Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Mauro Carvalho Chehab]

There was a recent discussion about the use/abuse of GFP_DMA flag when
allocating memories at LSF/MM 2018 (see Luis notes enclosed).

The idea seems to be to remove it, using CMA instead. Before doing that,
better to check if what we have on media is are valid use cases for it, or
if it is there just due to some misunderstanding (or because it was
copied from some other code).

Hans de Goede sent us today a patch stopping abuse at gspca, and I'm 
also posting today two other patches meant to stop abuse of it on USB
drivers. Still, there are 4 platform drivers using it:

	$ git grep -l -E "GFP_DMA\\b" drivers/media/
	drivers/media/platform/omap3isp/ispstat.c
	drivers/media/platform/sti/bdisp/bdisp-hw.c
	drivers/media/platform/sti/hva/hva-mem.c
	drivers/media/spi/cxd2880-spi.c

Could you please check if GFP_DMA is really needed there, or if it is
just because of some cut-and-paste from some other place?

Thanks!
Mauro


Em Thu, 26 Apr 2018 21:54:06 +0000
"Luis R. Rodriguez" <mcgrof@kernel.org> escreveu:

> Below are my notes on the ZONE_DMA discussion at LSF/MM 2018. There were some
> earlier discussion prior to my arrival to the session about moving around
> ZOME_DMA around, if someone has notes on that please share too :)
> 
> PS. I'm not subscribed to linux-mm
> 
>   Luis
> 
> Determining you don't need to support ZONE_DMA on x86 at run time
> =================================================================
> 
> In practice if you don't have a floppy device on x86, you don't need ZONE_DMA,
> in that case you dont need to support ZONE_DMA, however currently disabling it
> is only possible at compile time, and we won't know for sure until boot time if
> you have such a device. If you don't need ZONE_DMA though means we would not
> have to deal with slab allocators for them and special casings for it in a slew
> of places. In particular even kmalloc() has a branch which is always run if
> CONFIG_ZONE_DMA is enabled.
> 
> ZONE_DMA is needed for old devices that requires lower addresses since it allows
> allocations more reliably. There should be more devices that require this,
> not just floppy though.
> 
> Christoph Lameter added CONFIG_ZONE_DMA to disable ZONE_DMA at build time but
> most distributions enable this. If we could disable ZONE_DMA at run time once
> we know we don't have any device present requiring it we could get the same
> benefit of compiling without CONFIG_ZONE_DMA at run time.
> 
> It used to be that disabling CONFIG_ZONE_DMA could help with performance, we
> don't seem to have modern benchmarks over possible gains on removing it.
> Are the gains no longer expected to be significant? Very likely there are
> no performance gains. The assumption then is that the main advantage over
> being able to disable ZONE_DMA on x86 these days would be pure aesthetics, and
> having x86 work more like other architectures with allocations. Use of ZONE_DMA
> on drivers are also good signs these drivers are old, or may be deprecated.
> Perhaps some of these on x86 should be moved to staging.
> 
> Note that some architectures rely on ZONE_DMA as well, the above notes
> only applies to x86.
> 
> We can use certain kernel mechanisms to disable usage of x86 certain features
> at run time. Below are a few options:
> 
>   * x86 binary patching
>   * ACPI_SIG_FADT
>   * static keys
>   * compiler multiverse (at least the R&D gcc proof of concept is now complete)
> 
> Detecting legacy x86 devices with ACPI ACPI_SIG_FADT
