2006-01-11 21:46:18 +00:00
|
|
|
/*
|
|
|
|
* vSMPowered(tm) systems specific initialization
|
|
|
|
* Copyright (C) 2005 ScaleMP Inc.
|
|
|
|
*
|
|
|
|
* Use of this code is subject to the terms and conditions of the
|
|
|
|
* GNU general public license version 2. See "COPYING" or
|
|
|
|
* http://www.gnu.org/licenses/gpl.html
|
|
|
|
*
|
|
|
|
* Ravikiran Thirumalai <kiran@scalemp.com>,
|
|
|
|
* Shai Fultheim <shai@scalemp.com>
|
2008-02-11 19:16:04 +00:00
|
|
|
* Paravirt ops integration: Glauber de Oliveira Costa <gcosta@redhat.com>,
|
|
|
|
* Ravikiran Thirumalai <kiran@scalemp.com>
|
2006-01-11 21:46:18 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/pci_ids.h>
|
|
|
|
#include <linux/pci_regs.h>
|
2012-04-16 07:39:35 +00:00
|
|
|
#include <linux/smp.h>
|
2012-06-02 22:11:35 +00:00
|
|
|
#include <linux/irq.h>
|
2008-05-12 13:43:34 +00:00
|
|
|
|
|
|
|
#include <asm/apic.h>
|
2006-01-11 21:46:18 +00:00
|
|
|
#include <asm/pci-direct.h>
|
2006-10-12 19:17:52 +00:00
|
|
|
#include <asm/io.h>
|
2008-02-11 19:16:04 +00:00
|
|
|
#include <asm/paravirt.h>
|
2008-05-12 13:43:34 +00:00
|
|
|
#include <asm/setup.h>
|
2008-02-11 19:16:04 +00:00
|
|
|
|
2012-04-16 07:39:35 +00:00
|
|
|
#define TOPOLOGY_REGISTER_OFFSET 0x10
|
|
|
|
|
2009-03-24 06:14:29 +00:00
|
|
|
#if defined CONFIG_PCI && defined CONFIG_PARAVIRT
|
2008-02-11 19:16:04 +00:00
|
|
|
/*
|
|
|
|
* Interrupt control on vSMPowered systems:
|
|
|
|
* ~AC is a shadow of IF. If IF is 'on' AC should be 'off'
|
|
|
|
* and vice versa.
|
|
|
|
*/
|
|
|
|
|
2013-10-22 16:07:56 +00:00
|
|
|
asmlinkage unsigned long vsmp_save_fl(void)
|
2008-02-11 19:16:04 +00:00
|
|
|
{
|
|
|
|
unsigned long flags = native_save_fl();
|
|
|
|
|
|
|
|
if (!(flags & X86_EFLAGS_IF) || (flags & X86_EFLAGS_AC))
|
|
|
|
flags &= ~X86_EFLAGS_IF;
|
|
|
|
return flags;
|
|
|
|
}
|
x86/paravirt: add register-saving thunks to reduce caller register pressure
Impact: Optimization
One of the problems with inserting a pile of C calls where previously
there were none is that the register pressure is greatly increased.
The C calling convention says that the caller must expect a certain
set of registers may be trashed by the callee, and that the callee can
use those registers without restriction. This includes the function
argument registers, and several others.
This patch seeks to alleviate this pressure by introducing wrapper
thunks that will do the register saving/restoring, so that the
callsite doesn't need to worry about it, but the callee function can
be conventional compiler-generated code. In many cases (particularly
performance-sensitive cases) the callee will be in assembler anyway,
and need not use the compiler's calling convention.
Standard calling convention is:
arguments return scratch
x86-32 eax edx ecx eax ?
x86-64 rdi rsi rdx rcx rax r8 r9 r10 r11
The thunk preserves all argument and scratch registers. The return
register is not preserved, and is available as a scratch register for
unwrapped callee code (and of course the return value).
Wrapped function pointers are themselves wrapped in a struct
paravirt_callee_save structure, in order to get some warning from the
compiler when functions with mismatched calling conventions are used.
The most common paravirt ops, both statically and dynamically, are
interrupt enable/disable/save/restore, so handle them first. This is
particularly easy since their calls are handled specially anyway.
XXX Deal with VMI. What's their calling convention?
