This should be a slight improvement on Jinlong's previous version. Now it's not possible to trigger the error message from pm_runtime_put() by calling disable twice.
It's also similar to the original pre-breaking change version where pm_runtime_put() was only called if the device was actually disabled, but with one difference: Previously pm_runtime_put() was only called once for the last disable call, but because of the reference counting in pm_runtime, it should have been called once for each enable call. This meant that the clock would have never been disabled if there were ever multiple enable calls. This is now fixed.
The third commit is a refactor and doesn't need to be backported. I removed one of the atomic types because it didn't appear to be required. Maybe it was added for a reason which I'm not aware of, if so it should be pretty easy to drop that change.
James Clark (2): coresight: cti: Prevent negative values of enable count coresight: cti: Remove atomic type from enable_req_count
Mao Jinlong (1): coresight: cti: Add PM runtime call in enable_store
.../hwtracing/coresight/coresight-cti-core.c | 23 ++++++++++++------- .../hwtracing/coresight/coresight-cti-sysfs.c | 15 +++++++++--- drivers/hwtracing/coresight/coresight-cti.h | 2 +- 3 files changed, 28 insertions(+), 12 deletions(-)
base-commit: c767c34740132ffc478226864a7461493cdc2413
Writing 0 to the enable control repeatedly results in a negative value for enable_req_count. After this, writing 1 to the enable control appears to not work until the count returns to positive.
Change it so that it's impossible for enable_req_count to be < 0. Returning an error will also allow us to decide whether to call runtime_pm_put() or not in the following commit.
Signed-off-by: James Clark james.clark@arm.com --- drivers/hwtracing/coresight/coresight-cti-core.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-core.c b/drivers/hwtracing/coresight/coresight-cti-core.c index d2cf4f4848e1..838872f2484d 100644 --- a/drivers/hwtracing/coresight/coresight-cti-core.c +++ b/drivers/hwtracing/coresight/coresight-cti-core.c @@ -151,9 +151,16 @@ static int cti_disable_hw(struct cti_drvdata *drvdata) { struct cti_config *config = &drvdata->config; struct coresight_device *csdev = drvdata->csdev; + int ret = 0;
spin_lock(&drvdata->spinlock);
+ /* don't allow negative refcounts, return an error */ + if (!atomic_read(&drvdata->config.enable_req_count)) { + ret = -EINVAL; + goto cti_not_disabled; + } + /* check refcount - disable on 0 */ if (atomic_dec_return(&drvdata->config.enable_req_count) > 0) goto cti_not_disabled; @@ -171,12 +178,12 @@ static int cti_disable_hw(struct cti_drvdata *drvdata) coresight_disclaim_device_unlocked(csdev); CS_LOCK(drvdata->base); spin_unlock(&drvdata->spinlock); - return 0; + return ret;
/* not disabled this call */ cti_not_disabled: spin_unlock(&drvdata->spinlock); - return 0; + return ret; }
void cti_write_single_reg(struct cti_drvdata *drvdata, int offset, u32 value)
Hi James,
Thanks for fixing this.
On 06/01/2023 15:23, James Clark wrote:
Writing 0 to the enable control repeatedly results in a negative value for enable_req_count. After this, writing 1 to the enable control appears to not work until the count returns to positive.
Change it so that it's impossible for enable_req_count to be < 0.
I prefer rephrasing the following :
Returning an error will also allow us to decide whether to call runtime_pm_put() or not in the following commit.
"Return an error to indicate the disable request was invalid"
and don't mention the forward dependency.
May be also add ?
Fixes: 835d722ba10a ("coresight: cti: Initial CoreSight CTI Driver")
Signed-off-by: James Clark james.clark@arm.com
Rest looks fine to me.
