/ctrl/xc_core.c +2 -3
@@ -574,8 +574,7 @@ xc_domain_dumpcore_via_callback(xc_interface *xch, goto out; } - sts =xc_core_arch_map_p2m (xch, dinfo->guest_width, &info, live_shinfo, - &p2m, &dinfo->p2m_size); + sts =xc_core_arch_map_p2m (xch, dinfo, &info, live_shinfo, &p2m);if ( sts !=0 ) goto out;
@@ -945,7 +944,7 @@ out:if ( memory_map !=NULL )free (memory_map);if ( p2m !=NULL ) -munmap (p2m,PAGE_SIZE *P2M_FL_ENTRIES ); +munmap (p2m,PAGE_SIZE * dinfo->p2m_frames);if ( p2m_array !=NULL )free (p2m_array);if ( pfn_array !=NULL )
tools/libs/ctrl/xc_core.h +4 -4
@@ -138,14 +138,14 @@ int xc_core_arch_memory_map_get(xc_interface *xch, xc_dominfo_t *info, shared_info_any_t *live_shinfo, xc_core_memory_map_t **mapp, unsigned int *nr_entries); -int xc_core_arch_map_p2m (xc_interface *xch,unsigned int guest_width, +int xc_core_arch_map_p2m (xc_interface *xch,struct domain_info_context *dinfo,xc_dominfo_t *info,shared_info_any_t *live_shinfo, -xen_pfn_t **live_p2m,unsigned long *pfnp); +xen_pfn_t **live_p2m); -int xc_core_arch_map_p2m_writable (xc_interface *xch,unsigned int guest_width, +int xc_core_arch_map_p2m_writable (xc_interface *xch,struct domain_info_context *dinfo,xc_dominfo_t *info,shared_info_any_t *live_shinfo, -xen_pfn_t **live_p2m,unsigned long *pfnp); +xen_pfn_t **live_p2m);int xc_core_arch_get_scratch_gpfn (xc_interface *xch,uint32_t domid,xen_pfn_t *gpfn);
tools/libs/ctrl/xc_core_arm.c +7 -16
@@ -66,33 +66,24 @@ xc_core_arch_memory_map_get(xc_interface *xch, struct xc_core_arch_context *unusstatic int xc_core_arch_map_p2m_rw (xc_interface *xch,struct domain_info_context *dinfo,xc_dominfo_t *info, -shared_info_any_t *live_shinfo,xen_pfn_t **live_p2m, -unsigned long *pfnp,int rw) +shared_info_any_t *live_shinfo,xen_pfn_t **live_p2m,int rw) { errno =ENOSYS ;return -1 ; }int -xc_core_arch_map_p2m (xc_interface *xch,unsigned int guest_width,xc_dominfo_t *info, -shared_info_any_t *live_shinfo,xen_pfn_t **live_p2m, -unsigned long *pfnp) +xc_core_arch_map_p2m (xc_interface *xch,struct domain_info_context *dinfo,xc_dominfo_t *info, +shared_info_any_t *live_shinfo,xen_pfn_t **live_p2m) { -struct domain_info_context _dinfo = { .guest_width = guest_width }; -struct domain_info_context *dinfo = &_dinfo; -return xc_core_arch_map_p2m_rw (xch, dinfo, info, - live_shinfo, live_p2m, pfnp,0 ); +return xc_core_arch_map_p2m_rw (xch, dinfo, info, live_shinfo, live_p2m,0 ); }int -xc_core_arch_map_p2m_writable (xc_interface *xch,unsigned int guest_width,xc_dominfo_t *info, -shared_info_any_t *live_shinfo,xen_pfn_t **live_p2m, -unsigned long *pfnp) +xc_core_arch_map_p2m_writable (xc_interface *xch,struct domain_info_context *dinfo,xc_dominfo_t *info, +shared_info_any_t *live_shinfo,xen_pfn_t **live_p2m) { -struct domain_info_context _dinfo = { .guest_width = guest_width }; -struct domain_info_context *dinfo = &_dinfo; -return xc_core_arch_map_p2m_rw (xch, dinfo, info, - live_shinfo, live_p2m, pfnp,1 ); +return xc_core_arch_map_p2m_rw (xch, dinfo, info, live_shinfo, live_p2m,1 ); }int
tools/libs/ctrl/xc_core_x86.c +212 -44
308 lines not on this page · raw message
tools/libs/ctrl/xc_private.h +1 -0
@@ -79,6 +79,7 @@ struct iovec {struct domain_info_context {unsigned int guest_width; +unsigned int p2m_frames;unsigned long p2m_size; };
tools/libs/guest/xg_domain.c +6 -11
@@ -24,13 +24,9 @@int xc_unmap_domain_meminfo (xc_interface *xch,struct xc_domain_meminfo *minfo) { -struct domain_info_context _di = { .guest_width = minfo->guest_width, - .p2m_size = minfo->p2m_size}; -struct domain_info_context *dinfo = &_di; -free (minfo->pfn_type);if ( minfo->p2m_table ) -munmap (minfo->p2m_table,P2M_FL_ENTRIES *PAGE_SIZE ); +munmap (minfo->p2m_table, minfo->p2m_frames *PAGE_SIZE ); minfo->p2m_table =NULL ;return 0 ;
@@ -40,7 +36,6 @@ int xc_map_domain_meminfo(xc_interface *xch, uint32_t domid, struct xc_domain_meminfo *minfo) {struct domain_info_context _di; -struct domain_info_context *dinfo = &_di;xc_dominfo_t info;shared_info_any_t *live_shinfo;
@@ -96,16 +91,16 @@ int xc_map_domain_meminfo(xc_interface *xch, uint32_t domid,return -1 ; } -if (xc_core_arch_map_p2m_writable (xch, minfo->guest_width, &info, - live_shinfo, &minfo->p2m_table, - &minfo->p2m_size) ) +if (xc_core_arch_map_p2m_writable (xch, &_di, &info, + live_shinfo, &minfo->p2m_table) ) {PERROR ("Could not map the P2M table" );munmap (live_shinfo,PAGE_SIZE );return -1 ; }munmap (live_shinfo,PAGE_SIZE ); - _di.p2m_size = minfo->p2m_size; + minfo->p2m_size = _di.p2m_size; + minfo->p2m_frames = _di.p2m_frames;/* Make space and prepare for getting the PFN types */ minfo->pfn_type =calloc (sizeof (*minfo->pfn_type), minfo->p2m_size);
@@ -141,7 +136,7 @@ failed: }if ( minfo->p2m_table ) { -munmap (minfo->p2m_table,P2M_FL_ENTRIES *PAGE_SIZE ); +munmap (minfo->p2m_table, minfo->p2m_frames *PAGE_SIZE ); minfo->p2m_table =NULL ; }
Signature · 1 line
2.26.2
Re: [PATCH v2 2/6] tools/libs/ctrl: fix xc_core_arch_map_p2m() to support linear p2m table
Quoted 8 lines · Juergen Gross
The core of a pv linux guest produced via "xl dump-core" is not usable as since kernel 4.14 only the linear p2m table is kept if Xen indicates it is supporting that. Unfortunately
xc_core_arch_map_p2m()is still supporting the 3-level p2m tree only.Fix that by copying the functionality of
map_p2m()from libxenguest to libxenctrl.
