nilfs2: reduce build warnings with "make C=1"
Subsystems (2)
Ryusuke Konishi
maintainercover nilfs2: reduce build warnings with "make C=1"
Hi Andrew,
Please add these to the queue for the next merge window.
These two are cleanups that eliminate the warnings output by sparse (not including common one related to refcount_dec_and_lock). Neither is a bugfix.
Thanks,
Ryusuke Konishi
Ryusuke Konishi (2):
nilfs2: use integer type instead of enum req_op for event tracing
header
nilfs2: make superblock data array index computation sparse friendly
fs/nilfs2/mdt.c | 2 +- fs/nilfs2/the_nilfs.c | 20 ++++++++++++++++++-- include/trace/events/nilfs2.h | 4 ++-- 3 files changed, 21 insertions(+), 5 deletions(-)
Signature · 1 line
2.34.1
Ryusuke Konishi
maintainerPATCH 1/2 nilfs2: use integer type instead of enum req_op for event tracing header
The sparse check with "make C=1" outputs warnings regarding references to the header file "include/trace/events/nilfs2.h" for event tracing:
fs/nilfs2/segment.c: note: in included file (through
include/trace/trace_events.h, include/trace/define_trace.h, include/trace/events/nilfs2.h):
./include/trace/events/nilfs2.h:191:1: warning: cast to restricted
blk_opf_t
./include/trace/events/nilfs2.h:191:1: warning: restricted blk_opf_t
degrades to integer
./include/trace/events/nilfs2.h:191:1: warning: restricted blk_opf_t
degrades to integer
Fix this issue by reverting the type of the parameter related to the bio operation mode in the event tracing definition from "enum req_op" to "int".
In order to prevent sparse warnings on the caller side (where trace_nilfs2_mdt_submit_block() is used), also add a typecast to an argument passed to the tracepoint.
--- fs/nilfs2/mdt.c | 2 +- include/trace/events/nilfs2.h | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)
fs/nilfs2/mdt.c +1 -1
@@ -152,7 +152,7 @@ nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf, ret =0 ;trace_nilfs2_mdt_submit_block (inode, inode->i_ino, blkoff, - opf &REQ_OP_MASK ); + (__force int )(opf &REQ_OP_MASK )); out:get_bh (bh); *out_bh = bh;
include/trace/events/nilfs2.h +2 -2
@@ -192,7 +192,7 @@ TRACE_EVENT(nilfs2_mdt_submit_block, TP_PROTO(struct inode *inode, unsigned long ino, unsigned long blkoff, - enum req_op mode), + int mode), TP_ARGS(inode, ino, blkoff, mode),
@@ -200,7 +200,7 @@ TRACE_EVENT(nilfs2_mdt_submit_block, __field(struct inode *, inode) __field(unsigned long, ino) __field(unsigned long, blkoff) - __field(enum req_op, mode) + __field(int, mode) ), TP_fast_assign(
Signature · 1 line
2.34.1
Re: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
fs/nilfs2/mdt.c nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf, · line 155
Quoted 3 lines · Ryusuke Konishi
trace_nilfs2_mdt_submit_block(inode,
inode->i_ino, blkoff,fs/nilfs2/mdt.c +1
-1- opf & REQ_OP_MASK); + (__force int)(opf & REQ_OP_MASK));
Please keep the enum req_op type instead of casting that type away with "__force int".
Thanks,
Bart.
Ryusuke Konishi
maintainerRe: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
Quoted 6 lines · Bart Van Assche
fs/nilfs2/mdt.c nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf, · line 155
Quoted 3 lines · Ryusuke Konishi
trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff,fs/nilfs2/mdt.c +1
-1- opf & REQ_OP_MASK); + (__force int)(opf & REQ_OP_MASK));
Quoted 6 lines
Please keep the enum
req_optype instead of casting that type away with "__forceint".Thanks,
Bart.
Hi Bart,
No, this type cast is necessary to prevent the following sparse warning:
CC [M] fs/nilfs2/mdt.o
CHECK fs/nilfs2/mdt.cfs/nilfs2/mdt.c:155:43: warning: incorrect type in argument 4 (different base types)fs/nilfs2/mdt.c:155:43: expected int modefs/nilfs2/mdt.c:155:43: got restricted blk_opf_t
What we're doing here is just changing the event tracing type back to int, and keeping blk_opf_t and enum req_op in the rest of the code.
