xen: p2m: correctly initialize partial p2m leave
Subsystems (2)
PATCH 1/1 xen: p2m: correctly initialize partial p2m leave
There have been changes and code been moved around, so this is just a quick rebase of the change I tested on a 2.6.37 based kernel. The basic problem seem still valid, though.
Initially I thought of adding a cc to stable into the s-o-b, but the patch needs to be adapted anyway (I can supply that version if the way I fixed the issue looks ok).
Regards,
Stefan
Quoted 1 line
From
1e9c9514caf0399c88ae9288e6db8e3d1c4b4be5Mon Sep 17 00:00:00 2001
From: Stefan Bader <[email protected]>
Date: Thu, 20 Jan 2011 11:37:43 +0100
Subject: [PATCH] xen: p2m: correctly initialize partial p2m leave
After changing the p2m mapping to a tree by
commit 58e05027b530ff081ecea68e38de8d59db8f87e0
xen: convert p2m to a 3 level tree
and trying to boot a DomU with 615MB of memory, the following crash was observed in the dump:
kernel direct mapping tables up to 26f00000 @ 1ec4000-1fff000
BUG: unable to handle kernel NULL pointer dereference at (null) IP: [<c0107397>] xen_set_pte+0x27/0x60
*pdpt = 0000000000000000 *pde = 0000000000000000
Adding further debug statements showed that when trying to set up pfn=0x26700 the returned mapping was invalid.
pfn=0x266ff calling set_pte(0xc1fe77f8, 0x6b3003)
pfn=0x26700 calling set_pte(0xc1fe7800, 0x3)
Although the last_pfn obtained from the startup info is 0x26700, which should in turn not be hit, the additional 8MB which are added as extra memory normally seem to be ok. This lead to looking into the initial p2m tree construction, which uses the smaller value and assuming that there is other code handling the extra memory.
When the p2m tree is set up, the leaves are directly pointed to the array which the domain builder set up. But if the mapping is not on a boundary that fits into one p2m page, this will result in the last leaf being only partially valid. And as the invalid entries are not initialized in that case, things go badly wrong.
I am trying to fix that by checking whether the current leaf is a complete map and if not, allocate a completely new page and copy only the valid pointers there. This may not be the most efficient or elegant solution, but at least it seems to allow me booting DomUs with memory assignments all over the range.
--- arch/x86/xen/p2m.c | 20 +++++++++++++++++++- 1 files changed, 19 insertions(+), 1 deletions(-)
arch/x86/xen/p2m.c +19 -1
@@ -237,7 +237,25 @@ void __init xen_build_dynamic_phys_to_machine(void) p2m_top[topidx] = mid; } - p2m_top[topidx][mididx] = &mfn_list[pfn]; +/* +* As long as the mfn_list has enough entries to completely +* fill a p2m page, pointing into the array is ok. But if +* not the entries beyond the last pfn will be undefined. +* And guessing that the 'what-ever-there-is' does not take it +* too kindly when changing it to invalid markers, a new page +* is allocated, initialized and filled with the valid part. +*/ +if (unlikely (pfn +P2M_PER_PAGE > max_pfn)) { +unsigned long p2midx; +unsigned long **p2m =extend_brk (PAGE_SIZE ,PAGE_SIZE ); +p2m_init (p2m); + +for (p2midx =0 ; pfn + p2midx < max_pfn; p2midx++) { + p2m[p2midx] = mfn_list[pfn + p2midx]; + } + p2m_top[topidx][mididx] = p2m; + }else + p2m_top[topidx][mididx] = &mfn_list[pfn]; }m2p_override_init ();
Signature · 1 line
1.7.0.4
Quoted 3 lines · Stefan Bader
There have been changes and code been moved around, so this is just a quick rebase of the change I tested on a 2.6.37 based kernel. The basic problem seem still valid, though.
Nice catch..
Quoted 91 lines
Initially I thought of adding a cc to stable into the s-o-b, but the patch needs to be adapted anyway (I can supply that version if the way I fixed the issue looks ok).
