Thread (9 messages) flat view 9 messages, 2 authors, 2021-10-19

Re: [External] Re: [PATCH] bpf: use count for prealloc hashtab too

From: Chengming Zhou <hidden>
Date: 2021-10-18 05:49:51
Also in: bpf, lkml

在 2021/10/16 上午3:58, Alexei Starovoitov 写道:
On Fri, Oct 15, 2021 at 11:04 AM Chengming Zhou
[off-list ref] wrote:
quoted
We only use count for kmalloc hashtab not for prealloc hashtab, because
__pcpu_freelist_pop() return NULL when no more elem in pcpu freelist.

But the problem is that __pcpu_freelist_pop() will traverse all CPUs and
spin_lock for all CPUs to find there is no more elem at last.

We encountered bad case on big system with 96 CPUs that alloc_htab_elem()
would last for 1ms. This patch use count for prealloc hashtab too,
avoid traverse and spin_lock for all CPUs in this case.

Signed-off-by: Chengming Zhou <redacted>
It's not clear from the commit log what you're solving.
The atomic inc/dec in critical path of prealloc maps hurts performance.
That's why it's not used.
Thanks for the explanation, what I'm solving is when hash table hasn't free
elements, we don't need to call __pcpu_freelist_pop() to traverse and
spin_lock all CPUs. The ftrace output of this bad case is below:

 50)               |  htab_map_update_elem() {
 50)   0.329 us    |    _raw_spin_lock_irqsave();
 50)   0.063 us    |    lookup_elem_raw();
 50)               |    alloc_htab_elem() {
 50)               |      pcpu_freelist_pop() {
 50)   0.209 us    |        _raw_spin_lock();
 50)   0.264 us    |        _raw_spin_lock();
 50)   0.231 us    |        _raw_spin_lock();
 50)   0.168 us    |        _raw_spin_lock();
 50)   0.168 us    |        _raw_spin_lock();
 50)   0.300 us    |        _raw_spin_lock();
 50)   0.263 us    |        _raw_spin_lock();
 50)   0.304 us    |        _raw_spin_lock();
 50)   0.168 us    |        _raw_spin_lock();
 50)   0.177 us    |        _raw_spin_lock();
 50)   0.235 us    |        _raw_spin_lock();
 50)   0.162 us    |        _raw_spin_lock();
 50)   0.186 us    |        _raw_spin_lock();
 50)   0.185 us    |        _raw_spin_lock();
 50)   0.315 us    |        _raw_spin_lock();
 50)   0.172 us    |        _raw_spin_lock();
 50)   0.180 us    |        _raw_spin_lock();
 50)   0.173 us    |        _raw_spin_lock();
 50)   0.176 us    |        _raw_spin_lock();
 50)   0.261 us    |        _raw_spin_lock();
 50)   0.364 us    |        _raw_spin_lock();
 50)   0.180 us    |        _raw_spin_lock();
 50)   0.284 us    |        _raw_spin_lock();
 50)   0.226 us    |        _raw_spin_lock();
 50)   0.210 us    |        _raw_spin_lock();
 50)   0.237 us    |        _raw_spin_lock();
 50)   0.333 us    |        _raw_spin_lock();
 50)   0.295 us    |        _raw_spin_lock();
 50)   0.278 us    |        _raw_spin_lock();
 50)   0.260 us    |        _raw_spin_lock();
 50)   0.224 us    |        _raw_spin_lock();
 50)   0.447 us    |        _raw_spin_lock();
 50)   0.221 us    |        _raw_spin_lock();
 50)   0.320 us    |        _raw_spin_lock();
 50)   0.203 us    |        _raw_spin_lock();
 50)   0.213 us    |        _raw_spin_lock();
 50)   0.242 us    |        _raw_spin_lock();
 50)   0.230 us    |        _raw_spin_lock();
 50)   0.216 us    |        _raw_spin_lock();
 50)   0.525 us    |        _raw_spin_lock();
 50)   0.257 us    |        _raw_spin_lock();
 50)   0.235 us    |        _raw_spin_lock();
 50)   0.269 us    |        _raw_spin_lock();
 50)   0.368 us    |        _raw_spin_lock();
 50)   0.249 us    |        _raw_spin_lock();
 50)   0.217 us    |        _raw_spin_lock();
 50)   0.174 us    |        _raw_spin_lock();
 50)   0.173 us    |        _raw_spin_lock();
 50)   0.161 us    |        _raw_spin_lock();
 50)   0.282 us    |        _raw_spin_lock();
 50)   0.264 us    |        _raw_spin_lock();
 50)   0.160 us    |        _raw_spin_lock();
 50)   0.692 us    |        _raw_spin_lock();
 50)   0.185 us    |        _raw_spin_lock();
 50)   0.157 us    |        _raw_spin_lock();
 50)   0.168 us    |        _raw_spin_lock();
 50)   0.205 us    |        _raw_spin_lock();
 50)   0.189 us    |        _raw_spin_lock();
 50)   0.276 us    |        _raw_spin_lock();
 50)   0.171 us    |        _raw_spin_lock();
 50)   0.390 us    |        _raw_spin_lock();
 50)   0.164 us    |        _raw_spin_lock();
 50)   0.170 us    |        _raw_spin_lock();
 50)   0.188 us    |        _raw_spin_lock();
 50)   0.284 us    |        _raw_spin_lock();
 50)   0.191 us    |        _raw_spin_lock();
 50)   0.412 us    |        _raw_spin_lock();
 50)   0.285 us    |        _raw_spin_lock();
 50)   0.296 us    |        _raw_spin_lock();
 50)   0.315 us    |        _raw_spin_lock();
 50)   0.239 us    |        _raw_spin_lock();
 50)   0.225 us    |        _raw_spin_lock();
 50)   0.258 us    |        _raw_spin_lock();
 50)   0.228 us    |        _raw_spin_lock();
 50)   0.240 us    |        _raw_spin_lock();
 50)   0.297 us    |        _raw_spin_lock();
 50)   0.216 us    |        _raw_spin_lock();
 50)   0.213 us    |        _raw_spin_lock();
 50)   0.225 us    |        _raw_spin_lock();
 50)   0.223 us    |        _raw_spin_lock();
 50)   0.287 us    |        _raw_spin_lock();
 50)   0.258 us    |        _raw_spin_lock();
 50)   0.295 us    |        _raw_spin_lock();
 50)   0.262 us    |        _raw_spin_lock();
 50)   0.325 us    |        _raw_spin_lock();
 50)   0.203 us    |        _raw_spin_lock();
 50)   0.325 us    |        _raw_spin_lock();
 50)   0.255 us    |        _raw_spin_lock();
 50)   0.325 us    |        _raw_spin_lock();
 50)   0.216 us    |        _raw_spin_lock();
 50)   0.232 us    |        _raw_spin_lock();
 50)   0.804 us    |        _raw_spin_lock();
 50)   0.262 us    |        _raw_spin_lock();
 50)   0.242 us    |        _raw_spin_lock();
 50)   0.271 us    |        _raw_spin_lock();
 50)   0.175 us    |        _raw_spin_lock();
 50) + 61.026 us   |      }
 50) + 61.575 us   |    }
 50)   0.051 us    |    _raw_spin_unlock_irqrestore();
 50) + 64.863 us   |  }
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help