net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
In eth_init(), eth_get_dev() can return NULL. We should do sanity test on eth dev before calling its start function.
---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)
net/eth.c +4 -0
@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + } }while (old_current != current);return ret;
Signature · 1 line
1.8.2.1
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0
net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }
I'm not sure I get the point of this. We already have a check above...
current = eth_get_dev();
if (!current) {
printf("No ethernet found.\n");
return -ENODEV;
}Quoted 10 lines
net/eth.c +0
-1}while (old_current != current);return ret; --1.8.2.1
_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Joe,
Quoted 27 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
Quoted 6 lines
current = eth_get_dev();
if (!current) {
printf("No ethernet found.\n");
return -ENODEV;
}But this does not help. Each time eth_get_dev() is called, current can be NULL as driver's probe can fail.
Quoted 4 lines
net/eth.c int eth_init(void)
Quoted 4 lines
net/eth.c +0
-1}while (old_current != current);return ret; --
Regards,
Bin
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Bin,
Quoted 38 lines · Bin Meng
Hi Joe,
Quoted 27 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
Quoted 6 lines
current = eth_get_dev(); if (!current) { printf("No ethernet found.\n"); return -ENODEV; }
Quoted 2 lines
But this does not help. Each time
eth_get_dev()is called, current can be NULL as driver's probe can fail.
If that's the issue you are hitting it seems like you should attempt to skip the device instead of printing the message. It doesn't make sense to me to move to the next device and then print that there is no Ethernet.
Also, this is fundamental to the eth subsystem. You should add a unit test that fails in your case.
Quoted 7 lines
Quoted 4 lines
net/eth.c int eth_init(void)
Quoted 4 lines
net/eth.c +0
-1}while (old_current != current);return ret; --Regards,
Bin
Thanks,
-Joe
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Joe,
Quoted 41 lines · Joe Hershberger
Hi Bin,
Quoted 38 lines · Bin Meng
Hi Joe,
Quoted 33 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
current = eth_get_dev(); if (!current) { printf("No ethernet found.\n"); return -ENODEV; }
Quoted 7 lines
Quoted 2 lines
But this does not help. Each time
eth_get_dev()is called, current can be NULL as driver's probe can fail.If that's the issue you are hitting it seems like you should attempt to skip the device instead of printing the message. It doesn't make sense to me to move to the next device and then print that there is no Ethernet.
Do you mean we should not printf("No ethernet found.\n") and just break here?
If it fails, U-Boot just crashes as there is a NULL pointer. I am not sure if test case is able to handle this?
Quoted 9 lines
Also, this is fundamental to the eth subsystem. You should add a unit test that fails in your case.
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
net/eth.c +0
-1}while (old_current != current);return ret; --
Regards,
Bin
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Bin,
Quoted 48 lines · Bin Meng
Hi Joe,
Quoted 44 lines · Joe Hershberger
Hi Bin,
Quoted 40 lines · Bin Meng
Hi Joe,
Quoted 27 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
Quoted 6 lines
current = eth_get_dev(); if (!current) { printf("No ethernet found.\n"); return -ENODEV; }But this does not help. Each time
eth_get_dev()is called, current can be NULL as driver's probe can fail.
Quoted 6 lines
Quoted 4 lines
If that's the issue you are hitting it seems like you should attempt to skip the device instead of printing the message. It doesn't make sense to me to move to the next device and then print that there is no Ethernet.
Do you mean we should not
printf("No ethernet found.\n")and just break here?
I think you shouldn't break, but rather should have an if check around the top half of the loop. I.e.:
net/eth.c +17 -13
@@ -343,23 +343,27 @@ int eth_init(void) old_current = current; do { - debug("Trying %s\n", current->name); - - if (device_active(current)) { - ret = eth_get_ops(current)->start(current); - if (ret >= 0) { - struct eth_device_priv *priv = - current->uclass_priv; - - priv->state = ETH_STATE_ACTIVE; - return 0; + if (current) { + debug("Trying %s\n", current->name); + + if (device_active(current)) { + ret = eth_get_ops(current)->start(current); + if (ret >= 0) { + struct eth_device_priv *priv = + current->uclass_priv; + + priv->state = ETH_STATE_ACTIVE; + return 0; + } + } else { + ret = eth_errno; } + + debug("FAIL\n"); } else { - ret = eth_errno; + debug("PROBE FAIL\n"); } - debug("FAIL\n"); - /* * If ethrotate is enabled, this will change "current", * otherwise we will drop out of this while loop immediately
---
Note that I have not tested this, it's just what I'm thinking is more appropriate.
Quoted 2 lines
If it fails, U-Boot just crashes as there is a NULL pointer. I am not sure if test case is able to handle this?
