[PATCH] powerpc/papr_scm: Fix warning triggered by perf_stats_show()

[Vaibhav Jain]

Thanks for reviewing this patch Ira,


Ira Weiny <ira.weiny at intel.com> writes:

> On Thu, Sep 10, 2020 at 02:52:12PM +0530, Vaibhav Jain wrote:
>> A warning is reported by the kernel in case perf_stats_show() returns
>> an error code. The warning is of the form below:
>> 
>>  papr_scm ibm,persistent-memory:ibm,pmemory at 44100001:
>>  	  Failed to query performance stats, Err:-10
>>  dev_attr_show: perf_stats_show+0x0/0x1c0 [papr_scm] returned bad count
>>  fill_read_buffer: dev_attr_show+0x0/0xb0 returned bad count
>> 
>> On investigation it looks like that the compiler is silently truncating the
>> return value of drc_pmem_query_stats() from 'long' to 'int', since the
>> variable used to store the return code 'rc' is an 'int'. This
>> truncated value is then returned back as a 'ssize_t' back from
>> perf_stats_show() to 'dev_attr_show()' which thinks of it as a large
>> unsigned number and triggers this warning..
>> 
>> To fix this we update the type of variable 'rc' from 'int' to
>> 'ssize_t' that prevents the compiler from truncating the return value
>> of drc_pmem_query_stats() and returning correct signed value back from
>> perf_stats_show().
>> 
>> Fixes: 2d02bf835e573 ('powerpc/papr_scm: Fetch nvdimm performance
>>        stats from PHYP')
>> Signed-off-by: Vaibhav Jain <vaibhav at linux.ibm.com>
>> ---
>>  arch/powerpc/platforms/pseries/papr_scm.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>> 
>> diff --git a/arch/powerpc/platforms/pseries/papr_scm.c b/arch/powerpc/platforms/pseries/papr_scm.c
>> index a88a707a608aa..9f00b61676ab9 100644
>> --- a/arch/powerpc/platforms/pseries/papr_scm.c
>> +++ b/arch/powerpc/platforms/pseries/papr_scm.c
>> @@ -785,7 +785,8 @@ static int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc,
>>  static ssize_t perf_stats_show(struct device *dev,
>>  			       struct device_attribute *attr, char *buf)
>>  {
>> -	int index, rc;
>> +	int index;
>> +	ssize_t rc;
>
> I'm not sure this is really fixing everything here.

The issue is with the statement in perf_stats_show():

'return rc ? rc : seq_buf_used(&s);'

The function seq_buf_used() returns an 'unsigned int' and with 'rc'
typed as 'int', forces a promotion of the expression to 'unsigned int'
which causes a loss of signedness of 'rc' and compiler silently
assigns this unsigned value to the function return typed as 'signed
long'.

Making 'rc', a 'signed long' forces a promotion of the expresion to
'signed long' which preserves the signedness of 'rc' and will also be
compatible with the function return type.

>
> drc_pmem_query_stats() can return negative errno's.  Why are those not checked
> somewhere in perf_stats_show()?
>
For the specific invocation 'drc_pmem_query_stats(p, stats, 0)' we only
expect return value 'rc <=0' with '0' indicating a successful fetch of
nvdimm performance stats from hypervisor. Hence there are no explicit
checks for negative error codes in the functions as all return values
!=0 indicate an error.


> It seems like all this fix is handling is a > 0 return value: 'ret[0]' from
> line 289 in papr_scm.c...  Or something?
No, in case the arg 'num_stats' is '0' and 'buff_stats != NULL' the
variable 'size' is assigned a non-zero value hence that specific branch
you mentioned  is never taken. Instead in case of success
drc_pmem_query_stats() return '0' and in case of an error a negative
error code is returned.

>
> Worse yet drc_pmem_query_stats() is returning ssize_t which is a signed value.
> Therefore, it should not be returning -errno.  I'm surprised the static
> checkers did not catch that.
Didnt quite get the assertion here. The function is marked to return
'ssize_t' because we can return both +ve for success and -ve values to
indicate errors.

>
> I believe I caught similar errors with a patch series before which did not pay
> attention to variable types.
>
> Please audit this code for these types of errors and ensure you are really
> doing the correct thing when using the sysfs interface.  I'm pretty sure bad
> things will eventually happen (if they are not already) if you return some
> really big number to the sysfs core from *_show().
I think this problem is different compared to what you had previously pointed
to. The values returned from drc_pmem_query_stats() can be stored in an
'int' variable too, however it was the silent promotion of a signed type
to unsigned type was what caused this specific issue.

-- 
Cheers
~ Vaibhav