s390/uv: handle folios that cannot be split while dirty
Subsystems (5)
cover s390/uv: handle folios that cannot be split while dirty
From patch #3:
"
Currently, starting a PV VM on an iomap-based filesystem with large folio support, such as XFS, will not work. We'll be stuck in unpack_one()->gmap_make_secure(), because we can't seem to make progress splitting the large folio.
The problem is that we require a writable PTE but a writable PTE under such filesystems will imply a dirty folio.
So whenever we have a writable PTE, we'll have a dirty folio, and dirty iomap folios cannot currently get split, because split_folio()->split_huge_page_to_list_to_order()->filemap_release_folio() will fail in iomap_release_folio().
So we will not make any progress splitting such large folios. "
Let's fix one related problem during unpack first, to then handle such folios by triggering writeback before immediately trying to split them again.
This makes it work on XFS with large folios again.
Long-term, we should cleanly supporting splitting such folios even without writeback, but that's a bit harder to implement and not a quick fix.
David Hildenbrand (3):
s390/uv: don't return 0 from make_hva_secure() if the operation was
not successful
s390/uv: always return 0 from s390_wiggle_split_folio() if successful s390/uv: improve splitting of large folios that cannot be split while
dirty
arch/s390/kernel/uv.c | 85 ++++++++++++++++++++++++++++++++++++------- 1 file changed, 72 insertions(+), 13 deletions(-)
base-commit: 088d13246a4672bc03aec664675138e3f5bff68c
Signature · 1 line
2.49.0
PATCH 1/3 s390/uv: don't return 0 from make_hva_secure() if the operation was not successful
If s390_wiggle_split_folio() returns 0 because splitting a large folio succeeded, we will return 0 from make_hva_secure() even though a retry is required. Return -EAGAIN in that case.
Otherwise, we'll return 0 from gmap_make_secure(), and consequently from unpack_one(). In kvm_s390_pv_unpack(), we assume that unpacking succeeded and skip unpacking this page. Later on, we run into issues and fail booting the VM.
So far, this issue was only observed with follow-up patches where we split large pagecache XFS folios. Maybe it can also be triggered with shmem?
We'll cleanup s390_wiggle_split_folio() a bit next, to also return 0 if no split was required.
--- arch/s390/kernel/uv.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
arch/s390/kernel/uv.c +4 -1
@@ -393,8 +393,11 @@ int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_headerfolio_walk_end (&fw, vma);mmap_read_unlock (mm); -if (rc == -E2BIG || rc == -EBUSY ) +if (rc == -E2BIG || rc == -EBUSY ) { rc =s390_wiggle_split_folio (mm, folio, rc == -E2BIG ); +if (!rc) + rc = -EAGAIN ; + }folio_put (folio);return rc;
Signature · 1 line
2.49.0
Re: [PATCH v1 1/3] s390/uv: don't return 0 from make_hva_secure() if the operation was not successful
Quoted 4 lines · David Hildenbrand
If
s390_wiggle_split_folio()returns 0 because splitting a large folio succeeded, we will return 0 frommake_hva_secure()even though a retry is required. Return-EAGAINin that case.
Quoted 8 lines
Otherwise, we'll return 0 from
gmap_make_secure(), and consequently fromunpack_one(). Inkvm_s390_pv_unpack(), we assume that unpacking succeeded and skip unpacking this page. Later on, we run into issues and fail booting the VM.So far, this issue was only observed with follow-up patches where we split large pagecache XFS folios. Maybe it can also be triggered with shmem?
Yes! I can reproduce it when allocating pages outside of the qemu process.
$ echo force > /sys/kernel/mm/transparent_hugepage/shmem_enabled $ rm /dev/shm/vm-ram $ fallocate -l 4G /dev/shm/vm-ram $ /usr/libexec/qemu-kvm ... -object memory-backend-file,id=mem0,size=4g,share=on,mem-path=/dev/shm/vm-ram -M memory-backend=mem0
LOADPARM=[ ]
Using virtio-blk.
Using SCSI scheme.
