RE: [EXT] Re: [net v4 PATCH 1/1] net: mvpp2: clear BM pool before initialization

[Jenishkumar Patel [C]]

-----Original Message-----
From: Maxime Chevallier <maxime.chevallier@xxxxxxxxxxx> 
Sent: Friday, January 19, 2024 7:35 PM
To: Jenishkumar Patel [C] <jpatel2@xxxxxxxxxxx>
Cc: marcin.s.wojtas@xxxxxxxxx; linux@xxxxxxxxxxxxxxx; davem@xxxxxxxxxxxxx; edumazet@xxxxxxxxxx; kuba@xxxxxxxxxx; pabeni@xxxxxxxxxx; netdev@vger.kernelorg; linux-kernel@xxxxxxxxxxxxxxx; Antoine Tenart <atenart@xxxxxxxxxx>
Subject: [EXT] Re: [net v4 PATCH 1/1] net: mvpp2: clear BM pool before initialization

External Email

----------------------------------------------------------------------
Hello,

On Thu, 18 Jan 2024 19:59:14 -0800
Jenishkumar Maheshbhai Patel <jpatel2@xxxxxxxxxxx> wrote:

> Register value persist after booting the kernel using kexec which 
> results in kernel panic. Thus clear the BM pool registers before 
> initialisation to fix the issue.
> 
> Fixes: 3f518509dedc ("ethernet: Add new driver for Marvell Armada 375 
> network unit")
> Signed-off-by: Jenishkumar Maheshbhai Patel <jpatel2@xxxxxxxxxxx>
> ---
> v1-v2:
> -Move comments outside the loop
> -remove unrequired brances.
> v2-v3:
> -improve readability
> -correct register read API
> v3-v4:
> -optimize the code
> -improve readability

Thanks for taking the reviews into account, however ...

>  .../net/ethernet/marvell/mvpp2/mvpp2_main.c   | 27 ++++++++++++++++++-
>  1 file changed, 26 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c 
> b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
> index 820b1fabe297..23adf53c2aa1 100644
> --- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
> +++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
> @@ -614,12 +614,38 @@ static void mvpp23_bm_set_8pool_mode(struct mvpp2 *priv)
>  	mvpp2_write(priv, MVPP22_BM_POOL_BASE_ADDR_HIGH_REG, val);  }
>  
> +/* Cleanup pool before actual initialization in the OS */ static void 
> +mvpp2_bm_pool_cleanup(struct mvpp2 *priv, int pool_id) {
> +	unsigned int thread = mvpp2_cpu_to_thread(priv, get_cpu());
> +	u32 val;
> +	int i;
> +
> +	/* Drain the BM from all possible residues left by firmware */
> +	for (i = 0; i < MVPP2_BM_POOL_SIZE_MAX; i++)
> +		mvpp2_thread_read(priv, thread, MVPP2_BM_PHY_ALLOC_REG(pool_id));

.. I think you didn't answer Antoine's comment on that loop from the V2, regarding what this does exactly. From the other sites this is used, it seems to perform an allocation from the pool, can you clarify how safe it is to do so here, if for example the BM was never configured by the firmware beforehand and is therefore already in a Stopped state ?

Reading the register provides a pointer to buffer that is already allocated during BM initialization. When multiple reading is done on the register, it will drain all the pointers that are stored by previous firmware. Also reading this register does not perform any allocation as it is only performing register read operation, thus when the BM is not configured earlier then it will not lead to any stop state.

And are we not risking any leak if there was something in the pool that we don't release ?

The data on the pointer given by register read is written after the read operation is preformed, which means the pointer does not contain any data, thus there is no leak.

Thanks,

Maxime