| Message ID | 20230913113828.18421-4-jgross@suse.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 3e0377f529ce8c3755134f71864016905049d0ba |
| Headers | show |
| Series | xen: cleanup and fix lazy mode handling | expand |
On 9/13/23 7:38 AM, Juergen Gross wrote: > When running as a paravirtualized guest under Xen, Linux is using > "lazy mode" for issuing hypercalls which don't need to take immediate > effect in order to improve performance (examples are e.g. multiple > PTE changes). > > There are two different lazy modes defined: MMU and CPU lazy mode. > Today it is not possible to nest multiple lazy mode sections, even if > they are of the same kind. A recent change in memory management added > nesting of MMU lazy mode sections, resulting in a regression when > running as Xen PV guest. > > Technically there is no reason why nesting of multiple sections of the > same kind of lazy mode shouldn't be allowed. So add support for that > for fixing the regression. > > Fixes: bcc6cc832573 ("mm: add default definition of set_ptes()") > Signed-off-by: Juergen Gross <jgross@suse.com> For patches 2 and 3 Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
diff --git a/arch/x86/include/asm/xen/hypervisor.h b/arch/x86/include/asm/xen/hypervisor.h index ed05ce3df5c7..7048dfacc04b 100644 --- a/arch/x86/include/asm/xen/hypervisor.h +++ b/arch/x86/include/asm/xen/hypervisor.h @@ -72,10 +72,18 @@ enum xen_lazy_mode { }; DECLARE_PER_CPU(enum xen_lazy_mode, xen_lazy_mode); +DECLARE_PER_CPU(unsigned int, xen_lazy_nesting); static inline void enter_lazy(enum xen_lazy_mode mode) { - BUG_ON(this_cpu_read(xen_lazy_mode) != XEN_LAZY_NONE); + enum xen_lazy_mode old_mode = this_cpu_read(xen_lazy_mode); + + if (mode == old_mode) { + this_cpu_inc(xen_lazy_nesting); + return; + } + + BUG_ON(old_mode != XEN_LAZY_NONE); this_cpu_write(xen_lazy_mode, mode); } @@ -84,7 +92,10 @@ static inline void leave_lazy(enum xen_lazy_mode mode) { BUG_ON(this_cpu_read(xen_lazy_mode) != mode); - this_cpu_write(xen_lazy_mode, XEN_LAZY_NONE); + if (this_cpu_read(xen_lazy_nesting) == 0) + this_cpu_write(xen_lazy_mode, XEN_LAZY_NONE); + else + this_cpu_dec(xen_lazy_nesting); } enum xen_lazy_mode xen_get_lazy_mode(void); diff --git a/arch/x86/xen/enlighten_pv.c b/arch/x86/xen/enlighten_pv.c index 54b83825c4b6..bbbfdd495ebd 100644 --- a/arch/x86/xen/enlighten_pv.c +++ b/arch/x86/xen/enlighten_pv.c @@ -102,6 +102,7 @@ struct tls_descs { }; DEFINE_PER_CPU(enum xen_lazy_mode, xen_lazy_mode) = XEN_LAZY_NONE; +DEFINE_PER_CPU(unsigned int, xen_lazy_nesting); enum xen_lazy_mode xen_get_lazy_mode(void) {
When running as a paravirtualized guest under Xen, Linux is using "lazy mode" for issuing hypercalls which don't need to take immediate effect in order to improve performance (examples are e.g. multiple PTE changes). There are two different lazy modes defined: MMU and CPU lazy mode. Today it is not possible to nest multiple lazy mode sections, even if they are of the same kind. A recent change in memory management added nesting of MMU lazy mode sections, resulting in a regression when running as Xen PV guest. Technically there is no reason why nesting of multiple sections of the same kind of lazy mode shouldn't be allowed. So add support for that for fixing the regression. Fixes: bcc6cc832573 ("mm: add default definition of set_ptes()") Signed-off-by: Juergen Gross <jgross@suse.com> --- arch/x86/include/asm/xen/hypervisor.h | 15 +++++++++++++-- arch/x86/xen/enlighten_pv.c | 1 + 2 files changed, 14 insertions(+), 2 deletions(-)