| Message ID | 20100419045741.30276.23233.stgit@ppwaskie-hc2.jf.intel.com |
|---|---|
| State | RFC, archived |
| Delegated to: | David Miller |
| Headers | show |
From: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> Date: Sun, 18 Apr 2010 21:57:41 -0700 > This patch adds a callback function pointer to the irq_desc > structure, along with a registration function and a read-only > proc entry for each interrupt. > > This affinity_hint handle for each interrupt can be used by > underlying drivers that need a better mechanism to control > interrupt affinity. The underlying driver can register a > callback for the interrupt, which will allow the driver to > provide the CPU mask for the interrupt to anything that > requests it. The intent is to extend the userspace daemon, > irqbalance, to help hint to it a preferred CPU mask to balance > the interrupt into. > > Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> I'll leave it to the IRQ layer experts whether this is appropriate or not, it doesn't look too bad to me. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 18 Apr 2010, Peter P Waskiewicz Jr wrote: > +/** > + * struct irqaffinityhint - per interrupt affinity helper > + * @callback: device driver callback function > + * @dev: reference for the affected device > + * @irq: interrupt number > + */ > +struct irqaffinityhint { > + irq_affinity_hint_t callback; > + void *dev; > + int irq; > +}; Why do you need that extra data structure ? The device and the irq number are known, so all you need is the callback itself. So no need for allocating memory .... > +static ssize_t irq_affinity_hint_proc_write(struct file *file, > + const char __user *buffer, size_t count, loff_t *pos) > +{ > + /* affinity_hint is read-only from proc */ > + return -EOPNOTSUPP; > +} > + Why do you want a write function when the file is read only ? Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, 27 Apr 2010, Thomas Gleixner wrote: > On Sun, 18 Apr 2010, Peter P Waskiewicz Jr wrote: >> +/** >> + * struct irqaffinityhint - per interrupt affinity helper >> + * @callback: device driver callback function >> + * @dev: reference for the affected device >> + * @irq: interrupt number >> + */ >> +struct irqaffinityhint { >> + irq_affinity_hint_t callback; >> + void *dev; >> + int irq; >> +}; > > Why do you need that extra data structure ? The device and the irq > number are known, so all you need is the callback itself. So no need > for allocating memory .... When I register the function callback with the interrupt layer, I need to know what device structures to reference back in the driver. In other words, if I call into an underlying driver with just an interrupt number, then I have no way at getting at the dev structures (netdevice for me, plus my private adapter structures), unless I declare them globally (yuck). I had a different approach before this one where I assumed the device from the irq handler callback was safe to use for the device in this new callback. I didn't feel really great about that, since it's an implicit assumption that could cause things to go sideways really quickly. Let me know what you think either way. I'm certainly willing to make a change, I just don't know at this point what's the safest approach from what I currently have. > >> +static ssize_t irq_affinity_hint_proc_write(struct file *file, >> + const char __user *buffer, size_t count, loff_t *pos) >> +{ >> + /* affinity_hint is read-only from proc */ >> + return -EOPNOTSUPP; >> +} >> + > > Why do you want a write function when the file is read only ? It's leftover paranoia. I put it in early on, then changed the mode later. I can remove this function. I'll re-send something once we agree on how the code in your first comment should look. Thanks Thomas! -PJ -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
