Correct behaviour of irq affinity?
RFC 1/1 Correct behaviour of irq affinity?
The effect of setting desc->affinity (ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value with cpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.
2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.
If we should change it, here's what the patch looks like against x86 tip (cpu_mask_to_apicid_and already takes cpu_online_mask into account):
arch/x86/kernel/apic/io_apic.c +2 -2
@@ -592,10 +592,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ; -cpumask_and (desc->affinity, cfg->domain, mask); +cpumask_copy (desc->affinity, mask);set_extra_move_desc (desc, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->domain); } static void
Quoted 28 lines · Rusty Russell
The effect of setting
desc->affinity(ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value withcpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with
cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.If we should change it, here's what the patch looks like against x86 tip (
cpu_mask_to_apicid_andalready takescpu_online_maskinto account):arch/x86/kernel/apic/io_apic.c +2
-2@@ -592,10 +592,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask) if (assign_irq_vector(irq, cfg, mask)) return BAD_APICID; - cpumask_and(desc->affinity, cfg->domain, mask); + cpumask_copy(desc->affinity, mask); set_extra_move_desc(desc, mask); - return apic->cpu_mask_to_apicid_and(desc->affinity, cpu_online_mask); + return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain); }
Quoted 2 lines
static void
cfg->domain for logical flat: will be ALL_CPUS
for phys flat (aka bigsmp on 32bit) will be one cpu set mask.
so desc->affinity: for logical will be not changed, but set_desc_affinity() return will be changed. ( not add with cpu_online_mask anymore)
when mask is 0x0f
for phys flat, desc->affinity will be changed to 0x0f from 0x01/0x02/0x04/08, return set_desc_affinity is not changed. so /proc/irq/xx/smp_affinity will be changed. and it does reflect that actually affinity.
so this patch looks not right.
YH
Quoted 29 lines · On Tuesday 24 March 2009 17:51:43 Yinghai Lu wrote:
Quoted 28 lines · Rusty Russell
The effect of setting
desc->affinity(ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value withcpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with
cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.If we should change it, here's what the patch looks like against x86 tip (
cpu_mask_to_apicid_andalready takescpu_online_maskinto account):arch/x86/kernel/apic/io_apic.c +2
-2@@ -592,10 +592,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ; -cpumask_and (desc->affinity, cfg->domain, mask); +cpumask_copy (desc->affinity, mask);set_extra_move_desc (desc, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->domain); }
Quoted 8 lines
arch/x86/kernel/apic/io_apic.c set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask) · line 601
Quoted 2 lines
arch/x86/kernel/apic/io_apic.c +0
-0static void
cfg->domainfor logical flat: will beALL_CPUS
for phys flat (aka bigsmp on 32bit) will be one cpu set mask.so
desc->affinity: for logical will be not changed, butset_desc_affinity()return will be changed. ( not add withcpu_online_maskanymore)
No, internally cpu_mask_to_apicid_and() does and with cpu_online_mask already, eg in include/asm/bigsmp/apic.h:
static inline unsigned int cpu_mask_to_apicid_and(const struct cpumask *cpumask,
const struct cpumask *andmask)
{
int cpu;
/*
* We're using fixed IRQ delivery, can only return one phys APIC ID.
* May as well be the first.
*/
for_each_cpu_and(cpu, cpumask, andmask)
if (cpumask_test_cpu(cpu, cpu_online_mask))
break;
if (cpu < nr_cpu_ids)
return cpu_to_logical_apicid(cpu);
return BAD_APICID;
}Quoted 7 lines
when mask is
0x0f
for phys flat,desc->affinitywill be changed to0x0ffrom 0x01/0x02/0x04/08, returnset_desc_affinityis not changed. so/proc/irq/xx/smp_affinitywill be changed. and it does reflect that actually affinity.so this patch looks not right.
Only change should be that smp_affinity will reflect actual affinity, not affinity user set.
Thanks,
Rusty.
Quoted 66 lines · Rusty Russell
Quoted 29 lines · On Tuesday 24 March 2009 17:51:43 Yinghai Lu wrote:
Quoted 28 lines · Rusty Russell
The effect of setting
desc->affinity(ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value withcpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with
cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.If we should change it, here's what the patch looks like against x86 tip (
cpu_mask_to_apicid_andalready takescpu_online_maskinto account):arch/x86/kernel/apic/io_apic.c +2
-2@@ -592,10 +592,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ; -cpumask_and (desc->affinity, cfg->domain, mask); +cpumask_copy (desc->affinity, mask);set_extra_move_desc (desc, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->domain); }Quoted 8 lines
arch/x86/kernel/apic/io_apic.c set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask) · line 601
Quoted 2 lines
arch/x86/kernel/apic/io_apic.c +0
-0static void
cfg->domainfor logical flat: will beALL_CPUS
for phys flat (aka bigsmp on 32bit) will be one cpu set mask.so
desc->affinity: for logical will be not changed, butset_desc_affinity()return will be changed. ( not add withcpu_online_maskanymore)No, internally
cpu_mask_to_apicid_and()does and withcpu_online_maskalready, eg ininclude/asm/bigsmp/apic.h:static inline unsigned int cpu_mask_to_apicid_and(const struct cpumask *cpumask,
const struct cpumask *andmask){ int cpu; /* * We're using fixed IRQ delivery, can only return one phys APIC ID. * May as well be the first. */ for_each_cpu_and(cpu, cpumask, andmask) if (cpumask_test_cpu(cpu, cpu_online_mask)) break; if (cpu < nr_cpu_ids) return cpu_to_logical_apicid(cpu); return BAD_APICID; }Quoted 6 lines
when mask is
0x0f
for phys flat,desc->affinitywill be changed to0x0ffrom 0x01/0x02/0x04/08, returnset_desc_affinityis not changed. so/proc/irq/xx/smp_affinitywill be changed. and it does reflect that actually affinity.
Quoted 4 lines
Quoted 1 line
so this patch looks not right.
Only change should be that
smp_affinitywill reflect actual affinity, not affinity user set.
ok.
how about
static unsigned int flat_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
const struct cpumask *andmask)
{
unsigned long mask1 = cpumask_bits(cpumask)[0] & APIC_ALL_CPUS;
unsigned long mask2 = cpumask_bits(andmask)[0] & APIC_ALL_CPUS;
return mask1 & mask2;
}change it usedefault_cpu_mask_to_apicid_and ?
YH
Quoted 11 lines · Rusty Russell
The effect of setting
desc->affinity(ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value withcpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with
cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.If we should change it, here's what the patch looks like against x86 tip (
cpu_mask_to_apicid_andalready takescpu_online_maskinto account):
desc->affinity should be what the user requested, if it is at all possible to honor the user space request. YH the fact that we do not currently exercise the full freedom that user space gives us is irrelevant.
Further setting desc->affinity to the user space request is what x86_64 did before the grand merger.
Likewise desc->affinity & cfg->domain & cpu_online_map going into the selection of apic id, is what the code did before the grand merger, and what the code is currently doing. So logically that looks good.
YH has a point that several of the implementations of cpu_mask_to_apic_id do not take cpu_online_map into account and should probably be fixed. flat_cpu_mask_to_apicid was the one I could find.
Also now that I look at it there is one other bug in this routine that you have missed. set_extra_move_desc should be called before we set desc->affinity, as it compares that with the new value to see if we are going to be running on a new cpu, and if so we may need to reallocate irq_desc onto a new numa node. set_extra_move_desc looks a little fishy but it doesn't stand a chance if it is called with the wrong data.
Overall I like it.
Do you think you could fix those two issues and regenerate the patch?
Quoted 7 lines
arch/x86/kernel/apic/io_apic.c +0
-0@@ -592,10 +592,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ;
arch/x86/kernel/apic/io_apic.c set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask) · line 601
Quoted 10 lines
arch/x86/kernel/apic/io_apic.c +2
-2-cpumask_and (desc->affinity, cfg->domain, mask); +cpumask_copy (desc->affinity, mask);set_extra_move_desc (desc, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->domain); } static void
Eric
Quoted 8 lines · Eric W. Biederman
Also now that I look at it there is one other bug in this routine that you have missed.
set_extra_move_descshould be called before we setdesc->affinity, as it compares that with the new value to see if we are going to be running on a new cpu, and if so we may need to reallocateirq_desconto a new numa node.set_extra_move_desclooks a little fishy but it doesn't stand a chance if it is called with the wrong data.
you are right, that is introduce bycommit 22f65d31b25a320a5246592160bcb102d2791c45
Author: Mike Travis <[email protected]> Date: Tue Dec 16 17:33:56 2008 -0800
x86: Update io_apic.c to use new cpumask API
Impact: cleanup, consolidate patches, use new API
Consolidate the following into a single patch to adapt to new sparseirq code in arch/x86/kernel/io_apic.c, add allocation of cpumask_var_t's in domain and old_domain, and reduce further merge conflicts. Only one file (arch/x86/kernel/io_apic.c) is changed in all of these patches.