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-01-28 22:35:05 +00:00
|
|
|
PV_CALLEE_SAVE_REGS_THUNK(vsmp_save_fl);
|
2008-02-11 19:16:04 +00:00
|
|
|
|
2013-10-22 16:07:56 +00:00
|
|
|
__visible void vsmp_restore_fl(unsigned long flags)
|
2008-02-11 19:16:04 +00:00
|
|
|
{
|
|
|
|
if (flags & X86_EFLAGS_IF)
|
|
|
|
flags &= ~X86_EFLAGS_AC;
|
|
|
|
else
|
|
|
|
flags |= X86_EFLAGS_AC;
|
|
|
|
native_restore_fl(flags);
|
|
|
|
}
|
x86/paravirt: add register-saving thunks to reduce caller register pressure
Impact: Optimization
One of the problems with inserting a pile of C calls where previously
there were none is that the register pressure is greatly increased.
The C calling convention says that the caller must expect a certain
set of registers may be trashed by the callee, and that the callee can
use those registers without restriction. This includes the function
argument registers, and several others.
This patch seeks to alleviate this pressure by introducing wrapper
thunks that will do the register saving/restoring, so that the
callsite doesn't need to worry about it, but the callee function can
be conventional compiler-generated code. In many cases (particularly
performance-sensitive cases) the callee will be in assembler anyway,
and need not use the compiler's calling convention.
Standard calling convention is:
arguments return scratch
x86-32 eax edx ecx eax ?
x86-64 rdi rsi rdx rcx rax r8 r9 r10 r11
The thunk preserves all argument and scratch registers. The return
register is not preserved, and is available as a scratch register for
unwrapped callee code (and of course the return value).
Wrapped function pointers are themselves wrapped in a struct
paravirt_callee_save structure, in order to get some warning from the
compiler when functions with mismatched calling conventions are used.
The most common paravirt ops, both statically and dynamically, are
interrupt enable/disable/save/restore, so handle them first. This is
particularly easy since their calls are handled specially anyway.
XXX Deal with VMI. What's their calling convention?
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-01-28 22:35:05 +00:00
|
|
|
PV_CALLEE_SAVE_REGS_THUNK(vsmp_restore_fl);
|
2008-02-11 19:16:04 +00:00
|
|
|
|
2013-10-22 16:07:56 +00:00
|
|
|
asmlinkage void vsmp_irq_disable(void)
|
2008-02-11 19:16:04 +00:00
|
|
|
{
|
|
|
|
unsigned long flags = native_save_fl();
|
|
|
|
|
|
|
|
native_restore_fl((flags & ~X86_EFLAGS_IF) | X86_EFLAGS_AC);
|
|
|
|
}
|
x86/paravirt: add register-saving thunks to reduce caller register pressure
Impact: Optimization
One of the problems with inserting a pile of C calls where previously
there were none is that the register pressure is greatly increased.
The C calling convention says that the caller must expect a certain
set of registers may be trashed by the callee, and that the callee can
use those registers without restriction. This includes the function
argument registers, and several others.
This patch seeks to alleviate this pressure by introducing wrapper
thunks that will do the register saving/restoring, so that the
callsite doesn't need to worry about it, but the callee function can
be conventional compiler-generated code. In many cases (particularly
performance-sensitive cases) the callee will be in assembler anyway,
and need not use the compiler's calling convention.
Standard calling convention is:
arguments return scratch
x86-32 eax edx ecx eax ?
x86-64 rdi rsi rdx rcx rax r8 r9 r10 r11
The thunk preserves all argument and scratch registers. The return
register is not preserved, and is available as a scratch register for
unwrapped callee code (and of course the return value).
Wrapped function pointers are themselves wrapped in a struct
paravirt_callee_save structure, in order to get some warning from the
compiler when functions with mismatched calling conventions are used.
The most common paravirt ops, both statically and dynamically, are
interrupt enable/disable/save/restore, so handle them first. This is
particularly easy since their calls are handled specially anyway.
XXX Deal with VMI. What's their calling convention?
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-01-28 22:35:05 +00:00
|
|
|
PV_CALLEE_SAVE_REGS_THUNK(vsmp_irq_disable);
|
2008-02-11 19:16:04 +00:00
|
|
|
|
2013-10-22 16:07:56 +00:00
|
|
|
asmlinkage void vsmp_irq_enable(void)
|
2008-02-11 19:16:04 +00:00
|
|
|
{
|
|
|
|
unsigned long flags = native_save_fl();
|
|
|
|
|
|
|
|
native_restore_fl((flags | X86_EFLAGS_IF) & (~X86_EFLAGS_AC));
|
|
|
|
}
|
x86/paravirt: add register-saving thunks to reduce caller register pressure
Impact: Optimization
One of the problems with inserting a pile of C calls where previously
there were none is that the register pressure is greatly increased.
