i386, x86_64: fix the irqbalance quirk for E7520/E7320/E7525
Subsystems (3)
cover i386, x86_64: fix the irqbalance quirk for E7520/E7320/E7525
Mechanism of selecting physical mode in genapic when cpu hotplug is enabled on x86_64, broke the quirk(quirk_intel_irqbalance()) introduced for working around the transposing interrupt message errata in E7520/E7320/E7525 (revision ID 0x9 and below. errata #23 in
http://download.intel.com/design/chipsets/specupdt/30304203.pdf).
This errata requires the mode to be in logical flat, so that interrupts can be directed to more than one cpu(and thus use hardware IRQ balancing enabled by BIOS on these platforms).
Following four patches fixes this by moving the quirk to early quirk and forcing the x86_64 genapic selection to logical flat on these platforms.
Thanks to Shaohua for pointing out the breakage.
PATCH 1/4 add write_pci_config_byte() to direct PCI access routines
Add write_pci_config_byte() to direct PCI access routines
---
arch/i386/pci/early.c +7 -0
@@ -45,6 +45,13 @@ void write_pci_config(u8 bus, u8 slot, uoutl (val,0xcfc ); } +void write_pci_config_byte (u8 bus,u8 slot,u8 func,u8 offset,u8 val) +{ +PDprintk ("%x writing to %x: %x\n" , slot, offset, val); +outl (0x80000000 | (bus<<16 ) | (slot<<11 ) | (func<<8 ) | offset,0xcf8 ); +outb (val,0xcfc ); +} +int early_pci_allowed (void ) { return (pci_probe & (PCI_PROBE_CONF1|PCI_PROBE_NOEARLY)) ==
include/asm-x86_64/pci-direct.h +1 -0
@@ -10,6 +10,7 @@ extern u32 read_pci_config(u8 bus, u8 slextern u8 read_pci_config_byte (u8 bus,u8 slot,u8 func,u8 offset);extern u16 read_pci_config_16 (u8 bus,u8 slot,u8 func,u8 offset);extern void write_pci_config (u8 bus,u8 slot,u8 func,u8 offset,u32 val); +extern void write_pci_config_byte (u8 bus,u8 slot,u8 func,u8 offset,u8 val);extern int early_pci_allowed (void );
Re: [patch 2/4] introduce the mechanism of disabling cpu hotplug control
Add 'cpu_hotplug_no_control' and when set, the hotplug control file("online") will not be added under /sys/devices/system/cpu/cpuX/
Next patch doing PCI quirks will use this.
---
arch/i386/kernel/topology.c +8 -0
@@ -40,14 +40,22 @@ int arch_register_cpu(int num)* restrictions and assumptions in kernel. This basically * doesnt add a control file, one cannot attempt to offline * BSP. +* +* Also certain PCI quirks require to remove this control file +* for all CPU's. */ +#ifdef CONFIG_HOTPLUG_CPU +if (!num || cpu_hotplug_no_control) +#else if (!num) +#endif cpu_devices[num].cpu.no_control =1 ;return register_cpu (&cpu_devices[num].cpu, num); }#ifdef CONFIG_HOTPLUG_CPU +int cpu_hotplug_no_control;void arch_unregister_cpu (int num) {return unregister_cpu (&cpu_devices[num].cpu);
include/asm-i386/cpu.h +1 -0
@@ -13,6 +13,7 @@ struct i386_cpu {extern int arch_register_cpu (int num);#ifdef CONFIG_HOTPLUG_CPU extern void arch_unregister_cpu (int ); +extern int cpu_hotplug_no_control;#endif DECLARE_PER_CPU (int, cpu_state);
PATCH 3/4 x86_64: add genapic_force
Add genapic_force. Used by the next Intel quirks patch.
---
arch/x86_64/kernel/genapic.c +9 -1
@@ -33,7 +33,7 @@ extern struct genapic apic_flat;extern struct genapic apic_physflat;struct genapic *genapic = &apic_flat; - +struct genapic *genapic_force;/* * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode.