0006-x86-io_apic-change-irq_cfg-domain-old_domain-to.patch 0007-x86-io_apic-set_desc_affinity.patch 0008-x86-io_apic-send_cleanup_vector.patch 0009-x86-io_apic-eliminate-remaining-cpumask_ts-from-st.patch 0021-x86-final-cleanups-in-io_apic-to-use-new-cpumask-AP.patch
will send one patch for that.
YH
PATCH 1/1 x86: fix set_extra_move_desc calling
In reply to Eric W. BiedermanImpact: fix bug with desc moving when logical flat
Eric pointed out that we should compare with old affinity. acctually this bug is introduced by:
| commit 22f65d31b25a320a5246592160bcb102d2791c45 | | Author: Mike Travis <[email protected]> | Date: Tue Dec 16 17:33:56 2008 -0800 | | x86: Update io_apic.c to use new cpumask API | | Impact: cleanup, consolidate patches, use new API
so need to move that before...
--- arch/x86/kernel/apic/io_apic.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/io_apic.c +2 -1
@@ -592,8 +592,9 @@ set_desc_affinity(struct irq_desc *desc,if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ; -cpumask_and (desc->affinity, cfg->domain, mask); +/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); +cpumask_and (desc->affinity, cfg->domain, mask);return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); }
[tip:x86/apic] x86: fix set_extra_move_desc calling
Commit-ID: fa74c9073370e57fa28e02aff13f4d7b1806505c Gitweb: http://git.kernel.org/tip/fa74c9073370e57fa28e02aff13f4d7b1806505c Author: Yinghai Lu <[email protected]> AuthorDate: Tue, 24 Mar 2009 13:23:16 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Tue, 24 Mar 2009 22:12:10 +0100
x86: fix set_extra_move_desc calling
Impact: fix bug with irq-descriptor moving when logical flat
Rusty observed:
Quoted 8 lines
The effect of setting
desc->affinity(ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value withcpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with
cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.
Eric pointed out that desc->affinity should be what the user requested, if it is at all possible to honor the user space request.
This bug got introduced by commit 22f65d31b "x86: Update io_apic.c to use new cpumask API".
Fix it by moving the masking to before the descriptor moving ...
LKML-Reference: <[email protected]>
--- arch/x86/kernel/apic/io_apic.c | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/apic/io_apic.c +2 -1
@@ -592,8 +592,9 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ; -cpumask_and (desc->affinity, cfg->domain, mask); +/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); +cpumask_and (desc->affinity, cfg->domain, mask);return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); }
PATCH 1/3 x86: fix set_extra_move_desc calling
In reply to Yinghai LuImpact: fix bug with desc moving when logical flat
Eric pointed out that should compare with old desc. acctually this bug is introduced by:
| commit 22f65d31b25a320a5246592160bcb102d2791c45 | | Author: Mike Travis <[email protected]> | Date: Tue Dec 16 17:33:56 2008 -0800 | | x86: Update io_apic.c to use new cpumask API | | Impact: cleanup, consolidate patches, use new API
so need to move that before...
--- arch/x86/kernel/apic/io_apic.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/io_apic.c +2 -1
@@ -592,8 +592,9 @@ set_desc_affinity(struct irq_desc *desc,if (assign_irq_vector (irq, cfg, mask))return BAD_APICID ; -cpumask_and (desc->affinity, cfg->domain, mask); +/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); +cpumask_and (desc->affinity, cfg->domain, mask);return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); }
PATCH 2/3 x86: use default_cpu_mask_to_apicid for 64bit
Impact: cleanup
use online_mask directly with 64bit too.