I understand if you have enough reason to ignore the warnings, but Why do you have to keep enum req_op type instead of int for event tracing?
Regards,
Ryusuke Konishi
Ryusuke Konishi
maintainerRe: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
Quoted 26 lines · Ryusuke Konishi
Quoted 6 lines · Bart Van Assche
fs/nilfs2/mdt.c nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff, blk_opf_t opf, · line 155
Quoted 3 lines · Ryusuke Konishi
trace_nilfs2_mdt_submit_block(inode, inode->i_ino, blkoff,fs/nilfs2/mdt.c +1
-1- opf & REQ_OP_MASK); + (__force int)(opf & REQ_OP_MASK));Quoted 6 lines
Please keep the enum
req_optype instead of casting that type away with "__forceint".Thanks,
Bart.
Hi Bart,
No, this type cast is necessary to prevent the following sparse warning:
CC [M]
fs/nilfs2/mdt.o
CHECKfs/nilfs2/mdt.cfs/nilfs2/mdt.c:155:43: warning: incorrect type in argument 4 (different base types)fs/nilfs2/mdt.c:155:43: expected int modefs/nilfs2/mdt.c:155:43: got restrictedblk_opf_t
Quoted 8 lines
What we're doing here is just changing the event tracing type back to int, and keeping
blk_opf_tand enumreq_opin the rest of the code.I understand if you have enough reason to ignore the warnings, but Why do you have to keep enum
req_optype instead of int for event tracing?Regards,
Ryusuke Konishi
Hi Bart,
Sorry, I didn't realize you were digging into the issue and talking with the sparse and kbuild teams to resolve the issue.
Is there any hope for a solution?
If you haven't given up yet on solving the underlying problem, I would like to withdraw this patch.
Thanks,
Ryusuke Konishi
Re: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
Quoted 2 lines · Ryusuke Konishi
If you haven't given up yet on solving the underlying problem, I would like to withdraw this patch.
Has this untested change been considered?
include/trace/events/nilfs2.h +1 -1
@@ -214,7 +214,7 @@ TRACE_EVENT(nilfs2_mdt_submit_block, __entry->inode, __entry->ino, __entry->blkoff, - __entry->mode) + (__force u32)__entry->mode) );#endif /* _TRACE_NILFS2_H */
Ryusuke Konishi
maintainerRe: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
Quoted 12 lines · Bart Van Assche
Quoted 2 lines · Ryusuke Konishi
If you haven't given up yet on solving the underlying problem, I would like to withdraw this patch.
Has this untested change been considered?
include/trace/events/nilfs2.h +0
-0
include/trace/events/nilfs2.h TRACE_EVENT(nilfs2_mdt_submit_block, · line 220
Quoted 8 lines
include/trace/events/nilfs2.h +1
-1@@ -214,7 +214,7 @@ TRACE_EVENT(nilfs2_mdt_submit_block, __entry->inode, __entry->ino, __entry->blkoff, - __entry->mode) + (__force u32)__entry->mode) );#endif /* _TRACE_NILFS2_H */
No, I didn't think of that. There was no warning in TP_printk() declaration of the nilfs2_mdt_submit_block trace point.
If you suggested this as an alternative idea, unfortunately the following warnings are still output:
CC [M] fs/nilfs2/segment.o
CHECK fs/nilfs2/segment.cfs/nilfs2/segment.c: note: in included file (through include/trace/trace_events.h, include/trace/define_trace.h, include/trace/events/nilfs2.h):./include/trace/events/nilfs2.h:191:1: warning: cast to restricted blk_opf_t ./include/trace/events/nilfs2.h:191:1: warning: restricted blk_opf_t degrades to integer./include/trace/events/nilfs2.h:191:1: warning: restricted blk_opf_t degrades to integer
I also tried typecasting on the declaration header side of event tracing, but so far, the sparse warnings don't go away except for the patch I first proposed.
But, better suggestions or solutions to the underlying problem are welcome. (Again, should we put the patch on hold?)
Regards,
Ryusuke Konishi
Re: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
Quoted 3 lines · Ryusuke Konishi
I also tried typecasting on the declaration header side of event tracing, but so far, the sparse warnings don't go away except for the patch I first proposed.