So there are now two copies of the same logic? Is it possible to reduce it to only one?
Wei.
Re: [PATCH v2 2/6] tools/libs/ctrl: fix xc_core_arch_map_p2m() to support linear p2m table
Quoted 6 lines · Wei Liu
Quoted 5 lines · Juergen Gross
The core of a pv linux guest produced via "xl dump-core" is not usable as since kernel 4.14 only the linear p2m table is kept if Xen indicates it is supporting that. Unfortunately
xc_core_arch_map_p2m()is still supporting the 3-level p2m tree only.
Quoted 6 lines
Quoted 3 lines
Fix that by copying the functionality of
map_p2m()from libxenguest to libxenctrl.So there are now two copies of the same logic? Is it possible to reduce it to only one?
Yes. See the intro mail of the series.
I wanted to fix the issue first, before doing the major cleanup work.
Juergen
Wei Liu
Acked-byRe: [PATCH v2 2/6] tools/libs/ctrl: fix xc_core_arch_map_p2m() to support linear p2m table
Quoted 11 lines · Juergen Gross
Quoted 10 lines · Wei Liu
Quoted 8 lines · Juergen Gross
The core of a pv linux guest produced via "xl dump-core" is not usable as since kernel 4.14 only the linear p2m table is kept if Xen indicates it is supporting that. Unfortunately
xc_core_arch_map_p2m()is still supporting the 3-level p2m tree only.Fix that by copying the functionality of
map_p2m()from libxenguest to libxenctrl.
Quoted 6 lines
Quoted 2 lines
So there are now two copies of the same logic? Is it possible to reduce it to only one?
Yes. See the intro mail of the series.
I wanted to fix the issue first, before doing the major cleanup work.
PATCH v2 3/6 tools/libs/ctrl: use common p2m mapping code in xc_domain_resume_any()
In reply to Juergen GrossInstead of open coding the mapping of the p2m list use the already existing xc_core_arch_map_p2m() call, especially as the current code does not support guests with the linear p2m map. It should be noted that this code is needed for colo/remus only.
Switching to xc_core_arch_map_p2m() drops the need to bail out for bitness of tool stack and guest differing.
---
This might be a backport candidate
V2:
- add missing #include in ocaml stub (Andrew Cooper)
--- tools/libs/ctrl/xc_resume.c | 66 +++++++++-------------------- tools/ocaml/libs/xc/xenctrl_stubs.c | 1 + 2 files changed, 22 insertions(+), 45 deletions(-)
tools/libs/ctrl/xc_resume.c +21 -45
@@ -20,6 +20,7 @@#include <xen/foreign/x86_32.h> #include <xen/foreign/x86_64.h> #include <xen/hvm/params.h> +#include "xc_core.h" static int modify_returncode (xc_interface *xch,uint32_t domid) {
@@ -137,12 +138,10 @@ static int xc_domain_resume_any(xc_interface *xch, uint32_t domid)struct domain_info_context _dinfo = { .guest_width =0 , .p2m_size =0 };struct domain_info_context *dinfo = &_dinfo; -unsigned long mfn; +xen_pfn_t mfn, store_mfn, console_mfn;vcpu_guest_context_any_t ctxt; -start_info_t *start_info; -shared_info_t *shinfo =NULL ; -xen_pfn_t *p2m_frame_list_list =NULL ; -xen_pfn_t *p2m_frame_list =NULL ; +start_info_any_t *start_info; +shared_info_any_t *shinfo =NULL ;xen_pfn_t *p2m =NULL ;#endif
@@ -164,11 +163,6 @@ static int xc_domain_resume_any(xc_interface *xch, uint32_t domid)PERROR ("Could not get domain width" );return rc; } -if ( dinfo->guest_width !=sizeof (long ) ) - { -ERROR ("Cannot resume uncooperative cross-address-size guests" ); -return rc; - }/* Map the shared info frame */ shinfo = xc_map_foreign_range(xch, domid, PAGE_SIZE,
@@ -179,34 +173,8 @@ static int xc_domain_resume_any(xc_interface *xch, uint32_t domid) goto out; } - dinfo->p2m_size = shinfo->arch.max_pfn; - - p2m_frame_list_list = -xc_map_foreign_range (xch, domid,PAGE_SIZE ,PROT_READ , - shinfo->arch.pfn_to_mfn_frame_list_list); -if ( p2m_frame_list_list ==NULL ) - { -ERROR ("Couldn't map p2m_frame_list_list" ); - goto out; - } - - p2m_frame_list =xc_map_foreign_pages (xch, domid,PROT_READ , - p2m_frame_list_list, -P2M_FLL_ENTRIES ); -if ( p2m_frame_list ==NULL ) - { -ERROR ("Couldn't map p2m_frame_list" ); - goto out; - } - -/* Map all the frames of the pfn->mfn table. For migrate to succeed, -the guest must not change which frames are used for this purpose. -(its not clear why it would want to change them, and we'll be OK -from a safety POV anyhow. */ - p2m =xc_map_foreign_pages (xch, domid,PROT_READ , - p2m_frame_list, -P2M_FL_ENTRIES ); -if ( p2m ==NULL ) +/* Map the p2m list */ +if (xc_core_arch_map_p2m (xch, dinfo, &info, shinfo, &p2m) ) {ERROR ("Couldn't map p2m table" ); goto out;
