| Message ID | 20200612121133.GA1139533@mwanda (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | coresight: cti: Fix error handling in probe | expand |
On Fri, Jun 12, 2020 at 03:11:33PM +0300, Dan Carpenter wrote: > +static int cti_pm_setup(struct cti_drvdata *drvdata) > +{ > + int ret; > + > + if (drvdata->ctidev.cpu == -1) > + return 0; > + > + if (nr_cti_cpu) > + goto done; > + > + cpus_read_lock(); ^^^^^^^^^^^^^^^^ One thing which I do wonder is why we have locking here but not in the cti_pm_release() function. That was how the original code was so the patch doesn't change anything, but I am curious. > + ret = cpuhp_setup_state_nocalls_cpuslocked( > + CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, > + "arm/coresight_cti:starting", > + cti_starting_cpu, cti_dying_cpu); > + if (ret) { > + cpus_read_unlock(); > + return ret; > + } > + > + ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); > + cpus_read_unlock(); > + if (ret) { > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > + return ret; > + } > + > +done: > + nr_cti_cpu++; > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > + > + return 0; > +} > + > /* release PM registrations */ > static void cti_pm_release(struct cti_drvdata *drvdata) > { > - if (drvdata->ctidev.cpu >= 0) { > - if (--nr_cti_cpu == 0) { > - cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > + if (drvdata->ctidev.cpu == -1) > + return; > > - cpuhp_remove_state_nocalls( > - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > - } > - cti_cpu_drvdata[drvdata->ctidev.cpu] = NULL; > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > + if (--nr_cti_cpu == 0) { > + cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > } > } regards, dan carpenter
Hi Dan, Thanks for looking at this. I agree with the patch, other than the one change below. I have compiled and run on my DB410 system, against 5.8-rc1. On Fri, 12 Jun 2020 at 14:46, Dan Carpenter <dan.carpenter@oracle.com> wrote: > > There were a couple problems with error handling in the probe function: > 1) If the "drvdata" allocation failed then it lead to a NULL > dereference. > 2) On several error paths we decremented "nr_cti_cpu" before it was > incremented which lead to a reference counting bug. > > There were also some parts of the error handling which were not bugs but > were messy. The error handling was confusing to read. It printed some > unnecessary error messages. > > The simplest way to fix these problems was to create a cti_pm_setup() > function that did all the power management setup in one go. That way > when we call cti_pm_release() we don't have to deal with the > complications of a partially configured power management config. > > I reversed the "if (drvdata->ctidev.cpu >= 0)" condition in cti_pm_release() > so that it mirros the new cti_pm_setup() function. > > Fixes: 6a0953ce7de9 ("coresight: cti: Add CPU idle pm notifer to CTI devices") > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> > --- > Please note!!! I cannot compile this patch. Mike can you review it? > > drivers/hwtracing/coresight/coresight-cti.c | 96 ++++++++++++--------- > 1 file changed, 54 insertions(+), 42 deletions(-) > > diff --git a/drivers/hwtracing/coresight/coresight-cti.c b/drivers/hwtracing/coresight/coresight-cti.c > index 40387d58c8e7..d2da5bf9f552 100644 > --- a/drivers/hwtracing/coresight/coresight-cti.c > +++ b/drivers/hwtracing/coresight/coresight-cti.c > @@ -747,17 +747,50 @@ static int cti_dying_cpu(unsigned int cpu) > return 0; > } > > +static int cti_pm_setup(struct cti_drvdata *drvdata) > +{ > + int ret; > + > + if (drvdata->ctidev.cpu == -1) > + return 0; > + > + if (nr_cti_cpu) > + goto done; > + > + cpus_read_lock(); > + ret = cpuhp_setup_state_nocalls_cpuslocked( > + CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, > + "arm/coresight_cti:starting", > + cti_starting_cpu, cti_dying_cpu); > + if (ret) { > + cpus_read_unlock(); > + return ret; > + } > + > + ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); > + cpus_read_unlock(); > + if (ret) { > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > + return ret; > + } > + > +done: > + nr_cti_cpu++; > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > + > + return 0; > +} > + > /* release PM registrations */ > static void cti_pm_release(struct cti_drvdata *drvdata) > { > - if (drvdata->ctidev.cpu >= 0) { > - if (--nr_cti_cpu == 0) { > - cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > + if (drvdata->ctidev.cpu == -1) > + return; > > - cpuhp_remove_state_nocalls( > - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > - } > - cti_cpu_drvdata[drvdata->ctidev.cpu] = NULL; > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; This should remain as cti_cpu_drvdata[drvdata->ctidev.cpu] = NULL; here. We are reversing the assignment in cti_pm_setup(). > + if (--nr_cti_cpu == 0) { > + cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > } > } > > @@ -823,19 +856,14 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) > > /* driver data*/ > drvdata = devm_kzalloc(dev, sizeof(*drvdata), GFP_KERNEL); > - if (!drvdata) { > - ret = -ENOMEM; > - dev_info(dev, "%s, mem err\n", __func__); > - goto err_out; > - } > + if (!drvdata) > + return -ENOMEM; > > /* Validity for the resource is already checked by the AMBA core */ > base = devm_ioremap_resource(dev, res); > - if (IS_ERR(base)) { > - ret = PTR_ERR(base); > - dev_err(dev, "%s, remap err\n", __func__); > - goto err_out; > - } > + if (IS_ERR(base)) > + return PTR_ERR(base); > + > drvdata->base = base; > > dev_set_drvdata(dev, drvdata); > @@ -854,8 +882,7 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) > pdata = coresight_cti_get_platform_data(dev); > if (IS_ERR(pdata)) { > dev_err(dev, "coresight_cti_get_platform_data err\n"); > - ret = PTR_ERR(pdata); > - goto err_out; > + return PTR_ERR(pdata); > } > > /* default to powered - could change on PM notifications */ > @@ -867,35 +894,20 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) > drvdata->ctidev.cpu); > else > cti_desc.name = coresight_alloc_device_name(&cti_sys_devs, dev); > - if (!cti_desc.name) { > - ret = -ENOMEM; > - goto err_out; > - } > + if (!cti_desc.name) > + return -ENOMEM; > > /* setup CPU power management handling for CPU bound CTI devices. */ > - if (drvdata->ctidev.cpu >= 0) { > - cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > - if (!nr_cti_cpu++) { > - cpus_read_lock(); > - ret = cpuhp_setup_state_nocalls_cpuslocked( > - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, > - "arm/coresight_cti:starting", > - cti_starting_cpu, cti_dying_cpu); > - > - if (!ret) > - ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); > - cpus_read_unlock(); > - if (ret) > - goto err_out; > - } > - } > + ret = cti_pm_setup(drvdata); > + if (ret) > + return ret; > > /* create dynamic attributes for connections */ > ret = cti_create_cons_sysfs(dev, drvdata); > if (ret) { > dev_err(dev, "%s: create dynamic sysfs entries failed\n", > cti_desc.name); > - goto err_out; > + goto pm_release; > } > > /* set up coresight component description */ > @@ -908,7 +920,7 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) > drvdata->csdev = coresight_register(&cti_desc); > if (IS_ERR(drvdata->csdev)) { > ret = PTR_ERR(drvdata->csdev); > - goto err_out; > + goto pm_release; > } > > /* add to list of CTI devices */ > @@ -927,7 +939,7 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) > dev_info(&drvdata->csdev->dev, "CTI initialized\n"); > return 0; > > -err_out: > +pm_release: > cti_pm_release(drvdata); > return ret; > } > -- > 2.27.0 > Reviewed-by Mike Leach <mike.leach@linaro.org>
Hi Dan, On Fri, 12 Jun 2020 at 18:43, Dan Carpenter <dan.carpenter@oracle.com> wrote: > > On Fri, Jun 12, 2020 at 03:11:33PM +0300, Dan Carpenter wrote: > > +static int cti_pm_setup(struct cti_drvdata *drvdata) > > +{ > > + int ret; > > + > > + if (drvdata->ctidev.cpu == -1) > > + return 0; > > + > > + if (nr_cti_cpu) > > + goto done; > > + > > + cpus_read_lock(); > ^^^^^^^^^^^^^^^^ > One thing which I do wonder is why we have locking here but not in the > cti_pm_release() function. That was how the original code was so the > patch doesn't change anything, but I am curious. > Good point - the CTI PM code was modelled on the same code in the ETM drivers, which show the same pattern. Perhaps something we need to revisit in both drivers. Regards Mike > > + ret = cpuhp_setup_state_nocalls_cpuslocked( > > + CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, > > + "arm/coresight_cti:starting", > > + cti_starting_cpu, cti_dying_cpu); > > + if (ret) { > > + cpus_read_unlock(); > > + return ret; > > + } > > + > > + ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); > > + cpus_read_unlock(); > > + if (ret) { > > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > > + return ret; > > + } > > + > > +done: > > + nr_cti_cpu++; > > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > > + > > + return 0; > > +} > > + > > /* release PM registrations */ > > static void cti_pm_release(struct cti_drvdata *drvdata) > > { > > - if (drvdata->ctidev.cpu >= 0) { > > - if (--nr_cti_cpu == 0) { > > - cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > > + if (drvdata->ctidev.cpu == -1) > > + return; > > > > - cpuhp_remove_state_nocalls( > > - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > > - } > > - cti_cpu_drvdata[drvdata->ctidev.cpu] = NULL; > > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > > + if (--nr_cti_cpu == 0) { > > + cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > > } > > } > > regards, > dan carpenter >