> ----------------------------------------------------
> 
> We could expand on ACPI_SIG_FADT with more legacy devices. This mechanism was
> used to help determine if certain legacy x86 devices are present or not with
> paravirtualization. For instance:
> 
>   * ACPI_FADT_NO_VGA
>   * ACPI_FADT_NO_CMOS_RTC
> 
> CONFIG_ZONE_DMA
> ---------------
> 
> Christoph Lameter added CONFIG_ZONE_DMA through commit 4b51d66989218
> ("[PATCH] optional ZONE_DMA: optional ZONE_DMA in the VM") merged on
> v2.6.21.
> 
> On x86 ZONE_DMA is defined as follows:
> 
> config ZONE_DMA
>         bool "DMA memory allocation support" if EXPERT
>         default y
>         help
>           DMA memory allocation support allows devices with less than 32-bit
>           addressing to allocate within the first 16MB of address space.
>           Disable if no such devices will be used.
>                                                                                 
>           If unsure, say Y.
> 
> Most distributions enable CONFIG_ZONE_DMA.
> 
> Immediate impact of CONFIG_ZONE_DMA
> -----------------------------------
> 
> CONFIG_ZONE_DMA implicaates kmalloc() as follows:
> 
> struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
> {
> 	...
> #ifdef CONFIG_ZONE_DMA
> 	if (unlikely((flags & GFP_DMA)))
> 		return kmalloc_dma_caches[index];
> #endif
> 	...
> }
> 
> ZONE_DMA users
> ==============
> 
> Turns out there are much more users of ZONE_DMA than expected even on x86.
> 
> Explicit requirements for ZONE_DMA with gfp flags
> -------------------------------------------------
> 
> All drivers which explicitly use any of these flags implicate use
> of ZONE_DMA for allocations:
> 
>   * GFP_DMA
>   * __GFP_DMA
> 
> Implicit ZONE_DMA users
> -----------------------
> 
> There are a series of implicit users of ZONE_DMA which use helpers. These are,
> with details documented further below:
> 
>   * blk_queue_bounce()
>   * blk_queue_bounce_limit()
>   * dma_alloc_coherent_gfp_flags()
>   * dma_generic_alloc_coherent()
>   * intel_alloc_coherent()
>   * _regmap_raw_write()
>   * mempool_alloc_pages_isa()
> 
> x86 implicit and explicit ZONE_DMA users
> -----------------------------------------
> 
> We list below all x86 implicit and explicit ZONE_DMA users.
> 
> # Explicit x86 users of GFP_DMA or __GFP_DMA
> 
>   * drivers/iio/common/ssp_sensors - wonder if enabling this on x86 was a mistake.
>     Note that this needs SPI and SPI needs HAS_IOMEM. I only see HAS_IOMEM on
>     s390 ? But I do think the Intel Minnowboard has SPI, but doubt it has
>    the ssp sensor stuff.
> 
>  * drivers/input/rmi4/rmi_spi.c - same SPI question
>  * drivers/media/common/siano/ - make allyesconfig yields it enabled, but
>    not sure if this should ever be on x86
>  * drivers/media/platform/sti/bdisp/ - likewise
>   * drivers/media/platform/sti/hva/ - likewise
>   * drivers/media/usb/gspca/ - likewise
>   * drivers/mmc/host/wbsd.c - likewise
>   * drivers/mtd/nand/gpmi-nand/ - likewise
>   * drivers/net/can/spi/hi311x.c - likewise
>   * drivers/net/can/spi/mcp251x.c - likewise
>   * drivers/net/ethernet/agere/ - likewise
>   * drivers/net/ethernet/neterion/vxge/ - likewise
>   * drivers/net/ethernet/rocker/ - likewise
>   * drivers/net/usb/kalmia.c - likewise
>   * drivers/net/ethernet/neterion/vxge/ - likewise
>   * drivers/spi/spi-pic32-sqi.c - likewise
>   * drivers/spi/spi-sh-msiof.c - likewise
>   * drivers/spi/spi-ti-qspi.c - likewise
> 
>   * drivers/tty/serial/mxs-auart.c - likewise - MXS AUART support
