Impact: Reduce stack usage, use new cpumask API.
Convert misc driver functions to use struct cpumask.
To Do:
- Convert iucv_buffer_cpumask to cpumask_var_t.
Signed-off-by: Rusty Russell <[email protected]>
Signed-off-by: Mike Travis <[email protected]>
Cc: Ben Hutchings <[email protected]>
Cc: Dean Nelson <[email protected]>
Cc: Jeremy Fitzhardinge <[email protected]>
Cc: Robert Richter <[email protected]>
---
drivers/base/cpu.c | 2 +-
drivers/misc/sgi-xp/xpc_main.c | 2 +-
drivers/net/sfc/efx.c | 29 +++++++++++++++++------------
drivers/oprofile/buffer_sync.c | 22 ++++++++++++++++++----
drivers/oprofile/buffer_sync.h | 4 ++++
drivers/oprofile/oprof.c | 9 ++++++++-
drivers/xen/manage.c | 2 +-
net/iucv/iucv.c | 28 ++++++++++++++++++----------
8 files changed, 68 insertions(+), 30 deletions(-)
--- linux-2.6-for-ingo.orig/drivers/base/cpu.c
+++ linux-2.6-for-ingo/drivers/base/cpu.c
@@ -107,7 +107,7 @@ static SYSDEV_ATTR(crash_notes, 0400, sh
/*
* Print cpu online, possible, present, and system maps
*/
-static ssize_t print_cpus_map(char *buf, cpumask_t *map)
+static ssize_t print_cpus_map(char *buf, const struct cpumask *map)
{
int n = cpulist_scnprintf(buf, PAGE_SIZE-2, map);
--- linux-2.6-for-ingo.orig/drivers/misc/sgi-xp/xpc_main.c
+++ linux-2.6-for-ingo/drivers/misc/sgi-xp/xpc_main.c
@@ -318,7 +318,7 @@ xpc_hb_checker(void *ignore)
/* this thread was marked active by xpc_hb_init() */
- set_cpus_allowed_ptr(current, &cpumask_of_cpu(XPC_HB_CHECK_CPU));
+ set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU));
/* set our heartbeating to other partitions into motion */
xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
--- linux-2.6-for-ingo.orig/drivers/net/sfc/efx.c
+++ linux-2.6-for-ingo/drivers/net/sfc/efx.c
@@ -809,19 +809,18 @@ static void efx_fini_io(struct efx_nic *
/* Get number of RX queues wanted. Return number of online CPU
* packages in the expectation that an IRQ balancer will spread
* interrupts across them. */
-static int efx_wanted_rx_queues(void)
+static int efx_wanted_rx_queues(struct cpumask *scratch)
{
- cpumask_t core_mask;
int count;
int cpu;
- cpus_clear(core_mask);
+ cpumask_clear(scratch);
count = 0;
for_each_online_cpu(cpu) {
- if (!cpu_isset(cpu, core_mask)) {
+ if (!cpumask_test_cpu(cpu, scratch)) {
++count;
- cpus_or(core_mask, core_mask,
- topology_core_siblings(cpu));
+ cpumask_or(scratch, scratch,
+ topology_core_cpumask(cpu));
}
}
@@ -831,7 +830,7 @@ static int efx_wanted_rx_queues(void)
/* Probe the number and type of interrupts we are able to obtain, and
* the resulting numbers of channels and RX queues.
*/
-static void efx_probe_interrupts(struct efx_nic *efx)
+static void efx_probe_interrupts(struct efx_nic *efx, struct cpumask *scratch)
{
int max_channels =
min_t(int, efx->type->phys_addr_channels, EFX_MAX_CHANNELS);
@@ -845,7 +844,8 @@ static void efx_probe_interrupts(struct
* (or as specified by the rss_cpus module parameter).
* We will need one channel per interrupt.
