clk: renesas: Add CPG core wrapper for RZ/G2L SoC
Hello Lad Prabhakar,
The patch ef3c613ccd68: "clk: renesas: Add CPG core wrapper for RZ/G2L SoC" from Jun 9, 2021, leads to the following static checker warning:
drivers/clk/renesas/renesas-rzg2l-cpg.c:226 rzg2l_cpg_clk_src_twocell_get() warn: array off by one? 'priv->clks[clkidx]'
drivers/clk/renesas/renesas-rzg2l-cpg.c
209 static struct clk
210 *rzg2l_cpg_clk_src_twocell_get(struct of_phandle_args *clkspec,
211 void *data)
212 {
213 unsigned int clkidx = clkspec->args[1]; 214 struct rzg2l_cpg_priv *priv = data; 215 struct device *dev = priv->dev; 216 const char *type; 217 struct clk *clk;
218
219 switch (clkspec->args[0]) {
220 case CPG_CORE:
221 type = "core";
222 if (clkidx > priv->last_dt_core_clk) {
The ->last_dt_core_clk value comes from the device tree and I hate that we have to trust it. I haven't looked at the device tree and I only look at the code but based on the name "last_", I assume that in the device tree data this is set to either:
last_dt_core_clk = priv->num_core_clks + priv->num_mod_clks - 1;
Or maybe it's set so that:
last_dt_core_clk = priv->num_core_clks - 1;
So I think that it is not off by one (based on the naming scheme). But I would prefer that this code just used:
if (clkidx >= priv->num_core_clks)
Or:
if (clkidx >= priv->num_core_clks + priv->num_mod_clks)
223 dev_err(dev, "Invalid %s clock index %u\n", type, clkidx);
224 return ERR_PTR(-EINVAL);
225 }
226 clk = priv->clks[clkidx];
227 break;
228
229 case CPG_MOD:
230 type = "module";
231 if (clkidx > priv->num_mod_clks) {
^^^^^^^^^^^^^^^^^^^^^^^^^^^Smatch did not catch it, but this condition is definitely off by one. ;)
232 dev_err(dev, "Invalid %s clock index %u\n", type, 233 clkidx); 234 return ERR_PTR(-EINVAL); 235 } 236 clk = priv->clks[priv->num_core_clks + clkidx]; 237 break; 238 239 default: 240 dev_err(dev, "Invalid CPG clock type %u\n", clkspec->args[0]); 241 return ERR_PTR(-EINVAL); 242 } 243 244 if (IS_ERR(clk)) 245 dev_err(dev, "Cannot get %s clock %u: %ld", type, clkidx, 246 PTR_ERR(clk)); 247 else 248 dev_dbg(dev, "clock (%u, %u) is %pC at %lu Hz\n", 249 clkspec->args[0], clkspec->args[1], clk, 250 clk_get_rate(clk)); 251 return clk; 252 }
regards,
dan carpenter
Hi Dan,
Thank you for the review.
Quoted 43 lines
-----Original Message-----
From: Dan Carpenter <[email protected]>
Sent: 17 June 2021 14:37
To: Prabhakar Mahadev Lad <[email protected]>
Cc: [email protected]; [email protected]
Subject: [bug report] clk: renesas: Add CPG core wrapper for RZ/G2L SoCHello Lad Prabhakar,
The patch
ef3c613ccd68: "clk: renesas: Add CPG core wrapper for RZ/G2L SoC" from Jun 9, 2021, leads to the following static checker warning:drivers/clk/renesas/renesas-rzg2l-cpg.c:226 rzg2l_cpg_clk_src_twocell_get() warn: array off by one? 'priv->clks[clkidx]'
drivers/clk/renesas/renesas-rzg2l-cpg.c
209 static struct clk
210 *rzg2l_cpg_clk_src_twocell_get(structof_phandle_args*clkspec,
211 void *data)212 { 213 unsigned int clkidx = clkspec->args[1]; 214 struct rzg2l_cpg_priv *priv = data; 215 struct device *dev = priv->dev; 216 const char *type; 217 struct clk *clk; 218 219 switch (clkspec->args[0]) { 220 case CPG_CORE: 221 type = "core"; 222 if (clkidx > priv->last_dt_core_clk) {The ->last_dt_core_clk value comes from the device tree and I hate that we have to trust it. I haven't looked at the device tree and I only look at the code but based on the name "
last_", I assume that in the device tree data this is set to either:last_dt_core_clk = priv->num_core_clks + priv->num_mod_clks - 1;Or maybe it's set so that:
last_dt_core_clk = priv->num_core_clks - 1;
Quoted 8 lines
So I think that it is not off by one (based on the naming scheme). But I would prefer that this code just used:
if (clkidx >= priv->num_core_clks)Or:
if (clkidx >= priv->num_core_clks + priv->num_mod_clks)
last_dt_core_clk comes from header include/dt-bindings/clock/r9a07g044-cpg.h and used in the SoC specific file in case of RZ/G2L its assigned to " LAST_DT_CORE_CLK = R9A07G044_OSCCLK," in drivers/clk/renesas/r9a07g044-cpg.c index of the core clocks start from zero so this check will hold good here.
