Re: [PATCH v6 00/33] block layer: split block APIs in global state and I/O

[Kevin Wolf]

Am 08.02.2022 um 12:42 hat Emanuele Giuseppe Esposito geschrieben:
> 
> 
> On 07/02/2022 19:30, Kevin Wolf wrote:
> > Am 21.01.2022 um 18:05 hat Emanuele Giuseppe Esposito geschrieben:
> >> Each function in the GS API will have an assertion, checking
> >> that it is always running under BQL.
> >> I/O functions are instead thread safe (or so should be), meaning
> >> that they *can* run under BQL, but also in an iothread in another
> >> AioContext. Therefore they do not provide any assertion, and
> >> need to be audited manually to verify the correctness.
> > 
> > I wonder if we could actually do something to catch at least some kinds
> > of bugs. The first conclusion from thinking about it is that we probably
> > shouldn't open-code assert(qemu_in_main_thread()) everywhere, but have a
> > macro or inline function for each category to be called in each function.
> > 
> > So an IO_CODE() macro could increase a counter in the coroutine object
> > (that is decreased again at the end of the function with g_auto), and
> > then GLOBAL_STATE_CODE() could not only assert that we're holding the
> > BQL, but also that the counter is still 0, i.e. it is not (indirectly)
> > called by an I/O function.
> > 
> > We may want to enable this only in debug builds, but maybe still worth a
> > thought anyway?
> 
> I don't understand what is the point of the counter, do you want to use
> it as a boolean flag?

It would only be checked as a boolean flag, but it needs to be a counter
because of nesting where e.g. one I/O function calls another I/O
function.

> Would a single counter work in a multi-threaded context? Shouldn't we
> have it per-thread? And why you increase it only in coroutines?

I don't mean increasing it only in coroutine context, but having a
per-coroutine counter, including the leader coroutine which exists for
non-coroutine context in every thread.

Kevin