@@ -46,6 +46,14 @@ void __init clustered_apic_check(void)u8 cluster_cnt[NUM_APIC_CLUSTERS ];int max_apic =0 ; +/* genapic selection can be forced because of certain quirks and +* in future perhaps user can select this through cmdline +*/ +if (genapic_force) { + genapic = genapic_force; + goto print; + } +#if defined(CONFIG_ACPI )/* * Some x86_64 machines use physical APIC mode regardless of how many
include/asm-x86_64/genapic.h +1 -1
@@ -30,6 +30,6 @@ struct genapic { }; -extern struct genapic *genapic; +extern struct genapic *genapic, *genapic_force, apic_flat;#endif
PATCH 4/4 fix the irqbalance quirk for E7320/E7520/E7525
Move the irqbalance quirks for E7320/E7520/E7525(Errata 23 in
http://download.intel.com/design/chipsets/specupdt/30304203.pdf) to early quirks.
And add a PCI quirk for these platforms to check(which happens very late during the boot) if the APIC routing is indeed set to default flat mode.
This fixes the breakage(in x86_64) of this quirk due to cpu hotplug which selects physical mode instead of the logical flat(as needed for this errata workaround).
---
arch/i386/kernel/acpi/earlyquirk.c +21 -0
@@ -10,6 +10,7 @@ #include <linux/acpi.h>#include <asm/pci-direct.h> #include <asm/acpi.h> #include <asm/apic.h> +#include <asm/irq.h> #ifdef CONFIG_ACPI
@@ -43,6 +44,24 @@ #endifreturn 0 ; } +static void check_intel (void ) +{ +u16 vendor, device; + + vendor =read_pci_config_16 (0 ,0 ,0 ,PCI_VENDOR_ID ); + +if (vendor !=PCI_VENDOR_ID_INTEL ) +return ; + + device =read_pci_config_16 (0 ,0 ,0 ,PCI_DEVICE_ID ); +#ifdef CONFIG_SMP +if (device ==PCI_DEVICE_ID_INTEL_E7320_MCH || + device ==PCI_DEVICE_ID_INTEL_E7520_MCH || + device ==PCI_DEVICE_ID_INTEL_E7525_MCH ) +quirk_intel_irqbalance (); +#endif +} +void __init check_acpi_pci (void ) {int num, slot, func;
@@ -54,6 +73,8 @@ void __init check_acpi_pci(void)if (!early_pci_allowed ())return ; +check_intel (); +/* Poor man's PCI discovery */ for (num =0 ; num <32 ; num++) {for (slot =0 ; slot <32 ; slot++) {
arch/i386/kernel/quirks.c +40 -11
81 lines not on this page · raw message
arch/i386/kernel/smpboot.c +8 -0
@@ -52,6 +52,7 @@ #include <asm/tlbflush.h>#include <asm/desc.h> #include <asm/arch_hooks.h> #include <asm/nmi.h> +#include <asm/genapic.h> #include <mach_apic.h> #include <mach_wakecpu.h>
@@ -1461,6 +1462,13 @@ #endifcpu_set (cpu, smp_commenced_mask);while (!cpu_isset (cpu, cpu_online_map))cpu_relax (); + +if (num_online_cpus () >8 && genapic == &apic_default) { +printk (KERN_WARNING +"Default flat APIC routing can't be used with > 8 cpus\n" ); +BUG (); + } +return 0 ; }
arch/x86_64/kernel/early-quirks.c +13 -0
@@ -68,6 +68,18 @@ static void ati_bugs(void) } } +static void intel_bugs (void ) +{ +u16 device =read_pci_config_16 (0 ,0 ,0 ,PCI_DEVICE_ID ); + +#ifdef CONFIG_SMP +if (device ==PCI_DEVICE_ID_INTEL_E7320_MCH || + device ==PCI_DEVICE_ID_INTEL_E7520_MCH || + device ==PCI_DEVICE_ID_INTEL_E7525_MCH ) +quirk_intel_irqbalance (); +#endif +} +struct chipset {u16 vendor;void (*f)(void );
@@ -77,6 +89,7 @@ static struct chipset early_qrk[] = { { PCI_VENDOR_ID_NVIDIA, nvidia_bugs }, { PCI_VENDOR_ID_VIA, via_bugs }, { PCI_VENDOR_ID_ATI, ati_bugs }, + { PCI_VENDOR_ID_INTEL, intel_bugs}, {} };
arch/x86_64/kernel/smpboot.c +8 -0
@@ -60,6 +60,7 @@ #include <asm/nmi.h>#include <asm/irq.h> #include <asm/hw_irq.h> #include <asm/numa.h> +#include <asm/genapic.h> /* Number of siblings per CPU package */ int smp_num_siblings =1 ;