Regards,
StefanQuoted 1 line
From
1e9c9514caf0399c88ae9288e6db8e3d1c4b4be5Mon Sep 17 00:00:00 2001From: Stefan Bader <[email protected]>
Date: Thu, 20 Jan 2011 11:37:43 +0100
Subject: [PATCH] xen: p2m: correctly initialize partial p2m leaveAfter changing the p2m mapping to a tree by
commit
58e05027b530ff081ecea68e38de8d59db8f87e0
xen: convert p2m to a 3 level treeand trying to boot a DomU with 615MB of memory, the following crash was observed in the dump:
kernel direct mapping tables up to
26f00000@ 1ec4000-1fff000BUG: unable to handle kernel NULL pointer dereference at (null) IP: [<c0107397>] xen_set_pte+0x27/0x60*pdpt = 0000000000000000 *pde = 0000000000000000
Adding further debug statements showed that when trying to set up pfn=0x26700 the returned mapping was invalid.
pfn=0x266ff calling
set_pte(0xc1fe77f8, 0x6b3003)
pfn=0x26700 callingset_pte(0xc1fe7800, 0x3)Although the
last_pfnobtained from the startup info is0x26700, which should in turn not be hit, the additional 8MB which are added as extra memory normally seem to be ok. This lead to looking into the initial p2m tree construction, which uses the smaller value and assuming that there is other code handling the extra memory.When the p2m tree is set up, the leaves are directly pointed to the array which the domain builder set up. But if the mapping is not on a boundary that fits into one p2m page, this will result in the last leaf being only partially valid. And as the invalid entries are not initialized in that case, things go badly wrong.
I am trying to fix that by checking whether the current leaf is a complete map and if not, allocate a completely new page and copy only the valid pointers there. This may not be the most efficient or elegant solution, but at least it seems to allow me booting DomUs with memory assignments all over the range.
Signed-off-by Stefan Bader <[email protected]>--- arch/x86/xen/p2m.c | 20 +++++++++++++++++++- 1 files changed, 19 insertions(+), 1 deletions(-)arch/x86/xen/p2m.c +19
-1@@ -237,7 +237,25 @@ void __init xen_build_dynamic_phys_to_machine(void) p2m_top[topidx] = mid; } - p2m_top[topidx][mididx] = &mfn_list[pfn]; +/* +* As long as the mfn_list has enough entries to completely +* fill a p2m page, pointing into the array is ok. But if +* not the entries beyond the last pfn will be undefined. +* And guessing that the 'what-ever-there-is' does not take it +* too kindly when changing it to invalid markers, a new page +* is allocated, initialized and filled with the valid part. +*/ +if (unlikely (pfn +P2M_PER_PAGE > max_pfn)) { +unsigned long p2midx; +unsigned long **p2m =extend_brk (PAGE_SIZE ,PAGE_SIZE ); +p2m_init (p2m); + +for (p2midx =0 ; pfn + p2midx < max_pfn; p2midx++) { + p2m[p2midx] = mfn_list[pfn + p2midx]; + } + p2m_top[topidx][mididx] = p2m; + }else + p2m_top[topidx][mididx] = &mfn_list[pfn]; }m2p_override_init ();Signature · 1 line
1.7.0.4
Quoted 5 lines · Konrad Rzeszutek Wilk
Quoted 3 lines · Stefan Bader
There have been changes and code been moved around, so this is just a quick rebase of the change I tested on a 2.6.37 based kernel. The basic problem seem still valid, though.
Nice catch..
Indeed!
Quoted 42 lines
Quoted 42 lines
Initially I thought of adding a cc to stable into the s-o-b, but the patch needs to be adapted anyway (I can supply that version if the way I fixed the issue looks ok).
