Skip to contentSkip to conversation

coresight: tmc: Read TMC mode only when TMC hw is enabled

PATCH In review Sai Prakash Ranjan 4 participants 9 messages Apr 9, 2020 – Apr 16, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 coresight: tmc: Read TMC mode only when TMC hw is enabled

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

 Kernel panic - not syncing: Asynchronous SError Interrupt
 CPU: 7 PID: 2605 Comm: hexdump Tainted: G S                5.4.30 #122
 Call trace:
  dump_backtrace+0x0/0x188
  show_stack+0x20/0x2c
  dump_stack+0xdc/0x144
  panic+0x168/0x36c
  panic+0x0/0x36c
  arm64_serror_panic+0x78/0x84
  do_serror+0x130/0x138
  el1_error+0x84/0xf8
  tmc_read_prepare_etb+0x88/0xb8
  tmc_open+0x40/0xd8
  misc_open+0x120/0x158
  chrdev_open+0xb8/0x1a4
  do_dentry_open+0x268/0x3a0
  vfs_open+0x34/0x40
  path_openat+0x39c/0xdf4
  do_filp_open+0x90/0x10c
  do_sys_open+0x150/0x3e8
  __arm64_compat_sys_openat+0x28/0x34
  el0_svc_common+0xa8/0x160
  el0_svc_compat_handler+0x2c/0x38
  el0_svc_compat+0x8/0x10
Fixes 4525412a5046 ("coresight: tmc: making prepare/unprepare functions generic")
---
 drivers/hwtracing/coresight/coresight-tmc.c | 5 +++++
 drivers/hwtracing/coresight/coresight-tmc.h | 1 +
 2 files changed, 6 insertions(+)
drivers/hwtracing/coresight/coresight-tmc.c +5 -0
index 1cf82fa58289..7bae69748ab7 100644
@@ -62,11 +62,13 @@ void tmc_flush_and_stop(struct tmc_drvdata *drvdata)
 
 void tmc_enable_hw(struct tmc_drvdata *drvdata)
 {
+	drvdata->enable = true;
 	writel_relaxed(TMC_CTL_CAPT_EN, drvdata->base + TMC_CTL);
 }
 
 void tmc_disable_hw(struct tmc_drvdata *drvdata)
 {
+	drvdata->enable = false;
 	writel_relaxed(0x0, drvdata->base + TMC_CTL);
 }
 
@@ -102,6 +104,9 @@ static int tmc_read_prepare(struct tmc_drvdata *drvdata)
 {
 	int ret = 0;
 
+	if (!drvdata->enable)
+		return -EINVAL;
+
 	switch (drvdata->config_type) {
 	case TMC_CONFIG_TYPE_ETB:
 	case TMC_CONFIG_TYPE_ETF:
drivers/hwtracing/coresight/coresight-tmc.h +1 -0
index 71de978575f3..7c8712a6ed8d 100644
@@ -191,6 +191,7 @@ struct tmc_drvdata {
 	struct miscdevice	miscdev;
 	spinlock_t		spinlock;
 	pid_t			pid;
+	bool			enable;
 	bool			reading;
 	union {
 		char		*buf;		/* TMC ETB */
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
Tested-by on v1 1/1
#1 lore raw

Quoting Sai Prakash Ranjan (2020-04-09 04:35:38)

Quoted 33 lines

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

 Kernel panic - not syncing: Asynchronous SError Interrupt
 CPU: 7 PID: 2605 Comm: hexdump Tainted: G S                5.4.30 #122
 Call trace:
  dump_backtrace+0x0/0x188
  show_stack+0x20/0x2c
  dump_stack+0xdc/0x144
  panic+0x168/0x36c
  panic+0x0/0x36c
  arm64_serror_panic+0x78/0x84
  do_serror+0x130/0x138
  el1_error+0x84/0xf8
  tmc_read_prepare_etb+0x88/0xb8
  tmc_open+0x40/0xd8
  misc_open+0x120/0x158
  chrdev_open+0xb8/0x1a4
  do_dentry_open+0x268/0x3a0
  vfs_open+0x34/0x40
  path_openat+0x39c/0xdf4
  do_filp_open+0x90/0x10c
  do_sys_open+0x150/0x3e8
  __arm64_compat_sys_openat+0x28/0x34
  el0_svc_common+0xa8/0x160
  el0_svc_compat_handler+0x2c/0x38
  el0_svc_compat+0x8/0x10
Quoted 4 lines
Fixes 4525412a5046 ("coresight: tmc: making prepare/unprepare functions generic")

