xen/arch/x86/acpi/suspend.c | 14 +++++++++++++- xen/arch/x86/acpi/wakeup_prot.S | 13 +------------ xen/arch/x86/traps.c | 21 --------------------- xen/include/asm-x86/desc.h | 2 -- 4 files changed, 14 insertions(+), 36 deletions(-)
load_TR() is used exclusively in the resume path, but jumps through a lot of
unnecessary hoops.
As suspend/resume is strictly on CPU0 in idle context, the correct GDT to use
is boot_gdt, which means it doesn't need saving on suspend. Similarly, the
correct IDT to use can be derived, and the LDT is guaranteed to be NUL.
The TR is still correct in the GDT, but needs the busy bit clearing before we
can reload it.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Jan Beulich <JBeulich@suse.com>
CC: Wei Liu <wl@xen.org>
CC: Roger Pau Monné <roger.pau@citrix.com>
A slightly different option would be to call load_system_tables() rather than
opencoding part of it in restore_rest_processor_state(). However, that is
more setup than is necessary. Thoughts?
---
xen/arch/x86/acpi/suspend.c | 14 +++++++++++++-
xen/arch/x86/acpi/wakeup_prot.S | 13 +------------
xen/arch/x86/traps.c | 21 ---------------------
xen/include/asm-x86/desc.h | 2 --
4 files changed, 14 insertions(+), 36 deletions(-)
diff --git a/xen/arch/x86/acpi/suspend.c b/xen/arch/x86/acpi/suspend.c
index ba9d2e13a7..a6f2584645 100644
--- a/xen/arch/x86/acpi/suspend.c
+++ b/xen/arch/x86/acpi/suspend.c
@@ -41,7 +41,19 @@ void save_rest_processor_state(void)
void restore_rest_processor_state(void)
{
- load_TR();
+ unsigned int cpu = smp_processor_id();
+ seg_desc_t *gdt = per_cpu(gdt, cpu) - FIRST_RESERVED_GDT_ENTRY;
+ struct tss64 *tss = &per_cpu(tss_page, cpu).tss;
+ const struct desc_ptr idtr = {
+ .base = (unsigned long)idt_tables[cpu],
+ .limit = (IDT_ENTRIES * sizeof(idt_entry_t)) - 1,
+ };
+
+ _set_tssldt_desc(gdt + TSS_ENTRY, (unsigned long)tss,
+ sizeof(*tss) - 1, SYS_DESC_tss_avail);
+ lidt(&idtr);
+ ltr(TSS_SELECTOR);
+ lldt(0);
/* Recover syscall MSRs */
wrmsrl(MSR_LSTAR, saved_lstar);
diff --git a/xen/arch/x86/acpi/wakeup_prot.S b/xen/arch/x86/acpi/wakeup_prot.S
index 9e9fcc1ab6..74261cb4f1 100644
--- a/xen/arch/x86/acpi/wakeup_prot.S
+++ b/xen/arch/x86/acpi/wakeup_prot.S
@@ -34,10 +34,6 @@ ENTRY(do_suspend_lowlevel)
mov %ss, REF(saved_ss)
- sgdt REF(saved_gdt)
- sidt REF(saved_idt)
- sldt REF(saved_ldt)
-
mov %cr0, GREG(ax)
mov GREG(ax), REF(saved_cr0)
@@ -55,6 +51,7 @@ ENTRY(do_suspend_lowlevel)
ENTRY(__ret_point)
+ lgdt boot_gdtr(%rip)
/* mmu_cr4_features contains latest cr4 setting */
mov REF(mmu_cr4_features), GREG(ax)
@@ -66,10 +63,6 @@ ENTRY(__ret_point)
mov REF(saved_cr0), GREG(ax)
mov GREG(ax), %cr0
- lgdt REF(saved_gdt)
- lidt REF(saved_idt)
- lldt REF(saved_ldt)
-
mov REF(saved_ss), %ss
LOAD_GREG(sp)
@@ -129,9 +122,5 @@ DECLARE_GREG(13)
DECLARE_GREG(14)
DECLARE_GREG(15)
-saved_gdt: .quad 0,0
-saved_idt: .quad 0,0
-saved_ldt: .quad 0,0
-
saved_cr0: .quad 0
saved_cr3: .quad 0
diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
index 23069e25ec..b424687f85 100644
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -1901,27 +1901,6 @@ static void __init set_intr_gate(unsigned int n, void *addr)
__set_intr_gate(n, 0, addr);
}
-void load_TR(void)
-{
- struct tss64 *tss = &this_cpu(tss_page).tss;
- struct desc_ptr old_gdt, tss_gdt = {
- .base = (long)(this_cpu(gdt) - FIRST_RESERVED_GDT_ENTRY),
- .limit = LAST_RESERVED_GDT_BYTE
- };
-
- _set_tssldt_desc(
- this_cpu(gdt) + TSS_ENTRY - FIRST_RESERVED_GDT_ENTRY,
- (unsigned long)tss, sizeof(*tss) - 1, SYS_DESC_tss_avail);
- _set_tssldt_desc(
- this_cpu(compat_gdt) + TSS_ENTRY - FIRST_RESERVED_GDT_ENTRY,