@@ -228,8 +196,20 @@ static int xc_domain_resume_any(xc_interface *xch, uint32_t domid) goto out; } - start_info->store_mfn = p2m[start_info->store_mfn]; - start_info->console.domU.mfn = p2m[start_info->console.domU.mfn]; + store_mfn =GET_FIELD (start_info, store_mfn, dinfo->guest_width); + console_mfn =GET_FIELD (start_info, console.domU.mfn, dinfo->guest_width); +if ( dinfo->guest_width ==4 ) + { + store_mfn = ((uint32_t *)p2m)[store_mfn]; + console_mfn = ((uint32_t *)p2m)[console_mfn]; + } +else + { + store_mfn = ((uint64_t *)p2m)[store_mfn]; + console_mfn = ((uint64_t *)p2m)[console_mfn]; + } +SET_FIELD (start_info, store_mfn, store_mfn, dinfo->guest_width); +SET_FIELD (start_info, console.domU.mfn, console_mfn, dinfo->guest_width);munmap (start_info,PAGE_SIZE );#endif /* defined(__i386__) || defined(__x86_64__) */
@@ -250,11 +230,7 @@ static int xc_domain_resume_any(xc_interface *xch, uint32_t domid) out:#if defined(__i386__) || defined(__x86_64__)if (p2m) -munmap (p2m,P2M_FL_ENTRIES *PAGE_SIZE ); -if (p2m_frame_list) -munmap (p2m_frame_list,P2M_FLL_ENTRIES *PAGE_SIZE ); -if (p2m_frame_list_list) -munmap (p2m_frame_list_list,PAGE_SIZE ); +munmap (p2m, dinfo->p2m_frames *PAGE_SIZE );if (shinfo)munmap (shinfo,PAGE_SIZE );#endif
tools/ocaml/libs/xc/xenctrl_stubs.c +1 -0
@@ -32,6 +32,7 @@#define XC_WANT_COMPAT_MAP_FOREIGN_API #include <xenctrl.h> +#include <xenguest.h> #include <xen-tools/libs.h> #include "mmap_stubs.h"
Signature · 1 line
2.26.2
Re: [PATCH v2 3/6] tools/libs/ctrl: use common p2m mapping code in xc_domain_resume_any()
On 12 Apr 2021, at 16:22, Juergen Gross <[email protected]<mailto:[email protected]>> wrote:
Instead of open coding the mapping of the p2m list use the already existing xc_core_arch_map_p2m() call, especially as the current code does not support guests with the linear p2m map. It should be noted that this code is needed for colo/remus only.
[..]
tools/ocaml/libs/xc/xenctrl_stubs.c +0 -0
@@ -32,6 +32,7 @@
#define XC_WANT_COMPAT_MAP_FOREIGN_API
#include <xenctrl.h>
+#include <xenguest.h>
#include <xen-tools/libs.h>
#include "mmap_stubs.h"
—
2.26.2
Do we need this when there are no other changes in the file?
Re: [PATCH v2 3/6] tools/libs/ctrl: use common p2m mapping code in xc_domain_resume_any()
Quoted 20 lines · Christian Lindig
Quoted 20 lines · On 12 Apr 2021, at 16:22, Juergen Gross <[email protected]…
Instead of open coding the mapping of the p2m list use the already existing
xc_core_arch_map_p2m()call, especially as the current code does not support guests with the linear p2m map. It should be noted that this code is needed for colo/remus only.
[..]tools/ocaml/libs/xc/xenctrl_stubs.c +1
-0@@ -32,6 +32,7 @@#define XC_WANT_COMPAT_MAP_FOREIGN_API #include <xenctrl.h> +#include <xenguest.h> #include <xen-tools/libs.h>
Quoted 5 lines
Quoted 3 lines
tools/ocaml/libs/xc/xenctrl_stubs.c +0
-0#include "mmap_stubs.h" —
2.26.2Do we need this when there are no other changes in the file?
Yes. Some definitions moved from xenctrl.h into xenguest.h.
Quoted 2 lines
Acked-by Christian Lindig <[email protected] <mailto:[email protected]>>
Thanks,
Juergen
Re: [PATCH v2 3/6] tools/libs/ctrl: use common p2m mapping code in xc_domain_resume_any()
In reply to Juergen GrossQuoted 9 lines · Juergen Gross
Instead of open coding the mapping of the p2m list use the already existing
xc_core_arch_map_p2m()call, especially as the current code does not support guests with the linear p2m map. It should be noted that this code is needed for colo/remus only.Switching to
xc_core_arch_map_p2m()drops the need to bail out for bitness of tool stack and guest differing.Signed-off-by Juergen Gross <[email protected]>
PATCH v2 4/6 tools/libs: move xc_resume.c to libxenguest
In reply to Juergen GrossThe guest suspend functionality is already part of libxenguest. Move the resume functionality from libxenctrl to libxenguest, too.
