| Message ID | 20200624141429.382157-1-pbonzini@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [kvm-unit-tests] i386: setup segment registers before percpu areas | expand |
> On Jun 24, 2020, at 7:14 AM, Paolo Bonzini <pbonzini@redhat.com> wrote: > > The base of the percpu area is stored in the %gs base, and writing > to %gs destroys it. Move setup_segments earlier, before the %gs > base is written. > > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > x86/cstart.S | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/x86/cstart.S b/x86/cstart.S > index 5ad70b5..77dc34d 100644 > --- a/x86/cstart.S > +++ b/x86/cstart.S > @@ -106,6 +106,7 @@ MSR_GS_BASE = 0xc0000101 > .globl start > start: > mov $stacktop, %esp > + setup_segments > push %ebx > call setup_multiboot > call setup_libcflat > @@ -118,7 +119,6 @@ start: > > prepare_32: > lgdtl gdt32_descr > - setup_segments > > mov %cr4, %eax > bts $4, %eax // pse > — > 2.26.2 As I said in a different thread, this change breaks my setup. It is better not to make any assumption (or as few as possible) about the GDT content after boot and load the GDTR before setting up the segments. So I prefer to load the GDT before the segments. How about this change instead of yours? -- >8 -- From: Nadav Amit <namit@vmware.com> Date: Wed, 24 Jun 2020 19:50:36 +0000 Subject: [PATCH] x86: load gdt while loading segments Signed-off-by: Nadav Amit <namit@vmware.com> --- x86/cstart.S | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/x86/cstart.S b/x86/cstart.S index dd33d4d..1d8b8ac 100644 --- a/x86/cstart.S +++ b/x86/cstart.S @@ -95,6 +95,8 @@ MSR_GS_BASE = 0xc0000101 .endm .macro setup_segments + lgdtl gdt32_descr + mov $0x10, %ax mov %ax, %ds mov %ax, %es @@ -106,6 +108,8 @@ MSR_GS_BASE = 0xc0000101 .globl start start: mov $stacktop, %esp + setup_segments + push %ebx call setup_multiboot call setup_libcflat @@ -117,9 +121,6 @@ start: jmpl $8, $start32 prepare_32: - lgdtl gdt32_descr - setup_segments - mov %cr4, %eax bts $4, %eax // pse mov %eax, %cr4 --
On 24/06/20 21:58, Nadav Amit wrote: >> On Jun 24, 2020, at 7:14 AM, Paolo Bonzini <pbonzini@redhat.com> wrote: >> >> The base of the percpu area is stored in the %gs base, and writing >> to %gs destroys it. Move setup_segments earlier, before the %gs >> base is written. >> >> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> >> --- >> x86/cstart.S | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/x86/cstart.S b/x86/cstart.S >> index 5ad70b5..77dc34d 100644 >> --- a/x86/cstart.S >> +++ b/x86/cstart.S >> @@ -106,6 +106,7 @@ MSR_GS_BASE = 0xc0000101 >> .globl start >> start: >> mov $stacktop, %esp >> + setup_segments >> push %ebx >> call setup_multiboot >> call setup_libcflat >> @@ -118,7 +119,6 @@ start: >> >> prepare_32: >> lgdtl gdt32_descr >> - setup_segments >> >> mov %cr4, %eax >> bts $4, %eax // pse >> — >> 2.26.2 > > As I said in a different thread, this change breaks my setup. It is better > not to make any assumption (or as few as possible) about the GDT content > after boot and load the GDTR before setting up the segments. So I prefer to > load the GDT before the segments. How about this change instead of yours? Yes, this is better. Paolo > -- >8 -- > > From: Nadav Amit <namit@vmware.com> > Date: Wed, 24 Jun 2020 19:50:36 +0000 > Subject: [PATCH] x86: load gdt while loading segments > > Signed-off-by: Nadav Amit <namit@vmware.com> > --- > x86/cstart.S | 7 ++++--- > 1 file changed, 4 insertions(+), 3 deletions(-) > > diff --git a/x86/cstart.S b/x86/cstart.S > index dd33d4d..1d8b8ac 100644 > --- a/x86/cstart.S > +++ b/x86/cstart.S > @@ -95,6 +95,8 @@ MSR_GS_BASE = 0xc0000101 > .endm > > .macro setup_segments > + lgdtl gdt32_descr > + > mov $0x10, %ax > mov %ax, %ds > mov %ax, %es > @@ -106,6 +108,8 @@ MSR_GS_BASE = 0xc0000101 > .globl start > start: > mov $stacktop, %esp > + setup_segments > + > push %ebx > call setup_multiboot > call setup_libcflat > @@ -117,9 +121,6 @@ start: > jmpl $8, $start32 > > prepare_32: > - lgdtl gdt32_descr > - setup_segments > - > mov %cr4, %eax > bts $4, %eax // pse > mov %eax, %cr4 >
On 24/06/20 21:58, Nadav Amit wrote: > From: Nadav Amit <namit@vmware.com> > Date: Wed, 24 Jun 2020 19:50:36 +0000 > Subject: [PATCH] x86: load gdt while loading segments > > Signed-off-by: Nadav Amit <namit@vmware.com> > --- > x86/cstart.S | 7 ++++--- > 1 file changed, 4 insertions(+), 3 deletions(-) > > diff --git a/x86/cstart.S b/x86/cstart.S > index dd33d4d..1d8b8ac 100644 > --- a/x86/cstart.S > +++ b/x86/cstart.S > @@ -95,6 +95,8 @@ MSR_GS_BASE = 0xc0000101 > .endm > > .macro setup_segments > + lgdtl gdt32_descr > + > mov $0x10, %ax > mov %ax, %ds > mov %ax, %es > @@ -106,6 +108,8 @@ MSR_GS_BASE = 0xc0000101 > .globl start > start: > mov $stacktop, %esp > + setup_segments > + > push %ebx > call setup_multiboot > call setup_libcflat > @@ -117,9 +121,6 @@ start: > jmpl $8, $start32 > > prepare_32: > - lgdtl gdt32_descr > - setup_segments > - > mov %cr4, %eax > bts $4, %eax // pse > mov %eax, %cr4 > -- The GDT is already loaded elsewhere for APs, but the gist of the patch is good. I'll send v2 Paolo
diff --git a/x86/cstart.S b/x86/cstart.S index 5ad70b5..77dc34d 100644 --- a/x86/cstart.S +++ b/x86/cstart.S @@ -106,6 +106,7 @@ MSR_GS_BASE = 0xc0000101 .globl start start: mov $stacktop, %esp + setup_segments push %ebx call setup_multiboot call setup_libcflat @@ -118,7 +119,6 @@ start: prepare_32: lgdtl gdt32_descr - setup_segments mov %cr4, %eax bts $4, %eax // pse
The base of the percpu area is stored in the %gs base, and writing to %gs destroys it. Move setup_segments earlier, before the %gs base is written. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> --- x86/cstart.S | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)