The C calling convention says that the caller must expect a certain
set of registers may be trashed by the callee, and that the callee can
use those registers without restriction. This includes the function
argument registers, and several others.
This patch seeks to alleviate this pressure by introducing wrapper
thunks that will do the register saving/restoring, so that the
callsite doesn't need to worry about it, but the callee function can
be conventional compiler-generated code. In many cases (particularly
performance-sensitive cases) the callee will be in assembler anyway,
and need not use the compiler's calling convention.
Standard calling convention is:
arguments return scratch
x86-32 eax edx ecx eax ?
x86-64 rdi rsi rdx rcx rax r8 r9 r10 r11
The thunk preserves all argument and scratch registers. The return
register is not preserved, and is available as a scratch register for
unwrapped callee code (and of course the return value).
Wrapped function pointers are themselves wrapped in a struct
paravirt_callee_save structure, in order to get some warning from the
compiler when functions with mismatched calling conventions are used.
The most common paravirt ops, both statically and dynamically, are
interrupt enable/disable/save/restore, so handle them first. This is
particularly easy since their calls are handled specially anyway.
XXX Deal with VMI. What's their calling convention?
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-01-28 22:35:05 +00:00
|
|
|
PV_CALLEE_SAVE_REGS_THUNK(vsmp_irq_enable);
|
2008-02-11 19:16:04 +00:00
|
|
|
|
2008-09-23 05:58:47 +00:00
|
|
|
static unsigned __init_or_module vsmp_patch(u8 type, u16 clobbers, void *ibuf,
|
2008-02-11 19:16:04 +00:00
|
|
|
unsigned long addr, unsigned len)
|
|
|
|
{
|
|
|
|
switch (type) {
|
|
|
|
case PARAVIRT_PATCH(pv_irq_ops.irq_enable):
|
|
|
|
case PARAVIRT_PATCH(pv_irq_ops.irq_disable):
|
|
|
|
case PARAVIRT_PATCH(pv_irq_ops.save_fl):
|
|
|
|
case PARAVIRT_PATCH(pv_irq_ops.restore_fl):
|
|
|
|
return paravirt_patch_default(type, clobbers, ibuf, addr, len);
|
|
|
|
default:
|
|
|
|
return native_patch(type, clobbers, ibuf, addr, len);
|
|
|
|
}
|
|
|
|
|
|
|
|
}
|
2006-01-11 21:46:18 +00:00
|
|
|
|
2008-03-20 07:41:16 +00:00
|
|
|
static void __init set_vsmp_pv_ops(void)
|
2006-01-11 21:46:18 +00:00
|
|
|
{
|
2008-10-29 06:05:22 +00:00
|
|
|
void __iomem *address;
|
2008-02-11 19:16:03 +00:00
|
|
|
unsigned int cap, ctl, cfg;
|
2006-01-11 21:46:18 +00:00
|
|
|
|
|
|
|
/* set vSMP magic bits to indicate vSMP capable kernel */
|
2008-02-11 19:16:03 +00:00
|
|
|
cfg = read_pci_config(0, 0x1f, 0, PCI_BASE_ADDRESS_0);
|
|
|
|
address = early_ioremap(cfg, 8);
|
2006-01-11 21:46:18 +00:00
|
|
|
cap = readl(address);
|
|
|
|
ctl = readl(address + 4);
|
2008-01-30 12:30:24 +00:00
|
|
|
printk(KERN_INFO "vSMP CTL: capabilities:0x%08x control:0x%08x\n",
|
|
|
|
cap, ctl);
|
2012-06-02 22:11:35 +00:00
|
|
|
|
|
|
|
/* If possible, let the vSMP foundation route the interrupt optimally */
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
if (cap & ctl & BIT(8)) {
|
|
|
|
ctl &= ~BIT(8);
|
2012-06-14 15:43:08 +00:00
|
|
|
#ifdef CONFIG_PROC_FS
|
|
|
|
/* Don't let users change irq affinity via procfs */
|
2012-06-02 22:11:35 +00:00
|
|
|
no_irq_affinity = 1;
|
2012-06-14 15:43:08 +00:00
|
|
|
#endif
|
2012-06-02 22:11:35 +00:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2006-01-11 21:46:18 +00:00
|
|
|
if (cap & ctl & (1 << 4)) {
|
2008-03-20 07:43:16 +00:00
|
|
|
/* Setup irq ops and turn on vSMP IRQ fastpath handling */
|
x86/paravirt: add register-saving thunks to reduce caller register pressure
Impact: Optimization
One of the problems with inserting a pile of C calls where previously
there were none is that the register pressure is greatly increased.
