Message ID | 1493029017-31382-7-git-send-email-vladimir.murzin@arm.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Mon, Apr 24, 2017 at 11:16:56AM +0100, Vladimir Murzin wrote: > Now, we have dedicated non-cacheable region for consistent DMA > operations. However, that region can still be marked as bufferable by > MPU, so it'd be safer to have barriers by default. What do you actually want here? Your patch doesn't quite make sense, the commit description seems to indicate that you require this option to be set for V7M, but the patch says otherwise. > config ARM_DMA_MEM_BUFFERABLE > - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 > - default y if CPU_V6 || CPU_V6K || CPU_V7 > + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 This "if" conditional conditionalises the visibility of the option, it doesn't conditionalise the value. > + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M Taking both of these changes together what you end up with is an option presented to the user for "Use non-cacheable memory for DMA" which they can choose to disable. If you require this option to be set, that's incorrect - your modification to the default line is correct, but the first line is not. To achieve that, you want the if condition to evaluate false for V7M, thereby hiding the option from the user. In that case, the default value will always be assigned to the option. > help > Historically, the kernel has used strongly ordered mappings to > provide DMA coherent memory. With the advent of ARMv7, mapping > -- > 2.0.0 >
On Tue, May 23, 2017 at 10:01 PM, Russell King - ARM Linux <linux@armlinux.org.uk> wrote: > On Mon, Apr 24, 2017 at 11:16:56AM +0100, Vladimir Murzin wrote: >> Now, we have dedicated non-cacheable region for consistent DMA >> operations. However, that region can still be marked as bufferable by >> MPU, so it'd be safer to have barriers by default. > > What do you actually want here? Your patch doesn't quite make sense, > the commit description seems to indicate that you require this option > to be set for V7M, but the patch says otherwise. > >> config ARM_DMA_MEM_BUFFERABLE >> - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 >> - default y if CPU_V6 || CPU_V6K || CPU_V7 >> + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 > > This "if" conditional conditionalises the visibility of the option, > it doesn't conditionalise the value. > >> + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M > > Taking both of these changes together what you end up with is an option > presented to the user for "Use non-cacheable memory for DMA" which > they can choose to disable. > > If you require this option to be set, that's incorrect - your modification > to the default line is correct, but the first line is not. To achieve > that, you want the if condition to evaluate false for V7M, thereby hiding > the option from the user. In that case, the default value will always be > assigned to the option. I had the opposite comment in the previous version ;-) https://lkml.org/lkml/2017/4/19/185 I think the current patch is correct, but the description could still be clarified: On some of the beefier ARMv7-M machines (with DMA and write buffers) we want this enabled, while those that didn't need it until now also won't need it in the future. Arnd
On 23/05/17 21:33, Arnd Bergmann wrote: > On Tue, May 23, 2017 at 10:01 PM, Russell King - ARM Linux > <linux@armlinux.org.uk> wrote: >> On Mon, Apr 24, 2017 at 11:16:56AM +0100, Vladimir Murzin wrote: >>> Now, we have dedicated non-cacheable region for consistent DMA >>> operations. However, that region can still be marked as bufferable by >>> MPU, so it'd be safer to have barriers by default. >> >> What do you actually want here? Your patch doesn't quite make sense, >> the commit description seems to indicate that you require this option >> to be set for V7M, but the patch says otherwise. >> >>> config ARM_DMA_MEM_BUFFERABLE >>> - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 >>> - default y if CPU_V6 || CPU_V6K || CPU_V7 >>> + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 >> >> This "if" conditional conditionalises the visibility of the option, >> it doesn't conditionalise the value. >> >>> + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M >> >> Taking both of these changes together what you end up with is an option >> presented to the user for "Use non-cacheable memory for DMA" which >> they can choose to disable. >> >> If you require this option to be set, that's incorrect - your modification >> to the default line is correct, but the first line is not. To achieve >> that, you want the if condition to evaluate false for V7M, thereby hiding >> the option from the user. In that case, the default value will always be >> assigned to the option. > > I had the opposite comment in the previous version ;-) > https://lkml.org/lkml/2017/4/19/185 > > I think the current patch is correct, but the description could still be > clarified: On some of the beefier ARMv7-M machines (with DMA > and write buffers) we want this enabled, while those that didn't > need it until now also won't need it in the future. Ok. Do you want it go into commit message or option description or maybe both? Thanks Vladimir > > Arnd >
