Re: [PATCH] soc: xilinx: Fix a memory leak in zynqmp_pm_remove()

From: Christophe JAILLET
Date: Mon May 29 2023 - 09:53:36 EST

Le 29/05/2023 à 11:59, Dan Carpenter a écrit :
On Mon, May 29, 2023 at 09:53:24AM +0200, Christophe JAILLET wrote:
'rx_chan' is known to be NULL here.
Reverse the logic to free the mbox if it has been allocated.

Fixes: ffdbae28d9d1 ("drivers: soc: xilinx: Use mailbox IPI callback")
Signed-off-by: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
'rx_chan' may be NULL, but mbox_free_channel() handles it.
Maybe it is more informative to keep a (useless) "if (rx_chan)" to tell
that it may not be allocated.

On my machine, compilation fails with gcc (Ubuntu 12.1.0-2ubuntu1~22.04):

CC drivers/soc/xilinx/zynqmp_power.o
drivers/soc/xilinx/zynqmp_power.c: In function ‘zynqmp_pm_probe’:
drivers/soc/xilinx/zynqmp_power.c:193:12: error: ‘pm_api_version’ is used uninitialized [-Werror=uninitialized]
193 | if (pm_api_version < ZYNQMP_PM_VERSION)
| ^
drivers/soc/xilinx/zynqmp_power.c:187:13: note: ‘pm_api_version’ was declared here
187 | u32 pm_api_version;
| ^~~~~~~~~~~~~~
cc1: all warnings being treated as errors

I think that this warning is bogus and gcc is wrong.

But I don't know what to do with it :/
Anyway, it is un-realated to this patch.

I bet GCC is correct.

Do you have CONFIG_ZYNQMP_FIRMWARE enabled in your .config? This driver
can only be compiled with that enabled, but I've seen some of your
other patches depend on CONFIG_BROKEN so I think you're going outside of
the Kconfig rules.

dan carpenter

Ok, got it. This is for arm64.

make does not behave the same when you build a file or a directory.
Sometimes it is convenient, sometimes it is surprising.

On x86:
make -j7 drivers/soc/xilinx/zynqmp_power.o --> fail
make -j7 drivers/soc/xilinx/ --> build nothing because I'm not on arm64

export ARCH=arm64
./make.cross -j7 drivers/soc/xilinx/zynqmp_power.o --> OK

Thanks for the pointer.