Skip to contentSkip to conversation

Maple Tree: add new data structure

bug report Dan Carpenter 2 participants 2 messages Jul 19, 2022 Memory management

lore
Share & tools
lore
#0 lore raw

Hello Liam R. Howlett,

The patch 058c2f0f755f: "Maple Tree: add new data structure" from Jul 17, 2022, leads to the following Smatch static checker warning:

	lib/maple_tree.c:6969 mas_validate_limits()
	warn: add some parenthesis here? '!piv & (i != 0)'

lib/maple_tree.c

    6952 static void mas_validate_limits(struct ma_state *mas)
    6953 {
    6954         int i;
    6955         unsigned long prev_piv = 0;
    6956         enum maple_type type = mte_node_type(mas->node);
    6957         void __rcu **slots = ma_slots(mte_to_node(mas->node), type);
    6958         unsigned long *pivots = ma_pivots(mas_mn(mas), type);
    6959 
    6960         /* all limits are fine here. */
    6961         if (mte_is_root(mas->node))
    6962                 return;
    6963 
    6964         for (i = 0; i < mt_slots[type]; i++) {
    6965                 unsigned long piv;
    6966 
    6967                 piv = mas_safe_pivot(mas, pivots, i, type);
    6968 

--> 6969 if (!piv & (i != 0))

Sparse and Smatch both hate this. I think this code is correct and it should be:

		if ((!piv) & (i != 0))

However the vast majority of the time this warning indicates a precendence bug where what was intended was:

	if (!(piv & (i != 0)))

I guess it's not common to use booleans for bitwise ANDs...

    6970                         break;
    6971 
    6972                 if (!mte_is_leaf(mas->node)) {

regards,
dan carpenter

#1 lore raw
Quoted 31 lines · Dan Carpenter

Hello Liam R. Howlett,

The patch 058c2f0f755f: "Maple Tree: add new data structure" from Jul 17, 2022, leads to the following Smatch static checker warning:

	lib/maple_tree.c:6969 mas_validate_limits()
	warn: add some parenthesis here? '!piv & (i != 0)'

lib/maple_tree.c

    6952 static void mas_validate_limits(struct ma_state *mas)
    6953 {
    6954         int i;
    6955         unsigned long prev_piv = 0;
    6956         enum maple_type type = mte_node_type(mas->node);
    6957         void __rcu **slots = ma_slots(mte_to_node(mas->node), type);
    6958         unsigned long *pivots = ma_pivots(mas_mn(mas), type);
    6959 
    6960         /* all limits are fine here. */
    6961         if (mte_is_root(mas->node))
    6962                 return;
    6963 
    6964         for (i = 0; i < mt_slots[type]; i++) {
    6965                 unsigned long piv;
    6966 
    6967                 piv = mas_safe_pivot(mas, pivots, i, type);
    6968 

--> 6969 if (!piv & (i != 0))

Sparse and Smatch both hate this. I think this code is correct and it should be:

Quoted 6 lines
		if ((!piv) & (i != 0))

However the vast majority of the time this warning indicates a precendence bug where what was intended was:

	if (!(piv & (i != 0)))

Thanks! I caught this with C=1 (sparse). I've been working through another issue with Hugh that was more pressing. I'll send out a patch today. The double && will work here but it shouldn't actually make a difference.

Regards,
Liam

Quoted 10 lines

I guess it's not common to use booleans for bitwise ANDs...

    6970                         break;
    6971 
    6972                 if (!mte_is_leaf(mas->node)) {

regards,
dan carpenter