*/
- wanted_ints = rss_cpus ? rss_cpus : efx_wanted_rx_queues();
+ wanted_ints = rss_cpus ? rss_cpus :
+ efx_wanted_rx_queues(scratch);
efx->n_rx_queues = min(wanted_ints, max_channels);
for (i = 0; i < efx->n_rx_queues; i++)
@@ -923,24 +923,29 @@ static void efx_set_channels(struct efx_
static int efx_probe_nic(struct efx_nic *efx)
{
int rc;
+ cpumask_var_t scratch;
EFX_LOG(efx, "creating NIC\n");
+ if (!alloc_cpumask_var(&scratch, GFP_KERNEL))
+ return -ENOMEM;
+
/* Carry out hardware-type specific initialisation */
rc = falcon_probe_nic(efx);
if (rc)
- return rc;
+ goto out;
/* Determine the number of channels and RX queues by trying to hook
* in MSI-X interrupts. */
- efx_probe_interrupts(efx);
+ efx_probe_interrupts(efx, scratch);
efx_set_channels(efx);
/* Initialise the interrupt moderation settings */
efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec);
-
- return 0;
+out:
+ free_cpumask_var(scratch);
+ return rc;
}
static void efx_remove_nic(struct efx_nic *efx)
--- linux-2.6-for-ingo.orig/drivers/oprofile/buffer_sync.c
+++ linux-2.6-for-ingo/drivers/oprofile/buffer_sync.c
@@ -37,7 +37,7 @@
static LIST_HEAD(dying_tasks);
static LIST_HEAD(dead_tasks);
-static cpumask_t marked_cpus = CPU_MASK_NONE;
+static cpumask_var_t marked_cpus;
static DEFINE_SPINLOCK(task_mortuary);
static void process_task_mortuary(void);
@@ -524,10 +524,10 @@ static void mark_done(int cpu)
{
int i;
- cpu_set(cpu, marked_cpus);
+ cpumask_set_cpu(cpu, marked_cpus);
for_each_online_cpu(i) {
- if (!cpu_isset(i, marked_cpus))
+ if (!cpumask_test_cpu(i, marked_cpus))
return;
}
@@ -536,7 +536,7 @@ static void mark_done(int cpu)
*/
process_task_mortuary();
- cpus_clear(marked_cpus);
+ cpumask_clear(marked_cpus);
}
@@ -632,6 +632,20 @@ void sync_buffer(int cpu)
mutex_unlock(&buffer_mutex);
}
+int __init buffer_sync_init(void)
+{
+ if (!alloc_cpumask_var(&marked_cpus, GFP_KERNEL))
+ return -ENOMEM;
+
+ cpumask_copy(marked_cpus, cpu_none_mask);
+ return 0;
+}
+
+void __exit buffer_sync_cleanup(void)
+{
+ free_cpumask_var(marked_cpus);
+}
+
/* The function can be used to add a buffer worth of data directly to
* the kernel buffer. The buffer is assumed to be a circular buffer.
* Take the entries from index start and end at index end, wrapping
--- linux-2.6-for-ingo.orig/drivers/oprofile/buffer_sync.h
+++ linux-2.6-for-ingo/drivers/oprofile/buffer_sync.h
@@ -19,4 +19,8 @@ void sync_stop(void);
/* sync the given CPU's buffer */
void sync_buffer(int cpu);
+/* initialize/destroy the buffer system. */
+int buffer_sync_init(void);
+void buffer_sync_cleanup(void);
+
#endif /* OPROFILE_BUFFER_SYNC_H */
--- linux-2.6-for-ingo.orig/drivers/oprofile/oprof.c
+++ linux-2.6-for-ingo/drivers/oprofile/oprof.c
@@ -183,6 +183,10 @@ static int __init oprofile_init(void)
{
int err;
+ err = buffer_sync_init();
+ if (err)
+ return err;
+
err = oprofile_arch_init(&oprofile_ops);
if (err < 0 || timer) {
@@ -191,8 +195,10 @@ static int __init oprofile_init(void)
}
err = oprofilefs_register();
- if (err)
+ if (err) {
oprofile_arch_exit();
+ buffer_sync_cleanup();
+ }
return err;
}
@@ -202,6 +208,7 @@ static void __exit oprofile_exit(void)
{
oprofilefs_unregister();
oprofile_arch_exit();
+ buffer_sync_cleanup();
}
--- linux-2.6-for-ingo.orig/drivers/xen/manage.c
+++ linux-2.6-for-ingo/drivers/xen/manage.c
@@ -100,7 +100,7 @@ static void do_suspend(void)
/* XXX use normal device tree? */
xenbus_suspend();
- err = stop_machine(xen_suspend, &cancelled, &cpumask_of_cpu(0));
+ err = stop_machine(xen_suspend, &cancelled, cpumask_of(0));
if (err) {
printk(KERN_ERR "failed to start xen_suspend: %d\n", err);
goto out;
--- linux-2.6-for-ingo.orig/net/iucv/iucv.c
+++ linux-2.6-for-ingo/net/iucv/iucv.c
@@ -489,15 +489,14 @@ static void iucv_setmask_mp(void)
*
* Allow iucv interrupts on a single cpu.
