MySQL Bugs: #79652: Suspicious padding in srv_conc

Bug #79652	Suspicious padding in srv_conc_t
Submitted:	15 Dec 2015 19:17	Modified:	27 Aug 2018 11:56
Reporter:	Alexey Kopytov (OCA)	Email Updates:
Status:	No Feedback	Impact on me:	None
Category:	MySQL Server: InnoDB storage engine	Severity:	S5 (Performance)
Version:	5.7	OS:	Any
Assigned to:		CPU Architecture:	Any

Description:
The srv_conc_t is declared in srv0conc.cc as follows:

/** Variables tracking the active and waiting threads. */
struct srv_conc_t {
	char		pad[64 - (sizeof(ulint) + sizeof(lint))];

	/** Number of transactions that have declared_to_be_inside_innodb set.
	It used to be a non-error for this value to drop below zero temporarily.
	This is no longer true. We'll, however, keep the lint datatype to add
	assertions to catch any corner cases that we may have missed. */

	volatile lint	n_active;

	/** Number of OS threads waiting in the FIFO for permission to
	enter InnoDB */
	volatile lint	n_waiting;
};

Since there's only one global variable of the srv_conc_t type
(srv_conc), the purpose of pad member variable is unclear.

One possible purpose is preventing neighboring global variables like
srv_thread_concurrency from sharing the cache line with other members of
srv_conc. But the chosen size of the padding does not actually
guarantee that (consider a case when srv_thread_concurrency is located
at the start of a cache line -- it will share the cache line with the
n_active variable).

Another possible reason is preventing cache line sharing between the
n_active and n_waiting structure members, but then again, the padding
size does not actually guarantee that (consider a case when srv_conc is
located at the start of a cache line).

How to repeat:
Code analysis.

Hi Kaamos, my dear friend,

Thank you for your report.

I can fully see your problem. I would like to hear opinion on the solution(s).

Do you think that this definition would be OK:

char		pad[64];

Or would this one be sufficient:

char		pad[64 - sizeof(lint)];

Many thanks in advance.

No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".