On Wed, May 24, 2017 at 10:31 AM, Vladimir Murzin <vladimir.murzin@arm.com> wrote: > On 23/05/17 21:33, Arnd Bergmann wrote: >> On Tue, May 23, 2017 at 10:01 PM, Russell King - ARM Linux >> <linux@armlinux.org.uk> wrote: >>> On Mon, Apr 24, 2017 at 11:16:56AM +0100, Vladimir Murzin wrote: >>>> Now, we have dedicated non-cacheable region for consistent DMA >>>> operations. However, that region can still be marked as bufferable by >>>> MPU, so it'd be safer to have barriers by default. >>> >>> What do you actually want here? Your patch doesn't quite make sense, >>> the commit description seems to indicate that you require this option >>> to be set for V7M, but the patch says otherwise. >>> >>>> config ARM_DMA_MEM_BUFFERABLE >>>> - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 >>>> - default y if CPU_V6 || CPU_V6K || CPU_V7 >>>> + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 >>> >>> This "if" conditional conditionalises the visibility of the option, >>> it doesn't conditionalise the value. >>> >>>> + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M >>> >>> Taking both of these changes together what you end up with is an option >>> presented to the user for "Use non-cacheable memory for DMA" which >>> they can choose to disable. >>> >>> If you require this option to be set, that's incorrect - your modification >>> to the default line is correct, but the first line is not. To achieve >>> that, you want the if condition to evaluate false for V7M, thereby hiding >>> the option from the user. In that case, the default value will always be >>> assigned to the option. >> >> I had the opposite comment in the previous version ;-) >> https://lkml.org/lkml/2017/4/19/185 >> >> I think the current patch is correct, but the description could still be >> clarified: On some of the beefier ARMv7-M machines (with DMA >> and write buffers) we want this enabled, while those that didn't >> need it until now also won't need it in the future. > > Ok. Do you want it go into commit message or option description or maybe both? I'd say both. It would also be helpful to identify specifically which platforms require this, and then add a 'select ARM_DMA_MEM_BUFFERABLE' from the platform, as we do from Moxart. Arnd
On 24/05/17 09:36, Arnd Bergmann wrote: > On Wed, May 24, 2017 at 10:31 AM, Vladimir Murzin > <vladimir.murzin@arm.com> wrote: >> On 23/05/17 21:33, Arnd Bergmann wrote: >>> On Tue, May 23, 2017 at 10:01 PM, Russell King - ARM Linux >>> <linux@armlinux.org.uk> wrote: >>>> On Mon, Apr 24, 2017 at 11:16:56AM +0100, Vladimir Murzin wrote: >>>>> Now, we have dedicated non-cacheable region for consistent DMA >>>>> operations. However, that region can still be marked as bufferable by >>>>> MPU, so it'd be safer to have barriers by default. >>>> >>>> What do you actually want here? Your patch doesn't quite make sense, >>>> the commit description seems to indicate that you require this option >>>> to be set for V7M, but the patch says otherwise. >>>> >>>>> config ARM_DMA_MEM_BUFFERABLE >>>>> - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 >>>>> - default y if CPU_V6 || CPU_V6K || CPU_V7 >>>>> + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 >>>> >>>> This "if" conditional conditionalises the visibility of the option, >>>> it doesn't conditionalise the value. >>>> >>>>> + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M >>>> >>>> Taking both of these changes together what you end up with is an option >>>> presented to the user for "Use non-cacheable memory for DMA" which >>>> they can choose to disable. >>>> >>>> If you require this option to be set, that's incorrect - your modification >>>> to the default line is correct, but the first line is not. To achieve >>>> that, you want the if condition to evaluate false for V7M, thereby hiding >>>> the option from the user. In that case, the default value will always be >>>> assigned to the option. >>> >>> I had the opposite comment in the previous version ;-) >>> https://lkml.org/lkml/2017/4/19/185 >>> >>> I think the current patch is