cpufreq: tegra186: add OPP support and set bandwidth
Hello Aaron Kling,
Commit 71c46a6457e0 ("cpufreq: tegra186: add OPP support and set bandwidth") from Oct 21, 2025 (linux-next), leads to the following Smatch static checker warning:
drivers/cpufreq/tegra186-cpufreq.c:197 tegra186_cpufreq_init() error: uninitialized symbol 'freq_table'.
drivers/cpufreq/tegra186-cpufreq.c
174 static int tegra186_cpufreq_init(struct cpufreq_policy *policy)
175 {
176 struct tegra186_cpufreq_data *data = cpufreq_get_driver_data();
177 unsigned int cluster = data->cpus[policy->cpu].bpmp_cluster_id;
178 struct cpufreq_frequency_table *freq_table;
179 struct cpufreq_frequency_table *bpmp_lut;
180 u32 cpu;
181 int ret;
182
183 policy->cpuinfo.transition_latency = 300 * 1000;
184 policy->driver_data = NULL;
185
186 /* set same policy for all cpus in a cluster */
187 for (cpu = 0; cpu < ARRAY_SIZE(tegra186_cpus); cpu++) {
188 if (data->cpus[cpu].bpmp_cluster_id == cluster)
189 cpumask_set_cpu(cpu, policy->cpus);
190 }
191
192 bpmp_lut = data->clusters[cluster].bpmp_lut;
193
194 if (data->icc_dram_bw_scaling) {
195 ret = tegra_cpufreq_init_cpufreq_table(policy, bpmp_lut, &freq_table);
196 if (!ret) {--> 197 policy->freq_table = freq_table;
tegra_cpufreq_init_cpufreq_table() will return zero if there are no available opps. The freq_table will be uninitialized in that case. It's this path:
122 max_opps = dev_pm_opp_get_opp_count(cpu_dev);
123 if (max_opps <= 0) {
124 dev_err(cpu_dev, "Failed to add OPPs\n");
125 return max_opps;
126 }
198 return 0;
199 }
200 }
201
202 data->icc_dram_bw_scaling = false;
203 policy->freq_table = bpmp_lut;
204 pr_info("OPP tables missing from DT, EMC frequency scaling disabled\n");
205
206 return 0;
207 }regards,
dan carpenter
Quoted 57 lines · Dan Carpenter
Hello Aaron Kling,
Commit
71c46a6457e0("cpufreq: tegra186: add OPP support and set bandwidth") from Oct 21, 2025 (linux-next), leads to the following Smatch static checker warning:
drivers/cpufreq/tegra186-cpufreq.c:197tegra186_cpufreq_init()
error: uninitialized symbol 'freq_table'.
drivers/cpufreq/tegra186-cpufreq.c174 static int tegra186_cpufreq_init(struct cpufreq_policy *policy) 175 { 176 struct tegra186_cpufreq_data *data = cpufreq_get_driver_data(); 177 unsigned int cluster = data->cpus[policy->cpu].bpmp_cluster_id; 178 struct cpufreq_frequency_table *freq_table; 179 struct cpufreq_frequency_table *bpmp_lut; 180 u32 cpu; 181 int ret; 182 183 policy->cpuinfo.transition_latency = 300 * 1000; 184 policy->driver_data = NULL; 185 186 /* set same policy for all cpus in a cluster */ 187 for (cpu = 0; cpu < ARRAY_SIZE(tegra186_cpus); cpu++) { 188 if (data->cpus[cpu].bpmp_cluster_id == cluster) 189 cpumask_set_cpu(cpu, policy->cpus); 190 } 191 192 bpmp_lut = data->clusters[cluster].bpmp_lut; 193 194 if (data->icc_dram_bw_scaling) { 195 ret = tegra_cpufreq_init_cpufreq_table(policy, bpmp_lut, &freq_table); 196 if (!ret) {--> 197
policy->freq_table=freq_table;
tegra_cpufreq_init_cpufreq_table()will return zero if there are no available opps. Thefreq_tablewill be uninitialized in that case. It's this path:122 max_opps = dev_pm_opp_get_opp_count(cpu_dev); 123 if (max_opps <= 0) { 124 dev_err(cpu_dev, "Failed to add OPPs\n"); 125 return max_opps; 126 } 198 return 0; 199 } 200 } 201 202 data->icc_dram_bw_scaling = false; 203 policy->freq_table = bpmp_lut; 204 pr_info("OPP tables missing from DT, EMC frequency scaling disabled\n"); 205 206 return 0; 207 }
Quoted 2 lines
regards,
dan carpenter
I'm looking at dev_pm_opp_of_add_table_indexed which gets called before dev_pm_opp_get_opp_count inside tegra_cpufreq_init_cpufreq_table(). It states that it will return -ENODATA if the opp table is empty, which would then return a negative value from tegra_cpufreq_init_cpufreq_table(), which gets handled as expected. Is there a way for opp count to be zero that would get past this? The code could be better, but I'm not sure there's an operable fail case here.