HI Dan, Looked into this some more... On Wed, 17 Jun 2020 at 11:53, Mike Leach <mike.leach@linaro.org> wrote: > > Hi Dan, > > On Fri, 12 Jun 2020 at 18:43, Dan Carpenter <dan.carpenter@oracle.com> wrote: > > > > On Fri, Jun 12, 2020 at 03:11:33PM +0300, Dan Carpenter wrote: > > > +static int cti_pm_setup(struct cti_drvdata *drvdata) > > > +{ > > > + int ret; > > > + > > > + if (drvdata->ctidev.cpu == -1) > > > + return 0; > > > + > > > + if (nr_cti_cpu) > > > + goto done; > > > + > > > + cpus_read_lock(); > > ^^^^^^^^^^^^^^^^ > > One thing which I do wonder is why we have locking here but not in the > > cti_pm_release() function. That was how the original code was so the > > patch doesn't change anything, but I am curious. > > > > Good point - the CTI PM code was modelled on the same code in the ETM > drivers, which show the same pattern. > Perhaps something we need to revisit in both drivers. > The ETMv4 code calls into the hotplug API twice - so takes the lock and makes both calls while holding the lock - using the "_cpuslocked" call variant to render the pair of calls atomic from the CPUHP context point of view. CTI only calls once so does not really need to take the locks and could simply use the normal variant. In both cases the cpuhp_remove_state uses the normal variant, which takes the locks inside the api call. For the CTI there is certainly a case for simplification, i..e drop the "_cpuslocked" variant and remove the explicit taking of the locks. Something along the lines of.... ... if (nr_cti_cpu) goto done; ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); if (ret) return ret; ret = cpuhp_setup_state_nocalls(......); if (ret) { cpu_pm_unregister_notifier(....); return ret; } done: .... Regards Mike > Regards > > Mike > > > > + ret = cpuhp_setup_state_nocalls_cpuslocked( > > > + CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, > > > + "arm/coresight_cti:starting", > > > + cti_starting_cpu, cti_dying_cpu); > > > + if (ret) { > > > + cpus_read_unlock(); > > > + return ret; > > > + } > > > + > > > + ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); > > > + cpus_read_unlock(); > > > + if (ret) { > > > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > > > + return ret; > > > + } > > > + > > > +done: > > > + nr_cti_cpu++; > > > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > > > + > > > + return 0; > > > +} > > > + > > > /* release PM registrations */ > > > static void cti_pm_release(struct cti_drvdata *drvdata) > > > { > > > - if (drvdata->ctidev.cpu >= 0) { > > > - if (--nr_cti_cpu == 0) { > > > - cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > > > + if (drvdata->ctidev.cpu == -1) > > > + return; > > > > > > - cpuhp_remove_state_nocalls( > > > - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > > > - } > > > - cti_cpu_drvdata[drvdata->ctidev.cpu] = NULL; > > > + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; > > > + if (--nr_cti_cpu == 0) { > > > + cpu_pm_unregister_notifier(&cti_cpu_pm_nb); > > > + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); > > > } > > > } > > > > regards, > > dan carpenter > > > > > -- > Mike Leach > Principal Engineer, ARM Ltd. > Manchester Design Centre. UK
diff --git a/drivers/hwtracing/coresight/coresight-cti.c b/drivers/hwtracing/coresight/coresight-cti.c index 40387d58c8e7..d2da5bf9f552 100644 --- a/drivers/hwtracing/coresight/coresight-cti.c +++ b/drivers/hwtracing/coresight/coresight-cti.c @@ -747,17 +747,50 @@ static int cti_dying_cpu(unsigned int cpu) return 0; } +static int cti_pm_setup(struct cti_drvdata *drvdata) +{ + int ret; + + if (drvdata->ctidev.cpu == -1) + return 0; + + if (nr_cti_cpu) + goto done; + + cpus_read_lock(); + ret = cpuhp_setup_state_nocalls_cpuslocked( + CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, + "arm/coresight_cti:starting", + cti_starting_cpu, cti_dying_cpu); + if (ret) { + cpus_read_unlock(); + return