[dpdk-dev] [EXT] [PATCH] cryptodev: free memzone when releasing cryptodev

[Anoob Joseph]

Hi Akhil,

Please see inline.

Thanks,
Anoob

> -----Original Message-----
> From: Akhil Goyal <akhil.goyal at nxp.com>
> Sent: Friday, June 28, 2019 11:45 AM
> To: Anoob Joseph <anoobj at marvell.com>; Junxiao Shi
> <git at mail1.yoursunny.com>; dev at dpdk.org
> Cc: De Lara Guarch, Pablo <pablo.de.lara.guarch at intel.com>
> Subject: RE: [EXT] [dpdk-dev] [PATCH] cryptodev: free memzone when
> releasing cryptodev
> 
> Hi Anoob,
> > >
> > > --------------------------------------------------------------------
> > > -- When a cryptodev is created in a primary process,
> > > rte_cryptodev_data_alloc reserves a memzone.
> > > However, this memzone was not released when the cryptodev is
> uninitialized.
> > > After that, new cryptodev cannot be created due to memzone name
> conflict.
> > >
> > > This commit frees the memzone when a cryptodev is uninitialized,
> > > fixing this bug. This approach is chosen instead of keeping and
> > > reusing the old memzone, because the new cryptodev could belong to a
> different NUMA socket.
> > >
> > > Also, rte_cryptodev_data pointer is now properly recorded in
> > > cryptodev_globals.data array.
> > >
> > > Bugzilla ID: 105
> > >
> > > Signed-off-by: Junxiao Shi <git at mail1.yoursunny.com>
> > > ---
> > >  lib/librte_cryptodev/rte_cryptodev.c | 44
> > > +++++++++++++++++++++++++++++++-----
> > >  1 file changed, 38 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/lib/librte_cryptodev/rte_cryptodev.c
> > > b/lib/librte_cryptodev/rte_cryptodev.c
> > > index 00c2cf4..666dfea 100644
> > > --- a/lib/librte_cryptodev/rte_cryptodev.c
> > > +++ b/lib/librte_cryptodev/rte_cryptodev.c
> > > @@ -653,6 +653,31 @@ rte_cryptodev_data_alloc(uint8_t dev_id, struct
> > > rte_cryptodev_data **data,
> > >  	return 0;
> > >  }
> > >
> > > +static inline int
> > > +rte_cryptodev_data_free(uint8_t dev_id, struct rte_cryptodev_data
> > > +**data) {
> > > +	char mz_name[RTE_CRYPTODEV_NAME_MAX_LEN];

[Anoob] Shouldn't we use RTE_MEMZONE_NAMESIZE instead? I guess this is also coming from the existing code in rte_cryptodev_data_alloc(). May be we should fix that as well?
 
> > > +	const struct rte_memzone *mz;
> > > +	int n;
> > > +
> > > +	/* generate memzone name */
> > > +	n = snprintf(mz_name, sizeof(mz_name),
> "rte_cryptodev_data_%u",
> > > dev_id);
> > > +	if (n >= (int)sizeof(mz_name))
> > > +		return -EINVAL;
> >
> > [Anoob] Is the above check needed?
> I believe this being used while creating the memzone, so same logic is used
> while freeing it.
> Just to be safe.
> 

[Anoob] Thinking bit more, it seems like we are trying to capture a situation when the name is getting truncated because of insufficient buffer space. So it is safe to have I guess. But even in that case, 'n' will not be greater than the "size" field passed (which happens to be sizeof(mz_name) in our case).

My opinion is '==' might make more sense. But I leave that to your judgement. 
 
> >
> > > +
> > > +	mz = rte_memzone_lookup(mz_name);
> > > +	if (mz == NULL)
> > > +		return -ENOMEM;
> >
> > [Anoob] Is the return value correct? Shouldn't it be -EINVAL?
> >
> > @Akhil, thoughts?
> 
> 
> I believe ENOMEM is correct, as there is no memory associated with the
> cryptodev_data.

[Anoob] Agreed.

> 
> >
> > > +
> > > +	RTE_ASSERT(*data == mz->addr);
> > > +	*data = NULL;
> > > +
> > > +	if (rte_eal_process_type() == RTE_PROC_PRIMARY)
> > > +		return rte_memzone_free(mz);
> > > +
> > > +	return 0;
> > > +}
> > > +
> > >  static uint8_t
> > >  rte_cryptodev_find_free_device_index(void)
> > >  {
> > > @@ -687,16 +712,16 @@ rte_cryptodev_pmd_allocate(const char
> *name,
> > > int
> > > socket_id)
> > >  	cryptodev = rte_cryptodev_pmd_get_dev(dev_id);
> > >
> > >  	if (cryptodev->data == NULL) {
> > > -		struct rte_cryptodev_data *cryptodev_data =
> > > -				cryptodev_globals.data[dev_id];
> > > +		struct rte_cryptodev_data **cryptodev_data =
> > > +				&cryptodev_globals.data[dev_id];
> > >
> > > -		int retval = rte_cryptodev_data_alloc(dev_id,
> &cryptodev_data,
> > > +		int retval = rte_cryptodev_data_alloc(dev_id,
> cryptodev_data,
> > >  				socket_id);
> > >
> > > -		if (retval < 0 || cryptodev_data == NULL)
> > > +		if (retval < 0 || *cryptodev_data == NULL)
> > >  			return NULL;
> > >
> > > -		cryptodev->data = cryptodev_data;
> > > +		cryptodev->data = *cryptodev_data;
> > >
> > >  		strlcpy(cryptodev->data->name, name,
> > >  			RTE_CRYPTODEV_NAME_MAX_LEN);
> > > @@ -724,13 +749,20 @@ rte_cryptodev_pmd_release_device(struct
> > > rte_cryptodev *cryptodev)
> > >  	if (cryptodev == NULL)
> > >  		return -EINVAL;
> > >
> > > +	uint8_t dev_id = cryptodev->data->dev_id;
> > > +
> >
> > [Anoob] Variables need to be declared at the start of the function.
> > https://doc.dpdk.org/guides/contributing/coding_style.html
> >
> > >  	/* Close device only if device operations have been set */
> > >  	if (cryptodev->dev_ops) {
> > > -		ret = rte_cryptodev_close(cryptodev->data->dev_id);
> > > +		ret = rte_cryptodev_close(dev_id);
> > >  		if (ret < 0)
> > >  			return ret;
> > >  	}
> > >
> > > +	struct rte_cryptodev_data **cryptodev_data =
> > > &cryptodev_globals.data[dev_id];
> >
> > [Anoob] Same comment as above
> >
> > > +	ret = rte_cryptodev_data_free(dev_id, cryptodev_data);
> > > +	if (ret < 0)
> > > +		return ret;
> > > +
> > >  	cryptodev->attached = RTE_CRYPTODEV_DETACHED;
> > >  	cryptodev_globals.nb_devs--;
> > >  	return 0;
> > > --
> > > 2.7.4