>   * drivers/tty/synclink.c - likewise Microgate SyncLink card support
>   * drivers/uio/uio_pruss - Texas Instruments PRUSS driver
>   * drivers/usb/dwc2 - CONFIG_USB_DWC2_DUAL_ROLE - DesignWare USB2 DRD Core Support for dual role mode
>   * drivers/usb/gadget/udc/ USB_GR_UDC - Aeroflex Gaisler GRUSBDC USB Peripheral Controller Driver
>   * drivers/video/fbdev/da8xx-fb.c -  FB_DA8XX DA8xx/OMAP-L1xx/AM335x Framebuffer support
>   * drivers/video/fbdev/mb862xx/mb862xxfb_accel.c - CONFIG_FB_MB862XX - Fujitsu MB862xx GDC support
>   * drivers/video/fbdev/vermilion/vermilion.c - Intel LE80578 (Vermilion) support
> 
> Then we have a few drivers which we know we need on x86 but for these
> we could use a run time flip to enable ZONE_DMA.
> 
>   * drivers/net/ethernet/broadcom/b44.c - bleh, yeah and there are some work hw bug
>     work arounds for this, *but* again since its also odd, we could deal with this
>     at run time
>   * drivers/net/wimax/i2400m/ - ugh, who cares about this crap anyway nowadays, my
>    point being another run time oddity
>   * drivers/net/wireless/broadcom/b43legacy/ - ugh, same
>   * drivers/platform/x86/asus-wmi.c - ugh same
>   * drivers/platform/x86/dell-smbios.c - ugh same
> 
> Staging drivers are expected to have flaws, but worth noting.
> 
>   * drivers/staging/ - scattered drivers, rtlwifi/ is probably the only relevant one for x86                                                                                                  
> SCSI is *severely* affected:                                                                                                                                                                                   
>   * drivers/scsi/aacraid/ - crap Adaptec AACRAID support
>   * drivers/scsi/ch.c - SCSI media changer support...
>   * drivers/scsi/initio.c - Initio INI-9X00U/UW SCSI device driver...
>   * drivers/scsi/osst.c - CHR_DEV_OSST  - SCSI OnStream SC-x0 tape support...
>   * drivers/scsi/pmcraid.c - CONFIG_SCSI_PMCRAID - PMC SIERRA Linux MaxRAID adapter support
>   * drivers/scsi/snic/ - Cisco SNIC Driver
>   * drivers/mmc/core/mmc_test.c - MMC_TEST - MMC host test driver
>  * drivers/net/wireless/broadcom/b43/ - means we'd have to at least use
>    static keys
>              
> Larger blockers (now I see one reason why SCSI is a disaster):
> 
>   * drivers/scsi/hosts.c - scsi_host_alloc() always uses
>     __GFP_DMA if (sht->unchecked_isa_dma && privsize)
>     this could likely be adjusted or split off to other
>     callers where we know this to be true.
>  * drivers/scsi/scsi_scan.c - scsi_probe_and_add_lun() has a similar check
>   * drivers/scsi/sg.c - sg_build_indirect() has similar check
>   * drivers/scsi/sr.c - get_capabilities() *always* uses GFP_DMA
>     which is called on sr_probe() WTF
>     Don't drop this -- its cdrom
>   * drivers/scsi/sr_ioctl.c - seriously...
>   * drivers/scsi/sr_vendor.c - sr_cd_check() - checks if the CD is
>     multisession, asks for offset etc 
>   * drivers/scsi/st.c - SCSI tape support - on enlarge_buffer() this
>     call BTW is recursive..  called on st_open(), the struct
>     file_operations open()...
> 
> Larger blockers (SPI is also severely affected):
>   * drivers/spi/spi.c - spi_pump_messages() which processes spi message queue
> 
> Larger blockers:
> 
>   * drivers/tty/hvc/hvc_iucv.c - hyperv console
> 
> And finally a non-issue:
> 
>   * drivers/xen/swiotlb-xen.c - used on struct dma_map_ops