- (unsigned long)tss, sizeof(*tss) - 1, SYS_DESC_tss_busy);
-
- /* Switch to non-compat GDT (which has B bit clear) to execute LTR. */
- asm volatile (
- "sgdt %0; lgdt %2; ltr %w1; lgdt %0"
- : "=m" (old_gdt) : "rm" (TSS_SELECTOR), "m" (tss_gdt) : "memory" );
-}
-
static unsigned int calc_ler_msr(void)
{
switch ( boot_cpu_data.x86_vendor )
diff --git a/xen/include/asm-x86/desc.h b/xen/include/asm-x86/desc.h
index 603b9a9013..24db3e9510 100644
--- a/xen/include/asm-x86/desc.h
+++ b/xen/include/asm-x86/desc.h
@@ -216,8 +216,6 @@ DECLARE_PER_CPU(seg_desc_t *, compat_gdt);
DECLARE_PER_CPU(l1_pgentry_t, compat_gdt_l1e);
DECLARE_PER_CPU(bool, full_gdt_loaded);
-extern void load_TR(void);
-
static inline void lgdt(const struct desc_ptr *gdtr)
{
__asm__ __volatile__ ( "lgdt %0" :: "m" (*gdtr) : "memory" );
--
2.11.0
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-develOn 12.08.2019 20:21, Andrew Cooper wrote: > load_TR() is used exclusively in the resume path, but jumps through a lot of > unnecessary hoops. > > As suspend/resume is strictly on CPU0 in idle context, the correct GDT to use > is boot_gdt, which means it doesn't need saving on suspend. Similarly, the > correct IDT to use can be derived, and the LDT is guaranteed to be NUL. > > The TR is still correct in the GDT, but needs the busy bit clearing before we > can reload it. > > Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> > --- > CC: Jan Beulich <JBeulich@suse.com> > CC: Wei Liu <wl@xen.org> > CC: Roger Pau Monné <roger.pau@citrix.com> > > A slightly different option would be to call load_system_tables() rather than > opencoding part of it in restore_rest_processor_state(). However, that is > more setup than is necessary. Thoughts? This might indeed be better (despite the parts not needed on this path), as it would further centralize the handling. Of course only as long as there's no (significant) change needed to the function in order to be used here. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 28/08/2019 15:10, Jan Beulich wrote: > On 12.08.2019 20:21, Andrew Cooper wrote: >> load_TR() is used exclusively in the resume path, but jumps through a lot of >> unnecessary hoops. >> >> As suspend/resume is strictly on CPU0 in idle context, the correct GDT to use >> is boot_gdt, which means it doesn't need saving on suspend. Similarly, the >> correct IDT to use can be derived, and the LDT is guaranteed to be NUL. >> >> The TR is still correct in the GDT, but needs the busy bit clearing before we >> can reload it. >> >> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> >> --- >> CC: Jan Beulich <JBeulich@suse.com> >> CC: Wei Liu <wl@xen.org> >> CC: Roger Pau Monné <roger.pau@citrix.com> >> >> A slightly different option would be to call load_system_tables() rather than >> opencoding part of it in restore_rest_processor_state(). However, that is >> more setup than is necessary. Thoughts? > This might indeed be better (despite the parts not needed on this > path), as it would further centralize the handling. Of course only > as long as there's no (significant) change needed to the function > in order to be used here. No change at all. The AP's already do go through load_system_tables() on S3 resume. It is slightly suboptimal that we get two identical lgdt's this close together, but it is the only way of ending up entering C with __HYPERVISOR_CS. At some future point, I'll see how easy it is to arrange the trampoline GDT to be usable with __HYPERVISOR_CS, which would simplify the boot and resume paths. ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
© 2016 - 2024 Red Hat, Inc.