Re: [Patch V4 00/42] use irqdomain to dynamically allocate IRQ for IOAPIC
On Sat, Jun 21, 2014 at 2:08 PM, Thomas Gleixner wrote:
> Jiang,
>
> On Mon, 9 Jun 2014, Jiang Liu wrote:
>
>> On x86 platforms, IRQ number are statically allocated to IOAPIC pins at boot.
>> There are two issues with this design. First it causes trouble to IOAPIC
>> hotplug because we need to allocate a block of IRQ numbers for each IOAPIC.
>> Second it may waste IRQ nubmers even if some IOAPIC pins are not used because
>> IRQ numbers are statically assigned.
>>
>> This patchset tries to enable dynamic IRQ number allocation for IOAPIC
>> by adopting the irqdomain framework, it solves the two issues mentioned
>> above. It also simplifies the IOAPIC driver by consolidating ways to
>> program IOAPIC pins with the irqdomain map interface.
>>
>> We will enhance the IOAPIC driver core to support ACPI based IOAPIC hotplug
>> once the IOAPIC driver has been converted to irqdomain.
>
> I merged the whole lot. Thanks for doing this and reacting on all the
> review comments. Very nice cleanup work!
>
I got following warning on RHEL 7.0 shutdown path.
[58018.128900] [ cut here ]
[58018.134081] WARNING: CPU: 21 PID: 1 at fs/proc/generic.c:521
remove_proc_entry+0x150/0x1a0()
[58018.143596] remove_proc_entry: removing non-empty directory
'irq/23', leaking at least 'ehci_hcd:usb2'
[58018.154001] Modules linked in:
[58018.157435] CPU: 21 PID: 1 Comm: systemd-shutdow Tainted: G
W 3.16.0-rc2-yh-00420-g6d62036-dirty #36
[58018.168716] Hardware name: Oracle Corporation ORACLE SERVER
[58018.179703] 0009 885e759bfb60 81f76520
885e759bfba8
[58018.188033] 885e759bfb98 8109d03d 88edf9fb9f00
0002
[58018.196358] 885e759bfc6e 88f07b3f4330 88edf9fb9fb9
885e759bfbf8
[58018.204682] Call Trace:
[58018.207432] [] dump_stack+0x45/0x56
[58018.213186] [] warn_slowpath_common+0x7d/0xa0
[58018.219903] [] warn_slowpath_fmt+0x4c/0x50
[58018.226336] [] remove_proc_entry+0x150/0x1a0
[58018.232963] [] unregister_irq_proc+0xbf/0xd0
[58018.239589] [] free_desc+0x31/0xa0
[58018.245240] [] irq_free_descs+0x6c/0x80
[58018.251375] [] irq_dispose_mapping+0x53/0x60
[58018.258011] [] mp_unmap_irq+0xa0/0xc0
[58018.263960] [] acpi_unregister_gsi_ioapic+0x2f/0x40
[58018.271263] [] acpi_unregister_gsi+0x17/0x20
[58018.277896] [] acpi_pci_irq_disable+0x5e/0x68
[58018.284619] [] pcibios_disable_device+0x20/0x30
[58018.291536] [] do_pci_disable_device+0x52/0x60
[58018.298350] [] pci_disable_device+0x92/0xb0
[58018.304876] [] usb_hcd_pci_shutdown+0x3f/0x50
[58018.311593] [] pci_device_shutdown+0x36/0x70
[58018.318220] [] device_shutdown+0x135/0x190
[58018.324651] [] kernel_power_off+0x35/0x80
[58018.330980] [] SyS_reboot+0x233/0x260
[58018.336919] [] ? cpuacct_account_field+0x99/0xb0
[58018.343929] [] ? cpuacct_account_field+0x5/0xb0
[58018.350843] [] ? _raw_spin_unlock+0x27/0x30
[58018.357378] [] ? trace_hardirqs_on_caller+0x16/0x260
[58018.364774] [] ? trace_hardirqs_on+0xd/0x10
[58018.371302] [] ? syscall_trace_enter+0x17c/0x1c0
[58018.378312] [] tracesys+0xd4/0xd9
[58018.383867] ---[ end trace fade991312a6a2d9 ]---
[58018.389026] free irq_desc for 23
Like to suggest attached patch that skip mp_unregister_gsi on shutdown path.
