[PATCH v1 13/25] dma-buf: Use sequence counter with associated wound/wait mutex
christian.koenig at amd.com
Thu May 21 13:20:04 UTC 2020
Am 21.05.20 um 02:09 schrieb Ahmed S. Darwish:
> On Wed, May 20, 2020, Christian König wrote:
>> Am 19.05.20 um 23:45 schrieb Ahmed S. Darwish:
>>> A sequence counter write side critical section must be protected by some
>>> form of locking to serialize writers. If the serialization primitive is
>>> not disabling preemption implicitly, preemption has to be explicitly
>>> disabled before entering the sequence counter write side critical
>>> The dma-buf reservation subsystem uses plain sequence counters to manage
>>> updates to reservations. Writer serialization is accomplished through a
>>> wound/wait mutex.
>>> Acquiring a wound/wait mutex does not disable preemption, so this needs
>>> to be done manually before and after the write side critical section.
>>> Use the newly-added seqcount_ww_mutex_t instead:
>>> - It associates the ww_mutex with the sequence count, which enables
>>> lockdep to validate that the write side critical section is properly
>>> - It removes the need to explicitly add preempt_disable/enable()
>>> around the write side critical section because the write_begin/end()
>>> functions for this new data type automatically do this.
>>> If lockdep is disabled this ww_mutex lock association is compiled out
>>> and has neither storage size nor runtime overhead.
>> Mhm, is the dma_resv object the only user of this new seqcount_ww_mutex
>> variant ?
>> If yes we are trying to get rid of this sequence counter for quite some
>> time, so I would rather invest the additional time to finish this.
> In this patch series, each extra "seqcount with associated lock" data
> type costs us, exactly:
> - 1 typedef definition, seqcount_ww_mutex_t
> - 1 static initializer, SEQCNT_WW_MUTEX_ZERO()
> - 1 runtime initializer, seqcount_ww_mutex_init()
> Definitions for the typedef and the 2 initializers above are
> template-code one liners.
In this case I'm perfectly fine with this.
> The logic which automatically disables preemption upon entering a
> seqcount_ww_mutex_t write side critical section is also already shared
> with seqcount_mutex_t and any future, preemptible, associated lock.
> So, yes, dma-resv is the only user of seqcount_ww_mutex.
> But even in that case, given the one liner template code nature of
> seqcount_ww_mutex_t logic, it does not make sense to block the dma_resv
> and amdgpu change until at some point in the future the sequence counter
> is completely removed.
> **If and when** the sequence counter gets removed, please just remove
> the seqcount_ww_mutex_t data type with it. It will be extremely simple.
Completely agree, just wanted to prevent that we now add a lot of code
which gets removed again ~3 month from now.
> Ahmed S. Darwish
> Linutronix GmbH
More information about the amd-gfx