Regards,
StefanQuoted 1 line
From
1e9c9514caf0399c88ae9288e6db8e3d1c4b4be5Mon Sep 17 00:00:00 2001From: Stefan Bader <[email protected]>
Date: Thu, 20 Jan 2011 11:37:43 +0100
Subject: [PATCH] xen: p2m: correctly initialize partial p2m leaveAfter changing the p2m mapping to a tree by
commit
58e05027b530ff081ecea68e38de8d59db8f87e0
xen: convert p2m to a 3 level treeand trying to boot a DomU with 615MB of memory, the following crash was observed in the dump:
kernel direct mapping tables up to
26f00000@ 1ec4000-1fff000BUG: unable to handle kernel NULL pointer dereference at (null) IP: [<c0107397>] xen_set_pte+0x27/0x60*pdpt = 0000000000000000 *pde = 0000000000000000
Adding further debug statements showed that when trying to set up pfn=0x26700 the returned mapping was invalid.
pfn=0x266ff calling
set_pte(0xc1fe77f8, 0x6b3003)
pfn=0x26700 callingset_pte(0xc1fe7800, 0x3)Although the
last_pfnobtained from the startup info is0x26700, which should in turn not be hit, the additional 8MB which are added as extra memory normally seem to be ok. This lead to looking into the initial p2m tree construction, which uses the smaller value and assuming that there is other code handling the extra memory.When the p2m tree is set up, the leaves are directly pointed to the array which the domain builder set up. But if the mapping is not on a boundary that fits into one p2m page, this will result in the last leaf being only partially valid. And as the invalid entries are not initialized in that case, things go badly wrong.
Quoted 5 lines
Quoted 5 lines
I am trying to fix that by checking whether the current leaf is a complete map and if not, allocate a completely new page and copy only the valid pointers there. This may not be the most efficient or elegant solution, but at least it seems to allow me booting DomUs with memory assignments all over the range.
Since the p2m page is just a normal page that happens to have been initialized by the domain builder, I think we can just fill the tail of the page with INVALID_P2M_ENTRY in place, rather than having to allocate a new one.
J
Quoted 57 lines · Jeremy Fitzhardinge
Quoted 5 lines · Konrad Rzeszutek Wilk
Quoted 3 lines · Stefan Bader
There have been changes and code been moved around, so this is just a quick rebase of the change I tested on a 2.6.37 based kernel. The basic problem seem still valid, though.
Nice catch..
Indeed!
Quoted 42 lines
Quoted 42 lines
Initially I thought of adding a cc to stable into the s-o-b, but the patch needs to be adapted anyway (I can supply that version if the way I fixed the issue looks ok).
Regards,
StefanQuoted 1 line
From
1e9c9514caf0399c88ae9288e6db8e3d1c4b4be5Mon Sep 17 00:00:00 2001From: Stefan Bader <[email protected]>
Date: Thu, 20 Jan 2011 11:37:43 +0100
Subject: [PATCH] xen: p2m: correctly initialize partial p2m leaveAfter changing the p2m mapping to a tree by
commit
58e05027b530ff081ecea68e38de8d59db8f87e0
xen: convert p2m to a 3 level treeand trying to boot a DomU with 615MB of memory, the following crash was observed in the dump:
kernel direct mapping tables up to
26f00000@ 1ec4000-1fff000BUG: unable to handle kernel NULL pointer dereference at (null) IP: [<c0107397>] xen_set_pte+0x27/0x60*pdpt = 0000000000000000 *pde = 0000000000000000
Adding further debug statements showed that when trying to set up pfn=0x26700 the returned mapping was invalid.
pfn=0x266ff calling
set_pte(0xc1fe77f8, 0x6b3003)
pfn=0x26700 callingset_pte(0xc1fe7800, 0x3)Although the
last_pfnobtained from the startup info is0x26700, which should in turn not be hit, the additional 8MB which are added as extra memory normally seem to be ok. This lead to looking into the initial p2m tree construction, which uses the smaller value and assuming that there is other code handling the extra memory.When the p2m tree is set up, the leaves are directly pointed to the array which the domain builder set up. But if the mapping is not on a boundary that fits into one p2m page, this will result in the last leaf being only partially valid. And as the invalid entries are not initialized in that case, things go badly wrong.
Quoted 5 lines
Quoted 5 lines
I am trying to fix that by checking whether the current leaf is a complete map and if not, allocate a completely new page and copy only the valid pointers there. This may not be the most efficient or elegant solution, but at least it seems to allow me booting DomUs with memory assignments all over the range.
Quoted 6 lines
Since the p2m page is just a normal page that happens to have been initialized by the domain builder, I think we can just fill the tail of the page with
INVALID_P2M_ENTRYin place, rather than having to allocate a new one.J
That was exactly the detail I was not sure about and did not want to make assumptions after only spending a little bit digging around in the code. The safest assumption was to expect other data to be possibly located after the p2m array. Please feel free to clean up the code any time.
-Stefan
Quoted 3 lines · Stefan Bader
There have been changes and code been moved around, so this is just a quick rebase of the change I tested on a 2.6.37 based kernel. The basic problem seem still valid, though.
Yup.
Quoted 4 lines
Initially I thought of adding a cc to stable into the s-o-b, but the patch needs to be adapted anyway (I can supply that version if the way I fixed the issue looks ok).
OK, let me send this upstream to Linus for adaption. I fixed one compile warning but otherwsie it is the same. Look below for details.
And when that is done I would appreciate you sending a copy to stable.
Quoted 63 lines
Regards,
StefanQuoted 1 line
From
1e9c9514caf0399c88ae9288e6db8e3d1c4b4be5Mon Sep 17 00:00:00 2001From: Stefan Bader <[email protected]>
Date: Thu, 20 Jan 2011 11:37:43 +0100
Subject: [PATCH] xen: p2m: correctly initialize partial p2m leaveAfter changing the p2m mapping to a tree by
commit
58e05027b530ff081ecea68e38de8d59db8f87e0
xen: convert p2m to a 3 level treeand trying to boot a DomU with 615MB of memory, the following crash was observed in the dump:
kernel direct mapping tables up to
26f00000@ 1ec4000-1fff000BUG: unable to handle kernel NULL pointer dereference at (null) IP: [<c0107397>] xen_set_pte+0x27/0x60*pdpt = 0000000000000000 *pde = 0000000000000000
Adding further debug statements showed that when trying to set up pfn=0x26700 the returned mapping was invalid.
pfn=0x266ff calling
set_pte(0xc1fe77f8, 0x6b3003)
pfn=0x26700 callingset_pte(0xc1fe7800, 0x3)Although the
last_pfnobtained from the startup info is0x26700, which should in turn not be hit, the additional 8MB which are added as extra memory normally seem to be ok. This lead to looking into the initial p2m tree construction, which uses the smaller value and assuming that there is other code handling the extra memory.When the p2m tree is set up, the leaves are directly pointed to the array which the domain builder set up. But if the mapping is not on a boundary that fits into one p2m page, this will result in the last leaf being only partially valid. And as the invalid entries are not initialized in that case, things go badly wrong.
I am trying to fix that by checking whether the current leaf is a complete map and if not, allocate a completely new page and copy only the valid pointers there. This may not be the most efficient or elegant solution, but at least it seems to allow me booting DomUs with memory assignments all over the range.
Signed-off-by Stefan Bader <[email protected]>--- arch/x86/xen/p2m.c | 20 +++++++++++++++++++- 1 files changed, 19 insertions(+), 1 deletions(-)arch/x86/xen/p2m.c +1
-1@@ -237,7 +237,25 @@ void __init xen_build_dynamic_phys_to_machine(void) p2m_top[topidx] = mid; } - p2m_top[topidx][mididx] = &mfn_list[pfn]; +/*
arch/x86/xen/p2m.c void __init xen_build_dynamic_phys_to_machine(void) · line 250
Quoted 10 lines
arch/x86/xen/p2m.c +10
-0+* As long as the mfn_list has enough entries to completely +* fill a p2m page, pointing into the array is ok. But if +* not the entries beyond the last pfn will be undefined. +* And guessing that the 'what-ever-there-is' does not take it +* too kindly when changing it to invalid markers, a new page +* is allocated, initialized and filled with the valid part. +*/ +if (unlikely (pfn +P2M_PER_PAGE > max_pfn)) { +unsigned long p2midx; +unsigned long **p2m =extend_brk (PAGE_SIZE ,PAGE_SIZE );
unsigned long *p2m.