---
tools/include/xenctrl.h | 63 -------------------
tools/include/xenguest.h | 62 ++++++++++++++++++
tools/libs/ctrl/Makefile | 1 -
tools/libs/guest/Makefile | 1 +
.../{ctrl/xc_resume.c => guest/xg_resume.c} | 1 +
5 files changed, 64 insertions(+), 64 deletions(-)
rename tools/libs/{ctrl/xc_resume.c => guest/xg_resume.c} (99%)tools/include/xenctrl.h +0 -63
@@ -576,69 +576,6 @@ int xc_domain_destroy(xc_interface *xch, uint32_t domid); -/** -* This function resumes a suspended domain. The domain should have -* been previously suspended. -* -* Note that there are 'xc_domain_suspend' as suspending a domain -* is quite the endeavour. -* -* For the purpose of this explanation there are three guests: -* PV (using hypercalls for privilgied operations), HVM -* (fully hardware virtualized guests using emulated devices for everything), -* and PVHVM (PV aware with hardware virtualisation). -* -* HVM guest are the simplest - they suspend via S3 / S4 and resume from -* S3 / S4. Upon resume they have to re-negotiate with the emulated devices. -* -* PV and PVHVM communicate via hypercalls for suspend (and resume). -* For suspend the toolstack initiates the process by writing an value -* in XenBus "control/shutdown" with the string "suspend". -* -* The PV guest stashes anything it deems neccessary in 'struct -* start_info' in case of failure (PVHVM may ignore this) and calls -* the SCHEDOP_shutdown::SHUTDOWN_suspend hypercall (for PV as -* argument it passes the MFN to 'struct start_info'). -* -* And then the guest is suspended. -* -* The checkpointing or notifying a guest that the suspend failed or -* cancelled (in case of checkpoint) is by having the -* SCHEDOP_shutdown::SHUTDOWN_suspend hypercall return a non-zero -* value. -* -* The PV and PVHVM resume path are similar. For PV it would be -* similar to bootup - figure out where the 'struct start_info' is (or -* if the suspend was cancelled aka checkpointed - reuse the saved -* values). -* -* From here on they differ depending whether the guest is PV or PVHVM -* in specifics but follow overall the same path: -* - PV: Bringing up the vCPUS, -* - PVHVM: Setup vector callback, -* - Bring up vCPU runstates, -* - Remap the grant tables if checkpointing or setup from scratch, -* -* -* If the resume was not checkpointing (or if suspend was succesful) we would -* setup the PV timers and the different PV events. Lastly the PV drivers -* re-negotiate with the backend. -* -* This function would return before the guest started resuming. That is -* the guest would be in non-running state and its vCPU context would be -* in the the SCHEDOP_shutdown::SHUTDOWN_suspend hypercall return path -* (for PV and PVHVM). For HVM it would be in would be in QEMU emulated -* BIOS handling S3 suspend. -* -* @parm xch a handle to an open hypervisor interface -* @parm domid the domain id to resume -* @parm fast use cooperative resume (guest must support this) -* return 0 on success, -1 on failure -*/ -int xc_domain_resume (xc_interface *xch, -uint32_t domid, -int fast); -/** * This function will shutdown a domain. This is intended for use in * fully-virtualized domains where this operation is analogous to the
tools/include/xenguest.h +62 -0
68 lines not on this page · raw message
tools/libs/ctrl/Makefile +0 -1
@@ -20,7 +20,6 @@ SRCS-y += xc_rt.c SRCS-y += xc_tbuf.c SRCS-y += xc_pm.c SRCS-y += xc_cpu_hotplug.c -SRCS-y += xc_resume.c SRCS-y += xc_vm_event.c SRCS-y += xc_vmtrace.c SRCS-y += xc_monitor.c
tools/libs/guest/Makefile +1 -0
@@ -9,6 +9,7 @@ endif SRCS-y += xg_private.c SRCS-y += xg_domain.c SRCS-y += xg_suspend.c +SRCS-y += xg_resume.c ifeq ($(CONFIG_MIGRATE),y) SRCS-y += xg_sr_common.c SRCS-$(CONFIG_X86) += xg_sr_common_x86.c
tools/libs/guest/xg_resume.c renamed from tools/libs/ctrl/xc_resume.c +1 -0
@@ -14,6 +14,7 @@*/ #include "xc_private.h" +#include "xenguest.h" #if defined(__i386__) || defined(__x86_64__)
Signature · 1 line
2.26.2
Re: [PATCH v2 4/6] tools/libs: move xc_resume.c to libxenguest
Quoted 4 lines · Juergen Gross
The guest suspend functionality is already part of libxenguest. Move the resume functionality from libxenctrl to libxenguest, too.
Signed-off-by Juergen Gross <[email protected]>
PATCH v2 5/6 tools/libs: move xc_core* from libxenctrl to libxenguest
In reply to Juergen GrossThe functionality in xc_core* should be part of libxenguest instead of libxenctrl. Users are already either in libxenguest, or in xl. There is one single exception: xc_core_arch_auto_translated_physmap() is being used by xc_domain_memory_mapping(), which is used by qemu. So leave the xc_core_arch_auto_translated_physmap() functionality in libxenctrl.
This will make it easier to merge common functionality of xc_core* and xg_sr_save*.
---
tools/libs/ctrl/Makefile | 3 ---
tools/libs/ctrl/xc_domain.c | 2 --
tools/libs/ctrl/xc_private.h | 12 ++++++++++++
tools/libs/guest/Makefile | 3 +++
tools/libs/{ctrl/xc_core.c => guest/xg_core.c} | 2 +-
tools/libs/{ctrl/xc_core.h => guest/xg_core.h} | 5 ++---
.../libs/{ctrl/xc_core_arm.c => guest/xg_core_arm.c} | 8 +-------
.../libs/{ctrl/xc_core_arm.h => guest/xg_core_arm.h} | 0
.../libs/{ctrl/xc_core_x86.c => guest/xg_core_x86.c} | 8 +-------
.../libs/{ctrl/xc_core_x86.h => guest/xg_core_x86.h} | 0
tools/libs/guest/xg_dom_boot.c | 2 +-
tools/libs/guest/xg_domain.c | 2 +-
tools/libs/guest/xg_offline_page.c | 2 +-
tools/libs/guest/xg_resume.c | 2 +-
14 files changed, 24 insertions(+), 27 deletions(-)
rename tools/libs/{ctrl/xc_core.c => guest/xg_core.c} (99%)
