HWBKPT: Make bp_len type to u64 generic across the arch
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
Frederic Weisbecker
Acked-byQuoted 24 lines · Mahesh Salgaonkar
Change '
bp_len' type to__u64to make it work across the arch. The s390 architecture watch point length can be upto 2^64.reference:
http://lkml.org/lkml/2010/1/25/212Based on commit
6aa41f8b01301199af6c9febb24f3c1f5a0bc9d5Signed-off-by Mahesh Salgaonkar <[email protected]>---
include/linux/hw_breakpoint.h | 2 +- include/linux/perf_event.h | 6 ++---- kernel/hw_breakpoint.c | 2 +- kernel/perf_event.c | 2 +- 4 files changed, 5 insertions(+), 7 deletions(-)include/linux/hw_breakpoint.h +0
-0
include/linux/hw_breakpoint.h static inline int hw_breakpoint_type(struct perf_event *bp) · line 50
Quoted 8 lines
include/linux/hw_breakpoint.h +1
-1@@ -44,7 +44,7 @@ static inline int hw_breakpoint_type(struct perf_event *bp)return bp->attr.bp_type; } -static inline int hw_breakpoint_len (struct perf_event *bp) +static inline unsigned long hw_breakpoint_len (struct perf_event *bp) {return bp->attr.bp_len; }
This should return a u64, or gcc will warn us about loosing informations in 32 bits arch?
Quoted 6 lines
include/linux/perf_event.h +0
-0@@ -211,11 +211,9 @@ struct perf_event_attr {__u32 wakeup_watermark;/* bytes before wakeup */
include/linux/perf_event.h struct perf_event_attr { · line 217
Quoted 10 lines
include/linux/perf_event.h +2
-4}; -__u32 __reserved_2; - -__u64 bp_addr;__u32 bp_type; -__u32 bp_len; +__u64 bp_addr; +__u64 bp_len; };
Peter, what do you think about this new layout? Putting the bp_type right after the wakeup_* fields is going to remove the padding difference between 64 and 32 archs. That looks better than the __reserved_2 we had.
If this patch can make it for .33, it would be nice.
Quoted 22 lines
include/linux/perf_event.h +0
-0/* kernel/hw_breakpoint.c +1
-1@@ -324,8 +324,8 @@ EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);int modify_user_hw_breakpoint (struct perf_event *bp,struct perf_event_attr *attr) {u64 old_addr = bp->attr.bp_addr; +u64 old_len = bp->attr.bp_len;int old_type = bp->attr.bp_type; -int old_len = bp->attr.bp_len;int err =0 ;perf_event_disable (bp);kernel/perf_event.c +0
-0
kernel/perf_event.c static int perf_copy_attr(struct perf_event_attr __user *uattr, · line 4726
Quoted 8 lines
kernel/perf_event.c +1
-1@@ -4720,7 +4720,7 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,if (attr->type >=PERF_TYPE_MAX )return -EINVAL ; -if (attr->__reserved_1 || attr->__reserved_2) +if (attr->__reserved_1)return -EINVAL ;if (attr->sample_type & ~(PERF_SAMPLE_MAX -1 ))
Quoted 28 lines · Frederic Weisbecker
Quoted 27 lines · Mahesh Salgaonkar
Change '
bp_len' type to__u64to make it work across the arch. The s390 architecture watch point length can be upto 2^64.reference:
http://lkml.org/lkml/2010/1/25/212Based on commit
6aa41f8b01301199af6c9febb24f3c1f5a0bc9d5Signed-off-by Mahesh Salgaonkar<[email protected]>---
include/linux/hw_breakpoint.h | 2 +- include/linux/perf_event.h | 6 ++---- kernel/hw_breakpoint.c | 2 +- kernel/perf_event.c | 2 +- 4 files changed, 5 insertions(+), 7 deletions(-)include/linux/hw_breakpoint.h +0
-0@@ -44,7 +44,7 @@ static inline int hw_breakpoint_type(struct perf_event *bp)return bp->attr.bp_type; }
Quoted 11 lines
include/linux/hw_breakpoint.h static inline int hw_breakpoint_type(struct perf_event *bp) · line 50
Quoted 5 lines
include/linux/hw_breakpoint.h +1
-1-static inline int hw_breakpoint_len (struct perf_event *bp) +static inline unsigned long hw_breakpoint_len (struct perf_event *bp) {return bp->attr.bp_len; }This should return a u64, or gcc will warn us about loosing informations in 32 bits arch?
Yup, I even thought so, but then I figured out the function 'hw_breakpoint_addr()' also has same return type for returning 'attr.bp_addr' and kept the same type to be consistent. We may have to fix that also. What do you think?
Quoted 7 lines
Quoted 3 lines
Thanks for reviewing.
Quoted 40 lines · Mahesh Jagannath Salgaonkar
Quoted 28 lines · Frederic Weisbecker
Quoted 27 lines · Mahesh Salgaonkar
Change '
bp_len' type to__u64to make it work across the arch. The s390 architecture watch point length can be upto 2^64.reference:
http://lkml.org/lkml/2010/1/25/212Based on commit
6aa41f8b01301199af6c9febb24f3c1f5a0bc9d5Signed-off-by Mahesh Salgaonkar<[email protected]>---
include/linux/hw_breakpoint.h | 2 +- include/linux/perf_event.h | 6 ++---- kernel/hw_breakpoint.c | 2 +- kernel/perf_event.c | 2 +- 4 files changed, 5 insertions(+), 7 deletions(-)include/linux/hw_breakpoint.h +0
-0@@ -44,7 +44,7 @@ static inline int hw_breakpoint_type(struct perf_event *bp)return bp->attr.bp_type; }Quoted 11 lines
include/linux/hw_breakpoint.h static inline int hw_breakpoint_type(struct perf_event *bp) · line 50
Quoted 5 lines
include/linux/hw_breakpoint.h +1
-1-static inline int hw_breakpoint_len (struct perf_event *bp) +static inline unsigned long hw_breakpoint_len (struct perf_event *bp) {return bp->attr.bp_len; }This should return a u64, or gcc will warn us about loosing informations in 32 bits arch?
