| Message ID | 20200102131845.12992-7-t-kristo@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | remoteproc: updates for omap remoteproc support | expand |
On 1/2/20 8:18 AM, Tero Kristo wrote: > From: Suman Anna <s-anna@ti.com> > > The reserved memory nodes are not assigned to platform devices by > default in the driver core to avoid the lookup for every platform > device and incur a penalty as the real users are expected to be > only a few devices. > > OMAP remoteproc devices fall into the above category and the OMAP > remoteproc driver _requires_ specific CMA pools to be assigned > for each device at the moment to align on the location of the > vrings and vring buffers in the RTOS-side firmware images. So, Requires only at the moment due to firmware.. This sounds like some firmware images hard-coded their vring addresses instead of getting them dynamically as they should and we are hacking around that on the kernel side by giving them the addresses they use as carveouts. Should we rather make use of the IOMMU attached to the DSP to map any kernel address to the DSP where the firmware expects it? Or better yet fixup the firmwares. DRAM carveouts should be a last resort used only for when hardware really requires it. Andrew > use the of_reserved_mem_device_init/release() API appropriately > to assign the corresponding reserved memory region to the OMAP > remoteproc device. Note that only one region per device is > allowed by the framework. > > Signed-off-by: Suman Anna <s-anna@ti.com> > Signed-off-by: Tero Kristo <t-kristo@ti.com> > Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> > --- > drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++- > 1 file changed, 11 insertions(+), 1 deletion(-) > > diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c > index 9ca337f18ac2..8a6dd742a8b1 100644 > --- a/drivers/remoteproc/omap_remoteproc.c > +++ b/drivers/remoteproc/omap_remoteproc.c > @@ -17,6 +17,7 @@ > #include <linux/module.h> > #include <linux/err.h> > #include <linux/of_device.h> > +#include <linux/of_reserved_mem.h> > #include <linux/platform_device.h> > #include <linux/dma-mapping.h> > #include <linux/remoteproc.h> > @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev) > if (ret) > goto free_rproc; > > + ret = of_reserved_mem_device_init(&pdev->dev); > + if (ret) { > + dev_err(&pdev->dev, "device does not have specific CMA pool\n"); > + goto free_rproc; > + } > + > platform_set_drvdata(pdev, rproc); > > ret = rproc_add(rproc); > if (ret) > - goto free_rproc; > + goto release_mem; > > return 0; > > +release_mem: > + of_reserved_mem_device_release(&pdev->dev); > free_rproc: > rproc_free(rproc); > return ret; > @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev) > > rproc_del(rproc); > rproc_free(rproc); > + of_reserved_mem_device_release(&pdev->dev); > > return 0; > } >
On 07/01/2020 15:37, Andrew F. Davis wrote: > On 1/2/20 8:18 AM, Tero Kristo wrote: >> From: Suman Anna <s-anna@ti.com> >> >> The reserved memory nodes are not assigned to platform devices by >> default in the driver core to avoid the lookup for every platform >> device and incur a penalty as the real users are expected to be >> only a few devices. >> >> OMAP remoteproc devices fall into the above category and the OMAP >> remoteproc driver _requires_ specific CMA pools to be assigned >> for each device at the moment to align on the location of the >> vrings and vring buffers in the RTOS-side firmware images. So, > > > Requires only at the moment due to firmware.. > > This sounds like some firmware images hard-coded their vring addresses > instead of getting them dynamically as they should and we are hacking > around that on the kernel side by giving them the addresses they use as > carveouts. The firmwares are built on specific device addresses, this includes data + code. > Should we rather make use of the IOMMU attached to the DSP to map any > kernel address to the DSP where the firmware expects it? Or better yet > fixup the firmwares. Well, we do use IOMMU to map the corresponding memory area to specific device address. What this patch does, is to allocate a contiguous memory area for the remoteproc shared memories. Using completely random memory location would potentially fragment the remoteproc memory around page boundaries, resulting in a complex (read ineffective) IOMMU mapping. Also, we are going to need the reserved memory mechanism for the remoteproc anyways