Re: [PATCH v4 11/16] rpmsg: virtio: register the rpmsg_ctrl device

From: Arnaud POULIQUEN
Date: Thu Feb 18 2021 - 13:58:57 EST


Hi Dan,

On 2/18/21 1:27 PM, Dan Carpenter wrote:
> Hi Arnaud,
>
> url: https://github.com/0day-ci/linux/commits/Arnaud-Pouliquen/introduce-a-generic-IOCTL-interface-for-RPMsg-channels-management/20210217-214044
> base: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git f40ddce88593482919761f74910f42f4b84c004b
> config: x86_64-randconfig-m001-20210215 (attached as .config)
> compiler: gcc-9 (Debian 9.3.0-15) 9.3.0
>
> If you fix the issue, kindly add following tag as appropriate
> Reported-by: kernel test robot <lkp@xxxxxxxxx>
> Reported-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
>
> smatch warnings:
> drivers/rpmsg/virtio_rpmsg_bus.c:977 rpmsg_probe() error: uninitialized symbol 'vch'.
>
> vim +/vch +977 drivers/rpmsg/virtio_rpmsg_bus.c
>
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 845 static int rpmsg_probe(struct virtio_device *vdev)
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 846 {
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 847 vq_callback_t *vq_cbs[] = { rpmsg_recv_done, rpmsg_xmit_done };
> f7ad26ff952b3c Stefan Hajnoczi 2015-12-17 848 static const char * const names[] = { "input", "output" };
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 849 struct virtqueue *vqs[2];
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 850 struct virtproc_info *vrp;
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 851 struct virtio_rpmsg_channel *vch;
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 852 struct rpmsg_device *rpdev_ns = NULL, *rpdev_ctrl = NULL;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 853 void *bufs_va;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 854 int err = 0, i;
> b1b9891441fa33 Suman Anna 2014-09-16 855 size_t total_buf_space;
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 856 bool notify;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 857
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 858 vrp = kzalloc(sizeof(*vrp), GFP_KERNEL);
>
> You might want to consider updating this stuff to devm_kzalloc() which
> is trendy with the kids these days. It's cleaned up automatically when
> the module is released.
>
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 859 if (!vrp)
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 860 return -ENOMEM;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 861
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 862 vrp->vdev = vdev;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 863
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 864 idr_init(&vrp->endpoints);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 865 mutex_init(&vrp->endpoints_lock);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 866 mutex_init(&vrp->tx_lock);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 867 init_waitqueue_head(&vrp->sendq);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 868
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 869 /* We expect two virtqueues, rx and tx (and in this order) */
> 9b2bbdb2275884 Michael S. Tsirkin 2017-03-06 870 err = virtio_find_vqs(vdev, 2, vqs, vq_cbs, names, NULL);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 871 if (err)
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 872 goto free_vrp;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 873
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 874 vrp->rvq = vqs[0];
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 875 vrp->svq = vqs[1];
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 876
> b1b9891441fa33 Suman Anna 2014-09-16 877 /* we expect symmetric tx/rx vrings */
> b1b9891441fa33 Suman Anna 2014-09-16 878 WARN_ON(virtqueue_get_vring_size(vrp->rvq) !=
> b1b9891441fa33 Suman Anna 2014-09-16 879 virtqueue_get_vring_size(vrp->svq));
> b1b9891441fa33 Suman Anna 2014-09-16 880
> b1b9891441fa33 Suman Anna 2014-09-16 881 /* we need less buffers if vrings are small */
> b1b9891441fa33 Suman Anna 2014-09-16 882 if (virtqueue_get_vring_size(vrp->rvq) < MAX_RPMSG_NUM_BUFS / 2)
> b1b9891441fa33 Suman Anna 2014-09-16 883 vrp->num_bufs = virtqueue_get_vring_size(vrp->rvq) * 2;
> b1b9891441fa33 Suman Anna 2014-09-16 884 else
> b1b9891441fa33 Suman Anna 2014-09-16 885 vrp->num_bufs = MAX_RPMSG_NUM_BUFS;
> b1b9891441fa33 Suman Anna 2014-09-16 886
> f93848f9eeb0f8 Loic Pallardy 2017-03-28 887 vrp->buf_size = MAX_RPMSG_BUF_SIZE;
> f93848f9eeb0f8 Loic Pallardy 2017-03-28 888
> f93848f9eeb0f8 Loic Pallardy 2017-03-28 889 total_buf_space = vrp->num_bufs * vrp->buf_size;
> b1b9891441fa33 Suman Anna 2014-09-16 890