Quoted 4 lines
Yup, I even thought so, but then I figured out the function '
hw_breakpoint_addr()' also has same return type for returning 'attr.bp_addr' and kept the same type to be consistent. We may have to fix that also. What do you think?
True. I'll fix both then, thanks!
Peter Zijlstra
maintainerQuoted 8 lines · Frederic Weisbecker
Quoted 8 lines
include/linux/perf_event.h +0
-0@@ -211,11 +211,9 @@ struct perf_event_attr {__u32 wakeup_watermark;/* bytes before wakeup */ };
Quoted 16 lines
include/linux/perf_event.h struct perf_event_attr { · line 217
Quoted 8 lines
include/linux/perf_event.h +2
-4-__u32 __reserved_2; - -__u64 bp_addr;__u32 bp_type; -__u32 bp_len; +__u64 bp_addr; +__u64 bp_len; };Peter, what do you think about this new layout? Putting the
bp_typeright after the wakeup_* fields is going to remove the padding difference between 64 and 32 archs. That looks better than the__reserved_2we had.
Right, I think this works nicely in that all elements will be naturally aligned and not result in different layouts between 32/64 bit builds.
Quoted 1 line
If this patch can make it for .33, it would be nice.
It has to make .33, if it doesn't you're hosed because then the old layout is fixed in stone.
Quoted 26 lines · Peter Zijlstra
Quoted 8 lines · Frederic Weisbecker
Quoted 8 lines
include/linux/perf_event.h +0
-0@@ -211,11 +211,9 @@ struct perf_event_attr {__u32 wakeup_watermark;/* bytes before wakeup */ };Quoted 16 lines
include/linux/perf_event.h struct perf_event_attr { · line 217
Quoted 8 lines
include/linux/perf_event.h +2
-4-__u32 __reserved_2; - -__u64 bp_addr;__u32 bp_type; -__u32 bp_len; +__u64 bp_addr; +__u64 bp_len; };Peter, what do you think about this new layout? Putting the
bp_typeright after the wakeup_* fields is going to remove the padding difference between 64 and 32 archs. That looks better than the__reserved_2we had.
Quoted 8 lines
Right, I think this works nicely in that all elements will be naturally aligned and not result in different layouts between 32/64 bit builds.
Quoted 1 line
If this patch can make it for .33, it would be nice.
It has to make .33, if it doesn't you're hosed because then the old layout is fixed in stone.
Truly.
I'll send a pull request to Ingo quickly then.
Thanks.
[tip:perf/urgent] perf: Make bp_len type to u64 generic across the arch
In reply to a message not in the archiveCommit-ID: cd757645fbdc34a8343c04bb0e74e06fccc2cb10 Gitweb: http://git.kernel.org/tip/cd757645fbdc34a8343c04bb0e74e06fccc2cb10 Author: Mahesh Salgaonkar <[email protected]> AuthorDate: Sat, 30 Jan 2010 10:25:18 +0530 Committer: Frederic Weisbecker <[email protected]> CommitDate: Thu, 4 Feb 2010 01:07:12 +0100
perf: Make bp_len type to u64 generic across the arch
Change 'bp_len' type to __u64 to make it work across archs as the s390 architecture watch point length can be upto 2^64.
reference:
http://lkml.org/lkml/2010/1/25/212
This is an ABI change that is not backward compatible with the previous hardware breakpoint info layout integrated in this development cycle, a rebuilt of perf tools is necessary for versions based on 2.6.33-rc1 - 2.6.33-rc6 to work with a kernel based on this patch.
LKML-Reference: <[email protected]>
--- include/linux/hw_breakpoint.h | 2 +- include/linux/perf_event.h | 6 ++---- kernel/hw_breakpoint.c | 2 +- kernel/perf_event.c | 2 +- 4 files changed, 5 insertions(+), 7 deletions(-)
include/linux/hw_breakpoint.h +1 -1
@@ -44,7 +44,7 @@ static inline int hw_breakpoint_type(struct perf_event *bp)return bp->attr.bp_type; } -static inline int hw_breakpoint_len (struct perf_event *bp) +static inline unsigned long hw_breakpoint_len (struct perf_event *bp) {return bp->attr.bp_len; }
include/linux/perf_event.h +2 -4
@@ -211,11 +211,9 @@ struct perf_event_attr {__u32 wakeup_watermark;/* bytes before wakeup */ }; -__u32 __reserved_2; - -__u64 bp_addr;__u32 bp_type; -__u32 bp_len; +__u64 bp_addr; +__u64 bp_len; };/*
kernel/hw_breakpoint.c +1 -1
@@ -360,8 +360,8 @@ EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);int modify_user_hw_breakpoint (struct perf_event *bp,struct perf_event_attr *attr) {u64 old_addr = bp->attr.bp_addr; +u64 old_len = bp->attr.bp_len;int old_type = bp->attr.bp_type; -int old_len = bp->attr.bp_len;int err =0 ;perf_event_disable (bp);
kernel/perf_event.c +1 -1
@@ -4580,7 +4580,7 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,if (attr->type >=PERF_TYPE_MAX )return -EINVAL ; -if (attr->__reserved_1 || attr->__reserved_2) +if (attr->__reserved_1)return -EINVAL ;if (attr->sample_type & ~(PERF_SAMPLE_MAX -1 ))