later, as we are going to introduce the support for early-boot / late-attach. Bootloader would pass the used memory regions to the kernel via the reserved memory nodes in that case (unless we assume to use some hardcoded region, which would be worse than passing it via DT.) -Tero > > DRAM carveouts should be a last resort used only for when hardware > really requires it. > > Andrew > > >> use the of_reserved_mem_device_init/release() API appropriately >> to assign the corresponding reserved memory region to the OMAP >> remoteproc device. Note that only one region per device is >> allowed by the framework. >> >> Signed-off-by: Suman Anna <s-anna@ti.com> >> Signed-off-by: Tero Kristo <t-kristo@ti.com> >> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> >> --- >> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++- >> 1 file changed, 11 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c >> index 9ca337f18ac2..8a6dd742a8b1 100644 >> --- a/drivers/remoteproc/omap_remoteproc.c >> +++ b/drivers/remoteproc/omap_remoteproc.c >> @@ -17,6 +17,7 @@ >> #include <linux/module.h> >> #include <linux/err.h> >> #include <linux/of_device.h> >> +#include <linux/of_reserved_mem.h> >> #include <linux/platform_device.h> >> #include <linux/dma-mapping.h> >> #include <linux/remoteproc.h> >> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev) >> if (ret) >> goto free_rproc; >> >> + ret = of_reserved_mem_device_init(&pdev->dev); >> + if (ret) { >> + dev_err(&pdev->dev, "device does not have specific CMA pool\n"); >> + goto free_rproc; >> + } >> + >> platform_set_drvdata(pdev, rproc); >> >> ret = rproc_add(rproc); >> if (ret) >> - goto free_rproc; >> + goto release_mem; >> >> return 0; >> >> +release_mem: >> + of_reserved_mem_device_release(&pdev->dev); >> free_rproc: >> rproc_free(rproc); >> return ret; >> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev) >> >> rproc_del(rproc); >> rproc_free(rproc); >> + of_reserved_mem_device_release(&pdev->dev); >> >> return 0; >> } >> -- Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
On 1/7/20 9:25 AM, Tero Kristo wrote: > On 07/01/2020 15:37, Andrew F. Davis wrote: >> On 1/2/20 8:18 AM, Tero Kristo wrote: >>> From: Suman Anna <s-anna@ti.com> >>> >>> The reserved memory nodes are not assigned to platform devices by >>> default in the driver core to avoid the lookup for every platform >>> device and incur a penalty as the real users are expected to be >>> only a few devices. >>> >>> OMAP remoteproc devices fall into the above category and the OMAP >>> remoteproc driver _requires_ specific CMA pools to be assigned >>> for each device at the moment to align on the location of the >>> vrings and vring buffers in the RTOS-side firmware images. So, >> >> >> Requires only at the moment due to firmware.. >> >> This sounds like some firmware images hard-coded their vring addresses >> instead of getting them dynamically as they should and we are hacking >> around that on the kernel side by giving them the addresses they use as >> carveouts. > > The firmwares are built on specific device addresses, this includes data > + code. > >> Should we rather make use of the IOMMU attached to the DSP to map any >> kernel address to the DSP where the firmware expects it? Or better yet >> fixup the firmwares. > > Well, we do use IOMMU to map the corresponding memory area to specific > device address. What this patch does, is to allocate a contiguous memory > area for the remoteproc shared memories. Using completely random memory > location would potentially fragment the remoteproc memory around page > boundaries, resulting in a complex (read ineffective) IOMMU mapping. Complex is not always ineffective, this is what the (IO)MMUs are for, memory gets fragmented on page boundaries, they put it back together, that's part of modern computing despite its crazy complexity. Shying away from that and just using big static memory carveouts for usecases like this (that could otherwise work without them) will not scale. > Also, we are going to need the reserved memory mechanism for the > remoteproc anyways later, as we are going to introduce the support for > early-boot / late-attach. Bootloader would pass the used memory regions > to the kernel via the reserved memory nodes in that case (unless we > assume to use some hardcoded region, which would be worse than passing > it via DT.) This is a different case, I can see a more valid use here (although I'd