---

maintainer
#2 lore raw
In reply to Sai Prakash Ranjan

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

And also :

Quoted 53 lines
  Kernel panic - not syncing: Asynchronous SError Interrupt
  CPU: 7 PID: 2605 Comm: hexdump Tainted: G S                5.4.30 #122
  Call trace:
   dump_backtrace+0x0/0x188
   show_stack+0x20/0x2c
   dump_stack+0xdc/0x144
   panic+0x168/0x36c
   panic+0x0/0x36c
   arm64_serror_panic+0x78/0x84
   do_serror+0x130/0x138
   el1_error+0x84/0xf8
   tmc_read_prepare_etb+0x88/0xb8
   tmc_open+0x40/0xd8
   misc_open+0x120/0x158
   chrdev_open+0xb8/0x1a4
   do_dentry_open+0x268/0x3a0
   vfs_open+0x34/0x40
   path_openat+0x39c/0xdf4
   do_filp_open+0x90/0x10c
   do_sys_open+0x150/0x3e8
   __arm64_compat_sys_openat+0x28/0x34
   el0_svc_common+0xa8/0x160
   el0_svc_compat_handler+0x2c/0x38
   el0_svc_compat+0x8/0x10
Fixes 4525412a5046 ("coresight: tmc: making prepare/unprepare functions generic")
---
  drivers/hwtracing/coresight/coresight-tmc.c | 5 +++++
  drivers/hwtracing/coresight/coresight-tmc.h | 1 +
  2 files changed, 6 insertions(+)
drivers/hwtracing/coresight/coresight-tmc.c +2 -0
index 1cf82fa58289..7bae69748ab7 100644
@@ -62,11 +62,13 @@ void tmc_flush_and_stop(struct tmc_drvdata *drvdata)
 
  void tmc_enable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = true;
  	writel_relaxed(TMC_CTL_CAPT_EN, drvdata->base + TMC_CTL);
  }
 
  void tmc_disable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = false;
  	writel_relaxed(0x0, drvdata->base + TMC_CTL);
  }
 

drivers/hwtracing/coresight/coresight-tmc.c static int tmc_read_prepare(struct tmc_drvdata *drvdata) · line 109

Quoted 6 lines
drivers/hwtracing/coresight/coresight-tmc.c +3 -0
@@ -102,6 +104,9 @@ static int tmc_read_prepare(struct tmc_drvdata *drvdata)
  {
  	int ret = 0;
 
+	if (!drvdata->enable)
+		return -EINVAL;
+

Does this check always guarantee that the TMC is enabled when we actually get to reading the MODE ? This needs to be done under the spinlock.

Cheers
Suzuki

#3 lore raw

Hi Suzuki,

Quoted 11 lines · Suzuki K Poulose

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

Quoted 4 lines

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

I do not see any SError when I run the above command.

localhost ~ # cat /sys/bus/coresight/devices/tmc_etf0/mgmt/mode 0x0

And this is most likely due to

commit cd9e3474bb793dc ("coresight: add PM runtime calls to coresight_simple_func()")

Quoted 56 lines

And also :