ret; + } + + ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); + cpus_read_unlock(); + if (ret) { + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); + return ret; + } + +done: + nr_cti_cpu++; + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; + + return 0; +} + /* release PM registrations */ static void cti_pm_release(struct cti_drvdata *drvdata) { - if (drvdata->ctidev.cpu >= 0) { - if (--nr_cti_cpu == 0) { - cpu_pm_unregister_notifier(&cti_cpu_pm_nb); + if (drvdata->ctidev.cpu == -1) + return; - cpuhp_remove_state_nocalls( - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); - } - cti_cpu_drvdata[drvdata->ctidev.cpu] = NULL; + cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; + if (--nr_cti_cpu == 0) { + cpu_pm_unregister_notifier(&cti_cpu_pm_nb); + cpuhp_remove_state_nocalls(CPUHP_AP_ARM_CORESIGHT_CTI_STARTING); } } @@ -823,19 +856,14 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) /* driver data*/ drvdata = devm_kzalloc(dev, sizeof(*drvdata), GFP_KERNEL); - if (!drvdata) { - ret = -ENOMEM; - dev_info(dev, "%s, mem err\n", __func__); - goto err_out; - } + if (!drvdata) + return -ENOMEM; /* Validity for the resource is already checked by the AMBA core */ base = devm_ioremap_resource(dev, res); - if (IS_ERR(base)) { - ret = PTR_ERR(base); - dev_err(dev, "%s, remap err\n", __func__); - goto err_out; - } + if (IS_ERR(base)) + return PTR_ERR(base); + drvdata->base = base; dev_set_drvdata(dev, drvdata); @@ -854,8 +882,7 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) pdata = coresight_cti_get_platform_data(dev); if (IS_ERR(pdata)) { dev_err(dev, "coresight_cti_get_platform_data err\n"); - ret = PTR_ERR(pdata); - goto err_out; + return PTR_ERR(pdata); } /* default to powered - could change on PM notifications */ @@ -867,35 +894,20 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) drvdata->ctidev.cpu); else cti_desc.name = coresight_alloc_device_name(&cti_sys_devs, dev); - if (!cti_desc.name) { - ret = -ENOMEM; - goto err_out; - } + if (!cti_desc.name) + return -ENOMEM; /* setup CPU power management handling for CPU bound CTI devices. */ - if (drvdata->ctidev.cpu >= 0) { - cti_cpu_drvdata[drvdata->ctidev.cpu] = drvdata; - if (!nr_cti_cpu++) { - cpus_read_lock(); - ret = cpuhp_setup_state_nocalls_cpuslocked( - CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, - "arm/coresight_cti:starting", - cti_starting_cpu, cti_dying_cpu); - - if (!ret) - ret = cpu_pm_register_notifier(&cti_cpu_pm_nb); - cpus_read_unlock(); - if (ret) - goto err_out; - } - } + ret = cti_pm_setup(drvdata); + if (ret) + return ret; /* create dynamic attributes for connections */ ret = cti_create_cons_sysfs(dev, drvdata); if (ret) { dev_err(dev, "%s: create dynamic sysfs entries failed\n", cti_desc.name); - goto err_out; + goto pm_release; } /* set up coresight component description */ @@ -908,7 +920,7 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) drvdata->csdev = coresight_register(&cti_desc); if (IS_ERR(drvdata->csdev)) { ret = PTR_ERR(drvdata->csdev); - goto err_out; + goto pm_release; } /* add to list of CTI devices */ @@ -927,7 +939,7 @@ static int cti_probe(struct amba_device *adev, const struct amba_id *id) dev_info(&drvdata->csdev->dev, "CTI initialized\n"); return 0; -err_out: +pm_release: cti_pm_release(drvdata); return ret; }
There were a couple problems with error handling in the probe function: 1) If the "drvdata" allocation failed then it lead to a NULL dereference. 2) On several error paths we decremented "nr_cti_cpu" before it was incremented which lead to a reference counting bug. There were also some parts of the error handling which were not bugs but were messy. The error handling was confusing to read. It printed some unnecessary error messages. The simplest way to fix these problems was to create a cti_pm_setup() function that did all the power management setup in one go. That way when we call cti_pm_release() we don't have to deal with the complications of a partially configured power management config. I reversed the "if (drvdata->ctidev.cpu >= 0)" condition in cti_pm_release() so that it mirros the new cti_pm_setup() function. Fixes: 6a0953ce7de9 ("coresight: cti: Add CPU idle pm notifer to CTI devices") Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> --- Please note!!! I cannot compile this patch. Mike can you review it? drivers/hwtracing/coresight/coresight-cti.c | 96 ++++++++++++--------- 1 file changed, 54 insertions(+), 42 deletions(-)