Re: [RFC] [PATCH v2 1/3] scatterlist: Add support to clone scatterlist

[Franklin S Cooper Jr.]

On 07/06/2016 05:15 AM, Sekhar Nori wrote:
> On Monday 27 June 2016 08:24 PM, Franklin S Cooper Jr wrote:
> 
>> +/*
>> + * sg_table_clone - Duplicate an existing sg_table including chained sgl
> 
> This function should probably be called sg_clone_table() to be
> consistent with sg_alloc_table(), sg_free_table() etc.


Will fix.
> 
>> + * @orig_table:     Original sg_table to be duplicated
>> + * @len:            Total length of sg while taking chaining into account
>> + * @gfp_mask:       GFP allocation mask
>> + *
>> + * Description:
>> + *   Clone a sg_table along with chained sgl. This cloned copy may be
>> + *   modified in some ways while keeping the original table and sgl in tact.
>> + *   Also allow the cloned sgl copy to have a smaller length than the original
>> + *   which may reduce the sgl total sg entries.
>> + *
>> + * Returns:
>> + *   Pointer to new kmalloced sg_table, ERR_PTR() on error
>> + *
>> + */
>> +struct sg_table *sg_table_clone(struct sg_table *orig_table, u64 len,
>> +				gfp_t gfp_mask)
>> +{
>> +	struct sg_table	*table;
>> +
>> +	table = kmalloc(sizeof(struct sg_table), gfp_mask);
> 
> Can you use sg_alloc_table() to allocate the new table? The way you have
> it now, it looks like users will need to use kfree() to free a cloned
> table and use sg_free_table() otherwise. It will be nice if
> sg_free_table() can be used consistently.


Sg_alloc_table doesn't actually allocate a struct sg_table. It allocates
a single sgl or chained sgl. Drivers that use sg_table manually allocate
it via kmalloc. One change that might make sense is to not kmalloc
sg_table but instead have a user pass a pointer to a preallocated
instance of it. This way it will mimic how sg_table is typically handled.
> 
> Regards,
> Sekhar
>