rename tools/libs/{ctrl/xc_core.h => guest/xg_core.h} (97%)
rename tools/libs/{ctrl/xc_core_arm.c => guest/xg_core_arm.c} (96%)
rename tools/libs/{ctrl/xc_core_arm.h => guest/xg_core_arm.h} (100%)
rename tools/libs/{ctrl/xc_core_x86.c => guest/xg_core_x86.c} (99%)
rename tools/libs/{ctrl/xc_core_x86.h => guest/xg_core_x86.h} (100%)tools/libs/ctrl/Makefile +0 -3
@@ -2,9 +2,6 @@ XEN_ROOT = $(CURDIR)/../../.. include $(XEN_ROOT)/tools/Rules.mk SRCS-y += xc_altp2m.c -SRCS-y += xc_core.c -SRCS-$(CONFIG_X86) += xc_core_x86.c -SRCS-$(CONFIG_ARM) += xc_core_arm.c SRCS-y += xc_cpupool.c SRCS-y += xc_domain.c SRCS-y += xc_evtchn.c
tools/libs/ctrl/xc_domain.c +0 -2
@@ -19,8 +19,6 @@* Copyright (c) 2003, K A Fraser. */ -#include "xc_private.h" -#include "xc_core.h" #include "xc_private.h" #include <xen/memory.h> #include <xen/hvm/hvm_op.h>
tools/libs/ctrl/xc_private.h +12 -0
@@ -467,6 +467,18 @@ void *xc_vm_event_enable(xc_interface *xch, uint32_t domain_id, int param,int do_dm_op (xc_interface *xch,uint32_t domid,unsigned int nr_bufs, ...); +#if defined (__i386__) || defined (__x86_64__) +static inline int xc_core_arch_auto_translated_physmap (const xc_dominfo_t *info) +{ +return info->hvm; +} +#elif defined (__arm__) || defined(__aarch64__) +static inline int xc_core_arch_auto_translated_physmap (const xc_dominfo_t *info) +{ +return 1 ; +} +#endif +#endif /* __XC_PRIVATE_H__ */ /*
tools/libs/guest/Makefile +3 -0
@@ -24,6 +24,9 @@ SRCS-y += xg_offline_page.c else SRCS-y += xg_nomigrate.c endif +SRCS-y += xg_core.c +SRCS-$(CONFIG_X86) += xg_core_x86.c +SRCS-$(CONFIG_ARM) += xg_core_arm.c CFLAGS += -I$(XEN_libxenctrl)
tools/libs/guest/xg_core.c renamed from tools/libs/ctrl/xc_core.c +1 -1
@@ -61,7 +61,7 @@*/ #include "xc_private.h" -#include "xc_core.h" +#include "xg_core.h" #include <stdlib.h> #include <unistd.h>
tools/libs/guest/xg_core.h renamed from tools/libs/ctrl/xc_core.h +2 -3
@@ -131,7 +131,6 @@ struct xc_core_memory_map {uint64_t size; };typedef struct xc_core_memory_map xc_core_memory_map_t ; -int xc_core_arch_auto_translated_physmap (const xc_dominfo_t *info);struct xc_core_arch_context ; int xc_core_arch_memory_map_get(xc_interface *xch, struct xc_core_arch_context *arch_ctxt,
@@ -152,9 +151,9 @@ int xc_core_arch_get_scratch_gpfn(xc_interface *xch, uint32_t domid,#if defined (__i386__) || defined (__x86_64__) -# include "xc_core_x86.h" +# include "xg_core_x86.h" #elif defined (__arm__) || defined(__aarch64__) -# include "xc_core_arm.h" +# include "xg_core_arm.h" #else # error "unsupported architecture"#endif
tools/libs/guest/xg_core_arm.c renamed from tools/libs/ctrl/xc_core_arm.c +1 -7
@@ -17,7 +17,7 @@*/ #include "xc_private.h" -#include "xc_core.h" +#include "xg_core.h" #include <xen-tools/libs.h>
@@ -31,12 +31,6 @@ xc_core_arch_gpfn_may_present(struct xc_core_arch_context *arch_ctxt,return 0 ; } -int -xc_core_arch_auto_translated_physmap (const xc_dominfo_t *info) -{ -return 1 ; -} - int xc_core_arch_memory_map_get(xc_interface *xch, struct xc_core_arch_context *unused, xc_dominfo_t *info, shared_info_any_t *live_shinfo,
tools/libs/guest/xg_core_arm.h renamed from tools/libs/ctrl/xc_core_arm.h +0 -0
tools/libs/guest/xg_core_x86.c renamed from tools/libs/ctrl/xc_core_x86.c +1 -7
@@ -19,7 +19,7 @@#include <inttypes.h> #include "xc_private.h" -#include "xc_core.h" +#include "xg_core.h" #include <xen/hvm/e820.h> int
@@ -33,12 +33,6 @@ xc_core_arch_gpfn_may_present(struct xc_core_arch_context *arch_ctxt,return 1 ; } -int -xc_core_arch_auto_translated_physmap (const xc_dominfo_t *info) -{ -return info->hvm; -} - int xc_core_arch_memory_map_get(xc_interface *xch, struct xc_core_arch_context *unused, xc_dominfo_t *info, shared_info_any_t *live_shinfo,
tools/libs/guest/xg_core_x86.h renamed from tools/libs/ctrl/xc_core_x86.h +0 -0
tools/libs/guest/xg_dom_boot.c +1 -1
@@ -31,7 +31,7 @@#include <zlib.h> #include "xg_private.h" -#include "xc_core.h" +#include "xg_core.h" #include <xen/hvm/params.h> #include <xen/grant_table.h>
tools/libs/guest/xg_domain.c +1 -1
@@ -20,7 +20,7 @@*/ #include "xg_private.h" -#include "xc_core.h" +#include "xg_core.h" int xc_unmap_domain_meminfo (xc_interface *xch,struct xc_domain_meminfo *minfo) {
tools/libs/guest/xg_offline_page.c +1 -1
@@ -25,7 +25,7 @@#include <stdlib.h> #include <unistd.h> #include <sys/time.h> -#include <xc_core.h> +#include <xg_core.h> #include "xc_private.h" #include "xg_private.h"
tools/libs/guest/xg_resume.c +1 -1
@@ -21,7 +21,7 @@#include <xen/foreign/x86_32.h> #include <xen/foreign/x86_64.h> #include <xen/hvm/params.h> -#include "xc_core.h" +#include "xg_core.h" static int modify_returncode (xc_interface *xch,uint32_t domid) {
Signature · 1 line
2.26.2
Re: [PATCH v2 5/6] tools/libs: move xc_core* from libxenctrl to libxenguest
Quoted 11 lines · Juergen Gross
The functionality in xc_core* should be part of libxenguest instead of libxenctrl. Users are already either in libxenguest, or in xl. There is one single exception:
xc_core_arch_auto_translated_physmap()is being used byxc_domain_memory_mapping(), which is used by qemu. So leave thexc_core_arch_auto_translated_physmap()functionality in libxenctrl.This will make it easier to merge common functionality of xc_core* and xg_sr_save*.
Signed-off-by Juergen Gross <[email protected]>
PATCH v2 6/6 tools/libs/guest: make some definitions private to libxenguest
In reply to Juergen GrossThere are some definitions which are used in libxenguest only now. Move them from libxenctrl over to libxenguest.
Remove an unused macro.