B1;2005;0cPeter, On Tue, 27 Apr 2010, Peter P Waskiewicz Jr wrote: > On Tue, 27 Apr 2010, Thomas Gleixner wrote: > > On Sun, 18 Apr 2010, Peter P Waskiewicz Jr wrote: > > > +/** > > > + * struct irqaffinityhint - per interrupt affinity helper > > > + * @callback: device driver callback function > > > + * @dev: reference for the affected device > > > + * @irq: interrupt number > > > + */ > > > +struct irqaffinityhint { > > > + irq_affinity_hint_t callback; > > > + void *dev; > > > + int irq; > > > +}; > > > > Why do you need that extra data structure ? The device and the irq > > number are known, so all you need is the callback itself. So no need > > for allocating memory .... > > When I register the function callback with the interrupt layer, I need to > know what device structures to reference back in the driver. In other words, > if I call into an underlying driver with just an interrupt number, then I > have no way at getting at the dev structures (netdevice for me, plus my > private adapter structures), unless I declare them globally (yuck). Grr, I knew that I missed something. That'll teach me to review patches before the coffee has reached my brain cells :) > I had a different approach before this one where I assumed the device from > the irq handler callback was safe to use for the device in this new callback. > I didn't feel really great about that, since it's an implicit assumption that > could cause things to go sideways really quickly. > > Let me know what you think either way. I'm certainly willing to make a > change, I just don't know at this point what's the safest approach from what > I currently have. So you need a reference to your device, so what about the following: struct irq_affinity_hint; struct irq_affinity_hint { unsigned int (*callback)(unsigned int irq, struct irq_affinity_hint *hint, cpumask_var_t *mask); } Now you embed that struct into your device private data structure and you get the reference to it back in the callback function. No extra kmalloc/kfree, less code. One other thing I noticed, but forgot to comment on: > +static int irq_affinity_hint_proc_show(struct seq_file *m, void *v) > +{ > + struct irq_desc *desc = irq_to_desc((long)m->private); > + struct cpumask mask; > + unsigned int ret = 0; Why do we return 0, when there is no callback and no hint available ? > + We don't want to have cpumask enforced on stack. Please make that: cpumask_var_t mask; if (!alloc_cpumask_var(&mask, GFP_KERNEL)) return -ENOMEM; > + if (desc->hint && desc->hint->callback) { The access to desc-> needs to be protected with desc->lock. Otherwise you might race with a callback unregister. > + ret = desc->hint->callback(&mask, (long)m->private, > + desc->hint->dev); > + if (!ret) > + seq_cpumask(m, &mask); > + } > + > + seq_putc(m, '\n'); > + return ret; > +} Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, 2010-04-28 at 09:45 -0700, Thomas Gleixner wrote: > B1;2005;0cPeter, > > On Tue, 27 Apr 2010, Peter P Waskiewicz Jr wrote: > > On Tue, 27 Apr 2010, Thomas Gleixner wrote: > > > On Sun, 18 Apr 2010, Peter P Waskiewicz Jr wrote: > > > > +/** > > > > + * struct irqaffinityhint - per interrupt affinity helper > > > > + * @callback: device driver callback function > > > > + * @dev: reference for the affected device > > > > + * @irq: interrupt number > > > > + */ > > > > +struct irqaffinityhint { > > > > + irq_affinity_hint_t callback; > > > > + void *dev; > > > > + int irq; > > > > +}; > > > > > > Why do you need that extra data structure ? The device and the irq > > > number are known, so all you need is the callback itself. So no need > > > for allocating memory .... > > > > When I register the function callback with the interrupt layer, I need to > > know what device structures to reference back in the driver. In other words, > > if I call into an underlying driver with just an interrupt number, then I > > have no way at getting at the dev structures (netdevice for me, plus my > > private adapter structures), unless I declare them globally (yuck). > > Grr, I knew that I missed something. That'll teach me to review > patches before the coffee has reached my brain cells :) > > > I had a different approach before this one where I assumed the device from > > the irq handler callback was safe to use for the device in this new callback. > > I didn't feel really great about that, since it's an implicit assumption that > > could cause things to go sideways really quickly. > > > > Let me know what you think either way. I'm certainly willing to make a > > change, I just don't know at this point what's the safest approach from what > > I currently have. > > So you need a reference to your device, so what about the following: > > struct