Quoted 54 lines
  Kernel panic - not syncing: Asynchronous SError Interrupt
  CPU: 7 PID: 2605 Comm: hexdump Tainted: G S                5.4.30 
#122
  Call trace:
   dump_backtrace+0x0/0x188
   show_stack+0x20/0x2c
   dump_stack+0xdc/0x144
   panic+0x168/0x36c
   panic+0x0/0x36c
   arm64_serror_panic+0x78/0x84
   do_serror+0x130/0x138
   el1_error+0x84/0xf8
   tmc_read_prepare_etb+0x88/0xb8
   tmc_open+0x40/0xd8
   misc_open+0x120/0x158
   chrdev_open+0xb8/0x1a4
   do_dentry_open+0x268/0x3a0
   vfs_open+0x34/0x40
   path_openat+0x39c/0xdf4
   do_filp_open+0x90/0x10c
   do_sys_open+0x150/0x3e8
   __arm64_compat_sys_openat+0x28/0x34
   el0_svc_common+0xa8/0x160
   el0_svc_compat_handler+0x2c/0x38
   el0_svc_compat+0x8/0x10
Fixes 4525412a5046 ("coresight: tmc: making prepare/unprepare functions generic")
---
  drivers/hwtracing/coresight/coresight-tmc.c | 5 +++++
  drivers/hwtracing/coresight/coresight-tmc.h | 1 +
  2 files changed, 6 insertions(+)
drivers/hwtracing/coresight/coresight-tmc.c +2 -0
index 1cf82fa58289..7bae69748ab7 100644
@@ -62,11 +62,13 @@ void tmc_flush_and_stop(struct tmc_drvdata 
 *drvdata)
    void tmc_enable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = true;
  	writel_relaxed(TMC_CTL_CAPT_EN, drvdata->base + TMC_CTL);
  }
    void tmc_disable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = false;
  	writel_relaxed(0x0, drvdata->base + TMC_CTL);
  }
  @@ -102,6 +104,9 @@ static int tmc_read_prepare(struct tmc_drvdata 
 *drvdata)
  {
Quoted 9 lines

drivers/hwtracing/coresight/coresight-tmc.c static int tmc_read_prepare(struct tmc_drvdata *drvdata) · line 109

Quoted 4 lines
  	int ret = 0;
  +	if (!drvdata->enable)
drivers/hwtracing/coresight/coresight-tmc.c +2 -0
+		return -EINVAL;
+

Does this check always guarantee that the TMC is enabled when we actually get to reading the MODE ? This needs to be done under the spinlock.

Ok I will make this change.

Thanks,
Sai

Signature · 7 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#4 lore raw
Quoted 21 lines · Sai Prakash Ranjan

Hi Suzuki,

Quoted 11 lines · Suzuki K Poulose

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

Quoted 4 lines

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

I do not see any SError when I run the above command.

Quoted 7 lines

localhost ~ # cat /sys/bus/coresight/devices/tmc_etf0/mgmt/mode 0x0

And this is most likely due to

commit cd9e3474bb793dc ("coresight: add PM runtime calls to coresight_simple_func()")

Ok, so this is related to power management (you can ignore my question in the previous email).

Regarding function tmc_read_prepare_etb(), the best way to deal with this is probably make sure drvdata->mode != CS_MODE_DISABLED before reading TMC_MODE. If there is a buffer to read it will have been copied when the ETB was disabled and there won't be a need to access the HW.

Mathieu

Quoted 77 lines
Quoted 59 lines

And also :

Quoted 56 lines
  Kernel panic - not syncing: Asynchronous SError Interrupt
  CPU: 7 PID: 2605 Comm: hexdump Tainted: G S                5.4.30
#122
  Call trace:
   dump_backtrace+0x0/0x188
   show_stack+0x20/0x2c
   dump_stack+0xdc/0x144
   panic+0x168/0x36c
   panic+0x0/0x36c
   arm64_serror_panic+0x78/0x84
   do_serror+0x130/0x138
   el1_error+0x84/0xf8
   tmc_read_prepare_etb+0x88/0xb8
   tmc_open+0x40/0xd8
   misc_open+0x120/0x158
   chrdev_open+0xb8/0x1a4
   do_dentry_open+0x268/0x3a0
   vfs_open+0x34/0x40
   path_openat+0x39c/0xdf4
   do_filp_open+0x90/0x10c
   do_sys_open+0x150/0x3e8
   __arm64_compat_sys_openat+0x28/0x34
   el0_svc_common+0xa8/0x160
   el0_svc_compat_handler+0x2c/0x38
   el0_svc_compat+0x8/0x10
Fixes 4525412a5046 ("coresight: tmc: making prepare/unprepare

functions generic")