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 891 /* allocate coherent memory for the buffers */
> d999b622fcfb39 Loic Pallardy 2019-01-10 892 bufs_va = dma_alloc_coherent(vdev->dev.parent,
> b1b9891441fa33 Suman Anna 2014-09-16 893 total_buf_space, &vrp->bufs_dma,
> b1b9891441fa33 Suman Anna 2014-09-16 894 GFP_KERNEL);
> 3119b487e03650 Wei Yongjun 2013-04-29 895 if (!bufs_va) {
> 3119b487e03650 Wei Yongjun 2013-04-29 896 err = -ENOMEM;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 897 goto vqs_del;
> 3119b487e03650 Wei Yongjun 2013-04-29 898 }
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 899
> de4064af76537f Suman Anna 2018-10-23 900 dev_dbg(&vdev->dev, "buffers: va %pK, dma %pad\n",
> 8d95b322ba34b1 Anna, Suman 2016-08-12 901 bufs_va, &vrp->bufs_dma);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 902
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 903 /* half of the buffers is dedicated for RX */
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 904 vrp->rbufs = bufs_va;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 905
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 906 /* and half is dedicated for TX */
> b1b9891441fa33 Suman Anna 2014-09-16 907 vrp->sbufs = bufs_va + total_buf_space / 2;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 908
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 909 /* set up the receive buffers */
> b1b9891441fa33 Suman Anna 2014-09-16 910 for (i = 0; i < vrp->num_bufs / 2; i++) {
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 911 struct scatterlist sg;
> f93848f9eeb0f8 Loic Pallardy 2017-03-28 912 void *cpu_addr = vrp->rbufs + i * vrp->buf_size;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 913
> 9dd87c2af651b0 Loic Pallardy 2017-03-28 914 rpmsg_sg_init(&sg, cpu_addr, vrp->buf_size);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 915
> cee51d69a45b6c Rusty Russell 2013-03-20 916 err = virtqueue_add_inbuf(vrp->rvq, &sg, 1, cpu_addr,
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 917 GFP_KERNEL);
> 57e1a37347d31c Rusty Russell 2012-10-16 918 WARN_ON(err); /* sanity check; this can't really happen */
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 919 }
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 920
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 921 /* suppress "tx-complete" interrupts */
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 922 virtqueue_disable_cb(vrp->svq);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 923
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 924 vdev->priv = vrp;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 925
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 926 /* if supported by the remote processor, enable the name service */
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 927 if (virtio_has_feature(vdev, VIRTIO_RPMSG_F_NS)) {
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 928 vch = kzalloc(sizeof(*vch), GFP_KERNEL);
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 929 if (!vch) {
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 930 err = -ENOMEM;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 931 goto free_coherent;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 932 }
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 933
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 934 /* Link the channel to our vrp */
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 935 vch->vrp = vrp;
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 936
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 937 /* Assign public information to the rpmsg_device */
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 938 rpdev_ns = &vch->rpdev;
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 939 rpdev_ns->ops = &virtio_rpmsg_ops;
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 940 rpdev_ns->little_endian = virtio_is_little_endian(vrp->vdev);
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 941
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 942 rpdev_ns->dev.parent = &vrp->vdev->dev;
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 943 rpdev_ns->dev.release = virtio_rpmsg_release_device;
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 944
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 945 err = rpmsg_ns_register_device(rpdev_ns);
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 946 if (err)
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 947 goto free_coherent;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 948 }
>
> "vch" not initialized on else path. Also keep in mind that "rpdev_ctrl"
> is NULL at this point.
>
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 949
