| Message ID | 20220124153253.3548853-1-broonie@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | ASoC: Add bounds checking for written values | expand |
On 24. 01. 22 16:32, Mark Brown wrote: > This series adds verification that values written to registers are in > bounds for controls since the core doesn't validate for us. As discussed, those conditions should be optional to eventually catch the wrong applications. I don't see any benefit to report the range error back when there is value masking code already. The users will note when the unwanted values are written to the hardware, or not? Thanks, Jaroslav
On Mon, Jan 24, 2022 at 05:29:50PM +0100, Jaroslav Kysela wrote: > On 24. 01. 22 16:32, Mark Brown wrote: > > This series adds verification that values written to registers are in > > bounds for controls since the core doesn't validate for us. > As discussed, those conditions should be optional to eventually catch the > wrong applications. I don't see any benefit to report the range error back > when there is value masking code already. The users will note when the > unwanted values are written to the hardware, or not? In general I'd say that silent failures are harder to work with than returning an error at the point where we notice that there's a problem, assuming userspace is paying any attention to the error return at all of course. We certainly have quite a lot of existing put() methods that do return errors and it's not like the application isn't invoking undefined behaviour so I don't see a problem here.
On Mon, 24 Jan 2022 17:52:46 +0100, Mark Brown wrote: > > On Mon, Jan 24, 2022 at 05:29:50PM +0100, Jaroslav Kysela wrote: > > On 24. 01. 22 16:32, Mark Brown wrote: > > > > This series adds verification that values written to registers are in > > > bounds for controls since the core doesn't validate for us. > > > As discussed, those conditions should be optional to eventually catch the > > wrong applications. I don't see any benefit to report the range error back > > when there is value masking code already. The users will note when the > > unwanted values are written to the hardware, or not? > > In general I'd say that silent failures are harder to work with than > returning an error at the point where we notice that there's a problem, > assuming userspace is paying any attention to the error return at all of > course. We certainly have quite a lot of existing put() methods that do > return errors and it's not like the application isn't invoking undefined > behaviour so I don't see a problem here. I find also it's more useful to have the proper checks in general. Jaroslav, is you concern only about the compatibility of user-space? Or anything else? The compatibility is always certainly a slight issue; if this breaks really something useful and actually working stuff, we need to consider the workaround... thanks, Takashi
On 24. 01. 22 18:08, Takashi Iwai wrote: > On Mon, 24 Jan 2022 17:52:46 +0100, > Mark Brown wrote: >> >> On Mon, Jan 24, 2022 at 05:29:50PM +0100, Jaroslav Kysela wrote: >>> On 24. 01. 22 16:32, Mark Brown wrote: >> >>>> This series adds verification that values written to registers are in >>>> bounds for controls since the core doesn't validate for us. >> >>> As discussed, those conditions should be optional to eventually catch the >>> wrong applications. I don't see any benefit to report the range error back >>> when there is value masking code already. The users will note when the >>> unwanted values are written to the hardware, or not? >> >> In general I'd say that silent failures are harder to work with than >> returning an error at the point where we notice that there's a problem, >> assuming userspace is paying any attention to the error return at all of >> course. We certainly have quite a lot of existing put() methods that do >> return errors and it's not like the application isn't invoking undefined >> behaviour so I don't see a problem here. > > I find also it's more useful to have the proper checks in general. > > Jaroslav, is you concern only about the compatibility of user-space? > Or anything else? The compatibility is always certainly a slight > issue; if this breaks really something useful and actually working > stuff, we need to consider the workaround... My concern is only based on the fact that this code is normally not reachable. It only costs some CPU ticks and adds extra code to the drivers without any other benefit. The applications should not use out of range values and if they do, the behavior is not defined (the drivers should only avoid crashes). The code seems to be added only to make things consistent for the test applications. I don't think that it's worth to care only for this reason. What is the goal for this code? The result with the proposed code will be - the SoC core has the extra validation conditions. The user space can eventually add similar checks to detect the broken applications. Basically, I think that those checks should be marked as debug and they should be optional like we do for CONFIG_SND_CTL_VALIDATION and enable only the fast path by default. Jaroslav
On Mon, Jan 24, 2022 at 08:20:25PM +0100, Jaroslav Kysela wrote: > On 24. 01. 22 18:08, Takashi Iwai wrote: > > Jaroslav, is you concern only about the compatibility of user-space? > > Or anything else? The compatibility is always certainly a slight > > issue; if this breaks really something useful and actually working > > stuff, we need to consider the workaround... > My concern is only based on the fact that this code is normally not > reachable. It only costs some CPU ticks and adds extra code to the drivers > without any other benefit. The applications should not use out of range > values and if they do, the behavior is not defined (the drivers should only > avoid crashes). > The code seems to be added only to make things consistent for the test > applications. I don't think that it's worth to care only for this reason. > What is the goal for this code? The result with the proposed code will be - > the SoC core has the extra validation conditions. We need these checks all the time for the generic ASoC controls since the values for the controls are stored in the underlying device's register map so the out of range values will be written to the hardware, pushing it out of specified use. That's not a great idea in general and in extreme cases could result in physical damage to the system. The biggest risk I see here is around speaker drivers since they deal with high powers, even ignoring silicon requirements we also don't currently enforce platform maximums that the machine drivers specify - that feature was added after an inexperienced user burnt out the speakers in their Chromebook since the speakers in the system were rated for much lower powers than the CODEC was able to deliver. > Basically, I think that those checks should be marked as debug and they > should be optional like we do for CONFIG_SND_CTL_VALIDATION and enable only > the fast path by default. Note also that for everything using these helpers the underlying register map will be regmap based and with the possible exception of MMIO based regmaps the cost of writing out the new register value will be overwhelmingly greater than that of the bounds checks. It is extremely hard to envision a scenario in which even a pathological application would be able to observe a meaningful performance impact.
