[bug report] EDAC: Add a driver for the AMD Versal NET DDR controller

[bug report] EDAC: Add a driver for the AMD Versal NET DDR controller

[Dan Carpenter]

Hello Shubhrajyoti Datta,

Commit d5fe2fec6c40 ("EDAC: Add a driver for the AMD Versal NET DDR
controller") from Sep 8, 2025 (linux-next), leads to the following
Smatch static checker warning:

	drivers/edac/versalnet_edac.c:849 init_versalnet()
	warn: '_res' from device_register() not released on lines: 849.

drivers/edac/versalnet_edac.c
    761 static int init_versalnet(struct mc_priv *priv, struct platform_device *pdev)
    762 {
    763         u32 num_chans, rank, dwidth, config;
    764         struct edac_mc_layer layers[2];
    765         struct mem_ctl_info *mci;
    766         struct device *dev;
    767         enum dev_type dt;
    768         char *name;
    769         int rc, i;
    770 
    771         for (i = 0; i < NUM_CONTROLLERS; i++) {
    772                 config = priv->adec[CONF + i * ADEC_NUM];
    773                 num_chans = FIELD_GET(MC5_NUM_CHANS_MASK, config);
    774                 rank = 1 << FIELD_GET(MC5_RANK_MASK, config);
    775                 dwidth = FIELD_GET(MC5_BUS_WIDTH_MASK, config);
    776 
    777                 switch (dwidth) {
    778                 case XDDR5_BUS_WIDTH_16:
    779                         dt = DEV_X16;
    780                         break;
    781                 case XDDR5_BUS_WIDTH_32:
    782                         dt = DEV_X32;
    783                         break;
    784                 case XDDR5_BUS_WIDTH_64:
    785                         dt = DEV_X64;
    786                         break;
    787                 default:
    788                         dt = DEV_UNKNOWN;
    789                 }
    790 
    791                 if (dt == DEV_UNKNOWN)
    792                         continue;
    793 
    794                 /* Find the first enabled device and register that one. */
    795                 layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
    796                 layers[0].size = rank;
    797                 layers[0].is_virt_csrow = true;
    798                 layers[1].type = EDAC_MC_LAYER_CHANNEL;
    799                 layers[1].size = num_chans;
    800                 layers[1].is_virt_csrow = false;
    801 
    802                 rc = -ENOMEM;
    803                 mci = edac_mc_alloc(i, ARRAY_SIZE(layers), layers,
    804                                     sizeof(struct mc_priv));
    805                 if (!mci) {
    806                         edac_printk(KERN_ERR, EDAC_MC, "Failed memory allocation for MC%d\n", i);
    807                         goto err_alloc;
    808                 }
    809 
    810                 priv->mci[i] = mci;
    811                 priv->dwidth = dt;
    812 
    813                 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
    814                 dev->release = versal_edac_release;
    815                 name = kmalloc(32, GFP_KERNEL);
    816                 sprintf(name, "versal-net-ddrmc5-edac-%d", i);
    817                 dev->init_name = name;
    818                 rc = device_register(dev);
    819                 if (rc)
    820                         goto err_alloc;

This should do a edac_mc_free(mci) before the goto.

    821 
    822                 mci->pdev = dev;
    823 
    824                 platform_set_drvdata(pdev, priv);
    825 
    826                 mc_init(mci, dev);
    827                 rc = edac_mc_add_mc(mci);
    828                 if (rc) {
    829                         edac_printk(KERN_ERR, EDAC_MC, "Failed to register MC%d with EDAC core\n", i);
    830                         goto err_alloc;

I guess this should unregister and edac_mc_free mci before the goto.

    831                 }
    832         }
    833         return 0;
    834 
    835 err_alloc:
    836         while (i--) {
    837                 mci = priv->mci[i];
    838                 if (!mci)

No need for this check.  We know it's non-NULL.

    839                         continue;
    840 
    841                 if (mci->pdev) {

Unnecessary NULL check.

    842                         device_unregister(mci->pdev);
    843                         edac_mc_del_mc(mci->pdev);

I would have thought these would be in the other order but I don't know
if it matters.  I have a paragraph on unwinding from loops at the end of
my blog:

https://staticthinking.wordpress.com/2022/04/28/free-the-last-thing-style/

    844                 }
    845 
    846                 edac_mc_free(mci);
    847         }
    848 
--> 849         return rc;
    850 }

regards,
dan carpenter

Re: [bug report] EDAC: Add a driver for the AMD Versal NET DDR controller

[Borislav Petkov]

On Thu, Sep 18, 2025 at 09:55:52AM +0300, Dan Carpenter wrote:
> Hello Shubhrajyoti Datta,
> 
> Commit d5fe2fec6c40 ("EDAC: Add a driver for the AMD Versal NET DDR
> controller") from Sep 8, 2025 (linux-next), leads to the following
> Smatch static checker warning:
> 
> 	drivers/edac/versalnet_edac.c:849 init_versalnet()
> 	warn: '_res' from device_register() not released on lines: 849.

_res?

No _res there.

In any case, this mess is probably due to me trying to salvage a crap
situation already.

The proper fix is carving out the loop body into a __init_versalnet_mc() or so
function which does the allocation and everything along with goto labels for
error path for a *single* memory controller.

And then add another function which unwinds and frees everything, perhaps
__free_versalnet_mc().

Then this init_versalnet() function would only run the loop and call the
__init one. If the __init one returns an error, it would do the unwinding
using the __free one.

Sounds like a plan?

Would you like to give it a try and run your tool ontop to verify?

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette