Re: [PATCH] net: netcp: Fix crash in error path when DMA channel open fails

[Siddharth Vadapalli <s-vadapalli@ti.com>]

On 26/09/25 9:43 PM, Simon Horman wrote:
> On Fri, Sep 26, 2025 at 10:08:53AM -0500, Nishanth Menon wrote:
>> When knav_dma_open_channel() fails in netcp_setup_navigator_resources(),
>> the rx_channel field is set to an ERR_PTR value. Later, when
>> netcp_free_navigator_resources() is called in the error path, it attempts
>> to close this invalid channel pointer, causing a crash.
>>
>> Add a check for ERR values to handle the failure scenario.
>>
>> Fixes: 84640e27f230 ("net: netcp: Add Keystone NetCP core driver")
>> Signed-off-by: Nishanth Menon <nm@ti.com>
>> ---
>>
>> Seen on kci log for k2hk: https://dashboard.kernelci.org/log-viewer?itemId=ti%3A2eb55ed935eb42c292e02f59&org=ti&type=test&url=http%3A%2F%2Ffiles.kernelci.org%2F%2Fti%2Fmainline%2Fmaster%2Fv6.17-rc7-59-gbf40f4b87761%2Farm%2Fmulti_v7_defconfig%2BCONFIG_EFI%3Dy%2BCONFIG_ARM_LPAE%3Dy%2Bdebug%2Bkselftest%2Btinyconfig%2Fgcc-12%2Fbaseline-nfs-boot.nfs-k2hk-evm.txt.gz
>>
>>   drivers/net/ethernet/ti/netcp_core.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/ethernet/ti/netcp_core.c b/drivers/net/ethernet/ti/netcp_core.c
>> index 857820657bac..4ff17fd6caae 100644
>> --- a/drivers/net/ethernet/ti/netcp_core.c
>> +++ b/drivers/net/ethernet/ti/netcp_core.c
>> @@ -1549,7 +1549,7 @@ static void netcp_free_navigator_resources(struct netcp_intf *netcp)
>>   {
>>   	int i;
>>   
>> -	if (netcp->rx_channel) {
>> +	if (!IS_ERR(netcp->rx_channel)) {
>>   		knav_dma_close_channel(netcp->rx_channel);
>>   		netcp->rx_channel = NULL;
>>   	}
> 
> Hi Nishanth,
> 
> Thanks for your patch.
> 
> I expect that netcp_txpipe_close() has a similar problem too.
> 
> But I also think that using IS_ERR is not correct, because it seems to me
> that there are also cases where rx_channel can be NULL.

Could you please clarify where rx_channel is NULL? rx_channel is set by 
invoking knav_dma_open_channel(). Also, please refer to:
https://github.com/torvalds/linux/commit/5b6cb43b4d62
which specifically points out that knav_dma_open_channel() will not 
return NULL so the check for NULL isn't required.
> 
> I see that on error knav_dma_open_channel() always returns ERR_PTR(-EINVAL)
> (open coded as (void *)-EINVAL) on error. So I think a better approach
> would be to change knav_dma_open_channel() to return NULL, and update callers
> accordingly.

The commit referred to above made changes to the driver specifically due 
to the observation that knav_dma_open_channel() never returns NULL. 
Modifying knav_dma_open_channel() to return NULL will effectively result 
in having to undo the changes made by the commit.