.........................................................................................................................
qemu-kvm: KVM PV command 4 (KVM_PV_VERIFY) failed: header rc 102 rrc 1a IOCTL rc: -22 Protected boot has failed: 0xa02
Guest crashed on cpu 0: disabled-wait
PSW: 0x0002000080000000 0x0000000000004608
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH v1 1/3] s390/uv: don't return 0 from make_hva_secure() if the operation was not successful
In reply to David HildenbrandQuoted 28 lines · David Hildenbrand
If
s390_wiggle_split_folio()returns 0 because splitting a large folio succeeded, we will return 0 frommake_hva_secure()even though a retry is required. Return-EAGAINin that case.Otherwise, we'll return 0 from
gmap_make_secure(), and consequently fromunpack_one(). Inkvm_s390_pv_unpack(), we assume that unpacking succeeded and skip unpacking this page. Later on, we run into issues and fail booting the VM.So far, this issue was only observed with follow-up patches where we split large pagecache XFS folios. Maybe it can also be triggered with shmem?
We'll cleanup
s390_wiggle_split_folio()a bit next, to also return 0 if no split was required.Fixes d8dfda5af0be ("KVM: s390: pv: fix race when making a page secure")Signed-off-by David Hildenbrand <[email protected]>--- arch/s390/kernel/uv.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)arch/s390/kernel/uv.c +0
-0
arch/s390/kernel/uv.c int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_header · line 399
Quoted 8 lines
arch/s390/kernel/uv.c +3
-1@@ -393,8 +393,11 @@ int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_headerfolio_walk_end (&fw, vma);mmap_read_unlock (mm); -if (rc == -E2BIG || rc == -EBUSY ) +if (rc == -E2BIG || rc == -EBUSY ) { rc =s390_wiggle_split_folio (mm, folio, rc == -E2BIG ); +if (!rc) + rc = -EAGAIN ;
Why not just folio_put() then jump back to the beginning of the function to do the retry? This could avoid going all the way back to kvm_s390_unpack().
Quoted 6 lines
arch/s390/kernel/uv.c +1
-0+ }folio_put (folio);return rc;Signature · 1 line
2.49.0
--
Best Regards,
Yan, Zi
Re: [PATCH v1 1/3] s390/uv: don't return 0 from make_hva_secure() if the operation was not successful
Quoted 33 lines · Zi Yan
Quoted 31 lines · David Hildenbrand
If
s390_wiggle_split_folio()returns 0 because splitting a large folio succeeded, we will return 0 frommake_hva_secure()even though a retry is required. Return-EAGAINin that case.Otherwise, we'll return 0 from
gmap_make_secure(), and consequently fromunpack_one(). Inkvm_s390_pv_unpack(), we assume that unpacking succeeded and skip unpacking this page. Later on, we run into issues and fail booting the VM.So far, this issue was only observed with follow-up patches where we split large pagecache XFS folios. Maybe it can also be triggered with shmem?
We'll cleanup
s390_wiggle_split_folio()a bit next, to also return 0 if no split was required.Fixes d8dfda5af0be ("KVM: s390: pv: fix race when making a page secure")Signed-off-by David Hildenbrand <[email protected]>--- arch/s390/kernel/uv.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)arch/s390/kernel/uv.c +0
-0@@ -393,8 +393,11 @@ int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_headerfolio_walk_end (&fw, vma);mmap_read_unlock (mm);
Quoted 9 lines
arch/s390/kernel/uv.c int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_header · line 399
Quoted 5 lines
arch/s390/kernel/uv.c +3
-1-if (rc == -E2BIG || rc == -EBUSY ) +if (rc == -E2BIG || rc == -EBUSY ) { rc =s390_wiggle_split_folio (mm, folio, rc == -E2BIG ); +if (!rc) + rc = -EAGAIN ;Why not just
folio_put()then jump back to the beginning of the function to do the retry? This could avoid going all the way back tokvm_s390_unpack().
Hi, thanks for the review.
We had a pretty optimized version with such tricks before Claudio refactored it in:
commit 5cbe24350b7d8ef6d466a37d56b07ae643c622ca
Author: Claudio Imbrenda <[email protected]> Date: Thu Jan 23 15:46:17 2025 +0100
KVM: s390: move pv gmap functions into kvm
In particular, one relevant hunk was:
- switch (rc) {
- case -E2BIG:
- folio_lock(folio); - rc = split_folio(folio); - folio_unlock(folio); - folio_put(folio);
-
- switch (rc) {
- case 0:
- /* Splitting succeeded, try again immediately. */
- goto again;
- case -EAGAIN:
- /* Additional folio references. */
- if (drain_lru(&drain_lru_called))
- goto again;
- return -EAGAIN;
Claudio probably had a good reason to rewrite the code -- and I hope we'll be able to rip all of that out soon, so ...