>     xen_swiotlb_dma_ops alloc() for only to check if the caller
>     used it to se the dma_mask:                                                                                                                                   
>         dma_mask = dev->coherent_dma_mask;
>         if (!dma_mask)
>                 dma_mask = (gfp & GFP_DMA) ? DMA_BIT_MASK(24) : DMA_BIT_MASK(32);
> 
> That's the end of the review of all current explicit callers on x86.
> 
> # dma_alloc_coherent_gfp_flags() and dma_generic_alloc_coherent()
> 
> dma_alloc_coherent_gfp_flags() and dma_generic_alloc_coherent() set
> GFP_DMA if if (dma_mask <= DMA_BIT_MASK(24))
> 
> # blk_queue_bounce()
> 
> void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)           
> {
> 	...
> 	/*
> 	 * for non-isa bounce case, just check if the bounce pfn is equal
> 	 * to or bigger than the highest pfn in the system -- in that case,
> 	 * don't waste time iterating over bio segments
> 	 */
> 	if (!(q->bounce_gfp & GFP_DMA)) {
> 		if (q->limits.bounce_pfn >= blk_max_pfn)
> 			return;
> 		pool = page_pool;
> 	} else {
> 		BUG_ON(!isa_page_pool);
> 		pool = isa_page_pool;
> 	}
> ...
> }
> 
> # blk_queue_bounce_limit()
> 
> void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
> {
>         unsigned long b_pfn = max_addr >> PAGE_SHIFT;
>         int dma = 0;
>                                                                                 
>         q->bounce_gfp = GFP_NOIO;
> #if BITS_PER_LONG == 64
>         /*
>          * Assume anything <= 4GB can be handled by IOMMU.  Actually
>          * some IOMMUs can handle everything, but I don't know of a
>          * way to test this here.
>          */
>         if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
>                 dma = 1;
>         q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
> #else
>         if (b_pfn < blk_max_low_pfn)
>                 dma = 1;
>         q->limits.bounce_pfn = b_pfn;
> #endif
>         if (dma) {
>                 init_emergency_isa_pool();
>                 q->bounce_gfp = GFP_NOIO | GFP_DMA;
>                 q->limits.bounce_pfn = b_pfn;
>         }
> }
> 
> # dma_alloc_coherent_gfp_flags() and dma_generic_alloc_coherent()
> 
> dma_alloc_coherent_gfp_flags() sets GFP_DMA if if (dma_mask <= DMA_BIT_MASK(24)).                                                                                                             
> Likewise for dma_generic_alloc_coherent().
> 
> # intel_alloc_coherent()
> 
> intel_alloc_coherent() on drivers/iommu/intel-iommu.c also uses GFP_DMA                                                                                                                       
> for DMA_BIT_MASK(32), part of the struct dma_map_ops intel_dma_ops on alloc().                                                                                                                
> 
> # _regmap_raw_write()
> 
> _regmap_raw_write() seems to always use GFP_DMA for async writes.                                                                                                                             
> 
> # mempool_alloc_pages_isa()
> 
> It implies you use GFP_DMA.
> 
> Architectures removed which used ZONE_DMA
> -----------------------------------------
> 
> Although this topic pertains to x86, its worth mentioning that on the v4.17-rc1
> release 8 architectures were removed: blackfin, cris, frv, m32r, metag,
> mn10300, score, tile. Of these 8 architectures, 3 defined and used their own
> ZONE_DMA:
> 
>   * blackfin
>   * cris
>   * m32r