---
  drivers/hwtracing/coresight/coresight-tmc.c | 5 +++++
  drivers/hwtracing/coresight/coresight-tmc.h | 1 +
  2 files changed, 6 insertions(+)
drivers/hwtracing/coresight/coresight-tmc.c +2 -0
index 1cf82fa58289..7bae69748ab7 100644
@@ -62,11 +62,13 @@ void tmc_flush_and_stop(struct tmc_drvdata
 *drvdata)
    void tmc_enable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = true;
  	writel_relaxed(TMC_CTL_CAPT_EN, drvdata->base + TMC_CTL);
  }
    void tmc_disable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = false;
  	writel_relaxed(0x0, drvdata->base + TMC_CTL);
  }
  @@ -102,6 +104,9 @@ static int tmc_read_prepare(struct tmc_drvdata
 *drvdata)
  {
Quoted 9 lines

drivers/hwtracing/coresight/coresight-tmc.c static int tmc_read_prepare(struct tmc_drvdata *drvdata) · line 109

Quoted 4 lines
  	int ret = 0;
  +	if (!drvdata->enable)
drivers/hwtracing/coresight/coresight-tmc.c +2 -0
+		return -EINVAL;
+

Does this check always guarantee that the TMC is enabled when we actually get to reading the MODE ? This needs to be done under the spinlock.

Ok I will make this change.

Thanks,
Sai

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#5 lore raw

Hi Mathieu,

Quoted 30 lines · Mathieu Poirier
Quoted 21 lines · Sai Prakash Ranjan

Hi Suzuki,

Quoted 11 lines · Suzuki K Poulose

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

Quoted 4 lines

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

I do not see any SError when I run the above command.

Quoted 7 lines

localhost ~ # cat /sys/bus/coresight/devices/tmc_etf0/mgmt/mode 0x0

And this is most likely due to

commit cd9e3474bb793dc ("coresight: add PM runtime calls to coresight_simple_func()")

Quoted 8 lines

Ok, so this is related to power management (you can ignore my question in the previous email).

Regarding function tmc_read_prepare_etb(), the best way to deal with this is probably make sure drvdata->mode != CS_MODE_DISABLED before reading TMC_MODE. If there is a buffer to read it will have been copied when the ETB was disabled and there won't be a need to access the HW.

This works as well, thanks.

drivers/hwtracing/coresight/coresight-tmc-etf.c +5 -0
index d0cc3985b72a..7ffe05930984 100644
@@ -596,6 +596,11 @@ int tmc_read_prepare_etb(struct tmc_drvdata *drvdata)
                goto out;
        }
 
+       if (drvdata->mode == CS_MODE_DISABLED) {
+               ret = -EINVAL;
+               goto out;
+       }
+
        /* There is no point in reading a TMC in HW FIFO mode */
        mode = readl_relaxed(drvdata->base + TMC_MODE);
        if (mode != TMC_MODE_CIRCULAR_BUFFER) {

Thanks,
Sai

Signature · 7 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#6 lore raw
Quoted 55 lines · Sai Prakash Ranjan

Hi Mathieu,

Quoted 34 lines · Mathieu Poirier
Quoted 21 lines · Sai Prakash Ranjan

Hi Suzuki,

Quoted 11 lines · Suzuki K Poulose

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

Quoted 4 lines

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

I do not see any SError when I run the above command.

Quoted 7 lines

localhost ~ # cat /sys/bus/coresight/devices/tmc_etf0/mgmt/mode 0x0

And this is most likely due to

commit cd9e3474bb793dc ("coresight: add PM runtime calls to coresight_simple_func()")

Ok, so this is related to power management (you can ignore my question in the
previous email).

Quoted 8 lines

Regarding function tmc_read_prepare_etb(), the best way to deal with this is
probably make sure drvdata->mode != CS_MODE_DISABLED before reading TMC_MODE.
If there is a buffer to read it will have been copied when the ETB was disabled
and there won't be a need to access the HW.