How about this patch?
include/trace/events/nilfs2.h +5 -1
@@ -200,7 +200,11 @@ TRACE_EVENT(nilfs2_mdt_submit_block, __field(struct inode *, inode) __field(unsigned long, ino) __field(unsigned long, blkoff) - __field(enum req_op, mode) + /* + * Use field_struct to avoid is_signed_type() on the bitwise + * type enum req_op. + */ + __field_struct(enum req_op, mode) ), TP_fast_assign(
Thanks,
Bart.
Ryusuke Konishi
maintainerRe: [PATCH -mm 1/2] nilfs2: use integer type instead of enum req_op for event tracing header
Quoted 26 lines · Bart Van Assche
Quoted 3 lines · Ryusuke Konishi
I also tried typecasting on the declaration header side of event tracing, but so far, the sparse warnings don't go away except for the patch I first proposed.
How about this patch?
include/trace/events/nilfs2.h +5
-1@@ -200,7 +200,11 @@ TRACE_EVENT(nilfs2_mdt_submit_block, __field(struct inode *, inode) __field(unsigned long, ino) __field(unsigned long, blkoff) - __field(enum req_op, mode) + /* + * Use field_struct to avoid is_signed_type() on the bitwise + * type enum req_op. + */ + __field_struct(enum req_op, mode) ), TP_fast_assign(
Quoted 3 lines
Thanks,
Bart.
Great! This patch removed the sparse warnings.
It passed multiple build methods and did not break either tracing or nilfs2 functionality in operational testing.
I never thought of using __field_struct for enum type, but looking at the definition it looks like it can be used to just avoidis_signed_type() without any side effects. So, this patch looks
better than mine.
If you don't mind, could you complete this change as a patch by adding a commit message and related tags?
Thanks,
Ryusuke Konishi
Ryusuke Konishi
maintainerPATCH 2/2 nilfs2: make superblock data array index computation sparse friendly
In reply to Ryusuke KonishiUpon running sparse, "warning: dubious: x & !y" is output at an array index calculation within nilfs_load_super_block().
The calculation is not wrong, but to eliminate the sparse warning, replace it with an equivalent calculation.
Also, add a comment to make it easier to understand what the unintuitive array index calculation is doing and whether it's correct.
--- fs/nilfs2/the_nilfs.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-)
fs/nilfs2/the_nilfs.c +18 -2
@@ -592,7 +592,7 @@ static int nilfs_load_super_block(struct the_nilfs *nilfs,struct nilfs_super_block **sbp = nilfs->ns_sbp;struct buffer_head **sbh = nilfs->ns_sbh;u64 sb2off, devsize =bdev_nr_bytes (nilfs->ns_bdev); -int valid[2 ], swp =0 ; +int valid[2 ], swp =0 , older;if (devsize <NILFS_SEG_MIN_BLOCKS *NILFS_MIN_BLOCK_SIZE +4096 ) {nilfs_err (sb,"device size too small" );
@@ -648,9 +648,25 @@ static int nilfs_load_super_block(struct the_nilfs *nilfs,if (swp)nilfs_swap_super_block (nilfs); +/* +* Calculate the array index of the older superblock data. +* If one has been dropped, set index 0 pointing to the remaining one, +* otherwise set index 1 pointing to the old one (including if both +* are the same). +* +* Divided case valid[0] valid[1] swp -> older +* ------------------------------------------------------------- +* Both SBs are invalid 0 0 N/A (Error) +* SB1 is invalid 0 1 1 0 +* SB2 is invalid 1 0 0 0 +* SB2 is newer 1 1 1 0 +* SB2 is older or the same 1 1 0 1 +*/ + older = valid[1 ] ^ swp; + nilfs->ns_sbwcount =0 ; nilfs->ns_sbwtime =le64_to_cpu (sbp[0 ]->s_wtime); - nilfs->ns_prot_seq =le64_to_cpu (sbp[valid[1 ] & !swp]->s_last_seq); + nilfs->ns_prot_seq =le64_to_cpu (sbp[older]->s_last_seq); *sbpp = sbp[0 ];return 0 ; }
Signature · 1 line
2.34.1