... minimal changes until then :)
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH v1 1/3] s390/uv: don't return 0 from make_hva_secure() if the operation was not successful
Quoted 77 lines · David Hildenbrand
Quoted 33 lines · Zi Yan
Quoted 31 lines · David Hildenbrand
If
s390_wiggle_split_folio()returns 0 because splitting a large folio succeeded, we will return 0 frommake_hva_secure()even though a retry is required. Return-EAGAINin that case.Otherwise, we'll return 0 from
gmap_make_secure(), and consequently fromunpack_one(). Inkvm_s390_pv_unpack(), we assume that unpacking succeeded and skip unpacking this page. Later on, we run into issues and fail booting the VM.So far, this issue was only observed with follow-up patches where we split large pagecache XFS folios. Maybe it can also be triggered with shmem?
We'll cleanup
s390_wiggle_split_folio()a bit next, to also return 0 if no split was required.Fixes d8dfda5af0be ("KVM: s390: pv: fix race when making a page secure")Signed-off-by David Hildenbrand <[email protected]>--- arch/s390/kernel/uv.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)arch/s390/kernel/uv.c +0
-0@@ -393,8 +393,11 @@ int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_headerfolio_walk_end (&fw, vma);mmap_read_unlock (mm);Quoted 9 lines
arch/s390/kernel/uv.c int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_header · line 399
Quoted 5 lines
arch/s390/kernel/uv.c +3
-1-if (rc == -E2BIG || rc == -EBUSY ) +if (rc == -E2BIG || rc == -EBUSY ) { rc =s390_wiggle_split_folio (mm, folio, rc == -E2BIG ); +if (!rc) + rc = -EAGAIN ;Why not just
folio_put()then jump back to the beginning of the function to do the retry? This could avoid going all the way back tokvm_s390_unpack().Hi, thanks for the review.
We had a pretty optimized version with such tricks before Claudio refactored it in:
commit
5cbe24350b7d8ef6d466a37d56b07ae643c622caAuthor: Claudio Imbrenda <[email protected]> Date: Thu Jan 23 15:46:17 2025 +0100KVM: s390: move pv gmap functions into kvm
In particular, one relevant hunk was:
(quoted diff) +0
-16- switch (rc) { - case -E2BIG: - folio_lock(folio); - rc = split_folio(folio); - folio_unlock(folio); - folio_put(folio); - - switch (rc) { - case 0: - /* Splitting succeeded, try again immediately. */ - goto again; - case -EAGAIN: - /* Additional folio references. */ - if (drain_lru(&drain_lru_called)) - goto again; - return -EAGAIN;
Quoted 3 lines
Claudio probably had a good reason to rewrite the code -- and I hope we'll be able to rip all of that out soon, so ...
... minimal changes until then :)
Got it. Acked-by: Zi Yan <[email protected]>
--
Best Regards,
Yan, Zi
PATCH 2/3 s390/uv: always return 0 from s390_wiggle_split_folio() if successful
In reply to David HildenbrandLet's consistently return 0 if the operation was successful, and just detect ourselves whether splitting is required -- folio_test_large() is a cheap operation.
Update the documentation.
Should we simply always return -EAGAIN instead of 0, so we don't have to handle it in the caller? Not sure, staring at the documentation, this way looks a bit cleaner.