This works as well, thanks.

drivers/hwtracing/coresight/coresight-tmc-etf.c +0 -0
index d0cc3985b72a..7ffe05930984 100644

drivers/hwtracing/coresight/coresight-tmc-etf.c int tmc_read_prepare_etb(struct tmc_drvdata · line 604

Quoted 9 lines
drivers/hwtracing/coresight/coresight-tmc-etf.c +5 -0
@@ -596,6 +596,11 @@ int tmc_read_prepare_etb(struct tmc_drvdata
 *drvdata)
                 goto out;
         }
 
+       if (drvdata->mode == CS_MODE_DISABLED) {
+               ret = -EINVAL;
+               goto out;
+       }
+

We are back to your original solution where the ETB buffer can't be read if the ETB itself is not enabled. It _is_ possible to read the buffer of an ETB that has been disabled.

To fix this consider the following [1]. Take the block at line 607 and move it to line 598. As part of the if() condition at line 619, read the value of the TMC_MODE register and exit if not in circular mode. If it is in circular mode continue with disabling the hardware.

[1]. https://elixir.bootlin.com/linux/v5.7-rc1/source/drivers/hwtracing/coresight/coresight-tmc-etf.c

Quoted 17 lines
drivers/hwtracing/coresight/coresight-tmc-etf.c +0 -0
         /* There is no point in reading a TMC in HW FIFO mode */
         mode = readl_relaxed(drvdata->base + TMC_MODE);
         if (mode != TMC_MODE_CIRCULAR_BUFFER) {

Thanks,
Sai

--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#7 lore raw

Hi Mathieu,

Quoted 71 lines · Mathieu Poirier
Quoted 55 lines · Sai Prakash Ranjan

Hi Mathieu,

Quoted 34 lines · Mathieu Poirier
Quoted 21 lines · Sai Prakash Ranjan

Hi Suzuki,

Quoted 11 lines · Suzuki K Poulose

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

Quoted 4 lines

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

I do not see any SError when I run the above command.

Quoted 7 lines

localhost ~ # cat /sys/bus/coresight/devices/tmc_etf0/mgmt/mode 0x0

And this is most likely due to

commit cd9e3474bb793dc ("coresight: add PM runtime calls to coresight_simple_func()")

Ok, so this is related to power management (you can ignore my question in the
previous email).

Quoted 8 lines

Regarding function tmc_read_prepare_etb(), the best way to deal with this is
probably make sure drvdata->mode != CS_MODE_DISABLED before reading TMC_MODE.
If there is a buffer to read it will have been copied when the ETB was disabled
and there won't be a need to access the HW.

This works as well, thanks.

drivers/hwtracing/coresight/coresight-tmc-etf.c +0 -0
index d0cc3985b72a..7ffe05930984 100644

drivers/hwtracing/coresight/coresight-tmc-etf.c int tmc_read_prepare_etb(struct tmc_drvdata · line 604

Quoted 9 lines
drivers/hwtracing/coresight/coresight-tmc-etf.c +5 -0
@@ -596,6 +596,11 @@ int tmc_read_prepare_etb(struct tmc_drvdata
 *drvdata)
                 goto out;
         }
 
+       if (drvdata->mode == CS_MODE_DISABLED) {
+               ret = -EINVAL;
+               goto out;
+       }
+

We are back to your original solution where the ETB buffer can't be read if the ETB itself is not enabled. It _is_ possible to read the buffer of an ETB that has been disabled.

Quoted 8 lines

To fix this consider the following [1]. Take the block at line 607 and move it to line 598. As part of the if() condition at line 619, read the value of the TMC_MODE register and exit if not in circular mode. If it is in circular mode continue with disabling the hardware.