*/
-static void iucv_setmask_up(void)
+static void iucv_setmask_up(struct cpumask *cpumask)
{
- cpumask_t cpumask;
int cpu;
/* Disable all cpu but the first in cpu_irq_cpumask. */
- cpumask = iucv_irq_cpumask;
- cpu_clear(first_cpu(iucv_irq_cpumask), cpumask);
- for_each_cpu_mask_nr(cpu, cpumask)
+ cpumask_copy(cpumask, iucv_irq_cpumask);
+ cpumask_clear_cpu(cpumask_first(iucv_irq_cpumask), cpumask);
+ for_each_cpu(cpu, cpumask)
smp_call_function_single(cpu, iucv_block_cpu, NULL, 1);
}
@@ -555,7 +554,7 @@ static void iucv_disable(void)
static int __cpuinit iucv_cpu_notify(struct notifier_block *self,
unsigned long action, void *hcpu)
{
- cpumask_t cpumask;
+ cpumask_var_t cpumask;
long cpu = (long) hcpu;
switch (action) {
@@ -590,15 +589,20 @@ static int __cpuinit iucv_cpu_notify(str
break;
case CPU_DOWN_PREPARE:
case CPU_DOWN_PREPARE_FROZEN:
- cpumask = iucv_buffer_cpumask;
- cpu_clear(cpu, cpumask);
- if (cpus_empty(cpumask))
+ if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
+ return NOTIFY_BAD;
+ cpumask_copy(cpumask, &iucv_buffer_cpumask);
+ cpumask_clear_cpu(cpu, cpumask);
+ if (cpumask_empty(cpumask)) {
/* Can't offline last IUCV enabled cpu. */
+ free_cpumask_var(cpumask);
return NOTIFY_BAD;
+ }
smp_call_function_single(cpu, iucv_retrieve_cpu, NULL, 1);
if (cpus_empty(iucv_irq_cpumask))
smp_call_function_single(first_cpu(iucv_buffer_cpumask),
iucv_allow_cpu, NULL, 1);
+ free_cpumask_var(cpumask);
break;
}
return NOTIFY_OK;
@@ -683,9 +687,12 @@ static void iucv_cleanup_queue(void)
int iucv_register(struct iucv_handler *handler, int smp)
{
int rc;
+ cpumask_var_t scratch;
if (!iucv_available)
return -ENOSYS;
+ if (!alloc_cpumask_var(&scratch, GFP_KERNEL))
+ return -ENOMEM;
mutex_lock(&iucv_register_mutex);
if (!smp)
iucv_nonsmp_handler++;
@@ -694,7 +701,7 @@ int iucv_register(struct iucv_handler *h
if (rc)
goto out_mutex;
} else if (!smp && iucv_nonsmp_handler == 1)
- iucv_setmask_up();
+ iucv_setmask_up(scratch);
INIT_LIST_HEAD(&handler->paths);
spin_lock_bh(&iucv_table_lock);
@@ -703,6 +710,7 @@ int iucv_register(struct iucv_handler *h
rc = 0;
out_mutex:
mutex_unlock(&iucv_register_mutex);
+ free_cpumask_var(scratch);
return rc;
}
EXPORT_SYMBOL(iucv_register);
--
On Fri, 2008-12-19 at 08:01 -0800, Mike Travis wrote:
[...]
> --- linux-2.6-for-ingo.orig/drivers/net/sfc/efx.c
> +++ linux-2.6-for-ingo/drivers/net/sfc/efx.c
> @@ -809,19 +809,18 @@ static void efx_fini_io(struct efx_nic *
> /* Get number of RX queues wanted. Return number of online CPU
> * packages in the expectation that an IRQ balancer will spread
> * interrupts across them. */
> -static int efx_wanted_rx_queues(void)
> +static int efx_wanted_rx_queues(struct cpumask *scratch)
> {
> - cpumask_t core_mask;
> int count;
> int cpu;
>
> - cpus_clear(core_mask);
> + cpumask_clear(scratch);
> count = 0;
> for_each_online_cpu(cpu) {
> - if (!cpu_isset(cpu, core_mask)) {
> + if (!cpumask_test_cpu(cpu, scratch)) {
> ++count;
> - cpus_or(core_mask, core_mask,
> - topology_core_siblings(cpu));
> + cpumask_or(scratch, scratch,
> + topology_core_cpumask(cpu));
> }
> }
>
> @@ -831,7 +830,7 @@ static int efx_wanted_rx_queues(void)
> /* Probe the number and type of interrupts we are able to obtain, and
> * the resulting numbers of channels and RX queues.
> */
> -static void efx_probe_interrupts(struct efx_nic *efx)
> +static void efx_probe_interrupts(struct efx_nic *efx, struct cpumask *scratch)
> {
> int max_channels =
> min_t(int, efx->type->phys_addr_channels, EFX_MAX_CHANNELS);
> @@ -845,7 +844,8 @@ static void efx_probe_interrupts(struct
> * (or as specified by the rss_cpus module parameter).