irq_affinity_hint; > > struct irq_affinity_hint { > unsigned int (*callback)(unsigned int irq, struct irq_affinity_hint *hint, > cpumask_var_t *mask); > } > > Now you embed that struct into your device private data structure and > you get the reference to it back in the callback function. No extra > kmalloc/kfree, less code. Good idea! I'll roll that into my new version. > One other thing I noticed, but forgot to comment on: > > > +static int irq_affinity_hint_proc_show(struct seq_file *m, void *v) > > +{ > > + struct irq_desc *desc = irq_to_desc((long)m->private); > > + struct cpumask mask; > > + unsigned int ret = 0; > > Why do we return 0, when there is no callback and no hint available ? I initialized it to 0 to remove a compiler warning; I can put more thought into it and assign a more appropriate return value. > > + > > We don't want to have cpumask enforced on stack. Please make that: > > cpumask_var_t mask; > > if (!alloc_cpumask_var(&mask, GFP_KERNEL)) > return -ENOMEM; I'll roll this into my next version. > > + if (desc->hint && desc->hint->callback) { > > The access to desc-> needs to be protected with > desc->lock. Otherwise you might race with a callback unregister. Good point. I'll fix this. > > + ret = desc->hint->callback(&mask, (long)m->private, > > + desc->hint->dev); > > + if (!ret) > > + seq_cpumask(m, &mask); > > + } > > + > > + seq_putc(m, '\n'); > > + return ret; > > +} > > Thanks, > Thanks for the feedback. I'll have the updated patches for review soon. -PJ -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
B1;2005;0cPeter, On Thu, 29 Apr 2010, Peter P Waskiewicz Jr wrote: > On Wed, 2010-04-28 at 09:45 -0700, Thomas Gleixner wrote: > > So you need a reference to your device, so what about the following: > > > > struct irq_affinity_hint; > > > > struct irq_affinity_hint { > > unsigned int (*callback)(unsigned int irq, struct irq_affinity_hint *hint, > > cpumask_var_t *mask); > > } > > > > Now you embed that struct into your device private data structure and > > you get the reference to it back in the callback function. No extra > > kmalloc/kfree, less code. > > Good idea! I'll roll that into my new version. Thinking more about it, I wonder whether you have a cpu_mask in your driver/device private data anyway. I bet you have :) So it should be sufficient to set a pointer to that cpu_mask in irq_desc and get rid of the callback completely. Any access to desc->affinity_hint needs to be protected by desc->lock. For setting the pointer to a real mask resp. NULL that's fine. The copy which you need to do in the proc-read function is not going to introduce huge latencies either. Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, 2010-04-29 at 12:48 -0700, Thomas Gleixner wrote: > B1;2005;0cPeter, > > On Thu, 29 Apr 2010, Peter P Waskiewicz Jr wrote: > > On Wed, 2010-04-28 at 09:45 -0700, Thomas Gleixner wrote: > > > So you need a reference to your device, so what about the following: > > > > > > struct irq_affinity_hint; > > > > > > struct irq_affinity_hint { > > > unsigned int (*callback)(unsigned int irq, struct irq_affinity_hint *hint, > > > cpumask_var_t *mask); > > > } > > > > > > Now you embed that struct into your device private data structure and > > > you get the reference to it back in the callback function. No extra > > > kmalloc/kfree, less code. > > > > Good idea! I'll roll that into my new version. > > Thinking more about it, I wonder whether you have a cpu_mask in your > driver/device private data anyway. I bet you have :) Well, at this point we don't, but nothing says we can't. > So it should be sufficient to set a pointer to that cpu_mask in > irq_desc and get rid of the callback completely. So "register" would just assign the pointer, and "unregister" would make sure to NULL the mask pointer out. I like it. It'll sure clean things up too. > Any access to desc->affinity_hint needs to be protected by desc->lock. > For setting the pointer to a real mask resp. NULL that's fine. The > copy which you need to do in the proc-read function is not going to > introduce huge latencies either. Right. > Thanks, > > tglx Thanks for the additional inputs. Patches coming soon. -PJ -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, 29 Apr 2010, Peter P Waskiewicz Jr wrote: > On Thu, 2010-04-29 