Quoted 19 lines
arch/x86/xen/p2m.c +8
-0+p2m_init (p2m); + +for (p2midx =0 ; pfn + p2midx < max_pfn; p2midx++) { + p2m[p2midx] = mfn_list[pfn + p2midx]; + } + p2m_top[topidx][mididx] = p2m; + }else + p2m_top[topidx][mididx] = &mfn_list[pfn]; }m2p_override_init ();Signature · 7 lines
1.7.0.4 _______________________________________________ Xen-devel mailing list [email protected] http://lists.xensource.com/xen-devel
Quoted 7 lines · Konrad Rzeszutek Wilk
Quoted 3 lines · Stefan Bader
There have been changes and code been moved around, so this is just a quick rebase of the change I tested on a 2.6.37 based kernel. The basic problem seem still valid, though.
Yup.
Quoted 1 line
Quoted 9 lines
Quoted 3 lines
Initially I thought of adding a cc to stable into the s-o-b, but the patch needs to be adapted anyway (I can supply that version if the way I fixed the issue looks ok).
OK, let me send this upstream to Linus for adaption. I fixed one compile warning but otherwsie it is the same. Look below for details.
And when that is done I would appreciate you sending a copy to stable.
Sure, will do that as soon as the change hits 2.6.38 upstream.
-Stefan
Quoted 97 lines
Quoted 63 lines
Regards,
StefanQuoted 1 line
From
1e9c9514caf0399c88ae9288e6db8e3d1c4b4be5Mon Sep 17 00:00:00 2001From: Stefan Bader <[email protected]>
Date: Thu, 20 Jan 2011 11:37:43 +0100
Subject: [PATCH] xen: p2m: correctly initialize partial p2m leaveAfter changing the p2m mapping to a tree by
commit
58e05027b530ff081ecea68e38de8d59db8f87e0
xen: convert p2m to a 3 level treeand trying to boot a DomU with 615MB of memory, the following crash was observed in the dump:
kernel direct mapping tables up to
26f00000@ 1ec4000-1fff000BUG: unable to handle kernel NULL pointer dereference at (null) IP: [<c0107397>] xen_set_pte+0x27/0x60*pdpt = 0000000000000000 *pde = 0000000000000000
Adding further debug statements showed that when trying to set up pfn=0x26700 the returned mapping was invalid.
pfn=0x266ff calling
set_pte(0xc1fe77f8, 0x6b3003)
pfn=0x26700 callingset_pte(0xc1fe7800, 0x3)Although the
last_pfnobtained from the startup info is0x26700, which should in turn not be hit, the additional 8MB which are added as extra memory normally seem to be ok. This lead to looking into the initial p2m tree construction, which uses the smaller value and assuming that there is other code handling the extra memory.When the p2m tree is set up, the leaves are directly pointed to the array which the domain builder set up. But if the mapping is not on a boundary that fits into one p2m page, this will result in the last leaf being only partially valid. And as the invalid entries are not initialized in that case, things go badly wrong.
I am trying to fix that by checking whether the current leaf is a complete map and if not, allocate a completely new page and copy only the valid pointers there. This may not be the most efficient or elegant solution, but at least it seems to allow me booting DomUs with memory assignments all over the range.
Signed-off-by Stefan Bader <[email protected]>--- arch/x86/xen/p2m.c | 20 +++++++++++++++++++- 1 files changed, 19 insertions(+), 1 deletions(-)arch/x86/xen/p2m.c +1
-1@@ -237,7 +237,25 @@ void __init xen_build_dynamic_phys_to_machine(void) p2m_top[topidx] = mid; } - p2m_top[topidx][mididx] = &mfn_list[pfn]; +/* arch/x86/xen/p2m.c void __init xen_build_dynamic_phys_to_machine(void) · line 250
Quoted 10 lines
arch/x86/xen/p2m.c +10
-0+* As long as the mfn_list has enough entries to completely +* fill a p2m page, pointing into the array is ok. But if +* not the entries beyond the last pfn will be undefined. +* And guessing that the 'what-ever-there-is' does not take it +* too kindly when changing it to invalid markers, a new page +* is allocated, initialized and filled with the valid part. +*/ +if (unlikely (pfn +P2M_PER_PAGE > max_pfn)) { +unsigned long p2midx; +unsigned long **p2m =extend_brk (PAGE_SIZE ,PAGE_SIZE );unsigned long *p2m.