I think it's good to have the a test that hits your scenario. The bug fix will prevent the crash, so it's not like we expect it to crash, but it will lock down the desired behavior for this condition.
Quoted 11 lines
Quoted 8 lines
Also, this is fundamental to the eth subsystem. You should add a unit test that fails in your case.
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
net/eth.c +0
-1}while (old_current != current);return ret; --Regards,
Bin
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Joe,
Quoted 111 lines · Joe Hershberger
Hi Bin,
Quoted 48 lines · Bin Meng
Hi Joe,
Quoted 44 lines · Joe Hershberger
Hi Bin,
Quoted 40 lines · Bin Meng
Hi Joe,
Quoted 27 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
Quoted 6 lines
current = eth_get_dev(); if (!current) { printf("No ethernet found.\n"); return -ENODEV; }But this does not help. Each time
eth_get_dev()is called, current can be NULL as driver's probe can fail.Quoted 6 lines
Quoted 4 lines
If that's the issue you are hitting it seems like you should attempt to skip the device instead of printing the message. It doesn't make sense to me to move to the next device and then print that there is no Ethernet.
Do you mean we should not
printf("No ethernet found.\n")and just break here?I think you shouldn't break, but rather should have an if check around the top half of the loop. I.e.:
net/eth.c +17
-14@@ -343,23 +343,27 @@ int eth_init(void) old_current = current; do { - debug("Trying %s\n", current->name); - - if (device_active(current)) { - ret = eth_get_ops(current)->start(current); - if (ret >= 0) { - struct eth_device_priv *priv = - current->uclass_priv; - - priv->state = ETH_STATE_ACTIVE; - return 0; + if (current) { + debug("Trying %s\n", current->name); + + if (device_active(current)) { + ret = eth_get_ops(current)->start(current); + if (ret >= 0) { + struct eth_device_priv *priv = + current->uclass_priv; + + priv->state = ETH_STATE_ACTIVE; + return 0; + } + } else { + ret = eth_errno; } + + debug("FAIL\n"); } else { - ret = eth_errno; + debug("PROBE FAIL\n"); } - debug("FAIL\n"); - /* * If ethrotate is enabled, this will change "current", * otherwise we will drop out of this while loop immediately ---Note that I have not tested this, it's just what I'm thinking is more appropriate.
Quoted 7 lines
Quoted 2 lines
If it fails, U-Boot just crashes as there is a NULL pointer. I am not sure if test case is able to handle this?
I think it's good to have the a test that hits your scenario. The bug fix will prevent the crash, so it's not like we expect it to crash, but it will lock down the desired behavior for this condition.
I am afraid creating a test case to cover this scenario is not that easy. Checking function return value does not bring any harm. It makes our codes safer. In fact, during further debug today, I found another two places which does not check device_probe() return value. And it is indeed these two places which causes the subsequent failure here.
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
Also, this is fundamental to the eth subsystem. You should add a unit test that fails in your case.
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
net/eth.c +0
-1}while (old_current != current);return ret; --
Regards,
Bin
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Joe,
Quoted 123 lines · Bin Meng
Hi Joe,
Quoted 111 lines · Joe Hershberger
Hi Bin,
Quoted 48 lines · Bin Meng
Hi Joe,
Quoted 44 lines · Joe Hershberger
Hi Bin,
Quoted 40 lines · Bin Meng
Hi Joe,
Quoted 27 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
Quoted 6 lines
current = eth_get_dev(); if (!current) { printf("No ethernet found.\n"); return -ENODEV; }But this does not help. Each time
eth_get_dev()is called, current can be NULL as driver's probe can fail.Quoted 6 lines
Quoted 4 lines
If that's the issue you are hitting it seems like you should attempt to skip the device instead of printing the message. It doesn't make sense to me to move to the next device and then print that there is no Ethernet.
Do you mean we should not
printf("No ethernet found.\n")and just break here?I think you shouldn't break, but rather should have an if check around the top half of the loop. I.e.:
net/eth.c +17
-14@@ -343,23 +343,27 @@ int eth_init(void) old_current = current; do { - debug("Trying %s\n", current->name); - - if (device_active(current)) { - ret = eth_get_ops(current)->start(current); - if (ret >= 0) { - struct eth_device_priv *priv = - current->uclass_priv; - - priv->state = ETH_STATE_ACTIVE; - return 0; + if (current) { + debug("Trying %s\n", current->name); + + if (device_active(current)) { + ret = eth_get_ops(current)->start(current); + if (ret >= 0) { + struct eth_device_priv *priv = + current->uclass_priv; + + priv->state = ETH_STATE_ACTIVE; + return 0; + } + } else { + ret = eth_errno; } + + debug("FAIL\n"); } else { - ret = eth_errno; + debug("PROBE FAIL\n"); } - debug("FAIL\n"); - /* * If ethrotate is enabled, this will change "current", * otherwise we will drop out of this while loop immediately ---Note that I have not tested this, it's just what I'm thinking is more appropriate.