at 12:48 -0700, Thomas Gleixner wrote: > > Thinking more about it, I wonder whether you have a cpu_mask in your > > driver/device private data anyway. I bet you have :) > > Well, at this point we don't, but nothing says we can't. Somewhere you need to store that information in your driver, right ? > > So it should be sufficient to set a pointer to that cpu_mask in > > irq_desc and get rid of the callback completely. > > So "register" would just assign the pointer, and "unregister" would make > sure to NULL the mask pointer out. I like it. It'll sure clean things > up too. Yep, that'd be like the set_irq_chip() function. Just assign the pointer under desc->lock. Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, 2010-04-29 at 13:39 -0700, Thomas Gleixner wrote: > On Thu, 29 Apr 2010, Peter P Waskiewicz Jr wrote: > > On Thu, 2010-04-29 at 12:48 -0700, Thomas Gleixner wrote: > > > Thinking more about it, I wonder whether you have a cpu_mask in your > > > driver/device private data anyway. I bet you have :) > > > > Well, at this point we don't, but nothing says we can't. > > Somewhere you need to store that information in your driver, right ? Yes. But right now, storing a cpu_mask for an interrupt wouldn't buy us anything since we have no mechanism to make use of it today. :-) I'll be putting the cpu_mask entry in our q_vector structure, which is our abstraction of the MSI-X vector (it's where I have the hint struct right now in patch 2/2 for the ixgbe driver). It's a simple place to stick it. > > > So it should be sufficient to set a pointer to that cpu_mask in > > > irq_desc and get rid of the callback completely. > > > > So "register" would just assign the pointer, and "unregister" would make > > sure to NULL the mask pointer out. I like it. It'll sure clean things > > up too. > > Yep, that'd be like the set_irq_chip() function. Just assign the > pointer under desc->lock. > > Thanks, > > tglx -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h index 75f3f00..f2a7d0b 100644 --- a/include/linux/interrupt.h +++ b/include/linux/interrupt.h @@ -78,6 +78,8 @@ enum { }; typedef irqreturn_t (*irq_handler_t)(int, void *); +typedef unsigned int (*irq_affinity_hint_t)(cpumask_var_t, unsigned int, + void *); /** * struct irqaction - per interrupt action descriptor @@ -105,6 +107,18 @@ struct irqaction { unsigned long thread_flags; }; +/** + * struct irqaffinityhint - per interrupt affinity helper + * @callback: device driver callback function + * @dev: reference for the affected device + * @irq: interrupt number + */ +struct irqaffinityhint { + irq_affinity_hint_t callback; + void *dev; + int irq; +}; + extern irqreturn_t no_action(int cpl, void *dev_id); #ifdef CONFIG_GENERIC_HARDIRQS @@ -209,6 +223,9 @@ extern int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask); extern int irq_can_set_affinity(unsigned int irq); extern int irq_select_affinity(unsigned int irq); +extern int irq_register_affinity_hint(unsigned int irq, void *dev, + irq_affinity_hint_t callback); +extern int irq_unregister_affinity_hint(unsigned int irq); #else /* CONFIG_SMP */ static inline int irq_set_affinity(unsigned int irq, const struct cpumask *m) @@ -223,6 +240,16 @@ static inline int irq_can_set_affinity(unsigned int irq) static inline int irq_select_affinity(unsigned int irq) { return 0; } +static inline int irq_register_affinity_hint(unsigned int irq, void *dev, + irq_affinity_hint_t callback) +{ + return -EINVAL; +} + +static inline int irq_unregister_affinity_hint(unsigned int irq); +{ + return -EINVAL; +} #endif /* CONFIG_SMP && CONFIG_GENERIC_HARDIRQS */ #ifdef CONFIG_GENERIC_HARDIRQS diff --git a/include/linux/irq.h b/include/linux/irq.h index 707ab12..bd73e9b 100644 --- a/include/linux/irq.h +++ b/include/linux/irq.h @@ -206,6 +206,7 @@ struct irq_desc { struct proc_dir_entry *dir; #endif const char *name; + struct irqaffinityhint *hint; } ____cacheline_internodealigned_in_smp; extern void arch_init_copy_chip_data(struct irq_desc *old_desc, diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c index 704e488..3674b6a 100644 --- a/kernel/irq/manage.c +++ b/kernel/irq/manage.c @@ -138,6 +138,42 @@ int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask) return 0; } +int irq_register_affinity_hint(unsigned int irq, void *dev, + irq_affinity_hint_t callback) +{ + struct irq_desc *desc = irq_to_desc(irq); + unsigned long flags; + + raw_spin_lock_irqsave(&desc->lock, flags); + + if (!desc->hint) + desc->hint = kmalloc(sizeof(struct irqaffinityhint), + GFP_KERNEL); + desc->hint->callback = callback; + desc->hint->dev = dev; + desc->hint->irq = irq; + + raw_spin_unlock_irqrestore(&desc->lock, flags); + + return 0; +} +EXPORT_SYMBOL(irq_register_affinity_hint); + +int irq_unregister_affinity_hint(unsigned int irq) +{ + struct irq_desc *desc = irq_to_desc(irq); + unsigned long flags; + + raw_spin_lock_irqsave(&desc->lock, flags); + + kfree(desc->hint); + desc->hint = NULL; + + raw_spin_unlock_irqrestore(&desc->lock, flags); + + return 0; +} +EXPORT_SYMBOL(irq_unregister_affinity_hint); #ifndef CONFIG_AUTO_IRQ_AFFINITY /* * Generic version of the affinity autoselector. @@ -916,6 +952,9 @@ static struct irqaction *__free_irq(unsigned int irq, void *dev_id) desc->chip->disable(irq); } + /* make sure affinity_hint callback is cleaned up */ + kfree(desc->hint); + raw_spin_unlock_irqrestore(&desc->lock, flags); unregister_handler_proc(irq, action); diff --git a/kernel/irq/proc.c b/kernel/irq/proc.c index 7a6eb04..59110a3 100644 --- a/kernel/irq/proc.c +++ b/kernel/irq/proc.c @@ -32,6 +32,23 @@ static int irq_affinity_proc_show(struct seq_file *m, void *v) return 0; } +static int irq_affinity_hint_proc_show(struct seq_file *m, void *v) +{ + struct irq_desc *desc = irq_to_desc((long)m->private); + struct cpumask mask; + unsigned int ret = 0; + + if (desc->hint && desc->hint->callback) { + ret = desc->hint->callback(&mask, (long)m->private, + desc->hint->dev); + if (!ret) + seq_cpumask(m, &mask); + } + + seq_putc(m, '\n'); + return ret; +} + #ifndef is_affinity_mask_valid #define is_affinity_mask_valid(val) 1 #endif @@ -79,11 +96,23 @@ free_cpumask: return err; } +static ssize_t irq_affinity_hint_proc_write(struct file *file, + const char __user *buffer, size_t count, loff_t *pos) +{ + /* affinity_hint is read-only from proc */ + return -EOPNOTSUPP; +} + static int irq_affinity_proc_open(struct inode *inode, struct file *file) { return single_open(file, irq_affinity_proc_show, PDE(inode)->data); } +static int irq_affinity_hint_proc_open(struct inode *inode, struct file *file) +{ + return single_open(file, irq_affinity_hint_proc_show, PDE(inode)->data); +} + static const struct file_operations irq_affinity_proc_fops = { .open = irq_affinity_proc_open, .read = seq_read, @@ -92,6 +121,14 @@ static const struct file_operations irq_affinity_proc_fops = { .write = irq_affinity_proc_write, }; +static const struct file_operations irq_affinity_hint_proc_fops = { + .open = irq_affinity_hint_proc_open, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, + .write = irq_affinity_hint_proc_write, +}; + static int default_affinity_show(struct seq_file *m, void *v) { seq_cpumask(m, irq_default_affinity); @@ -231,6 +268,10 @@ void register_irq_proc(unsigned int irq, struct irq_desc *desc) /* create /proc/irq/<irq>/smp_affinity */ proc_create_data("smp_affinity", 0600, desc->dir, &irq_affinity_proc_fops, (void *)(long)irq); + + /* create /proc/irq/<irq>/affinity_hint */ + proc_create_data("affinity_hint", 0400, desc->dir, + &irq_affinity_hint_proc_fops, (void *)(long)irq); #endif proc_create_data("spurious", 0444, desc->dir,
This patch adds a callback function pointer to the irq_desc structure, along with a registration function and a read-only proc entry for each interrupt. This affinity_hint handle for each interrupt can be used by underlying drivers that need a better mechanism to control interrupt affinity. The underlying driver can register a callback for the interrupt, which will allow the driver to provide the CPU mask for the interrupt to anything that requests it. The intent is to extend the userspace daemon, irqbalance, to help hint to it a preferred CPU mask to balance the interrupt into. Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> --- include/linux/interrupt.h | 27 +++++++++++++++++++++++++++ include/linux/irq.h | 1 + kernel/irq/manage.c | 39 +++++++++++++++++++++++++++++++++++++++ kernel/irq/proc.c | 41 +++++++++++++++++++++++++++++++++++++++++ 4 files changed, 108 insertions(+), 0 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html