Quoted 19 lines
arch/x86/xen/p2m.c +8
-0+p2m_init (p2m); + +for (p2midx =0 ; pfn + p2midx < max_pfn; p2midx++) { + p2m[p2midx] = mfn_list[pfn + p2midx]; + } + p2m_top[topidx][mididx] = p2m; + }else + p2m_top[topidx][mididx] = &mfn_list[pfn]; }m2p_override_init ();Signature · 7 lines
1.7.0.4 _______________________________________________ Xen-devel mailing list [email protected] http://lists.xensource.com/xen-devel
Please always include an inline copy of a patch for easier review-by-reply, even if you also include an attachment because your mailer mangles patches.
Anyway, I suspect the following comment is obsoleted by Jeremy's "just do it in place" suggestion but:
Quoted 2 lines · Stefan Bader
[...]
+ unsigned long **p2m =extend_brk(PAGE_SIZE, PAGE_SIZE);
I think this would need to be matched by a corresponding RESERVE_BRK of some sort.
Ian.
Quoted 4 lines · Ian Campbell
Please always include an inline copy of a patch for easier review-by-reply, even if you also include an attachment because your mailer mangles patches.
Will try to remember. It easy to forget when the mailer used does show textual attachments inline.
Quoted 12 lines
Anyway, I suspect the following comment is obsoleted by Jeremy's "just do it in place" suggestion but:
Quoted 2 lines · Stefan Bader
[...]
+ unsigned long **p2m =extend_brk(PAGE_SIZE, PAGE_SIZE);I think this would need to be matched by a corresponding
RESERVE_BRKof some sort.Ian.
Currently the version extending brk is upstream. So we would need this going upstream as well...
arch/x86/xen/p2m.c +1 -0
@@ -61,6 +61,7 @@ static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_T RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PE RESERVE_BRK(p2m_mid_mfn, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MI +RESERVE_BRK(p2m_node, PAGE_SIZE); static inline unsigned p2m_top_index(unsigned long pfn) {
(code above not tested at all)
And for the "do it inline" this could possibly look like this? (I know I am repeating myself though the value at max_pfn always seemed to be 3L which seemed to be too consistent for coincidence)
arch/x86/xen/p2m.c +6 -12
@@ -241,21 +241,15 @@ void __init xen_build_dynamic_phys_to_machine(void)* As long as the mfn_list has enough entries to completely * fill a p2m page, pointing into the array is ok. But if * not the entries beyond the last pfn will be undefined. -* And guessing that the 'what-ever-there-is' does not take it -* too kindly when changing it to invalid markers, a new page -* is allocated, initialized and filled with the valid part. */ if (unlikely (pfn +P2M_PER_PAGE > max_pfn)) {unsigned long p2midx; -unsigned long *p2m =extend_brk (PAGE_SIZE ,PAGE_SIZE ); -p2m_init (p2m); - -for (p2midx =0 ; pfn + p2midx < max_pfn; p2midx++) { - p2m[p2midx] = mfn_list[pfn + p2midx]; - } - p2m_top[topidx][mididx] = p2m; - }else - p2m_top[topidx][mididx] = &mfn_list[pfn]; + + p2midx = max_pfn %P2M_PER_PAGE ; +for ( ; p2midx <P2M_PER_PAGE ; p2midx++) + mfn_list[pfn + p2midx] =INVALID_P2M_ENTRY ; + } + p2m_top[topidx][mididx] = &mfn_list[pfn]; }m2p_override_init ();
(again completely (not even compile) tested)
-Stefan