@@ -1167,6 +1168,13 @@ int __cpuinit __cpu_up(unsigned int cpu)while (!cpu_isset (cpu, cpu_online_map))cpu_relax (); + +if (num_online_cpus () >8 && genapic == &apic_flat) { +printk (KERN_WARNING +"flat APIC routing can't be used with > 8 cpus\n" ); +BUG (); + } + err =0 ;return err;
include/asm-i386/genapic.h +1 -1
@@ -122,6 +122,6 @@ #define APIC_INIT(aname, aprobe) { \APICFUNC (phys_pkg_id) \ } -extern struct genapic *genapic; +extern struct genapic *genapic, apic_default;#endif
include/asm-i386/irq.h +2 -0
@@ -37,6 +37,8 @@ #ifdef CONFIG_IRQBALANCEextern int irqbalance_disable (char *str);#endif +extern void quirk_intel_irqbalance (void ); +#ifdef CONFIG_HOTPLUG_CPU extern void fixup_irqs (cpumask_t map);#endif
include/asm-x86_64/proto.h +1 -0
@@ -88,6 +88,7 @@ extern void syscall32_cpu_init(void);extern void setup_node_bootmem (int nodeid,unsigned long start,unsigned long end);extern void early_quirks (void ); +extern void quirk_intel_irqbalance (void );extern void check_efer (void );extern int unhandled_signal (struct task_struct *tsk,int sig);
Re: [patch 2/4] introduce the mechanism of disabling cpu hotplug control
In reply to Siddha, Suresh BQuoted 5 lines · Suresh B
Add '
cpu_hotplug_no_control' and when set, the hotplug controlfile("online")will not be added under/sys/devices/system/cpu/cpuX/Next patch doing PCI quirks will use this.
I don't understand what this (ugly) patch has to do with the overall bugfix. We're fixing the APCI initialisation - what does that have to do with presenting cpu-hotplug files in sysfs?
Quoted 9 lines
---
arch/i386/kernel/topology.c +0
-0@@ -40,14 +40,22 @@ int arch_register_cpu(int num)* restrictions and assumptions in kernel. This basically * doesnt add a control file, one cannot attempt to offline
arch/i386/kernel/topology.c int arch_register_cpu(int num) · line 51
Quoted 10 lines
arch/i386/kernel/topology.c +7
-0* BSP. +* +* Also certain PCI quirks require to remove this control file +* for all CPU's. */ +#ifdef CONFIG_HOTPLUG_CPU +if (!num || cpu_hotplug_no_control) +#else if (!num) +#endif
This ifdef could be removed
Quoted 15 lines
arch/i386/kernel/topology.c +1
-0cpu_devices[num].cpu.no_control =1 ;return register_cpu (&cpu_devices[num].cpu, num); }#ifdef CONFIG_HOTPLUG_CPU +int cpu_hotplug_no_control;void arch_unregister_cpu (int num) {return unregister_cpu (&cpu_devices[num].cpu);include/asm-i386/cpu.h +0
-0
include/asm-i386/cpu.h struct i386_cpu { · line 16
Quoted 4 lines
include/asm-i386/cpu.h +1
-0@@ -13,6 +13,7 @@ struct i386_cpu {extern int arch_register_cpu (int num);#ifdef CONFIG_HOTPLUG_CPU extern void arch_unregister_cpu (int ); +extern int cpu_hotplug_no_control;
via:
#else
#define cpu_hotplug_no_control 1
here.
But does this variable _have_ to be a negative like this? The code would be simpler if it had the opposite sense and was called, say, cpu_hotplug_enable_control_file.
Are these patches considered 2.6.19 material? They look a bit big, ugly and scary for that.
Quoted 6 lines · Suresh B
Mechanism of selecting physical mode in genapic when cpu hotplug is enabled on
x86_64, broke the quirk(quirk_intel_irqbalance()) introduced for working around the transposing interrupt message errata in E7520/E7320/E7525 (revision ID0x9and below. errata #23 in
http://download.intel.com/design/chipsets/specupdt/30304203.pdf).
