lkml.org 
[lkml]   [2018]   [Jun]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v4] staging: rts5208: add error handling into rtsx_probe
On Tue, Jun 19, 2018 at 6:25 PM, Anton Vasilyev <vasilyev@ispras.ru> wrote:
> If rtsx_probe() fails to allocate dev->chip, then release_everything()
> will crash on uninitialized dev->cmnd_ready complete

Period is missed at the end.

>
> Patch adds error handling into rtsx_probe.

an error

>
> Found by Linux Driver Verification project (linuxtesting.org).
>

Reviewed-by: Andy Shevchenko <andy.shevhchenko@gmail.com>

Couple of comments below though.

> Signed-off-by: Anton Vasilyev <vasilyev@ispras.ru>
> ---
> v4: rename labels baced on Dan Carpenter's recommendation
> v3: fix subject and commit message
> v2: Add error handling into rtsx_probe based on Dan Carpenter's comment.
> I do not have corresponding hardware, so patch was tested by compilation only.
>
> I faced with inaccuracy at rtsx_remove() and original rtsx_probe():
> there is quiesce_and_remove_host() call with scsi_remove_host() inside,
> whereas release_everything() calls scsi_host_put() after this
> scsi_remove_host() call. This is strange for me.
> Also I do not know is it require to check result value of
> rtsx_init_chip() call on rtsx_probe().
> ---
> drivers/staging/rts5208/rtsx.c | 34 ++++++++++++++++++++++++----------
> 1 file changed, 24 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/staging/rts5208/rtsx.c b/drivers/staging/rts5208/rtsx.c
> index 70e0b8623110..233026ee5dd4 100644
> --- a/drivers/staging/rts5208/rtsx.c
> +++ b/drivers/staging/rts5208/rtsx.c
> @@ -879,7 +879,7 @@ static int rtsx_probe(struct pci_dev *pci,
> if (!dev->remap_addr) {
> dev_err(&pci->dev, "ioremap error\n");
> err = -ENXIO;
> - goto errout;
> + goto err_chip_free;
> }
>
> /*
> @@ -894,7 +894,7 @@ static int rtsx_probe(struct pci_dev *pci,
> if (!dev->rtsx_resv_buf) {
> dev_err(&pci->dev, "alloc dma buffer fail\n");
> err = -ENXIO;
> - goto errout;
> + goto err_addr_unmap;
> }
> dev->chip->host_cmds_ptr = dev->rtsx_resv_buf;
> dev->chip->host_cmds_addr = dev->rtsx_resv_buf_addr;
> @@ -915,7 +915,7 @@ static int rtsx_probe(struct pci_dev *pci,
>
> if (rtsx_acquire_irq(dev) < 0) {
> err = -EBUSY;
> - goto errout;
> + goto err_disable_msi;
> }
>
> pci_set_master(pci);
> @@ -935,14 +935,14 @@ static int rtsx_probe(struct pci_dev *pci,
> if (IS_ERR(th)) {
> dev_err(&pci->dev, "Unable to start control thread\n");
> err = PTR_ERR(th);
> - goto errout;
> + goto err_rtsx_release;
> }
> dev->ctl_thread = th;
>
> err = scsi_add_host(host, &pci->dev);
> if (err) {
> dev_err(&pci->dev, "Unable to add the scsi host\n");
> - goto errout;
> + goto err_complete_control_thread;
> }
>
> /* Start up the thread for delayed SCSI-device scanning */
> @@ -950,18 +950,16 @@ static int rtsx_probe(struct pci_dev *pci,
> if (IS_ERR(th)) {
> dev_err(&pci->dev, "Unable to start the device-scanning thread\n");
> complete(&dev->scanning_done);
> - quiesce_and_remove_host(dev);
> err = PTR_ERR(th);
> - goto errout;
> + goto err_stop_host;
> }
>
> /* Start up the thread for polling thread */
> th = kthread_run(rtsx_polling_thread, dev, "rtsx-polling");
> if (IS_ERR(th)) {
> dev_err(&pci->dev, "Unable to start the device-polling thread\n");
> - quiesce_and_remove_host(dev);
> err = PTR_ERR(th);
> - goto errout;
> + goto err_stop_host;
> }
> dev->polling_thread = th;
>
> @@ -970,9 +968,25 @@ static int rtsx_probe(struct pci_dev *pci,
> return 0;
>
> /* We come here if there are any problems */
> +err_stop_host:
> + quiesce_and_remove_host(dev);
> +err_complete_control_thread:
> + complete(&dev->cmnd_ready);
> + wait_for_completion(&dev->control_exit);
> +err_rtsx_release:
> + free_irq(dev->irq, (void *)dev);
> + rtsx_release_chip(dev->chip);
> +err_disable_msi:

> + dev->chip->host_cmds_ptr = NULL;
> + dev->chip->host_sg_tbl_ptr = NULL;

This seems superfluous, you are freeing entire struct few calls after.

> + if (dev->chip->msi_en)
> + pci_disable_msi(dev->pci);

This might be material for another improvement, PCI core tracks
MSI/MSI-X enable state.
So, the conditional with the local variable looks superfluous as well
(in some, rare, cases we indeed need something like this to basically
distinguish MSI from MSI-X, though I don't think it's a case here)

> +err_addr_unmap:
> + iounmap(dev->remap_addr);
> +err_chip_free:
> + kfree(dev->chip);
> errout:
> dev_err(&pci->dev, "%s failed\n", __func__);
> - release_everything(dev);
>
> return err;
> }
> --
> 2.17.1
>



--
With Best Regards,
Andy Shevchenko

\
 
 \ /
  Last update: 2018-06-19 19:13    [W:0.102 / U:1.304 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site