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 950 rpdev_ctrl = rpmsg_virtio_add_char_dev(vdev);
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 951 if (IS_ERR(rpdev_ctrl)) {
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 952 err = PTR_ERR(rpdev_ctrl);
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 953 goto free_coherent;
>
> I should create a Smatch warning for this as well.
>
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 954 }
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 955 /*
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 956 * Prepare to kick but don't notify yet - we can't do this before
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 957 * device is ready.
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 958 */
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 959 notify = virtqueue_kick_prepare(vrp->rvq);
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 960
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 961 /* From this point on, we can notify and get callbacks. */
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 962 virtio_device_ready(vdev);
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 963
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 964 /* tell the remote processor it can start sending messages */
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 965 /*
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 966 * this might be concurrent with callbacks, but we are only
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 967 * doing notify, not a full kick here, so that's ok.
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 968 */
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 969 if (notify)
> 71e4b8bf0482fc Michael S. Tsirkin 2015-03-12 970 virtqueue_notify(vrp->rvq);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 971
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 972 dev_info(&vdev->dev, "rpmsg host is online\n");
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 973
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 974 return 0;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 975
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 976 free_coherent:
> 950a7388f02bf7 Arnaud Pouliquen 2020-11-20 @977 kfree(vch);
> ^^^^^^^^^^
> Uninitalized.
>
> e3bba4363fe87b Arnaud Pouliquen 2021-02-17 978 kfree(to_virtio_rpmsg_channel(rpdev_ctrl));
>
> Now, let's say that "rpdev_ctrl" is NULL. That's fine because
> to_virtio_rpmsg_channel() is a no-op so it becomes kfree(NULL) which
> is a no-op. But why even have the to_virtio_rpmsg_channel() function
> if we are going to rely on it being a no-op?
>
> If "rpdev_ctrl" is an error pointer this will crash. The problem is we
> are freeing stuff that was not allocated. The fix for this is:
>
> 1) Free the most recent successful allocation.
> 2) The unwind code should mirror the allocation code.
> 3) Choose label names which say what the goto does. If the most recent
> allocation was "vch" the goto should be "goto free_vch;"
> 4) Every allocation function should have a matching free function. It's
> a layering violation to have to know that the internals of
> rpmsg_virtio_add_char_dev().
>
> So create function:
>
> void rpmsg_virtio_del_char_dev(struct rpmsg_device *rpdev_ctrl)
> {
> if (!rpdev_ctrl)
> return;
> kfree(to_virtio_rpmsg_channel(rpdev_ctrl));
> }
>
> The clean up code looks like this:
>
> return 0;
>
> free_vch:
> if (virtio_has_feature(vdev, VIRTIO_RPMSG_F_NS))
> kfree(vch);
> free_ctrl:
> rpmsg_virtio_del_char_dev(rpdev_ctrl);
> free_coherent:
> dma_free_coherent(vdev->dev.parent, total_buf_space,
> bufs_va, vrp->bufs_dma);
> vqs_del:
> vdev->config->del_vqs(vrp->vdev);
>
> Then just go through the function and as you read down the code keep
> track of the most recent successful allocation and goto the correct
> free label.

You're right, my error management is very bad here. I will fix this based on
your recommandation.

Thanks,
Arnaud

>
> d999b622fcfb39 Loic Pallardy 2019-01-10 979 dma_free_coherent(vdev->dev.parent, total_buf_space,
> eeb0074f36d1ab Fernando Guzman Lugo 2012-08-29 980 bufs_va, vrp->bufs_dma);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 981 vqs_del:
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 982 vdev->config->del_vqs(vrp->vdev);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 983 free_vrp:
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 984 kfree(vrp);
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 985 return err;
> bcabbccabffe73 Ohad Ben-Cohen 2011-10-20 986 }
>
> ---
> 0-DAY CI Kernel Test Service, Intel Corporation
> https://lists.01.org/hyperkitty/list/kbuild-all@xxxxxxxxxxxx
>