Suzuki
drivers/hwtracing/coresight/coresight-cti-core.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-core.c b/drivers/hwtracing/coresight/coresight-cti-core.c index d2cf4f4848e1..838872f2484d 100644 --- a/drivers/hwtracing/coresight/coresight-cti-core.c +++ b/drivers/hwtracing/coresight/coresight-cti-core.c @@ -151,9 +151,16 @@ static int cti_disable_hw(struct cti_drvdata *drvdata) { struct cti_config *config = &drvdata->config; struct coresight_device *csdev = drvdata->csdev;
- int ret = 0;
spin_lock(&drvdata->spinlock);
- /* don't allow negative refcounts, return an error */
- if (!atomic_read(&drvdata->config.enable_req_count)) {
ret = -EINVAL;goto cti_not_disabled;- }
- /* check refcount - disable on 0 */ if (atomic_dec_return(&drvdata->config.enable_req_count) > 0) goto cti_not_disabled;
@@ -171,12 +178,12 @@ static int cti_disable_hw(struct cti_drvdata *drvdata) coresight_disclaim_device_unlocked(csdev); CS_LOCK(drvdata->base); spin_unlock(&drvdata->spinlock);
- return 0;
- return ret;
/* not disabled this call */ cti_not_disabled: spin_unlock(&drvdata->spinlock);
- return 0;
- return ret; }
void cti_write_single_reg(struct cti_drvdata *drvdata, int offset, u32 value)
From: Mao Jinlong quic_jinlmao@quicinc.com
In commit 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") PM runtime calls are removed from cti_enable_hw/cti_disable_hw. When enabling CTI by writing enable sysfs node, clock for accessing CTI register won't be enabled. Device will crash due to register access issue. Add PM runtime call in enable_store to fix this issue.
Fixes: 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com [Change to only call pm_runtime_put if a disable happened] Signed-off-by: James Clark james.clark@arm.com --- drivers/hwtracing/coresight/coresight-cti-sysfs.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-sysfs.c b/drivers/hwtracing/coresight/coresight-cti-sysfs.c index 6d59c815ecf5..71e7a8266bb3 100644 --- a/drivers/hwtracing/coresight/coresight-cti-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-cti-sysfs.c @@ -108,10 +108,19 @@ static ssize_t enable_store(struct device *dev, if (ret) return ret;
- if (val) + if (val) { + ret = pm_runtime_resume_and_get(dev->parent); + if (ret) + return ret; ret = cti_enable(drvdata->csdev); - else + if (ret) + pm_runtime_put(dev->parent); + } else { ret = cti_disable(drvdata->csdev); + if (!ret) + pm_runtime_put(dev->parent); + } + if (ret) return ret; return size;
On 06/01/2023 15:23, James Clark wrote:
From: Mao Jinlong quic_jinlmao@quicinc.com
In commit 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") PM runtime calls are removed from cti_enable_hw/cti_disable_hw. When enabling CTI by writing enable sysfs node, clock for accessing CTI register won't be enabled. Device will crash due to register access issue. Add PM runtime call in enable_store to fix this issue.
Fixes: 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com [Change to only call pm_runtime_put if a disable happened] Signed-off-by: James Clark james.clark@arm.com
drivers/hwtracing/coresight/coresight-cti-sysfs.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-sysfs.c b/drivers/hwtracing/coresight/coresight-cti-sysfs.c index 6d59c815ecf5..71e7a8266bb3 100644 --- a/drivers/hwtracing/coresight/coresight-cti-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-cti-sysfs.c @@ -108,10 +108,19 @@ static ssize_t enable_store(struct device *dev, if (ret) return ret;
- if (val)
- if (val) {
ret = pm_runtime_resume_and_get(dev->parent);if (ret)return ret;
- else
if (ret)pm_runtime_put(dev->parent);- } else { ret = cti_disable(drvdata->csdev);
if (!ret)pm_runtime_put(dev->parent);- }
- if (ret) return ret; return size;
Looks good to me.
@Mao Jinolong,
Are you able to test the patches 1 & 2 and confirm they solve your issue ?
Suzuki
On 1/10/2023 12:47 AM, Suzuki K Poulose wrote:
On 06/01/2023 15:23, James Clark wrote:
From: Mao Jinlong quic_jinlmao@quicinc.com
In commit 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") PM runtime calls are removed from cti_enable_hw/cti_disable_hw. When enabling CTI by writing enable sysfs node, clock for accessing CTI register won't be enabled. Device will crash due to register access issue. Add PM runtime call in enable_store to fix this issue.