--- arch/x86/include/asm/apic.h | 20 ++++++++++---------- arch/x86/kernel/apic/apic_flat_64.c | 18 ++---------------- 2 files changed, 12 insertions(+), 26 deletions(-)
arch/x86/include/asm/apic.h +10 -10
@@ -489,10 +489,19 @@ static inline int default_apic_id_registreturn physid_isset (read_apic_id (), phys_cpu_present_map); } +static inline int default_phys_pkg_id (int cpuid_apic,int index_msb) +{ +return cpuid_apic >> index_msb; +} + +extern int default_apicid_to_node (int logical_apicid); + +#endif +static inline unsigned int default_cpu_mask_to_apicid (const struct cpumask *cpumask) { -return cpumask_bits (cpumask)[0 ]; +return cpumask_bits (cpumask)[0 ] &APIC_ALL_CPUS ; } static inline unsigned int
@@ -506,15 +515,6 @@ default_cpu_mask_to_apicid_and(const strreturn (unsigned int )(mask1 & mask2 & mask3); } -static inline int default_phys_pkg_id (int cpuid_apic,int index_msb) -{ -return cpuid_apic >> index_msb; -} - -extern int default_apicid_to_node (int logical_apicid); - -#endif -static inline unsigned long default_check_apicid_used (physid_mask_t bitmap,int apicid) {return physid_isset (apicid, bitmap);
arch/x86/kernel/apic/apic_flat_64.c +2 -16
@@ -159,20 +159,6 @@ static int flat_apic_id_registered(void)return physid_isset (read_xapic_id (), phys_cpu_present_map); } -static unsigned int flat_cpu_mask_to_apicid (const struct cpumask *cpumask) -{ -return cpumask_bits (cpumask)[0 ] &APIC_ALL_CPUS ; -} - -static unsigned int flat_cpu_mask_to_apicid_and (const struct cpumask *cpumask, -const struct cpumask *andmask) -{ -unsigned long mask1 =cpumask_bits (cpumask)[0 ] &APIC_ALL_CPUS ; -unsigned long mask2 =cpumask_bits (andmask)[0 ] &APIC_ALL_CPUS ; - -return mask1 & mask2; -} -static int flat_phys_pkg_id (int initial_apic_id,int index_msb) {return hard_smp_processor_id () >> index_msb;
@@ -213,8 +199,8 @@ struct apic apic_flat = { .set_apic_id = set_apic_id, .apic_id_mask = 0xFFu << 24, - .cpu_mask_to_apicid = flat_cpu_mask_to_apicid, - .cpu_mask_to_apicid_and = flat_cpu_mask_to_apicid_and, + .cpu_mask_to_apicid = default_cpu_mask_to_apicid, + .cpu_mask_to_apicid_and = default_cpu_mask_to_apicid_and, .send_IPI_mask = flat_send_IPI_mask, .send_IPI_mask_allbutself = flat_send_IPI_mask_allbutself,
[tip:x86/apic] x86: use default_cpu_mask_to_apicid for 64bit
Commit-ID: f56e5034121c4911a155ba907076ab920754626d Gitweb: http://git.kernel.org/tip/f56e5034121c4911a155ba907076ab920754626d Author: Yinghai Lu <[email protected]> AuthorDate: Tue, 24 Mar 2009 14:16:30 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Tue, 24 Mar 2009 22:28:38 +0100
x86: use default_cpu_mask_to_apicid for 64bit
Impact: cleanup
Use online_mask directly on 64bit too.
LKML-Reference: <[email protected]>
--- arch/x86/include/asm/apic.h | 20 ++++++++++---------- arch/x86/kernel/apic/apic_flat_64.c | 18 ++---------------- 2 files changed, 12 insertions(+), 26 deletions(-)
arch/x86/include/asm/apic.h +10 -10
@@ -489,10 +489,19 @@ static inline int default_apic_id_registered(void)return physid_isset (read_apic_id (), phys_cpu_present_map); } +static inline int default_phys_pkg_id (int cpuid_apic,int index_msb) +{ +return cpuid_apic >> index_msb; +} + +extern int default_apicid_to_node (int logical_apicid); + +#endif +static inline unsigned int default_cpu_mask_to_apicid (const struct cpumask *cpumask) { -return cpumask_bits (cpumask)[0 ]; +return cpumask_bits (cpumask)[0 ] &APIC_ALL_CPUS ; } static inline unsigned int
@@ -506,15 +515,6 @@ default_cpu_mask_to_apicid_and(const struct cpumask *cpumask,return (unsigned int )(mask1 & mask2 & mask3); } -static inline int default_phys_pkg_id (int cpuid_apic,int index_msb) -{ -return cpuid_apic >> index_msb; -} - -extern int default_apicid_to_node (int logical_apicid); - -#endif -static inline unsigned long default_check_apicid_used (physid_mask_t bitmap,int apicid) {return physid_isset (apicid, bitmap);
arch/x86/kernel/apic/apic_flat_64.c +2 -16