--- tools/libs/ctrl/xc_private.h | 32 -------------------------------- tools/libs/guest/xg_core.h | 2 +- tools/libs/guest/xg_core_x86.c | 16 +++++++++++++++- tools/libs/guest/xg_private.h | 16 +++++++++++++++- tools/libs/guest/xg_resume.c | 2 +- 5 files changed, 32 insertions(+), 36 deletions(-)
tools/libs/ctrl/xc_private.h +0 -32
@@ -65,38 +65,6 @@ struct iovec {#define ROUNDUP (_x,_w) (((unsigned long)(_x)+(1UL<<(_w))-1) & ~((1UL<<(_w))-1)) -#define GET_FIELD (_p, _f, _w) (((_w) == 8) ? ((_p)->x64._f) : ((_p)->x32._f)) - -#define SET_FIELD (_p, _f, _v, _w) do { \ - if ((_w) == 8) \ - (_p)->x64._f = (_v); \ - else \ - (_p)->x32._f = (_v); \ -} while (0) - -/* XXX SMH: following skanky macros rely on variable p2m_size being set */ -/* XXX TJD: also, "guest_width" should be the guest's sizeof(unsigned long) */ - -struct domain_info_context { -unsigned int guest_width; -unsigned int p2m_frames; -unsigned long p2m_size; -}; - -/* Number of xen_pfn_t in a page */ -#define FPP (PAGE_SIZE/(dinfo->guest_width)) - -/* Number of entries in the pfn_to_mfn_frame_list_list */ -#define P2M_FLL_ENTRIES (((dinfo->p2m_size)+(FPP*FPP)-1)/(FPP*FPP)) - -/* Number of entries in the pfn_to_mfn_frame_list */ -#define P2M_FL_ENTRIES (((dinfo->p2m_size)+FPP-1)/FPP) - -/* Size in bytes of the pfn_to_mfn_frame_list */ -#define P2M_GUEST_FL_SIZE ((P2M_FL_ENTRIES) * (dinfo->guest_width)) -#define P2M_TOOLS_FL_SIZE ((P2M_FL_ENTRIES) * \ - max_t(size_t, sizeof(xen_pfn_t), dinfo->guest_width)) -#define DECLARE_DOMCTL struct xen_domctl domctl#define DECLARE_SYSCTL struct xen_sysctl sysctl#define DECLARE_PHYSDEV_OP struct physdev_op physdev_op
tools/libs/guest/xg_core.h +1 -1
@@ -21,7 +21,7 @@#define XC_CORE_H #include "xen/version.h" -#include "xc_private.h" +#include "xg_private.h" #include "xen/libelf/elfstructs.h" /* section names */
tools/libs/guest/xg_core_x86.c +15 -1
@@ -18,10 +18,24 @@*/ #include <inttypes.h> -#include "xc_private.h" +#include "xg_private.h" #include "xg_core.h" #include <xen/hvm/e820.h> +/* Number of xen_pfn_t in a page */ +#define FPP (PAGE_SIZE/(dinfo->guest_width)) + +/* Number of entries in the pfn_to_mfn_frame_list_list */ +#define P2M_FLL_ENTRIES (((dinfo->p2m_size)+(FPP*FPP)-1)/(FPP*FPP)) + +/* Number of entries in the pfn_to_mfn_frame_list */ +#define P2M_FL_ENTRIES (((dinfo->p2m_size)+FPP-1)/FPP) + +/* Size in bytes of the pfn_to_mfn_frame_list */ +#define P2M_GUEST_FL_SIZE ((P2M_FL_ENTRIES) * (dinfo->guest_width)) +#define P2M_TOOLS_FL_SIZE ((P2M_FL_ENTRIES) * \ + max_t(size_t, sizeof(xen_pfn_t), dinfo->guest_width)) +int xc_core_arch_gpfn_may_present (struct xc_core_arch_context *arch_ctxt,unsigned long pfn)
tools/libs/guest/xg_private.h +15 -1
@@ -42,6 +42,21 @@#endif #endif +#define GET_FIELD (_p, _f, _w) (((_w) == 8) ? ((_p)->x64._f) : ((_p)->x32._f)) + +#define SET_FIELD (_p, _f, _v, _w) do { \ + if ((_w) == 8) \ + (_p)->x64._f = (_v); \ + else \ + (_p)->x32._f = (_v); \ +} while (0) + +struct domain_info_context { +unsigned int guest_width; +unsigned int p2m_frames; +unsigned long p2m_size; +}; +struct xc_dom_loader {char *name;/* Sadly the error returns from these functions are not consistent: */
@@ -139,7 +154,6 @@ static inline xen_pfn_t xc_pfn_to_mfn(xen_pfn_t pfn, xen_pfn_t *p2m,/* Masks for PTE<->PFN conversions */ #define MADDR_BITS_X86 ((dinfo->guest_width == 8) ? 52 : 44)#define MFN_MASK_X86 ((1ULL << (MADDR_BITS_X86 - PAGE_SHIFT_X86)) - 1) -#define MADDR_MASK_X86 (MFN_MASK_X86 << PAGE_SHIFT_X86)int pin_table (xc_interface *xch,unsigned int type,unsigned long mfn,uint32_t dom);
tools/libs/guest/xg_resume.c +1 -1
@@ -13,7 +13,7 @@* License along with this library; If not, see <http://www.gnu.org/licenses/>. */ -#include "xc_private.h" +#include "xg_private.h" #include "xenguest.h" #if defined(__i386__) || defined(__x86_64__)
Signature · 1 line
2.26.2
Re: [PATCH v2 6/6] tools/libs/guest: make some definitions private to libxenguest
Quoted 6 lines · Juergen Gross
There are some definitions which are used in libxenguest only now. Move them from libxenctrl over to libxenguest.
Remove an unused macro.
Signed-off-by Juergen Gross <[email protected]>
Ping?
Quoted 79 lines · Juergen Gross
There are some corners left which don't support the not so very new linear p2m list of pv guests, which has been introduced in Linux kernel 3.19 and which is mandatory for non-legacy versions of Xen since kernel 4.14.
This series adds support for the linear p2m list where it is missing (colo support and "xl dump-core").
In theory it should be possible to merge the p2m list mapping code from migration handling and core dump handling, but this needs quite some cleanup before this is possible.
The first three patches of this series are fixing real problems, so I've put them at the start of this series, especially in order to make backports easier.