--- arch/s390/kernel/uv.c | 22 ++++++++++++---------- 1 file changed, 12 insertions(+), 10 deletions(-)
arch/s390/kernel/uv.c +12 -10
@@ -324,34 +324,36 @@ static int make_folio_secure(struct mm_struct *mm, struct folio *folio, struct u }/** -* s390_wiggle_split_folio() - try to drain extra references to a folio and optionally split. +* s390_wiggle_split_folio() - try to drain extra references to a folio and +* split the folio if it is large. * @mm: the mm containing the folio to work on * @folio: the folio -* @split: whether to split a large folio * * Context: Must be called while holding an extra reference to the folio; * the mm lock should not be held. -* Return: 0 if the folio was split successfully; -* -EAGAIN if the folio was not split successfully but another attempt -* can be made, or if @split was set to false; -* -EINVAL in case of other errors. See split_folio(). +* Return: 0 if the operation was successful; +* -EAGAIN if splitting the large folio was not successful, +* but another attempt can be made; +* -EINVAL in case of other folio splitting errors. See split_folio(). */ -static int s390_wiggle_split_folio (struct mm_struct *mm,struct folio *folio,bool split) +static int s390_wiggle_split_folio (struct mm_struct *mm,struct folio *folio) {int rc;lockdep_assert_not_held (&mm->mmap_lock);folio_wait_writeback (folio);lru_add_drain_all (); -if (split) { + +if (folio_test_large (folio)) {folio_lock (folio); rc =split_folio (folio);folio_unlock (folio);if (rc != -EBUSY )return rc; +return -EAGAIN ; } -return -EAGAIN ; +return 0 ; }int make_hva_secure (struct mm_struct *mm,unsigned long hva,struct uv_cb_header *uvcb)
@@ -394,7 +396,7 @@ int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_headermmap_read_unlock (mm);if (rc == -E2BIG || rc == -EBUSY ) { - rc =s390_wiggle_split_folio (mm, folio, rc == -E2BIG ); + rc =s390_wiggle_split_folio (mm, folio);if (!rc) rc = -EAGAIN ; }
Signature · 1 line
2.49.0
Re: [PATCH v1 2/3] s390/uv: always return 0 from s390_wiggle_split_folio() if successful
Quoted 53 lines · David Hildenbrand
Let's consistently return 0 if the operation was successful, and just detect ourselves whether splitting is required --
folio_test_large()is a cheap operation.Update the documentation.
Should we simply always return
-EAGAINinstead of 0, so we don't have to handle it in the caller? Not sure, staring at the documentation, this way looks a bit cleaner.Signed-off-by David Hildenbrand <[email protected]>--- arch/s390/kernel/uv.c | 22 ++++++++++++---------- 1 file changed, 12 insertions(+), 10 deletions(-)arch/s390/kernel/uv.c +9
-8@@ -324,34 +324,36 @@ static int make_folio_secure(struct mm_struct *mm, struct folio *folio, struct u }/** -* s390_wiggle_split_folio() - try to drain extra references to a folio and optionally split. +* s390_wiggle_split_folio() - try to drain extra references to a folio and +* split the folio if it is large. * @mm: the mm containing the folio to work on * @folio: the folio -* @split: whether to split a large folio * * Context: Must be called while holding an extra reference to the folio; * the mm lock should not be held. -* Return: 0 if the folio was split successfully; -* -EAGAIN if the folio was not split successfully but another attempt -* can be made, or if @split was set to false; -* -EINVAL in case of other errors. See split_folio(). +* Return: 0 if the operation was successful; +* -EAGAIN if splitting the large folio was not successful, +* but another attempt can be made; +* -EINVAL in case of other folio splitting errors. See split_folio(). */ -static int s390_wiggle_split_folio (struct mm_struct *mm,struct folio *folio,bool split) +static int s390_wiggle_split_folio (struct mm_struct *mm,struct folio *folio) {int rc;lockdep_assert_not_held (&mm->mmap_lock);folio_wait_writeback (folio);lru_add_drain_all (); -if (split) { + +if (folio_test_large (folio)) {folio_lock (folio);
arch/s390/kernel/uv.c static int make_folio_secure(struct mm_struct *mm, struct folio *folio, struct u · line 357
Quoted 10 lines
arch/s390/kernel/uv.c +2
-1rc =split_folio (folio);folio_unlock (folio);if (rc != -EBUSY )return rc; +return -EAGAIN ; } -return -EAGAIN ; +return 0 ; }
I can see how this function is written to service as two purposes, trying to get rid of pcp ref of a folio and split a folio (to avoid the extra pcp ref from failing split, lru_add_drain_all() is called before split). Hope it will be refactored later.
