| Message ID | 20190710161733.39119-1-paul.durrant@citrix.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | xen/mm.h: add helper function to test-and-clear _PGC_allocated | expand |
On 10.07.2019 18:17, Paul Durrant wrote: > @@ -418,13 +417,7 @@ static void hvm_free_ioreq_mfn(struct hvm_ioreq_server *s, bool buf) > unmap_domain_page_global(iorp->va); > iorp->va = NULL; > > - /* > - * Check whether we need to clear the allocation reference before > - * dropping the explicit references taken by get_page_and_type(). > - */ > - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) > - put_page(page); > - > + clear_assignment_reference(page); > put_page_and_type(page); > } Is there a specific reason you drop the comment? It doesn't become less relevant than when it was added, does it? > --- a/xen/include/xen/mm.h > +++ b/xen/include/xen/mm.h > @@ -658,4 +658,15 @@ static inline void share_xen_page_with_privileged_guests( > share_xen_page_with_guest(page, dom_xen, flags); > } > > +static inline void clear_assignment_reference(struct page_info *page) I think the function should have 'page' in it's name. Perhaps page_deassign() / page_dealloc() are also misleading, but how about page_put_alloc() or page_put_alloc_ref()? > +{ > + /* > + * It is unsafe to clear _PGC_allocated without holding an additional > + * reference. > + */ > + ASSERT((page->count_info & PGC_count_mask) > 1); While this isn't really in line with our goal of wanting to limit damage also in release builds, I agree that there's no really good alternative here. Crashing the owner of the page wouldn't help much, and bailing from the function wouldn't necessarily be better either. Hence I think this would better be BUG_ON(). > + if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) > + put_page(page); > +} On the whole I have to admit I'm not entirely convinced the "open- coding" as you call it (to me it's not really open-coding as long as there is no helper) is such a bad thing here: Without the helper it is slightly more obvious at the use sites what's actually going on. But maybe that's indeed just me. Jan
> -----Original Message----- > From: Jan Beulich <JBeulich@suse.com> > Sent: 10 July 2019 23:53 > To: Paul Durrant <Paul.Durrant@citrix.com> > Cc: xen-devel@lists.xenproject.org; Julien Grall <julien.grall@arm.com>; Andrew Cooper > <Andrew.Cooper3@citrix.com>; Roger Pau Monne <roger.pau@citrix.com>; Volodymyr Babchuk > <Volodymyr_Babchuk@epam.com>; George Dunlap <George.Dunlap@citrix.com>; Ian Jackson > <Ian.Jackson@citrix.com>; Stefano Stabellini <sstabellini@kernel.org>; Konrad Rzeszutek Wilk > <konrad.wilk@oracle.com>; Tamas K Lengyel <tamas@tklengyel.com>; Tim (Xen.org) <tim@xen.org>; Wei Liu > <wl@xen.org> > Subject: Re: [Xen-devel] [PATCH] xen/mm.h: add helper function to test-and-clear _PGC_allocated > > On 10.07.2019 18:17, Paul Durrant wrote: > > @@ -418,13 +417,7 @@ static void hvm_free_ioreq_mfn(struct hvm_ioreq_server *s, bool buf) > > unmap_domain_page_global(iorp->va); > > iorp->va = NULL; > > > > - /* > > - * Check whether we need to clear the allocation reference before > > - * dropping the explicit references taken by get_page_and_type(). > > - */ > > - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) > > - put_page(page); > > - > > + clear_assignment_reference(page); > > put_page_and_type(page); > > } > > Is there a specific reason you drop the comment? It doesn't become > less relevant than when it was added, does it? Not sure, since what's actually going on is now internal to the function. If I change the function name to clear_allocation_reference() then I think the comment probably becomes extraneous. > > > --- a/xen/include/xen/mm.h > > +++ b/xen/include/xen/mm.h > > @@ -658,4 +658,15 @@ static inline void share_xen_page_with_privileged_guests( > > share_xen_page_with_guest(page, dom_xen, flags); > > } > > > > +static inline void clear_assignment_reference(struct page_info *page) > > I think the function should have 'page' in it's name. Perhaps > page_deassign() / page_dealloc() are