Also of note, the same effective flow exists in the tegra194 cpufreq driver, so if something does need to be fixed here, it should probably be applied there at the same time.
Sincerely,
Aaron
Quoted 58 lines · Aaron Kling
Quoted 57 lines · Dan Carpenter
Hello Aaron Kling,
Commit
71c46a6457e0("cpufreq: tegra186: add OPP support and set bandwidth") from Oct 21, 2025 (linux-next), leads to the following Smatch static checker warning:
drivers/cpufreq/tegra186-cpufreq.c:197tegra186_cpufreq_init()
error: uninitialized symbol 'freq_table'.
drivers/cpufreq/tegra186-cpufreq.c174 static int tegra186_cpufreq_init(struct cpufreq_policy *policy) 175 { 176 struct tegra186_cpufreq_data *data = cpufreq_get_driver_data(); 177 unsigned int cluster = data->cpus[policy->cpu].bpmp_cluster_id; 178 struct cpufreq_frequency_table *freq_table; 179 struct cpufreq_frequency_table *bpmp_lut; 180 u32 cpu; 181 int ret; 182 183 policy->cpuinfo.transition_latency = 300 * 1000; 184 policy->driver_data = NULL; 185 186 /* set same policy for all cpus in a cluster */ 187 for (cpu = 0; cpu < ARRAY_SIZE(tegra186_cpus); cpu++) { 188 if (data->cpus[cpu].bpmp_cluster_id == cluster) 189 cpumask_set_cpu(cpu, policy->cpus); 190 } 191 192 bpmp_lut = data->clusters[cluster].bpmp_lut; 193 194 if (data->icc_dram_bw_scaling) { 195 ret = tegra_cpufreq_init_cpufreq_table(policy, bpmp_lut, &freq_table); 196 if (!ret) {--> 197
policy->freq_table=freq_table;
tegra_cpufreq_init_cpufreq_table()will return zero if there are no available opps. Thefreq_tablewill be uninitialized in that case. It's this path:122 max_opps = dev_pm_opp_get_opp_count(cpu_dev); 123 if (max_opps <= 0) { 124 dev_err(cpu_dev, "Failed to add OPPs\n"); 125 return max_opps; 126 } 198 return 0; 199 } 200 } 201 202 data->icc_dram_bw_scaling = false; 203 policy->freq_table = bpmp_lut; 204 pr_info("OPP tables missing from DT, EMC frequency scaling disabled\n"); 205 206 return 0; 207 }
Quoted 7 lines
Quoted 2 lines
regards,
dan carpenterI'm looking at
dev_pm_opp_of_add_table_indexedwhich gets called beforedev_pm_opp_get_opp_countinsidetegra_cpufreq_init_cpufreq_table(). It states that it will return-ENODATAif the opp table is empty,
I don't see the comment you're referring to but the call to dev_pm_opp_get_opp_count() which comes next would prevent an empty table. So it's fine. We can leave this as-is. Thanks for reviewing it.
regards,
dan carpenter