Quoted 13 lines
arch/s390/kernel/uv.c +1
-1int make_hva_secure (struct mm_struct *mm,unsigned long hva,struct uv_cb_header *uvcb)@@ -394,7 +396,7 @@ int make_hva_secure(struct mm_struct *mm, unsigned long hva, struct uv_cb_headermmap_read_unlock (mm);if (rc == -E2BIG || rc == -EBUSY ) { - rc =s390_wiggle_split_folio (mm, folio, rc == -E2BIG ); + rc =s390_wiggle_split_folio (mm, folio);if (!rc) rc = -EAGAIN ; }Signature · 1 line
2.49.0
The changes look good to me. Acked-by: Zi Yan <[email protected]>
--
Best Regards,
Yan, Zi
PATCH 3/3 s390/uv: improve splitting of large folios that cannot be split while dirty
In reply to David HildenbrandCurrently, starting a PV VM on an iomap-based filesystem with large folio support, such as XFS, will not work. We'll be stuck in unpack_one()->gmap_make_secure(), because we can't seem to make progress splitting the large folio.
The problem is that we require a writable PTE but a writable PTE under such filesystems will imply a dirty folio.
So whenever we have a writable PTE, we'll have a dirty folio, and dirty iomap folios cannot currently get split, because split_folio()->split_huge_page_to_list_to_order()->filemap_release_folio() will fail in iomap_release_folio().
So we will not make any progress splitting such large folios.
Until dirty folios can be split more reliably, let's manually trigger writeback of the problematic folio using filemap_write_and_wait_range(), and retry the split immediately afterwards exactly once, before looking up the folio again.
Should this logic be part of split_folio()? Likely not; most split users don't have to split so eagerly to make any progress.
For now, this seems to affect xfs, zonefs and erofs, and this patch makes it work again (tested on xfs only).
While this could be considered a fix for 6795801366da ("xfs: Support large folios"), df2f9708ff1f ("zonefs: enable support for large folios") and ce529cc25b18 ("erofs: enable large folios for iomap mode"), before commit eef88fe45ac9 ("s390/uv: Split large folios in gmap_make_secure()"), we did not try splitting large folios at all. So it's all rather part of making SE compatible with file systems that support large folios. But to have some "Fixes:" tag, let's just use eef88fe45ac9.
Not CCing stable, because there are a lot of dependencies, and it simply not working is not critical in stable kernels.
--- arch/s390/kernel/uv.c | 66 +++++++++++++++++++++++++++++++++++++++---- 1 file changed, 60 insertions(+), 6 deletions(-)
arch/s390/kernel/uv.c +60 -6
@@ -15,6 +15,7 @@#include <linux/pagemap.h> #include <linux/swap.h> #include <linux/pagewalk.h> +#include <linux/backing-dev.h> #include <asm/facility.h> #include <asm/sections.h> #include <asm/uv.h>
@@ -338,22 +339,75 @@ static int make_folio_secure(struct mm_struct *mm, struct folio *folio, struct u*/ static int s390_wiggle_split_folio (struct mm_struct *mm,struct folio *folio) { -int rc; +int rc, tried_splits;lockdep_assert_not_held (&mm->mmap_lock);folio_wait_writeback (folio);lru_add_drain_all (); -if (folio_test_large (folio)) { +if (!folio_test_large (folio)) +return 0 ; + +for (tried_splits =0 ; tried_splits <2 ; tried_splits++) { +struct address_space *mapping; +loff_t lstart, lend; +struct inode *inode; +folio_lock (folio); rc =split_folio (folio); +if (rc != -EBUSY ) { +folio_unlock (folio); +return rc; + } + +/* +* Splitting with -EBUSY can fail for various reasons, but we +* have to handle one case explicitly for now: some mappings +* don't allow for splitting dirty folios; writeback will +* mark them clean again, including marking all page table +* entries mapping the folio read-only, to catch future write +* attempts. +* +* While the system should be writing back dirty folios in the +* background, we obtained this folio by looking up a writable +* page table entry. On these problematic mappings, writable +* page table entries imply dirty folios, preventing the +* split in the first place. +* +* To prevent a livelock when trigger writeback manually and +* letting the caller look up the folio again in the page +* table (turning it dirty), immediately try to split again. +* +* This is only a problem for some mappings (e.g., XFS); +* mappings that do