also misleading, but how > about page_put_alloc() or page_put_alloc_ref()? > Ok, I think page_put_alloc_ref() is most descriptive (particularly w.r.t. the above discussion). > > +{ > > + /* > > + * It is unsafe to clear _PGC_allocated without holding an additional > > + * reference. > > + */ > > + ASSERT((page->count_info & PGC_count_mask) > 1); > > While this isn't really in line with our goal of wanting to limit > damage also in release builds, I agree that there's no really good > alternative here. Crashing the owner of the page wouldn't help > much, and bailing from the function wouldn't necessarily be better > either. Hence I think this would better be BUG_ON(). Ok. > > > + if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) > > + put_page(page); > > +} > > On the whole I have to admit I'm not entirely convinced the "open- > coding" as you call it (to me it's not really open-coding as long as > there is no helper) is such a bad thing here: Without the helper it > is slightly more obvious at the use sites what's actually going on. > But maybe that's indeed just me. I still think a helper is better, but I'll add a comment to describe what it is doing. Paul > > Jan
On 15.07.2019 10:45, Paul Durrant wrote: >> From: Jan Beulich <JBeulich@suse.com> >> Sent: 10 July 2019 23:53 >> >> On 10.07.2019 18:17, Paul Durrant wrote: >>> @@ -418,13 +417,7 @@ static void hvm_free_ioreq_mfn(struct hvm_ioreq_server *s, bool buf) >>> unmap_domain_page_global(iorp->va); >>> iorp->va = NULL; >>> >>> - /* >>> - * Check whether we need to clear the allocation reference before >>> - * dropping the explicit references taken by get_page_and_type(). >>> - */ >>> - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) >>> - put_page(page); >>> - >>> + clear_assignment_reference(page); >>> put_page_and_type(page); >>> } >> >> Is there a specific reason you drop the comment? It doesn't become >> less relevant than when it was added, does it? > > Not sure, since what's actually going on is now internal to the function. > If I change the function name to clear_allocation_reference() then I > think the comment probably becomes extraneous. Well, the perspective I'm taking is that the ordering constraint wrt put_page_and_type() doesn't go away and is a relevant part of what the comment talks about. Jan
> -----Original Message----- > From: Jan Beulich <JBeulich@suse.com> > Sent: 15 July 2019 10:18 > To: Paul Durrant <Paul.Durrant@citrix.com> > Cc: JulienGrall <julien.grall@arm.com>; Andrew Cooper <Andrew.Cooper3@citrix.com>; George Dunlap > <George.Dunlap@citrix.com>; Ian Jackson <Ian.Jackson@citrix.com>; Roger Pau Monne > <roger.pau@citrix.com>; Volodymyr Babchuk <Volodymyr_Babchuk@epam.com>; Stefano Stabellini > <sstabellini@kernel.org>; xen-devel@lists.xenproject.org; Konrad Rzeszutek Wilk > <konrad.wilk@oracle.com>; Tamas K Lengyel <tamas@tklengyel.com>; Tim (Xen.org) <tim@xen.org>; Wei Liu > <wl@xen.org> > Subject: Re: [Xen-devel] [PATCH] xen/mm.h: add helper function to test-and-clear _PGC_allocated > > On 15.07.2019 10:45, Paul Durrant wrote: > >> From: Jan Beulich <JBeulich@suse.com> > >> Sent: 10 July 2019 23:53 > >> > >> On 10.07.2019 18:17, Paul Durrant wrote: > >>> @@ -418,13 +417,7 @@ static void hvm_free_ioreq_mfn(struct hvm_ioreq_server *s, bool buf) > >>> unmap_domain_page_global(iorp->va); > >>> iorp->va = NULL; > >>> > >>> - /* > >>> - * Check whether we need to clear the allocation reference before > >>> - * dropping the explicit references taken by get_page_and_type(). > >>> - */ > >>> - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) > >>> - put_page(page); > >>> - > >>> + clear_assignment_reference(page); > >>> put_page_and_type(page); > >>> } > >> > >> Is there a specific reason you drop the comment? It doesn't become > >> less relevant than when it was added, does it? > > > > Not sure, since what's actually going on is now internal to the function. > > If I change the function name to clear_allocation_reference() then I > > think the comment probably becomes extraneous. > > Well, the perspective I'm taking is that the ordering constraint > wrt put_page_and_type() doesn't go away and is a relevant part of > what the comment talks about. Ok. Would you be happy fixing the comment to your taste on commit then, as I'm not sure exactly what you want to say? Paul > > Jan
diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c index 4f44d5c742..78700d6f08 100644 --- a/xen/arch/arm/domain.c +++ b/xen/arch/arm/domain.c @@ -926,9 +926,7 @@ static int relinquish_memory(struct domain *d, struct page_list_head *list) */ continue; - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); - + clear_assignment_reference(page); put_page(page); if ( hypercall_preempt_check() ) diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c index 147f96a09e..c8c51d5f76 100644 --- a/xen/arch/x86/domain.c +++ b/xen/arch/x86/domain.c @@ -1939,8 +1939,7 @@ static int relinquish_memory( BUG(); } - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + clear_assignment_reference(page); /* * Forcibly invalidate top-most, still valid page tables at this point diff --git a/xen/arch/x86/hvm/ioreq.c b/xen/arch/x86/hvm/ioreq.c index 7a80cfb28b..129f9fddbc 100644 --- a/xen/arch/x86/hvm/ioreq.c +++ b/xen/arch/x86/hvm/ioreq.c @@ -398,8 +398,7 @@ static int hvm_alloc_ioreq_mfn(struct hvm_ioreq_server *s, bool buf) return 0; fail: - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + clear_assignment_reference(page); put_page_and_type(page); return -ENOMEM; @@ -418,13 +417,7 @@ static void hvm_free_ioreq_mfn(struct hvm_ioreq_server *s, bool buf) unmap_domain_page_global(iorp->va); iorp->va = NULL; - /* - * Check whether we need to clear the allocation reference before - * dropping the explicit references taken by get_page_and_type(). - */ - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); - + clear_assignment_reference(page); put_page_and_type(page); } diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c index df2c0130f1..9fe66a6d26 100644 --- a/xen/arch/x86/mm.c +++ b/xen/arch/x86/mm.c @@ -498,8 +498,7 @@ void share_xen_page_with_guest(struct page_info *page, struct domain *d, void free_shared_domheap_page(struct page_info *page) { - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + clear_assignment_reference(page); if ( !test_and_clear_bit(_PGC_xen_heap, &page->count_info) ) ASSERT_UNREACHABLE(); page->u.inuse.type_info = 0; diff --git a/xen/arch/x86/mm/mem_sharing.c b/xen/arch/x86/mm/mem_sharing.c index f16a3f5324..7a643aed53 100644 --- a/xen/arch/x86/mm/mem_sharing.c +++ b/xen/arch/x86/mm/mem_sharing.c @@ -1000,8 +1000,7 @@ static int share_pages(struct domain *sd, gfn_t sgfn, shr_handle_t sh, mem_sharing_page_unlock(firstpg); /* Free the client page */ - if(test_and_clear_bit(_PGC_allocated, &cpage->count_info)) - put_page(cpage); + clear_assignment_reference(cpage); put_page(cpage); /* We managed to free a domain page. */ @@ -1082,8 +1081,7 @@ int mem_sharing_add_to_physmap(struct domain *sd, unsigned long sgfn, shr_handle ret = -EOVERFLOW; goto err_unlock; } - if ( test_and_clear_bit(_PGC_allocated, &cpage->count_info) ) - put_page(cpage); + clear_assignment_reference(cpage); put_page(cpage); } } @@ -1177,8 +1175,7 @@ int __mem_sharing_unshare_page(struct domain *d, domain_crash(d); return -EOVERFLOW; } - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + clear_assignment_reference(page); put_page(page); } put_gfn(d, gfn); diff --git a/xen/arch/x86/mm/p2m-pod.c b/xen/arch/x86/mm/p2m-pod.c index 4313863066..2e22764950 100644 --- a/xen/arch/x86/mm/p2m-pod.c +++ b/xen/arch/x86/mm/p2m-pod.c @@ -274,9 +274,7 @@ p2m_pod_set_cache_target(struct p2m_domain *p2m, unsigned long pod_target, int p if ( test_and_clear_bit(_PGT_pinned, &(page+i)->u.inuse.type_info) ) put_page_and_type(page + i); - if ( test_and_clear_bit(_PGC_allocated, &(page+i)->count_info) ) - put_page(page + i); - + clear_assignment_reference(page + i); put_page(page + i); if ( preemptible && pod_target != p2m->pod.count && diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c index 4c9954867c..ce6859d51b 100644 --- a/xen/arch/x86/mm/p2m.c +++ b/xen/arch/x86/mm/p2m.c @@ -1609,8 +1609,7 @@ int p2m_mem_paging_evict(struct domain *d, unsigned long gfn_l) goto out_put; /* Decrement guest domain's ref count of the page */ - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + clear_assignment_reference(page); /* Remove mapping from p2m table */ ret = p2m_set_entry(p2m, gfn, INVALID_MFN, PAGE_ORDER_4K, diff --git a/xen/common/grant_table.c b/xen/common/grant_table.c index e6a0f30a4b..5ae85e3dad 100644 --- a/xen/common/grant_table.c +++ b/xen/common/grant_table.c @@ -1707,8 +1707,7 @@ gnttab_unpopulate_status_frames(struct domain *d, struct grant_table *gt) } BUG_ON(page_get_owner(pg) != d); - if ( test_and_clear_bit(_PGC_allocated, &pg->count_info) ) - put_page(pg); + clear_assignment_reference(pg); if ( pg->count_info & ~PGC_xen_heap ) { diff --git a/xen/common/memory.c b/xen/common/memory.c index 03db7bfa9e..ab19a4ca86 100644 --- a/xen/common/memory.c +++ b/xen/common/memory.c @@ -388,9 +388,8 @@ int guest_remove_page(struct domain *d, unsigned long gmfn) * For this purpose (and to match populate_physmap() behavior), the page * is kept allocated. */ - if ( !rc && !is_domain_direct_mapped(d) && - test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + if ( !rc && !is_domain_direct_mapped(d) ) + clear_assignment_reference(page); put_page(page); diff --git a/xen/common/xenoprof.c b/xen/common/xenoprof.c index 8a72e382e6..262d537074 100644 --- a/xen/common/xenoprof.c +++ b/xen/common/xenoprof.c @@ -173,8 +173,7 @@ unshare_xenoprof_page_with_guest(struct xenoprof *x) struct page_info *page = mfn_to_page(mfn_add(mfn, i)); BUG_ON(page_get_owner(page) != current->domain); - if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) - put_page(page); + clear_assignment_reference(page); } } diff --git a/xen/include/xen/mm.h b/xen/include/xen/mm.h index a57974ae51..1c36c74b8c 100644 --- a/xen/include/xen/mm.h +++ b/xen/include/xen/mm.h @@ -658,4 +658,15 @@ static inline void share_xen_page_with_privileged_guests( share_xen_page_with_guest(page, dom_xen, flags); } +static inline void clear_assignment_reference(struct page_info *page) +{ + /* + * It is unsafe to clear _PGC_allocated without holding an additional + * reference. + */ + ASSERT((page->count_info & PGC_count_mask) > 1); + if ( test_and_clear_bit(_PGC_allocated, &page->count_info) ) + put_page(page); +} + #endif /* __XEN_MM_H__ */
The _PGC_allocated flag is set on a page when it is assigned to a domain along with an initial reference count of 1. To clear this initial reference count it is necessary to test-and-clear _PGC_allocated and then only drop the reference if the test-and-clear succeeds. This is open- coded in many places. It is also unsafe to test-and-clear _PGC_allocated unless the caller holds an additional reference. This patch adds a helper function, clear_assignment_reference(), to replace all the open-coded test-and-clear/put_page occurrences and incorporates in that an ASSERTion that an additional page reference is held. Signed-off-by: Paul Durrant <paul.durrant@citrix.com> --- Cc: Stefano Stabellini <sstabellini@kernel.org> Cc: Julien Grall <julien.grall@arm.com> Cc: Volodymyr Babchuk <Volodymyr_Babchuk@epam.com> Cc: Andrew Cooper <andrew.cooper3@citrix.com> Cc: George Dunlap <George.Dunlap@eu.citrix.com> Cc: Ian Jackson <ian.jackson@eu.citrix.com> Cc: Jan Beulich <jbeulich@suse.com> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Tim Deegan <tim@xen.org> Cc: Wei Liu <wl@xen.org> Cc: "Roger Pau Monné" <roger.pau@citrix.com> Cc: Tamas K Lengyel <tamas@tklengyel.com> Cc: George Dunlap <george.dunlap@eu.citrix.com> --- xen/arch/arm/domain.c | 4 +--- xen/arch/x86/domain.c | 3 +-- xen/arch/x86/hvm/ioreq.c | 11 ++--------- xen/arch/x86/mm.c | 3 +-- xen/arch/x86/mm/mem_sharing.c | 9 +++------ xen/arch/x86/mm/p2m-pod.c | 4 +--- xen/arch/x86/mm/p2m.c | 3 +-- xen/common/grant_table.c | 3 +-- xen/common/memory.c | 5 ++--- xen/common/xenoprof.c | 3 +-- xen/include/xen/mm.h | 11 +++++++++++ 11 files changed, 25 insertions(+), 34 deletions(-)