The C calling convention says that the caller must expect a certain
set of registers may be trashed by the callee, and that the callee can
use those registers without restriction. This includes the function
argument registers, and several others.
This patch seeks to alleviate this pressure by introducing wrapper
thunks that will do the register saving/restoring, so that the
callsite doesn't need to worry about it, but the callee function can
be conventional compiler-generated code. In many cases (particularly
performance-sensitive cases) the callee will be in assembler anyway,
and need not use the compiler's calling convention.
Standard calling convention is:
arguments return scratch
x86-32 eax edx ecx eax ?
x86-64 rdi rsi rdx rcx rax r8 r9 r10 r11
The thunk preserves all argument and scratch registers. The return
register is not preserved, and is available as a scratch register for
unwrapped callee code (and of course the return value).
Wrapped function pointers are themselves wrapped in a struct
paravirt_callee_save structure, in order to get some warning from the
compiler when functions with mismatched calling conventions are used.
The most common paravirt ops, both statically and dynamically, are
interrupt enable/disable/save/restore, so handle them first. This is
particularly easy since their calls are handled specially anyway.
XXX Deal with VMI. What's their calling convention?
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2009-01-28 22:35:05 +00:00
|
|
|
pv_irq_ops.irq_disable = PV_CALLEE_SAVE(vsmp_irq_disable);
|
|
|
|
pv_irq_ops.irq_enable = PV_CALLEE_SAVE(vsmp_irq_enable);
|
|
|
|
pv_irq_ops.save_fl = PV_CALLEE_SAVE(vsmp_save_fl);
|
|
|
|
pv_irq_ops.restore_fl = PV_CALLEE_SAVE(vsmp_restore_fl);
|
2008-03-20 07:43:16 +00:00
|
|
|
pv_init_ops.patch = vsmp_patch;
|
2006-01-11 21:46:18 +00:00
|
|
|
ctl &= ~(1 << 4);
|
|
|
|
}
|
2012-06-02 22:11:35 +00:00
|
|
|
writel(ctl, address + 4);
|
|
|
|
ctl = readl(address + 4);
|
|
|
|
pr_info("vSMP CTL: control set to:0x%08x\n", ctl);
|
2006-01-11 21:46:18 +00:00
|
|
|
|
2008-02-11 19:16:03 +00:00
|
|
|
early_iounmap(address, 8);
|
2008-03-20 07:41:16 +00:00
|
|
|
}
|
|
|
|
#else
|
|
|
|
static void __init set_vsmp_pv_ops(void)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2009-03-24 06:14:29 +00:00
|
|
|
#ifdef CONFIG_PCI
|
2008-03-24 21:48:36 +00:00
|
|
|
static int is_vsmp = -1;
|
2008-03-20 07:41:16 +00:00
|
|
|
|
2008-03-24 21:48:36 +00:00
|
|
|
static void __init detect_vsmp_box(void)
|
2008-03-20 07:41:16 +00:00
|
|
|
{
|
2008-03-24 21:48:36 +00:00
|
|
|
is_vsmp = 0;
|
2008-03-20 07:41:16 +00:00
|
|
|
|
|
|
|
if (!early_pci_allowed())
|
2008-03-24 21:48:36 +00:00
|
|
|
return;
|
2008-03-20 07:41:16 +00:00
|
|
|
|
2008-03-24 21:48:36 +00:00
|
|
|
/* Check if we are running on a ScaleMP vSMPowered box */
|
2008-03-21 08:55:06 +00:00
|
|
|
if (read_pci_config(0, 0x1f, 0, PCI_VENDOR_ID) ==
|
|
|
|
(PCI_VENDOR_ID_SCALEMP | (PCI_DEVICE_ID_SCALEMP_VSMP_CTL << 16)))
|
2008-03-24 21:48:36 +00:00
|
|
|
is_vsmp = 1;
|
|
|
|
}
|
2008-03-20 07:41:16 +00:00
|
|
|
|