not support writeback (e.g., shmem) do not +* apply. +*/ +if (!folio_test_dirty (folio) ||folio_test_anon (folio) || + !folio->mapping || !mapping_can_writeback (folio->mapping)) { +folio_unlock (folio); +break ; + } + +/* +* Ideally, we'd only trigger writeback on this exact folio. But +* there is no easy way to do that, so we'll stabilize the +* mapping while we still hold the folio lock, so we can drop +* the folio lock to trigger writeback on the range currently +* covered by the folio instead. +*/ + mapping = folio->mapping; + lstart =folio_pos (folio); + lend = lstart +folio_size (folio) -1 ; + inode =igrab (mapping->host);folio_unlock (folio); -if (rc != -EBUSY ) -return rc; -return -EAGAIN ; +if (unlikely (!inode)) +break ; + +filemap_write_and_wait_range (mapping, lstart, lend); +iput (mapping->host); } -return 0 ; +return -EAGAIN ; }int make_hva_secure (struct mm_struct *mm,unsigned long hva,struct uv_cb_header *uvcb)
Signature · 1 line
2.49.0
Quoted 23 lines · On Fri, 16 May 2025 14:39:43 +0200 David Hildenbrand <[email protected]> wrote:
From patch #3:
"
Currently, starting a PV VM on an iomap-based filesystem with large folio support, such as XFS, will not work. We'll be stuck in unpack_one()->gmap_make_secure(), because we can't seem to make progress splitting the large folio.The problem is that we require a writable PTE but a writable PTE under such filesystems will imply a dirty folio.
So whenever we have a writable PTE, we'll have a dirty folio, and dirty iomap folios cannot currently get split, because split_folio()->split_huge_page_to_list_to_order()->filemap_release_folio() will fail in
iomap_release_folio().So we will not make any progress splitting such large folios. "
Let's fix one related problem during unpack first, to then handle such folios by triggering writeback before immediately trying to split them again.
Quoted 5 lines
This makes it work on XFS with large folios again.
Long-term, we should cleanly supporting splitting such folios even without writeback, but that's a bit harder to implement and not a quick fix.
yet another layer of duck tape
I really dislike the current interaction between secure execution and I/O, I hope I can get a cleaner solution as soon as possible
meanwhile, let's keep the boat afloat; whole series:
David: thanks for fixing this mess!
Quoted 26 lines
Cc Christian Borntraeger <[email protected]>Cc Janosch Frank <[email protected]>Cc Claudio Imbrenda <[email protected]>Cc David Hildenbrand <[email protected]>Cc Heiko Carstens <[email protected]>Cc Vasily Gorbik <[email protected]>Cc Alexander Gordeev <[email protected]>Cc Sven Schnelle <[email protected]>Cc Thomas Huth <[email protected]>Cc Matthew Wilcox <[email protected]>Cc Zi Yan <[email protected]>Cc Sebastian Mitterle <[email protected]>David Hildenbrand (3):
s390/uv: don't return 0 frommake_hva_secure()if the operation was
not successful
s390/uv: always return 0 froms390_wiggle_split_folio()if successful s390/uv: improve splitting of large folios that cannot be split while
dirtyarch/s390/kernel/uv.c | 85 ++++++++++++++++++++++++++++++++++++------- 1 file changed, 72 insertions(+), 13 deletions(-)base-commit:
088d13246a4672bc03aec664675138e3f5bff68c
Quoted 28 lines · Claudio Imbrenda
Quoted 25 lines · On Fri, 16 May 2025 14:39:43 +0200 David Hildenbrand <[email protected]> wrote:
From patch #3:
"
Currently, starting a PV VM on an iomap-based filesystem with large folio support, such as XFS, will not work. We'll be stuck in unpack_one()->gmap_make_secure(), because we can't seem to make progress splitting the large folio.The problem is that we require a writable PTE but a writable PTE under such filesystems will imply a dirty folio.
So whenever we have a writable PTE, we'll have a dirty folio, and dirty iomap folios cannot currently get split, because split_folio()->split_huge_page_to_list_to_order()->filemap_release_folio() will fail in
iomap_release_folio().So we will not make any progress splitting such large folios. "
Let's fix one related problem during unpack first, to then handle such folios by triggering writeback before immediately trying to split them again.
This makes it work on XFS with large folios again.