argue passing bootloader generated software configuration like this to kernel is a gray area for DT, but I'll leave that for our DT maintainer friends). At very least can we make the reserved memory node optional here? DSP/IPU Firmware can/should be made to work without it. Andrew > > -Tero > >> >> DRAM carveouts should be a last resort used only for when hardware >> really requires it. >> >> Andrew >> >> >>> use the of_reserved_mem_device_init/release() API appropriately >>> to assign the corresponding reserved memory region to the OMAP >>> remoteproc device. Note that only one region per device is >>> allowed by the framework. >>> >>> Signed-off-by: Suman Anna <s-anna@ti.com> >>> Signed-off-by: Tero Kristo <t-kristo@ti.com> >>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> >>> --- >>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++- >>> 1 file changed, 11 insertions(+), 1 deletion(-) >>> >>> diff --git a/drivers/remoteproc/omap_remoteproc.c >>> b/drivers/remoteproc/omap_remoteproc.c >>> index 9ca337f18ac2..8a6dd742a8b1 100644 >>> --- a/drivers/remoteproc/omap_remoteproc.c >>> +++ b/drivers/remoteproc/omap_remoteproc.c >>> @@ -17,6 +17,7 @@ >>> #include <linux/module.h> >>> #include <linux/err.h> >>> #include <linux/of_device.h> >>> +#include <linux/of_reserved_mem.h> >>> #include <linux/platform_device.h> >>> #include <linux/dma-mapping.h> >>> #include <linux/remoteproc.h> >>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct >>> platform_device *pdev) >>> if (ret) >>> goto free_rproc; >>> + ret = of_reserved_mem_device_init(&pdev->dev); >>> + if (ret) { >>> + dev_err(&pdev->dev, "device does not have specific CMA >>> pool\n"); >>> + goto free_rproc; >>> + } >>> + >>> platform_set_drvdata(pdev, rproc); >>> ret = rproc_add(rproc); >>> if (ret) >>> - goto free_rproc; >>> + goto release_mem; >>> return 0; >>> +release_mem: >>> + of_reserved_mem_device_release(&pdev->dev); >>> free_rproc: >>> rproc_free(rproc); >>> return ret; >>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct >>> platform_device *pdev) >>> rproc_del(rproc); >>> rproc_free(rproc); >>> + of_reserved_mem_device_release(&pdev->dev); >>> return 0; >>> } >>> > > -- > Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. > Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
On 1/7/20 8:37 AM, Andrew F. Davis wrote: > On 1/7/20 9:25 AM, Tero Kristo wrote: >> On 07/01/2020 15:37, Andrew F. Davis wrote: >>> On 1/2/20 8:18 AM, Tero Kristo wrote: >>>> From: Suman Anna <s-anna@ti.com> >>>> >>>> The reserved memory nodes are not assigned to platform devices by >>>> default in the driver core to avoid the lookup for every platform >>>> device and incur a penalty as the real users are expected to be >>>> only a few devices. >>>> >>>> OMAP remoteproc devices fall into the above category and the OMAP >>>> remoteproc driver _requires_ specific CMA pools to be assigned >>>> for each device at the moment to align on the location of the >>>> vrings and vring buffers in the RTOS-side firmware images. So, >>> >>> >>> Requires only at the moment due to firmware.. >>> >>> This sounds like some firmware images hard-coded their vring addresses >>> instead of getting them dynamically as they should and we are hacking >>> around that on the kernel side by giving them the addresses they use as >>> carveouts. >> >> The firmwares are built on specific device addresses, this includes data >> + code. >> >>> Should we rather make use of the IOMMU attached to the DSP to map any >>> kernel address to the DSP where the firmware expects it? Or better yet >>> fixup the firmwares. >> >> Well, we do use IOMMU to map the corresponding memory area to specific >> device address. What this patch does, is to allocate a contiguous memory >> area for the remoteproc shared memories. Using completely random memory >> location would potentially fragment the remoteproc memory around page >> boundaries, resulting in a complex (read ineffective) IOMMU mapping. > > > Complex is not always ineffective, this is what the (IO)MMUs are for, > memory gets fragmented on page boundaries, they put it back together, > that's part of modern computing despite its crazy complexity. Shying > away from that and just using big static memory carveouts for usecases > like this (that could otherwise work without them) will not scale. Not sure what your definition of static carveout is, but we are really using device-specific CMA pool here, and