On Fri, Sep 04, 2026 at 10:58:06AM +0800, David Yang wrote:
On Fri, Sep 4, 2026 at 12:25 AM Andrew Lunn [off-list ref] wrote:
quoted
On Thu, Sep 03, 2026 at 10:35:08PM +0800, David Yang wrote:
quoted
64-bit variables might not be atomic on 32-bit architectures, and could
lead to load/store tearing. Use u64_stats_t to ensure consistency.
Can this patch also be broken up. It should be possible to use
u64_stats_read() without changing the locking? So one patch which only
does u64_stats_read(). You should then be able to add the syncp
code. And lastly the mutex changes. That then allows you to explain
why the mutex changes are safe in the commit message.
Andrew
u64_stats_read are meant to be used with syncp. This sounds like an
incorrect partial patch, works only because of the internal
implementation of u64_stats_t.
What is making the current code correct? The mutex? So long as you
have the mutex, the code is still correct?
You can say in the commit message this is an intermediary step doing
the mechanical work of introducing u64_stats_t. The following patches
will then change the locking to be more efficient.
Andrew