2008-03-24 21:48:36 +00:00
|
|
|
int is_vsmp_box(void)
|
|
|
|
{
|
|
|
|
if (is_vsmp != -1)
|
|
|
|
return is_vsmp;
|
|
|
|
else {
|
|
|
|
WARN_ON_ONCE(1);
|
|
|
|
return 0;
|
|
|
|
}
|
2008-03-20 07:41:16 +00:00
|
|
|
}
|
|
|
|
|
2009-03-24 06:14:29 +00:00
|
|
|
#else
|
|
|
|
static void __init detect_vsmp_box(void)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
int is_vsmp_box(void)
|
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
2012-04-16 07:39:35 +00:00
|
|
|
|
|
|
|
static void __init vsmp_cap_cpus(void)
|
|
|
|
{
|
|
|
|
#if !defined(CONFIG_X86_VSMP) && defined(CONFIG_SMP)
|
|
|
|
void __iomem *address;
|
|
|
|
unsigned int cfg, topology, node_shift, maxcpus;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* CONFIG_X86_VSMP is not configured, so limit the number CPUs to the
|
|
|
|
* ones present in the first board, unless explicitly overridden by
|
|
|
|
* setup_max_cpus
|
|
|
|
*/
|
|
|
|
if (setup_max_cpus != NR_CPUS)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Read the vSMP Foundation topology register */
|
|
|
|
cfg = read_pci_config(0, 0x1f, 0, PCI_BASE_ADDRESS_0);
|
|
|
|
address = early_ioremap(cfg + TOPOLOGY_REGISTER_OFFSET, 4);
|
|
|
|
if (WARN_ON(!address))
|
|
|
|
return;
|
|
|
|
|
|
|
|
topology = readl(address);
|
|
|
|
node_shift = (topology >> 16) & 0x7;
|
|
|
|
if (!node_shift)
|
|
|
|
/* The value 0 should be decoded as 8 */
|
|
|
|
node_shift = 8;
|
|
|
|
maxcpus = (topology & ((1 << node_shift) - 1)) + 1;
|
|
|
|
|
|
|
|
pr_info("vSMP CTL: Capping CPUs to %d (CONFIG_X86_VSMP is unset)\n",
|
|
|
|
maxcpus);
|
|
|
|
setup_max_cpus = maxcpus;
|
|
|
|
early_iounmap(address, 4);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2012-06-02 22:11:34 +00:00
|
|
|
static int apicid_phys_pkg_id(int initial_apic_id, int index_msb)
|
|
|
|
{
|
|
|
|
return hard_smp_processor_id() >> index_msb;
|
|
|
|
}
|
|
|
|
|
2012-06-02 22:11:35 +00:00
|
|
|
/*
|
|
|
|
* In vSMP, all cpus should be capable of handling interrupts, regardless of
|
|
|
|
* the APIC used.
|
|
|
|
*/
|
2012-06-25 20:38:28 +00:00
|
|
|
static void fill_vector_allocation_domain(int cpu, struct cpumask *retmask,
|
|
|
|
const struct cpumask *mask)
|
2012-06-02 22:11:35 +00:00
|
|
|
{
|
|
|
|
cpumask_setall(retmask);
|
|
|
|
}
|
|
|
|
|
2012-06-02 22:11:34 +00:00
|
|
|
static void vsmp_apic_post_init(void)
|
|
|
|
{
|
|
|
|
/* need to update phys_pkg_id */
|
|
|
|
apic->phys_pkg_id = apicid_phys_pkg_id;
|
2012-06-02 22:11:35 +00:00
|
|
|
apic->vector_allocation_domain = fill_vector_allocation_domain;
|
2012-06-02 22:11:34 +00:00
|
|
|
}
|
|
|
|
|
2008-03-20 07:41:16 +00:00
|
|
|
void __init vsmp_init(void)
|
|
|
|
{
|
2008-03-24 21:48:36 +00:00
|
|
|
detect_vsmp_box();
|
2008-03-20 07:41:16 +00:00
|
|
|
if (!is_vsmp_box())
|
|
|
|
return;
|
|
|
|
|
2012-06-02 22:11:34 +00:00
|
|
|
x86_platform.apic_post_init = vsmp_apic_post_init;
|
|
|
|
|
2012-04-16 07:39:35 +00:00
|
|
|
vsmp_cap_cpus();
|
|
|
|
|
2008-03-20 07:41:16 +00:00
|
|
|
set_vsmp_pv_ops();
|
2008-02-11 19:16:02 +00:00
|
|
|
return;
|
2006-01-11 21:46:18 +00:00
|
|
|
}
|