Quoted 8 lines
This errata requires the mode to be in logical flat, so that interrupts can be directed to more than one cpu(and thus use hardware IRQ balancing enabled by BIOS on these platforms).
Following four patches fixes this by moving the quirk to early quirk and forcing the
x86_64genapic selection to logical flat on these platforms.Thanks to Shaohua for pointing out the breakage.
It blew up with the first config I tried
(http://userweb.kernel.org/~akpm/config-vmm.txt):
arch/i386/kernel/built-in.o: In function `verify_quirk_intel_irqbalance':arch/i386/kernel/quirks.c:19: undefined reference to genapic' arch/i386/kernel/quirks.c:19: undefined reference to apic_default' arch/i386/kernel/built-in.o: In function `__cpu_up':arch/i386/kernel/smpboot.c:1487: undefined reference to genapic' arch/i386/kernel/smpboot.c:1487: undefined reference to apic_default'
The dependencies in the code which you're touching here are really really complex and fragile. One needs to review the change very carefully and test it exhaustively.
Re: [patch 2/4] introduce the mechanism of disabling cpu hotplug control
In reply to Andrew MortonQuoted 6 lines · Andrew Morton
Quoted 3 lines · Suresh B
Add '
cpu_hotplug_no_control' and when set, the hotplug controlfile("online")will not be added under/sys/devices/system/cpu/cpuX/
Quoted 6 lines
Quoted 2 lines
Next patch doing PCI quirks will use this.
I don't understand what this (ugly) patch has to do with the overall bugfix. We're fixing the APCI initialisation - what does that have to do with presenting cpu-hotplug files in sysfs?
This patch is adding a mechanism which will not export the cpu hotplug control file. And the quirk will use this from preventing the users doing cpu hotplug. On these platforms, we need to have atleast 2 cpus online to workaround the errata.
In future we can use this mechanism to disable cpu hotplug during bootup.
Quoted 3 lines
But does this variable _have_ to be a negative like this? The code would be simpler if it had the opposite sense and was called, say,
cpu_hotplug_enable_control_file.
I wanted to add something like disable_cpu_hotplug.. but suspend code seems to be already using.. Will clean this up.
Quoted 2 lines
Are these patches considered 2.6.19 material? They look a bit big, ugly and scary for that.
Though no one reported failures so far, it would be good to get included in 2.6.19. If it is too late and sounds scary, we can consider in -stable after spending sometime in -mm..
thanks,
suresh
Quoted 10 lines · Andrew Morton
arch/i386/kernel/built-in.o: In function `verify_quirk_intel_irqbalance':arch/i386/kernel/quirks.c:19: undefined reference togenapic' arch/i386/kernel/quirks.c:19: undefined reference toapic_default'arch/i386/kernel/built-in.o: In function `__cpu_up':arch/i386/kernel/smpboot.c:1487: undefined reference togenapic' arch/i386/kernel/smpboot.c:1487: undefined reference toapic_default'The dependencies in the code which you're touching here are really really complex and fragile. One needs to review the change very carefully and test it exhaustively.
oops. I did compile and boot with 6 x86/x86_64 configs.. Seems something is broken in my git test tree. Let me fix and getback to you.
thanks,
suresh
Re: [patch 2/4] introduce the mechanism of disabling cpu hotplug control
In reply to Siddha, Suresh BQuoted 1 line · Suresh B
I wanted to add something like
disable_cpu_hotplug
My point is, `enable_cpu_hotplug' is nicer
if (enable_cpu_hotplug) ... if (!enable_cpu_hotplug) ...
versus
if (disable_cpu_hotplug) ... if (!disable_cpu_hotplug) ... /* My brain hurts */
Re: [patch 2/4] introduce the mechanism of disabling cpu hotplug control
Quoted 6 lines · Andrew Morton
Quoted 1 line · Suresh B
I wanted to add something like
disable_cpu_hotplugMy point is, `enable_cpu_hotplug' is nicer
Yep. I got it and hence my "will clean this up" assurance :)
This is all coming from the no_control' member in cpu structure and I will change that to something like hotpluggable'. That will make the patch slightly big but def clean.