@@ -159,20 +159,6 @@ static int flat_apic_id_registered(void)return physid_isset (read_xapic_id (), phys_cpu_present_map); } -static unsigned int flat_cpu_mask_to_apicid (const struct cpumask *cpumask) -{ -return cpumask_bits (cpumask)[0 ] &APIC_ALL_CPUS ; -} - -static unsigned int flat_cpu_mask_to_apicid_and (const struct cpumask *cpumask, -const struct cpumask *andmask) -{ -unsigned long mask1 =cpumask_bits (cpumask)[0 ] &APIC_ALL_CPUS ; -unsigned long mask2 =cpumask_bits (andmask)[0 ] &APIC_ALL_CPUS ; - -return mask1 & mask2; -} -static int flat_phys_pkg_id (int initial_apic_id,int index_msb) {return hard_smp_processor_id () >> index_msb;
@@ -213,8 +199,8 @@ struct apic apic_flat = { .set_apic_id = set_apic_id, .apic_id_mask = 0xFFu << 24, - .cpu_mask_to_apicid = flat_cpu_mask_to_apicid, - .cpu_mask_to_apicid_and = flat_cpu_mask_to_apicid_and, + .cpu_mask_to_apicid = default_cpu_mask_to_apicid, + .cpu_mask_to_apicid_and = default_cpu_mask_to_apicid_and, .send_IPI_mask = flat_send_IPI_mask, .send_IPI_mask_allbutself = flat_send_IPI_mask_allbutself,
Re: [tip:x86/apic] x86: use default_cpu_mask_to_apicid for 64bit
build failure:
arch/x86/kernel/apic/io_apic.c:600: error: ‘struct irq_cfg’ has no member named ‘cpu_domain’
Ingo
PATCH 3/3 x86: Correct behaviour of irq affinity -v2
From: Rusty Russell <[email protected]>
Impact: get correct smp_affinity as user requested
The effect of setting desc->affinity (ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value with cpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.
2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.
If we should change it, here's what the patch looks like against x86 tip (cpu_mask_to_apicid_and already takes cpu_online_mask into account)
--- arch/x86/kernel/apic/io_apic.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
arch/x86/kernel/apic/io_apic.c +3 -2
@@ -594,9 +594,10 @@ set_desc_affinity(struct irq_desc *desc,/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); -cpumask_and (desc->affinity, cfg->domain, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +cpumask_copy (desc->affinity, mask); + +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->domain); } static void
PATCH 3/3 x86: Correct behaviour of irq affinity
In reply to Yinghai LuFrom: Rusty Russell <[email protected]>
Impact: get correct smp_affinity as user requested
The effect of setting desc->affinity (ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value with cpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.
2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with cfg->domain. Is this right, or should we just set it to what the user said? Or as now, indicate that we're restricting it.
If we should change it, here's what the patch looks like against x86 tip (cpu_mask_to_apicid_and already takes cpu_online_mask into account)
--- arch/x86/kernel/apic/io_apic.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
arch/x86/kernel/apic/io_apic.c +3 -2
@@ -594,9 +594,10 @@ set_desc_affinity(struct irq_desc *desc,/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); -cpumask_and (desc->affinity, cfg->domain, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +cpumask_copy (desc->affinity, mask); + +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->cpu_domain); } static void
[tip:x86/apic] x86: Correct behaviour of irq affinity
Commit-ID: 85a328f27b927e3f42563c76e1a6992068100433 Gitweb: http://git.kernel.org/tip/85a328f27b927e3f42563c76e1a6992068100433 Author: Rusty Russell <[email protected]> AuthorDate: Tue, 24 Mar 2009 14:17:19 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Tue, 24 Mar 2009 22:28:39 +0100
x86: Correct behaviour of irq affinity
Impact: get correct smp_affinity as user requested
The effect of setting desc->affinity (ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value with cpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.
2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with cfg->domain.
We should just set it to what the user said - if possible.