[1].
https://elixir.bootlin.com/linux/v5.7-rc1/source/drivers/hwtracing/coresight/coresight-tmc-etf.c

Thanks, got it now. Posted v2 - https://lore.kernel.org/patchwork/patch/1226022/

-Sai

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#8 lore raw
In reply to Suzuki K Poulose
Quoted 11 lines · Suzuki K Poulose

Hi Sai,

Quoted 7 lines · Sai Prakash Ranjan

Reading TMC mode register in tmc_read_prepare_etb without enabling the TMC hardware leads to async exceptions like the one in the call trace below. This can happen if the user tries to read the TMC etf data via device node without setting up source and the sink first which enables the TMC hardware in the path. So make sure that the TMC is enabled before we try to read TMC data.

Quoted 3 lines

So, one can trigger the same SError by simply :

$ cat /sys/bus/coresight/device/tmc_etb0/mgmt/mode

A true TMC-ETB is a rarity nowadays... What boards was this tested on? I don't know if it is how the IP behaves or how things are connected on Sai's specific board but doing echo'ing tmc_etf0/mgmt/mode on my Juno R0 doesn't give rise to any problems. That certainly doesn't mean it can't happen on another platform.

Quoted 51 lines

And also :

Quoted 47 lines
  Kernel panic - not syncing: Asynchronous SError Interrupt
  CPU: 7 PID: 2605 Comm: hexdump Tainted: G S                5.4.30 #122
  Call trace:
   dump_backtrace+0x0/0x188
   show_stack+0x20/0x2c
   dump_stack+0xdc/0x144
   panic+0x168/0x36c
   panic+0x0/0x36c
   arm64_serror_panic+0x78/0x84
   do_serror+0x130/0x138
   el1_error+0x84/0xf8
   tmc_read_prepare_etb+0x88/0xb8
   tmc_open+0x40/0xd8
   misc_open+0x120/0x158
   chrdev_open+0xb8/0x1a4
   do_dentry_open+0x268/0x3a0
   vfs_open+0x34/0x40
   path_openat+0x39c/0xdf4
   do_filp_open+0x90/0x10c
   do_sys_open+0x150/0x3e8
   __arm64_compat_sys_openat+0x28/0x34
   el0_svc_common+0xa8/0x160
   el0_svc_compat_handler+0x2c/0x38
   el0_svc_compat+0x8/0x10
Fixes 4525412a5046 ("coresight: tmc: making prepare/unprepare functions generic")
---
  drivers/hwtracing/coresight/coresight-tmc.c | 5 +++++
  drivers/hwtracing/coresight/coresight-tmc.h | 1 +
  2 files changed, 6 insertions(+)
drivers/hwtracing/coresight/coresight-tmc.c +2 -0
index 1cf82fa58289..7bae69748ab7 100644
@@ -62,11 +62,13 @@ void tmc_flush_and_stop(struct tmc_drvdata *drvdata)
  void tmc_enable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = true;
  	writel_relaxed(TMC_CTL_CAPT_EN, drvdata->base + TMC_CTL);
  }
  void tmc_disable_hw(struct tmc_drvdata *drvdata)
  {
+	drvdata->enable = false;
Quoted 8 lines

drivers/hwtracing/coresight/coresight-tmc.c static int tmc_read_prepare(struct tmc_drvdata *drvdata) · line 108

Quoted 8 lines
drivers/hwtracing/coresight/coresight-tmc.c +3 -0
  	writel_relaxed(0x0, drvdata->base + TMC_CTL);
  }
@@ -102,6 +104,9 @@ static int tmc_read_prepare(struct tmc_drvdata *drvdata)
  {
  	int ret = 0;
+	if (!drvdata->enable)
+		return -EINVAL;
+

Proceeding this way would mean that ETB and ETF internal memory buffers can't be read unless the device is enabled and collecting trace. That is not practical because 1) it changes the way the sysfs interface works on all boards where there isn't a problem and 2) it makes it very difficult to capture the correct data.

When the device is __not__ enabled, does reading any of the registers under "/sys/bus/coresight/devices/tmc_etbX/mgmt/" also cause a problem or is it just 'mode'?

Thanks,
Mathieu

Quoted 10 lines

Does this check always guarantee that the TMC is enabled when we actually get to reading the MODE ? This needs to be done under the spinlock.

Cheers
Suzuki