thanks,
suresh
PATCH 1/1 change the 'no_control' field to 'hotpluggable' in the struct cpu
Quoted 5 lines · Suresh B
Quoted 4 lines · Andrew Morton
Quoted 1 line · Suresh B
I wanted to add something like
disable_cpu_hotplug
Quoted 7 lines
Quoted 1 line
My point is, `enable_cpu_hotplug' is nicer
Yep. I got it and hence my "will clean this up" assurance :)
This is all coming from the
no_control' member in cpu structure and I will change that to something likehotpluggable'. That will make the patch slightly big but def clean.
Andrew, Appended the cleanup patch which goes on top of your -mm tree. Please apply. Thanks.
---
Change the 'no_control' field in the cpu struct to a more positive and better term 'hotpluggable'. And change(/cleanup) the logic accordingly.
---
arch/i386/kernel/topology.c +2 -2
@@ -44,8 +44,8 @@ int arch_register_cpu(int num)* Also certain PCI quirks require not to enable hotplug control * for all CPU's. */ -if (!num || !enable_cpu_hotplug) - cpu_devices[num].cpu.no_control =1 ; +if (num && enable_cpu_hotplug) + cpu_devices[num].cpu.hotpluggable =1 ;return register_cpu (&cpu_devices[num].cpu, num); }
arch/ia64/kernel/topology.c +4 -4
@@ -31,11 +31,11 @@ int arch_register_cpu(int num) {#if defined (CONFIG_ACPI ) && defined (CONFIG_HOTPLUG_CPU )/* -* If CPEI cannot be re-targetted, and this is -* CPEI target, then dont create the control file +* If CPEI can be re-targetted or if this is not +* CPEI target, then it is hotpluggable */ -if (!can_cpei_retarget () &&is_cpu_cpei_target (num)) - sysfs_cpus[num].cpu.no_control =1 ; +if (can_cpei_retarget () || !is_cpu_cpei_target (num)) + sysfs_cpus[num].cpu.hotpluggable =1 ;map_cpu_to_node (num, node_cpuid[num].nid);#endif
arch/powerpc/kernel/sysfs.c +4 -4
@@ -240,7 +240,7 @@ static void unregister_cpu_online(unsignstruct cpu *c = &per_cpu (cpu_devices, cpu);struct sys_device *s = &c->sysdev; -BUG_ON (c->no_control); +BUG_ON (!c->hotpluggable);#ifndef CONFIG_PPC_ISERIES if (cpu_has_feature (CPU_FTR_SMT ))
@@ -360,10 +360,10 @@ static int __init topology_init(void)* CPU. For instance, the boot cpu might never be valid * for hotplugging. */ -if (!ppc_md.cpu_die) - c->no_control =1 ; +if (ppc_md.cpu_die) + c->hotpluggable =1 ; -if (cpu_online (cpu) || (c->no_control ==0 )) { +if (cpu_online (cpu) || c->hotpluggable) {register_cpu (c, cpu);sysdev_create_file (&c->sysdev, &attr_physical_id);
drivers/base/cpu.c +3 -3
@@ -103,8 +103,8 @@ #endif/* * register_cpu - Setup a driverfs device for a CPU. -* @cpu - Callers can set the cpu->no_control field to 1, to indicate not to -* generate a control file in sysfs for this CPU. +* @cpu - cpu->hotpluggable field set to 1 will generate a control file in +* sysfs for this CPU. * @num - CPU number to use when creating the device. * * Initialize and register the CPU device.
@@ -118,7 +118,7 @@ int __devinit register_cpu(struct cpu *c error =sysdev_register (&cpu->sysdev); -if (!error && !cpu->no_control) +if (!error && cpu->hotpluggable)register_cpu_control (cpu);if (!error) cpu_sys_devices[num] = &cpu->sysdev;
include/linux/cpu.h +1 -1
@@ -27,7 +27,7 @@ #include <asm/semaphore.h>struct cpu {int node_id;/* The node which contains the CPU */ -int no_control;/* Should the sysfs control file be created? */ +int hotpluggable;/* creates sysfs control file if hotpluggable */ struct sys_device sysdev; };