> * We will need one channel per interrupt.
> */
> - wanted_ints = rss_cpus ? rss_cpus : efx_wanted_rx_queues();
> + wanted_ints = rss_cpus ? rss_cpus :
> + efx_wanted_rx_queues(scratch);
> efx->n_rx_queues = min(wanted_ints, max_channels);
>
> for (i = 0; i < efx->n_rx_queues; i++)
> @@ -923,24 +923,29 @@ static void efx_set_channels(struct efx_
> static int efx_probe_nic(struct efx_nic *efx)
> {
> int rc;
> + cpumask_var_t scratch;
>
> EFX_LOG(efx, "creating NIC\n");
>
> + if (!alloc_cpumask_var(&scratch, GFP_KERNEL))
> + return -ENOMEM;
> +
> /* Carry out hardware-type specific initialisation */
> rc = falcon_probe_nic(efx);
> if (rc)
> - return rc;
> + goto out;
>
> /* Determine the number of channels and RX queues by trying to hook
> * in MSI-X interrupts. */
> - efx_probe_interrupts(efx);
> + efx_probe_interrupts(efx, scratch);
>
> efx_set_channels(efx);
>
> /* Initialise the interrupt moderation settings */
> efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec);
> -
> - return 0;
> +out:
> + free_cpumask_var(scratch);
> + return rc;
> }
>
> static void efx_remove_nic(struct efx_nic *efx)
[...]
I assume you're moving the allocation up to efx_probe_nic() because
efx_wanted_rx_queues() and efx_probe_interrupts() cannot return failure.
It really isn't worth exposing that detail up the call chain though. I
think it's acceptable for efx_wanted_rx_queues() to log an error message
and return 1 in the exceedingly unlikely case that the allocation fails.
Ben.
--
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
On Sunday 21 December 2008 09:52:39 Ben Hutchings wrote:
> I assume you're moving the allocation up to efx_probe_nic() because
> efx_wanted_rx_queues() and efx_probe_interrupts() cannot return failure.
> It really isn't worth exposing that detail up the call chain though. I
> think it's acceptable for efx_wanted_rx_queues() to log an error message
> and return 1 in the exceedingly unlikely case that the allocation fails.
OK, fair call. I was trying trying hard not to break anything.
How's this?
cpumask: convert drivers/net/sfc
Remove a cpumask from the stack. Ben Hutchings indicated that printing
a warning and returning 1 was acceptable for the corner case where allocation
fails.
Signed-off-by: Rusty Russell <[email protected]>
Cc: Ben Hutchings <[email protected]>
diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c
--- a/drivers/net/sfc/efx.c
+++ b/drivers/net/sfc/efx.c
@@ -854,20 +854,27 @@ static void efx_fini_io(struct efx_nic *
* interrupts across them. */
static int efx_wanted_rx_queues(void)
{
- cpumask_t core_mask;
+ cpumask_var_t core_mask;
int count;
int cpu;
- cpus_clear(core_mask);
+ if (!alloc_cpumask_var(&core_mask, GFP_KERNEL)) {
+ printk(KERN_WARNING
+ "efx.c: allocation failure, irq balancing hobbled\n");
+ return 1;
+ }
+
+ cpumask_clear(core_mask);
count = 0;
for_each_online_cpu(cpu) {
- if (!cpu_isset(cpu, core_mask)) {
+ if (!cpumask_test_cpu(cpu, core_mask)) {
++count;
- cpumask_or(&core_mask, &core_mask,
+ cpumask_or(core_mask, core_mask,
topology_core_cpumask(cpu));
}
}
+ free_cpumask_var(core_mask);
return count;
}
On Fri, Dec 19, 2008 at 08:01:51AM -0800, Mike Travis wrote:
> Impact: Reduce stack usage, use new cpumask API.
>
> Convert misc driver functions to use struct cpumask.
>
> To Do:
> - Convert iucv_buffer_cpumask to cpumask_var_t.
>
> Signed-off-by: Rusty Russell <[email protected]>
> Signed-off-by: Mike Travis <[email protected]>
[...]
> --- linux-2.6-for-ingo.orig/drivers/misc/sgi-xp/xpc_main.c
> +++ linux-2.6-for-ingo/drivers/misc/sgi-xp/xpc_main.c
> @@ -318,7 +318,7 @@ xpc_hb_checker(void *ignore)
>
> /* this thread was marked active by xpc_hb_init() */
>
> - set_cpus_allowed_ptr(current, &cpumask_of_cpu(XPC_HB_CHECK_CPU));
> + set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU));
>
> /* set our heartbeating to other partitions into motion */
> xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
Acked-by: Dean Nelson <[email protected]>