Message ID | 20220929050426.955139-1-dmitry.torokhov@gmail.com (mailing list archive) |
---|---|
State | Changes Requested |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | [net-next,1/2] nfc: s3fwrn5: fix order of freeing resources | expand |
Context | Check | Description |
---|---|---|
netdev/tree_selection | success | Clearly marked for net-next |
netdev/apply | fail | Patch does not apply to net-next |
On 29/09/2022 07:04, Dmitry Torokhov wrote: > Caution needs to be exercised when mixing together regular and managed > resources. In case of this driver devm_request_threaded_irq() should > not be used, because it will result in the interrupt being freed too > late, and there being a chance that it fires up at an inopportune > moment and reference already freed data structures. Non-devm was so far recommended only for IRQF_SHARED, not for regular ones. Otherwise you have to fix half of Linux kernel drivers... why is s3fwrn5 special? Please use scripts/get_maintainers.pl to Cc also netdev folks. > > Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> > --- > drivers/nfc/s3fwrn5/i2c.c | 7 ++++--- > 1 file changed, 4 insertions(+), 3 deletions(-) > Best regards, Krzysztof
On 29/09/2022 09:26, Krzysztof Kozlowski wrote: > On 29/09/2022 07:04, Dmitry Torokhov wrote: >> Caution needs to be exercised when mixing together regular and managed >> resources. In case of this driver devm_request_threaded_irq() should >> not be used, because it will result in the interrupt being freed too >> late, and there being a chance that it fires up at an inopportune >> moment and reference already freed data structures. > > Non-devm was so far recommended only for IRQF_SHARED, not for regular > ones. Otherwise you have to fix half of Linux kernel drivers... why is > s3fwrn5 special? > > Please use scripts/get_maintainers.pl to Cc also netdev folks. Ah, they are not printed for NFC drivers. So never mind last comment. Best regards, Krzysztof
On September 29, 2022 12:27:19 AM PDT, Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> wrote: >On 29/09/2022 09:26, Krzysztof Kozlowski wrote: >> On 29/09/2022 07:04, Dmitry Torokhov wrote: >>> Caution needs to be exercised when mixing together regular and managed >>> resources. In case of this driver devm_request_threaded_irq() should >>> not be used, because it will result in the interrupt being freed too >>> late, and there being a chance that it fires up at an inopportune >>> moment and reference already freed data structures. >> >> Non-devm was so far recommended only for IRQF_SHARED, not for regular >> ones. If we are absolutely sure there is no possibility of interrupts firing then devm should be ok, but it is much safer not to use it. Or use custom devm actions to free non-managed resources. >> Otherwise you have to fix half of Linux kernel drivers... Yes, if they are doing the wrong thing. >> why is s3fwrn5 special? >> I simply happened to look at it. > >> Please use scripts/get_maintainers.pl to Cc also netdev folks. > >Ah, they are not printed for NFC drivers. So never mind last comment. > >Best regards, >Krzysztof > Thanks.
