Re: [PATCH] Bluetooth: btintel_pcie: Fix the error handling path of btintel_pcie_probe()

[Christophe JAILLET <christophe.jaillet@wanadoo.fr>]

Le 24/05/2024 à 21:39, Luiz Augusto von Dentz a écrit :
> Hi Christophe,
> 
> On Mon, May 20, 2024 at 3:42 AM Christophe JAILLET
> <christophe.jaillet@wanadoo.fr> wrote:
>>
>> Some resources freed in the remove function are not handled by the error
>> handling path of the probe.
>>
>> Add the needed function calls.
>>
>> Fixes: c2b636b3f788 ("Bluetooth: btintel_pcie: Add support for PCIe transport")
>> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
>> ---
>> Compile tested only.
>> Maybe incomplete.
>> ---
>>   drivers/bluetooth/btintel_pcie.c | 20 ++++++++++++++------
>>   1 file changed, 14 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/bluetooth/btintel_pcie.c b/drivers/bluetooth/btintel_pcie.c
>> index 5b6805d87fcf..d572576d0dbc 100644
>> --- a/drivers/bluetooth/btintel_pcie.c
>> +++ b/drivers/bluetooth/btintel_pcie.c
>> @@ -1280,17 +1280,17 @@ static int btintel_pcie_probe(struct pci_dev *pdev,
>>
>>          err = btintel_pcie_config_pcie(pdev, data);
>>          if (err)
>> -               goto exit_error;
>> +               goto exit_destroy_worqueue;
>>
>>          pci_set_drvdata(pdev, data);
>>
>>          err = btintel_pcie_alloc(data);
>>          if (err)
>> -               goto exit_error;
>> +               goto exit_free_irq_vectors;
>>
>>          err = btintel_pcie_enable_bt(data);
>>          if (err)
>> -               goto exit_error;
>> +               goto exit_free_pcie;
>>
>>          /* CNV information (CNVi and CNVr) is in CSR */
>>          data->cnvi = btintel_pcie_rd_reg32(data, BTINTEL_PCIE_CSR_HW_REV_REG);
>> @@ -1299,17 +1299,25 @@ static int btintel_pcie_probe(struct pci_dev *pdev,
>>
>>          err = btintel_pcie_start_rx(data);
>>          if (err)
>> -               goto exit_error;
>> +               goto exit_free_pcie;
>>
>>          err = btintel_pcie_setup_hdev(data);
>>          if (err)
>> -               goto exit_error;
>> +               goto exit_free_pcie;
>>
>>          bt_dev_dbg(data->hdev, "cnvi: 0x%8.8x cnvr: 0x%8.8x", data->cnvi,
>>                     data->cnvr);
>>          return 0;
>>
>> -exit_error:
>> +exit_free_pcie:
>> +       btintel_pcie_free(data);
>> +
>> +exit_free_irq_vectors:
>> +       pci_free_irq_vectors(pdev);
>> +
>> +exit_destroy_worqueue:
>> +       destroy_workqueue(data->workqueue);
>> +
> 
> This looks a bit messy, perhaps we should really be calling
> btintel_pcie_remove instead and adapt it to check if a field has been
> initialized or not then proceed to free/cleanup/etc.
> 

Not sure it would be that easy / readable.

It would look like something like:
static void btintel_pcie_remove(struct pci_dev *pdev)
{
	struct btintel_pcie_data *data;

	data = pci_get_drvdata(pdev);

	btintel_pcie_reset_bt(data);
	for (int i = 0; i < data->alloc_vecs; i++) {
		struct msix_entry *msix_entry;

		msix_entry = &data->msix_entries[i];
		free_irq(msix_entry->vector, msix_entry);
	}

	if (data->alloc_vecs)
		pci_free_irq_vectors(pdev);

	btintel_pcie_release_hdev(data);

	flush_work(&data->rx_work);

	if (data->workqueue)
		destroy_workqueue(data->workqueue);

	if (data->dma_pool)
		btintel_pcie_free(data);

	pci_clear_master(pdev);

	pci_set_drvdata(pdev, NULL);
}

The added tests don't always look related to the function call just 
after it :

   - data->alloc_vecs vs pci_free_irq_vectors(), ok why not

   - data->dma_pool vs btintel_pcie_free() does not look that really 
obvious.


There is also another issue in the remove function. We call free_irq() 
on irq allocated with devm_request_threaded_irq().

I'll try to see if more managed resources usage and/or some 
devm_add_action_or_reset() could help.

CJ

>>          /* reset device before exit */
>>          btintel_pcie_reset_bt(data);
>>
>> --
>> 2.45.1
>>
> 
>