Quoted 7 lines
Quoted 2 lines
If it fails, U-Boot just crashes as there is a NULL pointer. I am not sure if test case is able to handle this?
I think it's good to have the a test that hits your scenario. The bug fix will prevent the crash, so it's not like we expect it to crash, but it will lock down the desired behavior for this condition.
Quoted 6 lines
I am afraid creating a test case to cover this scenario is not that easy. Checking function return value does not bring any harm. It makes our codes safer. In fact, during further debug today, I found another two places which does not check
device_probe()return value. And it is indeed these two places which causes the subsequent failure here.
OK, I've found a way in the DM test codes to trigger this fault, but unfortunately by creating this test case I've found another potential issue. I will send a v2 patch for all of these.
Regards,
Bin
PATCH 1/1 net: eth: Do sanity test on eth dev before eth_get_ops(dev)->start
Hi Bin,
Quoted 133 lines · Bin Meng
Hi Joe,
Quoted 123 lines · Bin Meng
Hi Joe,
Quoted 111 lines · Joe Hershberger
Hi Bin,
Quoted 48 lines · Bin Meng
Hi Joe,
Quoted 44 lines · Joe Hershberger
Hi Bin,
Quoted 40 lines · Bin Meng
Hi Joe,
Quoted 27 lines · Joe Hershberger
Hi Bin,
Quoted 14 lines · Bin Meng
In
eth_init(),eth_get_dev()can return NULL. We should do sanity test on eth dev before calling its start function.Signed-off-by Bin Meng <[email protected]>---
net/eth.c | 4 ++++ 1 file changed, 4 insertions(+)net/eth.c +0
-0net/eth.c int eth_init(void) · line 376
Quoted 7 lines
net/eth.c +4
-0@@ -370,6 +370,10 @@ int eth_init(void)eth_try_another (0 );/* This will ensure the new "current" attempted to probe */ current =eth_get_dev (); +if (!current) { +printf ("No ethernet found.\n" ); +break ; + }I'm not sure I get the point of this. We already have a check above...
Quoted 6 lines
current = eth_get_dev(); if (!current) { printf("No ethernet found.\n"); return -ENODEV; }But this does not help. Each time
eth_get_dev()is called, current can be NULL as driver's probe can fail.Quoted 6 lines
Quoted 4 lines
If that's the issue you are hitting it seems like you should attempt to skip the device instead of printing the message. It doesn't make sense to me to move to the next device and then print that there is no Ethernet.
Do you mean we should not
printf("No ethernet found.\n")and just break here?I think you shouldn't break, but rather should have an if check around the top half of the loop. I.e.:
net/eth.c +17
-14@@ -343,23 +343,27 @@ int eth_init(void) old_current = current; do { - debug("Trying %s\n", current->name); - - if (device_active(current)) { - ret = eth_get_ops(current)->start(current); - if (ret >= 0) { - struct eth_device_priv *priv = - current->uclass_priv; - - priv->state = ETH_STATE_ACTIVE; - return 0; + if (current) { + debug("Trying %s\n", current->name); + + if (device_active(current)) { + ret = eth_get_ops(current)->start(current); + if (ret >= 0) { + struct eth_device_priv *priv = + current->uclass_priv; + + priv->state = ETH_STATE_ACTIVE; + return 0; + } + } else { + ret = eth_errno; } + + debug("FAIL\n"); } else { - ret = eth_errno; + debug("PROBE FAIL\n"); } - debug("FAIL\n"); - /* * If ethrotate is enabled, this will change "current", * otherwise we will drop out of this while loop immediately ---Note that I have not tested this, it's just what I'm thinking is more appropriate.
Quoted 7 lines
Quoted 2 lines
If it fails, U-Boot just crashes as there is a NULL pointer. I am not sure if test case is able to handle this?
I think it's good to have the a test that hits your scenario. The bug fix will prevent the crash, so it's not like we expect it to crash, but it will lock down the desired behavior for this condition.
Quoted 6 lines
I am afraid creating a test case to cover this scenario is not that easy. Checking function return value does not bring any harm. It makes our codes safer. In fact, during further debug today, I found another two places which does not check
device_probe()return value. And it is indeed these two places which causes the subsequent failure here.
Quoted 3 lines
OK, I've found a way in the DM test codes to trigger this fault, but unfortunately by creating this test case I've found another potential issue. I will send a v2 patch for all of these.
Thanks for looking into this.
Cheers,
-Joe