Quoted 11 lines
223 dev_err(dev, "Invalid %s clock index %u\n", type, clkidx);
224 return ERR_PTR(-EINVAL);
225 }
226 clk = priv->clks[clkidx];
227 break;
228
229 case CPG_MOD:
230 type = "module";
231 if (clkidx > priv->num_mod_clks) {
^^^^^^^^^^^^^^^^^^^^^^^^^^^Quoted 2 lines
Smatch did not catch it, but this condition is definitely off by one. ;)
Good catch this definitely needs to be if (clkidx > (priv->num_mod_clks - 1)
Cheers,
Prabhakar
Quoted 11 lines · Prabhakar Mahadev Lad
Quoted 11 lines
223 dev_err(dev, "Invalid %s clock index %u\n", type, clkidx); 224 return ERR_PTR(-EINVAL); 225 } 226 clk = priv->clks[clkidx]; 227 break; 228 229 case CPG_MOD: 230 type = "module"; 231 if (clkidx > priv->num_mod_clks) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^
Quoted 3 lines
Quoted 2 lines
Smatch did not catch it, but this condition is definitely off by one. ;)
Good catch this definitely needs to be if (clkidx > (
priv->num_mod_clks- 1)
The size - 1 format is riskier because there is a potential for underflow.
Imagine that in the future priv->num_mod_clks is zero. "priv->num_mod_clks - 1" is now UINT_MAX and any value of "clkidx" is accepted. In this case, you know that the value of num_mod_clks if 57 but it took me some time to figure that out and ensure that it couldn't be zero.
regards,
dan carpenter
Hi Dan,
Quoted 19 lines
-----Original Message-----
From: Dan Carpenter <[email protected]>
Sent: 17 June 2021 15:43
To: Prabhakar Mahadev Lad <[email protected]>
Cc: [email protected]; [email protected]
Subject: Re: [bug report] clk: renesas: Add CPG core wrapper for RZ/G2L SoCQuoted 11 lines · Prabhakar Mahadev Lad
Quoted 11 lines
223 dev_err(dev, "Invalid %s clock index %u\n", type, clkidx); 224 return ERR_PTR(-EINVAL); 225 } 226 clk = priv->clks[clkidx]; 227 break; 228 229 case CPG_MOD: 230 type = "module"; 231 if (clkidx > priv->num_mod_clks) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Smatch did not catch it, but this condition is definitely off by one. ;)
Good catch this definitely needs to be if (clkidx > (
priv->num_mod_clks- 1)The size - 1 format is riskier because there is a potential for underflow.
Agreed, >= check should do the trick.
Cheers,
Prabhakar
Quoted 8 lines
Imagine that in the future
priv->num_mod_clksis zero.
"priv->num_mod_clks- 1" is nowUINT_MAXand any value of "clkidx" is accepted. In this case, you know that the value ofnum_mod_clksif 57 but it took me some time to figure that out and ensure that it couldn't be zero.regards,
dan carpenter