diff options
author | Maciej Żenczykowski <maze@google.com> | 2023-06-20 19:40:22 +0000 |
---|---|---|
committer | Cherrypicker Worker <android-build-cherrypicker-worker@google.com> | 2023-10-09 11:08:38 +0000 |
commit | 53d77df5114329ca04155c4be7c284acede78b6c (patch) | |
tree | 5ee46399332940970bb6cb3ef48096b8e41dfa25 | |
parent | dbe70cc96aa42c4c762df60c9604a8b0286ec358 (diff) | |
download | bpf-53d77df5114329ca04155c4be7c284acede78b6c.tar.gz |
add a comment about page size vs bpf ring buffer sizing
As requested by Patrick on:
https://android-review.git.corp.google.com/c/platform/system/bpf/+/2628969
bpfLoader: automatically bump ringbuffer size up to page size
Test: N/A, comment only
Signed-off-by: Maciej Żenczykowski <maze@google.com>
(cherry picked from https://android-review.googlesource.com/q/commit:28f01bb027e5daf9793ed9a436bf735942cd423e)
Merged-In: Ibbce6a72e5e25bf18bb89bca2f00d332ef824f73
Change-Id: Ibbce6a72e5e25bf18bb89bca2f00d332ef824f73
-rw-r--r-- | libbpf_android/Loader.cpp | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/libbpf_android/Loader.cpp b/libbpf_android/Loader.cpp index 231fe39..83f6a88 100644 --- a/libbpf_android/Loader.cpp +++ b/libbpf_android/Loader.cpp @@ -707,6 +707,9 @@ static bool mapMatchesExpectations(const unique_fd& fd, const string& mapName, if (type == BPF_MAP_TYPE_DEVMAP || type == BPF_MAP_TYPE_DEVMAP_HASH) desired_map_flags |= BPF_F_RDONLY_PROG; + // The .h file enforces that this is a power of two, and page size will + // also always be a power of two, so this logic is actually enough to + // force it to be a multiple of the page size, as required by the kernel. unsigned int desired_max_entries = mapDef.max_entries; if (type == BPF_MAP_TYPE_RINGBUF) { if (desired_max_entries < page_size) desired_max_entries = page_size; @@ -849,6 +852,9 @@ static int createMaps(const char* elfPath, ifstream& elfFile, vector<unique_fd>& type = BPF_MAP_TYPE_HASH; } + // The .h file enforces that this is a power of two, and page size will + // also always be a power of two, so this logic is actually enough to + // force it to be a multiple of the page size, as required by the kernel. unsigned int max_entries = md[i].max_entries; if (type == BPF_MAP_TYPE_RINGBUF) { if (max_entries < page_size) max_entries = page_size; |