Thanks
Yinghai
---
arch/x86/kernel/acpi/boot.c |6 ++
1 file changed, 6 insertions(+)
Index: linux-2.6/arch/x86/kernel/acpi/boot.c
===
--- linux-2.6.orig/arch/x86/kernel/acpi/boot.c
+++ linux-2.6/arch/x86/kernel/acpi/boot.c
@@ -659,6 +659,12 @@ static int acpi_register_gsi_ioapic(stru
static void acpi_unregister_gsi_ioapic(u32 gsi)
{
#ifdef CONFIG_X86_IO_APIC
+ /* should not worry about shutdown path */
+ if (system_state == SYSTEM_HALT ||
+ system_state == SYSTEM_POWER_OFF ||
+ system_state == SYSTEM_RESTART)
+ return;
+
mp_unregister_gsi(gsi);
#endif
}
Re: [Patch V4 00/42] use irqdomain to dynamically allocate IRQ for IOAPIC
On Sun, Jun 22, 2014 at 10:39:46PM +0800, Feng Tang wrote: > Hi, > > On Sun, Jun 22, 2014 at 04:42:27PM +0800, Ingo Molnar wrote: > > > > JFYI, with these patches applied x86 allyes32config fails with: > > > > arch/x86/platform/intel-mid/device_libs/platform_wdt.c:45:3: error: > > implicit declaration of function ‘io_apic_set_pci_routing’ > > In some intel-mid platforms (Medfield/Merrifield etc), the IOAPIC > may have more than 80 pins, and some devices like a i2c based touch > screen has its interrupt line directly connected to IOAPIC (this > platform_wdt.c here should be the same case). > > Since these devices are not PCI device or ACPI device, they can't > use the general PCI/ACPI xxx_enable_dev APIs which implicitly set > up the ioapic entry, but use this "io_apic_set_pci_routing" > directly. Just quickly read Jiang's patch set, guess these driver should be ok by using Jiang's new cleaner API "mp_set_gsi_attr()" to replace these io_apic_set_pci_routing(). Thanks, Feng -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [Patch V4 00/42] use irqdomain to dynamically allocate IRQ for IOAPIC
Hi, On Sun, Jun 22, 2014 at 04:42:27PM +0800, Ingo Molnar wrote: > > JFYI, with these patches applied x86 allyes32config fails with: > > arch/x86/platform/intel-mid/device_libs/platform_wdt.c:45:3: error: > implicit declaration of function ‘io_apic_set_pci_routing’ In some intel-mid platforms (Medfield/Merrifield etc), the IOAPIC may have more than 80 pins, and some devices like a i2c based touch screen has its interrupt line directly connected to IOAPIC (this platform_wdt.c here should be the same case). Since these devices are not PCI device or ACPI device, they can't use the general PCI/ACPI xxx_enable_dev APIs which implicitly set up the ioapic entry, but use this "io_apic_set_pci_routing" directly. Thanks, Feng > > Thanks, > > Ingo > -- > To unsubscribe from this list: send the line "unsubscribe linux-acpi" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [Patch V4 00/42] use irqdomain to dynamically allocate IRQ for IOAPIC
JFYI, with these patches applied x86 allyes32config fails with: arch/x86/platform/intel-mid/device_libs/platform_wdt.c:45:3: error: implicit declaration of function ‘io_apic_set_pci_routing’ Thanks, Ingo -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [Patch V4 00/42] use irqdomain to dynamically allocate IRQ for IOAPIC
Jiang, On Mon, 9 Jun 2014, Jiang Liu wrote: > On x86 platforms, IRQ number are statically allocated to IOAPIC pins at boot. > There are two issues with this design. First it causes trouble to IOAPIC > hotplug because we need to allocate a block of IRQ numbers for each IOAPIC. > Second it may waste IRQ nubmers even if some IOAPIC pins are not used because > IRQ numbers are statically assigned. > > This patchset tries to enable dynamic IRQ number allocation for IOAPIC > by adopting the irqdomain framework, it solves the two issues mentioned > above. It also simplifies the IOAPIC driver by consolidating ways to > program IOAPIC pins with the irqdomain map interface. > > We will enhance the IOAPIC driver core to support ACPI based IOAPIC hotplug > once the IOAPIC driver has been converted to irqdomain. I merged the whole lot. Thanks for doing this and reacting on all the review comments. Very nice cleanup work! Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