The other three patches are only the first steps of cleanup. The main work done here is to concentrate all p2m mapping in libxenguest instead of having one implementation in each of libxenguest and libxenctrl.
Merging the two implementations should be rather easy, but this will require to touch many lines of code, as the migration handling variant seems to be more mature, but it is using the migration stream specific structures heavily. So I'd like to have some confirmation that my way to clean this up is the right one.
My idea would be to add the data needed for p2m mapping to struct
domain_info_contextand replace the related fields in structxc_sr_contextwith a structdomain_info_context. Modifying the interface ofxc_core_arch_map_p2m()to take most current parameters via structdomain_info_contextwould then enable migration coding to usexc_core_arch_map_p2m()for mapping the p2m.xc_core_arch_map_p2m()should look basically like the current migration p2m mapping code afterwards.Any comments to that plan?
Changes in V2:
- added missing #include in ocaml stubJuergen Gross (6):
tools/libs/guest: fixmax_pfnsetting inmap_p2m()tools/libs/ctrl: fixxc_core_arch_map_p2m()to support linear p2m
table
tools/libs/ctrl: use common p2m mapping code inxc_domain_resume_any()tools/libs: movexc_resume.cto libxenguest
tools/libs: move xc_core* from libxenctrl to libxenguesttools/libs/guest: make some definitions private to libxenguesttools/include/xenctrl.h | 63 --- tools/include/xenguest.h | 63 +++ tools/libs/ctrl/Makefile | 4 - tools/libs/ctrl/xc_core_x86.c | 223 ---------- tools/libs/ctrl/xc_domain.c | 2 - tools/libs/ctrl/xc_private.h | 43 +- tools/libs/guest/Makefile | 4 + .../libs/{ctrl/xc_core.c => guest/xg_core.c} | 7 +- .../libs/{ctrl/xc_core.h => guest/xg_core.h} | 15 +- .../xc_core_arm.c => guest/xg_core_arm.c} | 31 +- .../xc_core_arm.h => guest/xg_core_arm.h} | 0 tools/libs/guest/xg_core_x86.c | 399 ++++++++++++++++++ .../xc_core_x86.h => guest/xg_core_x86.h} | 0 tools/libs/guest/xg_dom_boot.c | 2 +- tools/libs/guest/xg_domain.c | 19 +- tools/libs/guest/xg_offline_page.c | 2 +- tools/libs/guest/xg_private.h | 16 +- .../{ctrl/xc_resume.c => guest/xg_resume.c} | 69 +-- tools/libs/guest/xg_sr_save_x86_pv.c | 2 +- tools/ocaml/libs/xc/xenctrl_stubs.c | 1 + 20 files changed, 545 insertions(+), 420 deletions(-)delete mode 100644
tools/libs/ctrl/xc_core_x86.c
renametools/libs/{ctrl/xc_core.c=>guest/xg_core.c} (99%) renametools/libs/{ctrl/xc_core.h=>guest/xg_core.h} (92%) renametools/libs/{ctrl/xc_core_arm.c=>guest/xg_core_arm.c} (72%) renametools/libs/{ctrl/xc_core_arm.h=>guest/xg_core_arm.h} (100%) create mode 100644tools/libs/guest/xg_core_x86.c
renametools/libs/{ctrl/xc_core_x86.h=>guest/xg_core_x86.h} (100%) renametools/libs/{ctrl/xc_resume.c=>guest/xg_resume.c} (80%)
Quoted 1 line · Juergen Gross
Ping?
Now each patch has an Ack by Wei. Could the series be either applied or get more comments, please?
Juergen
Quoted 51 lines · Juergen Gross
Quoted 49 lines
There are some corners left which don't support the not so very new linear p2m list of pv guests, which has been introduced in Linux kernel 3.19 and which is mandatory for non-legacy versions of Xen since kernel 4.14.
This series adds support for the linear p2m list where it is missing (colo support and "xl dump-core").
In theory it should be possible to merge the p2m list mapping code from migration handling and core dump handling, but this needs quite some cleanup before this is possible.
The first three patches of this series are fixing real problems, so I've put them at the start of this series, especially in order to make backports easier.
The other three patches are only the first steps of cleanup. The main work done here is to concentrate all p2m mapping in libxenguest instead of having one implementation in each of libxenguest and libxenctrl.
Merging the two implementations should be rather easy, but this will require to touch many lines of code, as the migration handling variant seems to be more mature, but it is using the migration stream specific structures heavily. So I'd like to have some confirmation that my way to clean this up is the right one.
My idea would be to add the data needed for p2m mapping to struct
domain_info_contextand replace the related fields in structxc_sr_contextwith a structdomain_info_context. Modifying the interface ofxc_core_arch_map_p2m()to take most current parameters via structdomain_info_contextwould then enable migration coding to usexc_core_arch_map_p2m()for mapping the p2m.xc_core_arch_map_p2m()should look basically like the current migration p2m mapping code afterwards.Any comments to that plan?