Fixes: 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com [Change to only call pm_runtime_put if a disable happened] Signed-off-by: James Clark james.clark@arm.com
drivers/hwtracing/coresight/coresight-cti-sysfs.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-sysfs.c b/drivers/hwtracing/coresight/coresight-cti-sysfs.c index 6d59c815ecf5..71e7a8266bb3 100644 --- a/drivers/hwtracing/coresight/coresight-cti-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-cti-sysfs.c @@ -108,10 +108,19 @@ static ssize_t enable_store(struct device *dev, if (ret) return ret; - if (val) + if (val) { + ret = pm_runtime_resume_and_get(dev->parent); + if (ret) + return ret; ret = cti_enable(drvdata->csdev); - else + if (ret) + pm_runtime_put(dev->parent); + } else { ret = cti_disable(drvdata->csdev); + if (!ret) + pm_runtime_put(dev->parent); + }
if (ret) return ret; return size;
Looks good to me.
@Mao Jinolong,
Are you able to test the patches 1 & 2 and confirm they solve your issue ?
Hi Suzuki,
Tested from my side. Patch 1 & 2 can solve the issue when enable CTI by writing 1 to enable sysfs node.
Thanks Jinlong Mao
Suzuki
On 10/01/2023 05:56, Jinlong Mao wrote:
On 1/10/2023 12:47 AM, Suzuki K Poulose wrote:
On 06/01/2023 15:23, James Clark wrote:
From: Mao Jinlong quic_jinlmao@quicinc.com
In commit 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") PM runtime calls are removed from cti_enable_hw/cti_disable_hw. When enabling CTI by writing enable sysfs node, clock for accessing CTI register won't be enabled. Device will crash due to register access issue. Add PM runtime call in enable_store to fix this issue.
Fixes: 6746eae4bbad ("coresight: cti: Fix hang in cti_disable_hw()") Signed-off-by: Mao Jinlong quic_jinlmao@quicinc.com [Change to only call pm_runtime_put if a disable happened] Signed-off-by: James Clark james.clark@arm.com
drivers/hwtracing/coresight/coresight-cti-sysfs.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-sysfs.c b/drivers/hwtracing/coresight/coresight-cti-sysfs.c index 6d59c815ecf5..71e7a8266bb3 100644 --- a/drivers/hwtracing/coresight/coresight-cti-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-cti-sysfs.c @@ -108,10 +108,19 @@ static ssize_t enable_store(struct device *dev, if (ret) return ret; - if (val) + if (val) { + ret = pm_runtime_resume_and_get(dev->parent); + if (ret) + return ret; ret = cti_enable(drvdata->csdev); - else + if (ret) + pm_runtime_put(dev->parent); + } else { ret = cti_disable(drvdata->csdev); + if (!ret) + pm_runtime_put(dev->parent); + }
if (ret) return ret; return size;
Looks good to me.
@Mao Jinolong,
Are you able to test the patches 1 & 2 and confirm they solve your issue ?
Hi Suzuki,
Tested from my side. Patch 1 & 2 can solve the issue when enable CTI by writing 1 to enable sysfs node.
Thanks, I added your tested-by tag on patches 1 + 2.
enable_req_count is only ever accessed inside the spinlock, so to avoid confusion that there are concurrent accesses and simplify the code, change it to an int.
One access outside of the spinlock is in enable_show() which appears to allow partially written data to be displayed between enable_req_count, powered and enabled so move this one inside the spin lock too.
