Re: [PATCH 1/4] soc: fsl: qbman: FSL_DPAA depends on COMPILE_TEST

From: Breno Leitao
Date: Mon Jul 08 2024 - 15:08:44 EST


Hello Vladimir,

On Mon, Jul 08, 2024 at 04:37:46PM +0300, Vladimir Oltean wrote:
> On Thu, Jun 27, 2024 at 11:40:24AM -0700, Breno Leitao wrote:

> > > > 454 | static int dpaa_set_coalesce(struct net_device *dev,
> > > > | ^
> > > > 1 warning generated.
> > >
> > > Arrays of NR_CPUS elements are what it probably doesn't like?

> > Can it use the number of online CPUs instead of NR_CPUS?

> I don't see how, given that variable length arrays are something which
> should be avoided in the kernel?

I thought about a patch like the following (compile tested only). What
do you think?

Author: Breno Leitao <leitao@xxxxxxxxxx>
Date: Mon Jul 8 11:57:33 2024 -0700

net: dpaa: Allocate only for online CPUs in dpaa_set_coalesce

Currently, dpaa_set_coalesce allocates a boolean for every possible CPU
(NR_CPUS). This approach is suboptimal and causes failures in COMPILE_TEST.
For reference, see:
https://lore.kernel.org/all/202406261920.l5pzM1rj-lkp@xxxxxxxxx/

Modify the allocation to consider only online CPUs instead of
NR_CPUs. This change reduces the function's memory footprint and resolves
the COMPILE_TEST issues.

Signed-off-by: Breno Leitao <leitao@xxxxxxxxxx>

diff --git a/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
index 5bd0b36d1feb..7202a5310045 100644
--- a/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
+++ b/drivers/net/ethernet/freescale/dpaa/dpaa_ethtool.c
@@ -457,7 +457,7 @@ static int dpaa_set_coalesce(struct net_device *dev,
struct netlink_ext_ack *extack)
{
const cpumask_t *cpus = qman_affine_cpus();
- bool needs_revert[NR_CPUS] = {false};
+ bool *needs_revert;
struct qman_portal *portal;
u32 period, prev_period;
u8 thresh, prev_thresh;
@@ -466,6 +466,11 @@ static int dpaa_set_coalesce(struct net_device *dev,
period = c->rx_coalesce_usecs;
thresh = c->rx_max_coalesced_frames;

+ needs_revert = kmalloc_array(num_possible_cpus(), sizeof(bool), GFP_KERNEL);
+ if (!needs_revert)
+ return -ENOMEM;
+ memset(needs_revert, 0, num_online_cpus() * sizeof(bool));
+
/* save previous values */
portal = qman_get_affine_portal(smp_processor_id());
qman_portal_get_iperiod(portal, &prev_period);
@@ -498,6 +503,7 @@ static int dpaa_set_coalesce(struct net_device *dev,
qman_dqrr_set_ithresh(portal, prev_thresh);
}

+ kfree(needs_revert);
return res;
}