rely on RSC_CARVEOUTs from the resource table to allocate the memory from that pool. Obviously, this cannot be a CMA pool and has to be a static carveout for early-boot scenarios. Note that our OMAP IOMMUs are very simple, they only can handle 32-bit physical addresses, and actually cannot add any memory attributes, and that is actually handled by another sub-module managed and controlled by the remote processor. So, this does place some constraints in using a generic CMA pool. regards Suman > > >> Also, we are going to need the reserved memory mechanism for the >> remoteproc anyways later, as we are going to introduce the support for >> early-boot / late-attach. Bootloader would pass the used memory regions >> to the kernel via the reserved memory nodes in that case (unless we >> assume to use some hardcoded region, which would be worse than passing >> it via DT.) > > > This is a different case, I can see a more valid use here (although I'd > argue passing bootloader generated software configuration like this to > kernel is a gray area for DT, but I'll leave that for our DT maintainer > friends). > > At very least can we make the reserved memory node optional here? > DSP/IPU Firmware can/should be made to work without it. > > Andrew > > >> >> -Tero >> >>> >>> DRAM carveouts should be a last resort used only for when hardware >>> really requires it. >>> >>> Andrew >>> >>> >>>> use the of_reserved_mem_device_init/release() API appropriately >>>> to assign the corresponding reserved memory region to the OMAP >>>> remoteproc device. Note that only one region per device is >>>> allowed by the framework. >>>> >>>> Signed-off-by: Suman Anna <s-anna@ti.com> >>>> Signed-off-by: Tero Kristo <t-kristo@ti.com> >>>> Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> >>>> --- >>>> drivers/remoteproc/omap_remoteproc.c | 12 +++++++++++- >>>> 1 file changed, 11 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/remoteproc/omap_remoteproc.c >>>> b/drivers/remoteproc/omap_remoteproc.c >>>> index 9ca337f18ac2..8a6dd742a8b1 100644 >>>> --- a/drivers/remoteproc/omap_remoteproc.c >>>> +++ b/drivers/remoteproc/omap_remoteproc.c >>>> @@ -17,6 +17,7 @@ >>>> #include <linux/module.h> >>>> #include <linux/err.h> >>>> #include <linux/of_device.h> >>>> +#include <linux/of_reserved_mem.h> >>>> #include <linux/platform_device.h> >>>> #include <linux/dma-mapping.h> >>>> #include <linux/remoteproc.h> >>>> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct >>>> platform_device *pdev) >>>> if (ret) >>>> goto free_rproc; >>>> + ret = of_reserved_mem_device_init(&pdev->dev); >>>> + if (ret) { >>>> + dev_err(&pdev->dev, "device does not have specific CMA >>>> pool\n"); >>>> + goto free_rproc; >>>> + } >>>> + >>>> platform_set_drvdata(pdev, rproc); >>>> ret = rproc_add(rproc); >>>> if (ret) >>>> - goto free_rproc; >>>> + goto release_mem; >>>> return 0; >>>> +release_mem: >>>> + of_reserved_mem_device_release(&pdev->dev); >>>> free_rproc: >>>> rproc_free(rproc); >>>> return ret; >>>> @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct >>>> platform_device *pdev) >>>> rproc_del(rproc); >>>> rproc_free(rproc); >>>> + of_reserved_mem_device_release(&pdev->dev); >>>> return 0; >>>> } >>>> >> >> -- >> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki. >> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
diff --git a/drivers/remoteproc/omap_remoteproc.c b/drivers/remoteproc/omap_remoteproc.c index 9ca337f18ac2..8a6dd742a8b1 100644 --- a/drivers/remoteproc/omap_remoteproc.c +++ b/drivers/remoteproc/omap_remoteproc.c @@ -17,6 +17,7 @@ #include <linux/module.h> #include <linux/err.h> #include <linux/of_device.h> +#include <linux/of_reserved_mem.h> #include <linux/platform_device.h> #include <linux/dma-mapping.h> #include <linux/remoteproc.h> @@ -480,14 +481,22 @@ static int omap_rproc_probe(struct platform_device *pdev) if (ret) goto free_rproc; + ret = of_reserved_mem_device_init(&pdev->dev); + if (ret) { + dev_err(&pdev->dev, "device does not have specific CMA pool\n"); + goto free_rproc; + } + platform_set_drvdata(pdev, rproc); ret = rproc_add(rproc); if (ret) - goto free_rproc; + goto release_mem; return 0; +release_mem: + of_reserved_mem_device_release(&pdev->dev); free_rproc: rproc_free(rproc); return ret; @@ -499,6 +508,7 @@ static int omap_rproc_remove(struct platform_device *pdev) rproc_del(rproc); rproc_free(rproc); + of_reserved_mem_device_release(&pdev->dev); return 0; }