correct, but the description could still be >>> clarified: On some of the beefier ARMv7-M machines (with DMA >>> and write buffers) we want this enabled, while those that didn't >>> need it until now also won't need it in the future. >> >> Ok. Do you want it go into commit message or option description or maybe both? > > I'd say both. It would also be helpful to identify specifically which platforms > require this, and then add a 'select ARM_DMA_MEM_BUFFERABLE' from > the platform, as we do from Moxart. > I'm a bit confused here. In case we want to control it on platform level via 'select ARM_DMA_MEM_BUFFERABLE' wouldn't we need to 'default n' for CPU_V7M? IIUC, Moxart needs to select this option because it is neither CPU_V6(K) or CPU_V7. Thanks Vladimir > Arnd >
On Wed, May 24, 2017 at 11:08 AM, Vladimir Murzin <vladimir.murzin@arm.com> wrote: > On 24/05/17 09:36, Arnd Bergmann wrote: >> On Wed, May 24, 2017 at 10:31 AM, Vladimir Murzin >> <vladimir.murzin@arm.com> wrote: >>> On 23/05/17 21:33, Arnd Bergmann wrote: >>>> On Tue, May 23, 2017 at 10:01 PM, Russell King - ARM Linux >>>> <linux@armlinux.org.uk> wrote: >>>>> On Mon, Apr 24, 2017 at 11:16:56AM +0100, Vladimir Murzin wrote: >>>>>> Now, we have dedicated non-cacheable region for consistent DMA >>>>>> operations. However, that region can still be marked as bufferable by >>>>>> MPU, so it'd be safer to have barriers by default. >>>>> >>>>> What do you actually want here? Your patch doesn't quite make sense, >>>>> the commit description seems to indicate that you require this option >>>>> to be set for V7M, but the patch says otherwise. >>>>> >>>>>> config ARM_DMA_MEM_BUFFERABLE >>>>>> - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 >>>>>> - default y if CPU_V6 || CPU_V6K || CPU_V7 >>>>>> + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 >>>>> >>>>> This "if" conditional conditionalises the visibility of the option, >>>>> it doesn't conditionalise the value. >>>>> >>>>>> + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M >>>>> >>>>> Taking both of these changes together what you end up with is an option >>>>> presented to the user for "Use non-cacheable memory for DMA" which >>>>> they can choose to disable. >>>>> >>>>> If you require this option to be set, that's incorrect - your modification >>>>> to the default line is correct, but the first line is not. To achieve >>>>> that, you want the if condition to evaluate false for V7M, thereby hiding >>>>> the option from the user. In that case, the default value will always be >>>>> assigned to the option. >>>> >>>> I had the opposite comment in the previous version ;-) >>>> https://lkml.org/lkml/2017/4/19/185 >>>> >>>> I think the current patch is correct, but the description could still be >>>> clarified: On some of the beefier ARMv7-M machines (with DMA >>>> and write buffers) we want this enabled, while those that didn't >>>> need it until now also won't need it in the future. >>> >>> Ok. Do you want it go into commit message or option description or maybe both? >> >> I'd say both. It would also be helpful to identify specifically which platforms >> require this, and then add a 'select ARM_DMA_MEM_BUFFERABLE' from >> the platform, as we do from Moxart. >> > > I'm a bit confused here. In case we want to control it on platform level via > 'select ARM_DMA_MEM_BUFFERABLE' wouldn't we need to 'default n' for CPU_V7M? > IIUC, Moxart needs to select this option because it is neither CPU_V6(K) or > CPU_V7. It depends: If we want to control it purely by platform, then we don't need to patch anything here, just add the 'select' and be done with it, leaving the default to 'n' for ARMv7-M. If there are platforms on which you might reasonably make the option user-visible (e.g. you only need it when you actually want to use one of the DMA masters, but most configurations don't), then having the 'default y' might still be appropriate, to default to the safe setting on platforms that don't have the 'select' but letting users turn it off when they know what they are doing. Arnd
diff --git a/arch/arm/mm/Kconfig b/arch/arm/mm/Kconfig index d731f28..7e357c6 100644 --- a/arch/arm/mm/Kconfig +++ b/arch/arm/mm/Kconfig @@ -1049,8 +1049,8 @@ config ARM_L1_CACHE_SHIFT default 5 config ARM_DMA_MEM_BUFFERABLE - bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K) && !CPU_V7 - default y if CPU_V6 || CPU_V6K || CPU_V7 + bool "Use non-cacheable memory for DMA" if (CPU_V6 || CPU_V6K || CPU_V7M) && !CPU_V7 + default y if CPU_V6 || CPU_V6K || CPU_V7 || CPU_V7M help Historically, the kernel has used strongly ordered mappings to provide DMA coherent memory. With the advent of ARMv7, mapping