On 29/09/2022 09:37, Dmitry Torokhov wrote: > On September 29, 2022 12:27:19 AM PDT, Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> wrote: >> On 29/09/2022 09:26, Krzysztof Kozlowski wrote: >>> On 29/09/2022 07:04, Dmitry Torokhov wrote: >>>> Caution needs to be exercised when mixing together regular and managed >>>> resources. In case of this driver devm_request_threaded_irq() should >>>> not be used, because it will result in the interrupt being freed too >>>> late, and there being a chance that it fires up at an inopportune >>>> moment and reference already freed data structures. >>> >>> Non-devm was so far recommended only for IRQF_SHARED, not for regular >>> ones. > > If we are absolutely sure there is no possibility of interrupts firing then devm > should be ok, but it is much safer not to use it. Or use custom devm actions > to free non-managed resources. I am not sure and the pattern itself is a bit risky, I agree. However the driver calls s3fwrn5_remove() which then calls s3fwrn5_phy_power_ctrl() which cuts the power via GPIO pin. I assume that the hardware should stop generating interrupts at this point. > >>> Otherwise you have to fix half of Linux kernel drivers... > > Yes, if they are doing the wrong thing. What I meant, that this pattern appears pretty often. If we agree that this driver has a risky pattern (hardware might not be off after remove() callback), then we should maybe document it somewhere and include it in usual reviews. Best regards, Krzysztof
On September 29, 2022 12:42:08 AM PDT, Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> wrote: >On 29/09/2022 09:37, Dmitry Torokhov wrote: >> On September 29, 2022 12:27:19 AM PDT, Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> wrote: >>> On 29/09/2022 09:26, Krzysztof Kozlowski wrote: >>>> On 29/09/2022 07:04, Dmitry Torokhov wrote: >>>>> Caution needs to be exercised when mixing together regular and managed >>>>> resources. In case of this driver devm_request_threaded_irq() should >>>>> not be used, because it will result in the interrupt being freed too >>>>> late, and there being a chance that it fires up at an inopportune >>>>> moment and reference already freed data structures. >>>> >>>> Non-devm was so far recommended only for IRQF_SHARED, not for regular >>>> ones. >> >> If we are absolutely sure there is no possibility of interrupts firing then devm >> should be ok, but it is much safer not to use it. Or use custom devm actions >> to free non-managed resources. > >I am not sure and the pattern itself is a bit risky, I agree. However >the driver calls s3fwrn5_remove() which then calls >s3fwrn5_phy_power_ctrl() which cuts the power via GPIO pin. I assume >that the hardware should stop generating interrupts at this point. Ok, fair enough. The GPIO is mandatory, so I guess the chip will be disabled. > >> >>>> Otherwise you have to fix half of Linux kernel drivers... >> >> Yes, if they are doing the wrong thing. > >What I meant, that this pattern appears pretty often. If we agree that >this driver has a risky pattern (hardware might not be off after >remove() callback), then we should maybe document it somewhere and >include it in usual reviews. I have been saying that mixing managed and non managed resource is dangerous for years. Disabling clocks before shutting off interrupts, freeing memory, etc, is not safe. The best approach with devm is all or nothing. Thanks.
diff --git a/drivers/nfc/s3fwrn5/i2c.c b/drivers/nfc/s3fwrn5/i2c.c index f824dc7099ce..fb36860df81c 100644 --- a/drivers/nfc/s3fwrn5/i2c.c +++ b/drivers/nfc/s3fwrn5/i2c.c @@ -231,9 +231,9 @@ static int s3fwrn5_i2c_probe(struct i2c_client *client, if (ret < 0) goto disable_clk; - ret = devm_request_threaded_irq(&client->dev, phy->i2c_dev->irq, NULL, - s3fwrn5_i2c_irq_thread_fn, IRQF_ONESHOT, - S3FWRN5_I2C_DRIVER_NAME, phy); + ret = request_threaded_irq(phy->i2c_dev->irq, + NULL, s3fwrn5_i2c_irq_thread_fn, + IRQF_ONESHOT, S3FWRN5_I2C_DRIVER_NAME, phy); if (ret) goto s3fwrn5_remove; @@ -250,6 +250,7 @@ static void s3fwrn5_i2c_remove(struct i2c_client *client) { struct s3fwrn5_i2c_phy *phy = i2c_get_clientdata(client); + free_irq(phy->i2c_dev->irq, phy); s3fwrn5_remove(phy->common.ndev); clk_disable_unprepare(phy->clk); }
Caution needs to be exercised when mixing together regular and managed resources. In case of this driver devm_request_threaded_irq() should not be used, because it will result in the interrupt being freed too late, and there being a chance that it fires up at an inopportune moment and reference already freed data structures. Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> --- drivers/nfc/s3fwrn5/i2c.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)