Changes in V2:
- added missing #include in ocaml stubJuergen Gross (6):
tools/libs/guest: fixmax_pfnsetting inmap_p2m()tools/libs/ctrl: fixxc_core_arch_map_p2m()to support linear p2m table
tools/libs/ctrl: use common p2m mapping code inxc_domain_resume_any()tools/libs: movexc_resume.cto libxenguest
tools/libs: move xc_core* from libxenctrl to libxenguesttools/libs/guest: make some definitions private to libxenguest
Quoted 6 lines
Quoted 6 lines
tools/include/xenctrl.h | 63 --- tools/include/xenguest.h | 63 +++ tools/libs/ctrl/Makefile | 4 - tools/libs/ctrl/xc_core_x86.c | 223 ---------- tools/libs/ctrl/xc_domain.c | 2 - tools/libs/ctrl/xc_private.h |
43 +-
Quoted 3 lines
Quoted 3 lines
tools/libs/guest/Makefile | 4 +
.../libs/{ctrl/xc_core.c=> guest/xg_core.c} | 7 +-.../libs/{ctrl/xc_core.h=> guest/xg_core.h} | 15
+- >> .../xc_core_arm.c => guest/xg_core_arm.c} | 31 +- >> .../xc_core_arm.h => guest/xg_core_arm.h} | 0 >> tools/libs/guest/xg_core_x86.c | 399 ++++++++++++++++++ >> .../xc_core_x86.h => guest/xg_core_x86.h} | 0 >> tools/libs/guest/xg_dom_boot.c | 2 +- >> tools/libs/guest/xg_domain.c |
19 +-
Quoted 3 lines
Quoted 3 lines
tools/libs/guest/xg_offline_page.c | 2 +- tools/libs/guest/xg_private.h | 16 +-
.../{ctrl/xc_resume.c=> guest/xg_resume.c} |
69 +--
Quoted 7 lines
Quoted 7 lines
tools/libs/guest/xg_sr_save_x86_pv.c | 2 +- tools/ocaml/libs/xc/xenctrl_stubs.c | 1 +
20 files changed, 545 insertions(+), 420 deletions(-)delete mode 100644
tools/libs/ctrl/xc_core_x86.crenametools/libs/{ctrl/xc_core.c=>guest/xg_core.c} (99%) renametools/libs/{ctrl/xc_core.h=>guest/xg_core.h} (92%) renametools/libs/{ctrl/xc_core_arm.c=>guest/xg_core_arm.c}
(72%)
Quoted 1 line
Quoted 1 line
rename
tools/libs/{ctrl/xc_core_arm.h=>guest/xg_core_arm.h}
(100%)
Quoted 2 lines
Quoted 2 lines
create mode 100644
tools/libs/guest/xg_core_x86.crenametools/libs/{ctrl/xc_core_x86.h=>guest/xg_core_x86.h}
(100%)
Quoted 3 lines
Quoted 2 lines
rename
tools/libs/{ctrl/xc_resume.c=>guest/xg_resume.c} (80%)
Quoted 5 lines · Juergen Gross
Quoted 1 line · Juergen Gross
Ping?
Now each patch has an Ack by Wei. Could the series be either applied or get more comments, please?
And another PING.
Do I need to setup a cron job pinging each day to get a reaction?
Juergen
Quoted 96 lines
Juergen
Quoted 51 lines · Juergen Gross
Quoted 49 lines
There are some corners left which don't support the not so very new linear p2m list of pv guests, which has been introduced in Linux kernel 3.19 and which is mandatory for non-legacy versions of Xen since kernel 4.14.
This series adds support for the linear p2m list where it is missing (colo support and "xl dump-core").
In theory it should be possible to merge the p2m list mapping code from migration handling and core dump handling, but this needs quite some cleanup before this is possible.
The first three patches of this series are fixing real problems, so I've put them at the start of this series, especially in order to make backports easier.
The other three patches are only the first steps of cleanup. The main work done here is to concentrate all p2m mapping in libxenguest instead of having one implementation in each of libxenguest and libxenctrl.
Merging the two implementations should be rather easy, but this will require to touch many lines of code, as the migration handling variant seems to be more mature, but it is using the migration stream specific structures heavily. So I'd like to have some confirmation that my way to clean this up is the right one.
My idea would be to add the data needed for p2m mapping to struct
domain_info_contextand replace the related fields in structxc_sr_contextwith a structdomain_info_context. Modifying the interface ofxc_core_arch_map_p2m()to take most current parameters via structdomain_info_contextwould then enable migration coding to usexc_core_arch_map_p2m()for mapping the p2m.xc_core_arch_map_p2m()should look basically like the current migration p2m mapping code afterwards.Any comments to that plan?
Changes in V2:
- added missing #include in ocaml stubJuergen Gross (6):
tools/libs/guest: fixmax_pfnsetting inmap_p2m()tools/libs/ctrl: fixxc_core_arch_map_p2m()to support linear p2m table
tools/libs/ctrl: use common p2m mapping code inxc_domain_resume_any()tools/libs: movexc_resume.cto libxenguest
tools/libs: move xc_core* from libxenctrl to libxenguesttools/libs/guest: make some definitions private to libxenguestQuoted 6 lines
Quoted 6 lines
tools/include/xenctrl.h | 63 --- tools/include/xenguest.h | 63 +++ tools/libs/ctrl/Makefile | 4 - tools/libs/ctrl/xc_core_x86.c | 223 ---------- tools/libs/ctrl/xc_domain.c | 2 - tools/libs/ctrl/xc_private.h |
43 +-
Quoted 3 lines
Quoted 3 lines
tools/libs/guest/Makefile | 4 +
.../libs/{ctrl/xc_core.c=> guest/xg_core.c} | 7 +-.../libs/{ctrl/xc_core.h=> guest/xg_core.h} | 15+- >> .../xc_core_arm.c => guest/xg_core_arm.c} | 31 +- >> .../xc_core_arm.h => guest/xg_core_arm.h} | 0 >> tools/libs/guest/xg_core_x86.c | 399 ++++++++++++++++++ >> .../xc_core_x86.h => guest/xg_core_x86.h} | 0 >> tools/libs/guest/xg_dom_boot.c | 2+- >> tools/libs/guest/xg_domain.c |19 +-
Quoted 3 lines
Quoted 3 lines
tools/libs/guest/xg_offline_page.c | 2 +- tools/libs/guest/xg_private.h | 16 +-
.../{ctrl/xc_resume.c=> guest/xg_resume.c} |69 +--
Quoted 7 lines
Quoted 7 lines
tools/libs/guest/xg_sr_save_x86_pv.c | 2 +- tools/ocaml/libs/xc/xenctrl_stubs.c | 1 +
20 files changed, 545 insertions(+), 420 deletions(-)delete mode 100644
tools/libs/ctrl/xc_core_x86.crenametools/libs/{ctrl/xc_core.c=>guest/xg_core.c} (99%) renametools/libs/{ctrl/xc_core.h=>guest/xg_core.h} (92%) renametools/libs/{ctrl/xc_core_arm.c=>guest/xg_core_arm.c}(72%)
Quoted 1 line
Quoted 1 line
rename
tools/libs/{ctrl/xc_core_arm.h=>guest/xg_core_arm.h}(100%)
Quoted 2 lines
Quoted 2 lines
create mode 100644
tools/libs/guest/xg_core_x86.crenametools/libs/{ctrl/xc_core_x86.h=>guest/xg_core_x86.h}(100%)
Quoted 3 lines
Quoted 2 lines
rename
tools/libs/{ctrl/xc_resume.c=>guest/xg_resume.c} (80%)