(cpu_mask_to_apicid_and already takes cpu_online_mask into account)
LKML-Reference: <[email protected]>
--- arch/x86/kernel/apic/io_apic.c | 5 +++-- 1 files changed, 3 insertions(+), 2 deletions(-)
arch/x86/kernel/apic/io_apic.c +3 -2
@@ -594,9 +594,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); -cpumask_and (desc->affinity, cfg->domain, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +cpumask_copy (desc->affinity, mask); + +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->cpu_domain); } static void
[tip:x86/apic] x86: Correct behaviour of irq affinity
In reply to Yinghai LuCommit-ID: e06b1b56f9bfcc91e1f175fe8d8bf3e35dafa080 Gitweb: http://git.kernel.org/tip/e06b1b56f9bfcc91e1f175fe8d8bf3e35dafa080 Author: Rusty Russell <[email protected]> AuthorDate: Tue, 24 Mar 2009 14:17:19 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Wed, 25 Mar 2009 18:48:29 +0100
x86: Correct behaviour of irq affinity
Impact: get correct smp_affinity as user requested
The effect of setting desc->affinity (ie. from userspace via sysfs) has varied over time. In 2.6.27, the 32-bit code anded the value with cpu_online_map, and both 32 and 64-bit did that anding whenever a cpu was unplugged.
2.6.29 consolidated this into one routine (and fixed hotplug) but introduced another variation: anding the affinity with cfg->domain.
We should just set it to what the user said - if possible.
(cpu_mask_to_apicid_and already takes cpu_online_mask into account)
LKML-Reference: <[email protected]>
--- arch/x86/kernel/apic/io_apic.c | 5 +++-- 1 files changed, 3 insertions(+), 2 deletions(-)
arch/x86/kernel/apic/io_apic.c +3 -2
@@ -594,9 +594,10 @@ set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)/* check that before desc->addinity get updated */ set_extra_move_desc (desc, mask); -cpumask_and (desc->affinity, cfg->domain, mask); -return apic->cpu_mask_to_apicid_and (desc->affinity, cpu_online_mask); +cpumask_copy (desc->affinity, mask); + +return apic->cpu_mask_to_apicid_and (desc->affinity, cfg->domain); } static void
Quoted 4 lines · On Tuesday 24 March 2009 23:09:37 Eric W. Biederman wrote:
desc->affinityshould be what the user requested, if it is at all possible to honor the user space request. YH the fact that we do not currently exercise the full freedom that user space gives us is irrelevant.
Yep, OK.
Quoted 3 lines
YH has a point that several of the implementations of
cpu_mask_to_apic_iddo not takecpu_online_mapinto account and should probably be fixed.flat_cpu_mask_to_apicidwas the one I could find.
Also the numaq apic.h. I'll do an audit and send a patch.
Quoted 7 lines
Also now that I look at it there is one other bug in this routine that you have missed.
set_extra_move_descshould be called before we setdesc->affinity, as it compares that with the new value to see if we are going to be running on a new cpu, and if so we may need to reallocateirq_desconto a new numa node.set_extra_move_desclooks a little fishy but it doesn't stand a chance if it is called with the wrong data.
Yes, agree with Yinghai's fix. I'll re-spin my patch on top of his.
Thanks for looking at this!
Rusty.
Quoted 5 lines · On Wednesday 25 March 2009 11:03:08 Rusty Russell wrote:
Quoted 2 lines · On Tuesday 24 March 2009 23:09:37 Eric W. Biederman wrote:
cpu_mask_to_apic_iddo not takecpu_online_mapinto account and should probably be fixed.flat_cpu_mask_to_apicidwas the one I could find.Also the numaq
apic.h. I'll do an audit and send a patch.
Actually, numaq is just weird, so I'll assume it's correct. I couldn't find any others.
Quoted 1 line
Yes, agree with Yinghai's fix. I'll re-spin my patch on top of his.
But he already did it while I slept in.
Thanks Yinghai!
Rusty.
Quoted 4 lines · Rusty Russell
Quoted 3 lines · On Wednesday 25 March 2009 11:03:08 Rusty Russell wrote:
Quoted 2 lines · On Tuesday 24 March 2009 23:09:37 Eric W. Biederman wrote:
cpu_mask_to_apic_iddo not takecpu_online_mapinto account and should probably be fixed.flat_cpu_mask_to_apicidwas the one I could find.
Quoted 9 lines
Quoted 1 line
Also the numaq
apic.h. I'll do an audit and send a patch.Actually, numaq is just weird, so I'll assume it's correct. I couldn't find any others.
Quoted 1 line
Yes, agree with Yinghai's fix. I'll re-spin my patch on top of his.
But he already did it while I slept in.
waiting for ingo to put that three into tip/master:
[PATCH 1/3] x86: fix set_extra_move_desc calling [PATCH 2/3] x86: use default_cpu_mask_to_apicid for 64bit [PATCH 3/3] x86: Correct behaviour of irq affinity -v2
YH