user space takes responsibility of not calling reserve and cache op at same
time. Given that promise, some times it is better for performance to avoid zap
when concurrent access by cpu and device is guaranteed in some other way.
Bug
200092803
Change-Id: I20d7e2106fd6114712b63991204d5e556912af21
Signed-off-by: Sri Krishna chowdary <schowdary@nvidia.com>
Reviewed-on: http://git-master/r/776097
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
if (!handle)
return -EINVAL;
- if (handle->userflags & NVMAP_HANDLE_CACHE_SYNC_AT_RESERVE)
- goto put_handle;
-
down_read(¤t->mm->mmap_sem);
vma = find_vma(current->active_mm, (unsigned long)op->addr);
false);
out:
up_read(¤t->mm->mmap_sem);
-put_handle:
nvmap_handle_put(handle);
return err;
}
goto free_mem;
}
- /* skip cache op when NVMAP_HANDLE_CACHE_SYNC_AT_RESERVE is specified */
- if (count && !is_reserve_ioctl) {
- err = 0;
- goto free_mem;
- }
-
/*
* When NVMAP_HANDLE_CACHE_SYNC_AT_RESERVE is specified, mix can cause
* cache WB_INV at unreserve op on iovmm handles increasing overhead.