On 24. 01. 22 21:16, Mark Brown wrote: > On Mon, Jan 24, 2022 at 08:20:25PM +0100, Jaroslav Kysela wrote: >> On 24. 01. 22 18:08, Takashi Iwai wrote: > >>> Jaroslav, is you concern only about the compatibility of user-space? >>> Or anything else? The compatibility is always certainly a slight >>> issue; if this breaks really something useful and actually working >>> stuff, we need to consider the workaround... > >> My concern is only based on the fact that this code is normally not >> reachable. It only costs some CPU ticks and adds extra code to the drivers >> without any other benefit. The applications should not use out of range >> values and if they do, the behavior is not defined (the drivers should only >> avoid crashes). > >> The code seems to be added only to make things consistent for the test >> applications. I don't think that it's worth to care only for this reason. >> What is the goal for this code? The result with the proposed code will be - >> the SoC core has the extra validation conditions. > > We need these checks all the time for the generic ASoC controls since > the values for the controls are stored in the underlying device's > register map so the out of range values will be written to the hardware, > pushing it out of specified use. That's not a great idea in general and > in extreme cases could result in physical damage to the system. The > biggest risk I see here is around speaker drivers since they deal with > high powers, even ignoring silicon requirements we also don't currently > enforce platform maximums that the machine drivers specify - that > feature was added after an inexperienced user burnt out the speakers in > their Chromebook since the speakers in the system were rated for much > lower powers than the CODEC was able to deliver. It may be good to add this to the comment for two related patches. It's not obvious from the first glance. So one condition is a must. >> Basically, I think that those checks should be marked as debug and they >> should be optional like we do for CONFIG_SND_CTL_VALIDATION and enable only >> the fast path by default. > > Note also that for everything using these helpers the underlying > register map will be regmap based and with the possible exception of > MMIO based regmaps the cost of writing out the new register value will > be overwhelmingly greater than that of the bounds checks. It is > extremely hard to envision a scenario in which even a pathological > application would be able to observe a meaningful performance impact. It's true that your patches are touching the universal code. I stop complain for now. I just don't like to add chunks of the extra code just to resolve something which is not eventually a problem. Jaroslav
On Mon, 24 Jan 2022 15:32:50 +0000, Mark Brown wrote: > This series adds verification that values written to registers are in > bounds for controls since the core doesn't validate for us. > > Mark Brown (3): > ASoC: ops: Reject out of bounds values in snd_soc_put_volsw() > ASoC: ops: Reject out of bounds values in snd_soc_put_volsw_sx() > ASoC: ops: Reject out of bounds values in snd_soc_put_xr_sx() > > [...] Applied to https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git for-linus Thanks! [1/3] ASoC: ops: Reject out of bounds values in snd_soc_put_volsw() commit: 817f7c9335ec01e0f5e8caffc4f1dcd5e458a4c0 [2/3] ASoC: ops: Reject out of bounds values in snd_soc_put_volsw_sx() commit: 4f1e50d6a9cf9c1b8c859d449b5031cacfa8404e [3/3] ASoC: ops: Reject out of bounds values in snd_soc_put_xr_sx() commit: 4cf28e9ae6e2e11a044be1bcbcfa1b0d8675fe4d All being well this means that it will be integrated into the linux-next tree (usually sometime in the next 24 hours) and sent to Linus during the next merge window (or sooner if it is a bug fix), however if problems are discovered then the patch may be dropped or reverted. You may get further e-mails resulting from automated or manual testing and review of the tree, please engage with people reporting problems and send followup patches addressing any issues that are reported if needed. If any updates are required or you are submitting further changes they should be sent as incremental updates against current git, existing patches will not be replaced. Please add any relevant lists and maintainers to the CCs when replying to this mail. Thanks, Mark