Skip to contentSkip to conversation

Commit ecee9... (Program net device MAC addresses after initializing) breakage

U-Boot Graeme Russ 3 participants 7 messages Nov 6, 2010 – Nov 13, 2010 Network drivers

lore
Share & tools
lore
#0 lore raw

Hi All,

I finally got a few spare hours to do some U-Boot hacking and to my dismay found the build for my board (eNET) had broken. It builds OK, but crashed during Ethernet initialisation. I tried winding back to a last known good commit without much luck.

I realised I had upgraded to Ubuntu 10.10 lately, and thought maybe a tool-chain change was the culprit (was running gcc-4.4.1 and binutils 2.19 from source, now running Ubuntu gcc 4.4.5 and Binutils 2.20)

But, I think that commit ecee9324d73555e744593f3e0d387bec4c566f55 may be 'not quite right'. My board uses the rtl8139 driver, and the following patch gets my board booting again:

drivers/net/rtl8139.c +1 -0
index db8a727..3646148 100644
@@ -229,6 +229,7 @@ int rtl8139_initialize(bd_t *bis)
                dev->halt = rtl_disable;
                dev->send = rtl_transmit;
                dev->recv = rtl_poll;
+               dev->write_hwaddr = NULL;
 #ifdef CONFIG_MCAST_TFTP
                dev->mcast = rtl_bcast_addr;
 #endif

Just before this code we see that dev is malloc'd:

dev = (struct eth_device *)malloc(sizeof *dev);

So there is no guarantee that dev is NULL'd.

Has anyone else run into similar problems?

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

Regards,

Graeme

#1 lore raw
Quoted 30 lines · Graeme Russ

Hi All,

I finally got a few spare hours to do some U-Boot hacking and to my dismay found the build for my board (eNET) had broken. It builds OK, but crashed during Ethernet initialisation. I tried winding back to a last known good commit without much luck.

I realised I had upgraded to Ubuntu 10.10 lately, and thought maybe a tool-chain change was the culprit (was running gcc-4.4.1 and binutils 2.19 from source, now running Ubuntu gcc 4.4.5 and Binutils 2.20)

But, I think that commit ecee9324d73555e744593f3e0d387bec4c566f55 may be 'not quite right'. My board uses the rtl8139 driver, and the following patch gets my board booting again:

drivers/net/rtl8139.c +1 -0
index db8a727..3646148 100644
@@ -229,6 +229,7 @@ int rtl8139_initialize(bd_t *bis)
               dev->halt = rtl_disable;
               dev->send = rtl_transmit;
               dev->recv = rtl_poll;
+               dev->write_hwaddr = NULL;
 #ifdef CONFIG_MCAST_TFTP
               dev->mcast = rtl_bcast_addr;
 #endif

Just before this code we see that dev is malloc'd:

Quoted 8 lines

dev = (struct eth_device *)malloc(sizeof *dev);

So there is no guarantee that dev is NULL'd.

Has anyone else run into similar problems?

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

I believe we are seeing similar behavior w/e1000 driver.

- k

#2 lore raw
In reply to Graeme Russ
Quoted 13 lines · On Saturday, November 06, 2010 18:13:08 Graeme Russ wrote:

I finally got a few spare hours to do some U-Boot hacking and to my dismay found the build for my board (eNET) had broken. It builds OK, but crashed during Ethernet initialisation. I tried winding back to a last known good commit without much luck.

I realised I had upgraded to Ubuntu 10.10 lately, and thought maybe a tool-chain change was the culprit (was running gcc-4.4.1 and binutils 2.19 from source, now running Ubuntu gcc 4.4.5 and Binutils 2.20)

But, I think that commit ecee9324d73555e744593f3e0d387bec4c566f55 may be 'not quite right'. My board uses the rtl8139 driver, and the following patch gets my board booting again:

Quoted 5 lines

Just before this code we see that dev is malloc'd:

dev = (struct eth_device *)malloc(sizeof *dev);

So there is no guarantee that dev is NULL'd.

rtl8139 is broken. it should be clearing its memory.

someone posted a whole bunch of patches to memset() net drivers ...

Quoted 2 lines

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

i dont think anyone posted a patch. it would make sense though to generalize the zalloc() code since some places are already doing it.
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20101109/42fb1c9c/attachment.pgp

#3 lore raw
Quoted 7 lines · Mike Frysinger
Quoted 2 lines · On Saturday, November 06, 2010 18:13:08 Graeme Russ wrote:

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

i dont think anyone posted a patch. it would make sense though to generalize the zalloc() code since some places are already doing it.

Actually, I think we should be using calloc() more than we are

Regards,

Graeme

#4 lore raw
Quoted 9 lines · On Friday, November 12, 2010 05:55:55 Graeme Russ wrote:
Quoted 6 lines · Mike Frysinger
Quoted 2 lines · On Saturday, November 06, 2010 18:13:08 Graeme Russ wrote:

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

i dont think anyone posted a patch. it would make sense though to generalize the zalloc() code since some places are already doing it.

Actually, I think we should be using calloc() more than we are

if we want to define zalloc() to calloc(), then sure. but forcing people to use calloc() with a size=1 is silly. the API is not nearly as obvious to passing observers as zalloc().
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20101112/25905c2e/attachment.pgp

#5 lore raw
Quoted 6 lines · Mike Frysinger
Quoted 5 lines · On Friday, November 12, 2010 05:55:55 Graeme Russ wrote:
Quoted 4 lines · Mike Frysinger
Quoted 2 lines · On Saturday, November 06, 2010 18:13:08 Graeme Russ wrote:

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

i dont think anyone posted a patch. it would make sense though to generalize the zalloc() code since some places are already doing it.

Actually, I think we should be using calloc() more than we are

if we want to define zalloc() to calloc(), then sure. but forcing people to use calloc() with a size=1 is silly. the API is not nearly as obvious to passing observers as zalloc().

Hmm, U-Boot has a zalloc in lib/gunzip.c which does not zero memory

void *zalloc(void *x, unsigned items, unsigned size)

{
	void *p;

	size *= items;
	size = (size + ZALLOC_ALIGNMENT - 1) & ~(ZALLOC_ALIGNMENT - 1);

	p = malloc (size);

	return (p);
}

and mtd has kzalloc defined thus:

#define kzalloc(size, flags) calloc(size, 1)

So there is some confusion surrounding 'zalloc' anyway ;)

Regards,

Graeme

#6 lore raw
Quoted 10 lines · On Friday, November 12, 2010 20:51:12 Graeme Russ wrote:
Quoted 9 lines · Mike Frysinger
Quoted 8 lines · On Friday, November 12, 2010 05:55:55 Graeme Russ wrote:
Quoted 6 lines · Mike Frysinger
Quoted 2 lines · On Saturday, November 06, 2010 18:13:08 Graeme Russ wrote:

I saw discussion a little while ago regarding implementing a version of malloc that returns cleared memory - did this gain any traction?

i dont think anyone posted a patch. it would make sense though to generalize the zalloc() code since some places are already doing it.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

Actually, I think we should be using calloc() more than we are

if we want to define zalloc() to calloc(), then sure. but forcing people to use calloc() with a size=1 is silly. the API is not nearly as obvious to passing observers as zalloc().

Hmm, U-Boot has a zalloc in lib/gunzip.c which does not zero memory

yes, that's why i said we should generalize things
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
Url : http://lists.denx.de/pipermail/u-boot/attachments/20101112/e64b75aa/attachment.pgp