Signed-off-by: James Clark james.clark@arm.com --- drivers/hwtracing/coresight/coresight-cti-core.c | 14 +++++++------- drivers/hwtracing/coresight/coresight-cti-sysfs.c | 2 +- drivers/hwtracing/coresight/coresight-cti.h | 2 +- 3 files changed, 9 insertions(+), 9 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-cti-core.c b/drivers/hwtracing/coresight/coresight-cti-core.c index 838872f2484d..277c890a1f1f 100644 --- a/drivers/hwtracing/coresight/coresight-cti-core.c +++ b/drivers/hwtracing/coresight/coresight-cti-core.c @@ -107,12 +107,12 @@ static int cti_enable_hw(struct cti_drvdata *drvdata) cti_write_all_hw_regs(drvdata);
config->hw_enabled = true; - atomic_inc(&drvdata->config.enable_req_count); + drvdata->config.enable_req_count++; spin_unlock_irqrestore(&drvdata->spinlock, flags); return rc;
cti_state_unchanged: - atomic_inc(&drvdata->config.enable_req_count); + drvdata->config.enable_req_count++;
/* cannot enable due to error */ cti_err_not_enabled: @@ -129,7 +129,7 @@ static void cti_cpuhp_enable_hw(struct cti_drvdata *drvdata) config->hw_powered = true;
/* no need to do anything if no enable request */ - if (!atomic_read(&drvdata->config.enable_req_count)) + if (!drvdata->config.enable_req_count) goto cti_hp_not_enabled;
/* try to claim the device */ @@ -156,13 +156,13 @@ static int cti_disable_hw(struct cti_drvdata *drvdata) spin_lock(&drvdata->spinlock);
/* don't allow negative refcounts, return an error */ - if (!atomic_read(&drvdata->config.enable_req_count)) { + if (!drvdata->config.enable_req_count) { ret = -EINVAL; goto cti_not_disabled; }
/* check refcount - disable on 0 */ - if (atomic_dec_return(&drvdata->config.enable_req_count) > 0) + if (--drvdata->config.enable_req_count > 0) goto cti_not_disabled;
/* no need to do anything if disabled or cpu unpowered */ @@ -239,7 +239,7 @@ static void cti_set_default_config(struct device *dev, /* Most regs default to 0 as zalloc'ed except...*/ config->trig_filter_enable = true; config->ctigate = GENMASK(config->nr_ctm_channels - 1, 0); - atomic_set(&config->enable_req_count, 0); + config->enable_req_count = 0; }
/* @@ -696,7 +696,7 @@ static int cti_cpu_pm_notify(struct notifier_block *nb, unsigned long cmd, drvdata->config.hw_enabled = false;
/* check enable reference count to enable HW */ - if (atomic_read(&drvdata->config.enable_req_count)) { + if (drvdata->config.enable_req_count) { /* check we can claim the device as we re-power */ if (coresight_claim_device(csdev)) goto cti_notify_exit; diff --git a/drivers/hwtracing/coresight/coresight-cti-sysfs.c b/drivers/hwtracing/coresight/coresight-cti-sysfs.c index 71e7a8266bb3..e528cff9d4e2 100644 --- a/drivers/hwtracing/coresight/coresight-cti-sysfs.c +++ b/drivers/hwtracing/coresight/coresight-cti-sysfs.c @@ -84,8 +84,8 @@ static ssize_t enable_show(struct device *dev, bool enabled, powered; struct cti_drvdata *drvdata = dev_get_drvdata(dev->parent);
- enable_req = atomic_read(&drvdata->config.enable_req_count); spin_lock(&drvdata->spinlock); + enable_req = drvdata->config.enable_req_count; powered = drvdata->config.hw_powered; enabled = drvdata->config.hw_enabled; spin_unlock(&drvdata->spinlock); diff --git a/drivers/hwtracing/coresight/coresight-cti.h b/drivers/hwtracing/coresight/coresight-cti.h index acf7b545e6b9..8b106b13a244 100644 --- a/drivers/hwtracing/coresight/coresight-cti.h +++ b/drivers/hwtracing/coresight/coresight-cti.h @@ -141,7 +141,7 @@ struct cti_config { int nr_trig_max;
/* cti enable control */ - atomic_t enable_req_count; + int enable_req_count; bool hw_enabled; bool hw_powered;
-
James Clark -
Jinlong Mao -
Suzuki K Poulose