IGMP - Timer bug in igmp_start_timer()

[forkineye]

Issue

igmp_start_timer() sometimes assigns excessive times to group->timer, causing IGMP memberships to expire. The problem is related to the usage of r_rand() via the LWIP_RAND() macro in the following function: https://github.com/esp8266/Arduino/blob/master/tools/sdk/lwip/src/core/ipv4/igmp.c#L703. group->timer should never exceed 100. Debug output in the function produces the following:

os_printf("igmp_start_timer: %8X / %u / %u \n", &group->group_address, group->timer, max_time);
igmp_start_timer: 3FFF55B4 / 65515 / 100 
igmp_start_timer: 3FFF559C / 65 / 100 
igmp_start_timer: 3FFF5584 / 65493 / 100 
igmp_start_timer: 3FFF556C / 65503 / 100 
igmp_start_timer: 3FFF5554 / 43 / 100 
igmp_start_timer: 3FFF5054 / 65531 / 100 

Fix

Assigning the output of LWIP_RAND() to a uint32_t variable and using that in place for the modulus fixes the issue.

Also, using RANDOM_REG32 in place of r_rand() works as well:

#define LWIP_RAND() *((volatile uint32_t *)(0x3FF20E44))  // RANDOM_REG32

Not sure why its breaking as is, but the above two fixes work for me.

[forkineye]

Using os_random() instead of r_randr() is the best solution I've found. PR submitted #3499

[forkineye]

Thanks for merging the PR!

[gjt211]

Is it possible for someone to provide some symptoms of the fault this fixed if possible?
Thanks.

[forkineye]

The symptom was that devices would stop receiving multicast traffic after a while if associated with an AP that had IGMP snooping enabled. I noticed it while testing a 720 pixel matrix being driven via multicast in my ESPixelStick project - https://github.com/forkineye/ESPixelStick. I tested the fix for 48hrs streaming E1.31 data at 50ms refresh rates.