Quoted 8 lines
Quoted 3 lines
Long-term, we should cleanly supporting splitting such folios even without writeback, but that's a bit harder to implement and not a quick fix.
yet another layer of duck tape
I really dislike the current interaction between secure execution and I/O, I hope I can get a cleaner solution as soon as possible
I'll be more than happy to review such a series -- hoping we can just support large folios naturally :)
Quoted 4 lines
Reviewed-by Claudio Imbrenda <[email protected]>
Thanks!
Quoted 1 line
David: thanks for fixing this mess!
NP; I had a prototype of patch #3 for a long time. But after rebasing on top of your work I saw these weird validation errors and just couldn't find the issue. And I only saw them with patch #3 on ordinary pagecache folios, not with shmem, which severely confused.
... gave it another try today after ~1month and almost immediately spotted the issue.
Some things just need time :)
Signature · 3 lines
Cheers, David / dhildenb
Claudio Imbrenda
maintainerQuoted 23 lines · On Fri, 16 May 2025 14:39:43 +0200 David Hildenbrand <[email protected]> wrote:
From patch #3:
"
Currently, starting a PV VM on an iomap-based filesystem with large folio support, such as XFS, will not work. We'll be stuck in unpack_one()->gmap_make_secure(), because we can't seem to make progress splitting the large folio.The problem is that we require a writable PTE but a writable PTE under such filesystems will imply a dirty folio.
So whenever we have a writable PTE, we'll have a dirty folio, and dirty iomap folios cannot currently get split, because split_folio()->split_huge_page_to_list_to_order()->filemap_release_folio() will fail in
iomap_release_folio().So we will not make any progress splitting such large folios. "
Let's fix one related problem during unpack first, to then handle such folios by triggering writeback before immediately trying to split them again.
Quoted 5 lines
This makes it work on XFS with large folios again.
Long-term, we should cleanly supporting splitting such folios even without writeback, but that's a bit harder to implement and not a quick fix.
picked for 6.16, I think it will survive the CI without issues, since I assume you tested this thoroughly
Quoted 26 lines
Cc Christian Borntraeger <[email protected]>Cc Janosch Frank <[email protected]>Cc Claudio Imbrenda <[email protected]>Cc David Hildenbrand <[email protected]>Cc Heiko Carstens <[email protected]>Cc Vasily Gorbik <[email protected]>Cc Alexander Gordeev <[email protected]>Cc Sven Schnelle <[email protected]>Cc Thomas Huth <[email protected]>Cc Matthew Wilcox <[email protected]>Cc Zi Yan <[email protected]>Cc Sebastian Mitterle <[email protected]>David Hildenbrand (3):
s390/uv: don't return 0 frommake_hva_secure()if the operation was
not successful
s390/uv: always return 0 froms390_wiggle_split_folio()if successful s390/uv: improve splitting of large folios that cannot be split while
dirtyarch/s390/kernel/uv.c | 85 ++++++++++++++++++++++++++++++++++++------- 1 file changed, 72 insertions(+), 13 deletions(-)base-commit:
088d13246a4672bc03aec664675138e3f5bff68c
Quoted 26 lines · Claudio Imbrenda
Quoted 23 lines · On Fri, 16 May 2025 14:39:43 +0200 David Hildenbrand <[email protected]> wrote:
From patch #3:
"
Currently, starting a PV VM on an iomap-based filesystem with large folio support, such as XFS, will not work. We'll be stuck in unpack_one()->gmap_make_secure(), because we can't seem to make progress splitting the large folio.The problem is that we require a writable PTE but a writable PTE under such filesystems will imply a dirty folio.
So whenever we have a writable PTE, we'll have a dirty folio, and dirty iomap folios cannot currently get split, because split_folio()->split_huge_page_to_list_to_order()->filemap_release_folio() will fail in
iomap_release_folio().So we will not make any progress splitting such large folios. "
Let's fix one related problem during unpack first, to then handle such folios by triggering writeback before immediately trying to split them again.
Quoted 8 lines
Quoted 5 lines
This makes it work on XFS with large folios again.
Long-term, we should cleanly supporting splitting such folios even without writeback, but that's a bit harder to implement and not a quick fix.
picked for 6.16, I think it will survive the CI without issues, since I assume you tested this thoroughly
I did test what was known to be broken, but our QE did not run a bigger test on it. So giving it some soaking time + waiting for a bit for more review might be a good idea!
Signature · 3 lines
Cheers, David / dhildenb