Stefan Reinauer stepan@coresystems.de writes:
JasonZhao@viatech.com.cn wrote:
Hardwaremain()-> dev_configure();-> void root_dev_read_resources(device_t root) ->
void compact_resources(device_t dev)->
memmove(resource, resource + 1, dev->resources - i);
The third parameter should be (dev->resources-I)* sizeof(*resource)
Dear Jason,
thank you very much for your finding!
It is indeed a bug and should be fixed.
I am surprised we did not hit this before, as this potentially wipes out whole resource entries. (unless the number of resources is bigger than the size of a resource in bytes)
But only if there are holes in the resource entries. Not sure why/how these occur. Eric, your word!
Looking at the code I'm fairly certain that we never or practically never exercise it. To create a hole it would require someone to call new_resource and then never setup the resource. compact_resources looks to be there just in case that happens, rather than to handle an actual case that someone has hit.
We have a small static pool of 24 resources on each device, and a struct resource looks to be 64 bytes.
Eric