Thanks,
Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Laurent Pinchart]

Hi Mauro,

On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:
> There was a recent discussion about the use/abuse of GFP_DMA flag when
> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> 
> The idea seems to be to remove it, using CMA instead. Before doing that,
> better to check if what we have on media is are valid use cases for it, or
> if it is there just due to some misunderstanding (or because it was
> copied from some other code).
> 
> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> also posting today two other patches meant to stop abuse of it on USB
> drivers. Still, there are 4 platform drivers using it:
> 
> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> 	drivers/media/platform/omap3isp/ispstat.c
> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> 	drivers/media/platform/sti/hva/hva-mem.c
> 	drivers/media/spi/cxd2880-spi.c
> 
> Could you please check if GFP_DMA is really needed there, or if it is
> just because of some cut-and-paste from some other place?

I started looking at that for the omap3isp driver but Sakari beat me at 
submitting a patch. GFP_DMA isn't needed for omap3isp.

-- 
Regards,

Laurent Pinchart

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Mauro Carvalho Chehab]

Em Mon, 07 May 2018 16:26:08 +0300
Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:

> Hi Mauro,
> 
> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:
> > There was a recent discussion about the use/abuse of GFP_DMA flag when
> > allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> > 
> > The idea seems to be to remove it, using CMA instead. Before doing that,
> > better to check if what we have on media is are valid use cases for it, or
> > if it is there just due to some misunderstanding (or because it was
> > copied from some other code).
> > 
> > Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> > also posting today two other patches meant to stop abuse of it on USB
> > drivers. Still, there are 4 platform drivers using it:
> > 
> > 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> > 	drivers/media/platform/omap3isp/ispstat.c
> > 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> > 	drivers/media/platform/sti/hva/hva-mem.c
> > 	drivers/media/spi/cxd2880-spi.c
> > 
> > Could you please check if GFP_DMA is really needed there, or if it is
> > just because of some cut-and-paste from some other place?  
> 
> I started looking at that for the omap3isp driver but Sakari beat me at 
> submitting a patch. GFP_DMA isn't needed for omap3isp.
> 
Thank you both for looking into it.

Regards,
Mauro



Thanks,
Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Fabien DESSENNE]

On 07/05/18 17:19, Mauro Carvalho Chehab wrote:
> Em Mon, 07 May 2018 16:26:08 +0300
> Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:
>
>> Hi Mauro,
>>
>> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:
>>> There was a recent discussion about the use/abuse of GFP_DMA flag when
>>> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
>>>
>>> The idea seems to be to remove it, using CMA instead. Before doing that,
>>> better to check if what we have on media is are valid use cases for it, or
>>> if it is there just due to some misunderstanding (or because it was
>>> copied from some other code).
>>>
>>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
>>> also posting today two other patches meant to stop abuse of it on USB
>>> drivers. Still, there are 4 platform drivers using it:
>>>
>>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
>>> 	drivers/media/platform/omap3isp/ispstat.c
>>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
>>> 	drivers/media/platform/sti/hva/hva-mem.c

Hi Mauro,

The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run 
on ARM platforms, not on x86.
Since this thread deals with x86 & DMA trouble, I am not sure that we 
actually have a problem for the sti drivers.

There are some other sti drivers that make use of this GFP_DMA flag 
(drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.

Nevertheless I can see that the media sti drivers depend on COMPILE_TEST 
(which is not the case for the DRM ones).
Would it be an acceptable solution to remove the COMPILE_TEST dependency?

BR

Fabien

>>> 	drivers/media/spi/cxd2880-spi.c
>>>
>>> Could you please check if GFP_DMA is really needed there, or if it is
>>> just because of some cut-and-paste from some other place?
>> I started looking at that for the omap3isp driver but Sakari beat me at
>> submitting a patch. GFP_DMA isn't needed for omap3isp.
>>
> Thank you both for looking into it.
>
> Regards,
> Mauro
>
>
>
> Thanks,
> Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Mauro Carvalho Chehab]

Hi Fabien,

Em Mon, 14 May 2018 08:00:37 +0000
Fabien DESSENNE <fabien.dessenne@st.com> escreveu:

> On 07/05/18 17:19, Mauro Carvalho Chehab wrote:
> > Em Mon, 07 May 2018 16:26:08 +0300
> > Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:
> >  
> >> Hi Mauro,
> >>
> >> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:  
> >>> There was a recent discussion about the use/abuse of GFP_DMA flag when
> >>> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> >>>
> >>> The idea seems to be to remove it, using CMA instead. Before doing that,
> >>> better to check if what we have on media is are valid use cases for it, or
> >>> if it is there just due to some misunderstanding (or because it was
> >>> copied from some other code).
> >>>
> >>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> >>> also posting today two other patches meant to stop abuse of it on USB
> >>> drivers. Still, there are 4 platform drivers using it:
> >>>
> >>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> >>> 	drivers/media/platform/omap3isp/ispstat.c
> >>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> >>> 	drivers/media/platform/sti/hva/hva-mem.c  
> 
> Hi Mauro,
> 
> The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run 
> on ARM platforms, not on x86.
> Since this thread deals with x86 & DMA trouble, I am not sure that we 
> actually have a problem for the sti drivers.
> 
> There are some other sti drivers that make use of this GFP_DMA flag 
> (drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.
> 
> Nevertheless I can see that the media sti drivers depend on COMPILE_TEST 
> (which is not the case for the DRM ones).
> Would it be an acceptable solution to remove the COMPILE_TEST dependency?

This has nothing to do with either x86 or COMPILE_TEST. The thing is
that there's a plan for removing GFP_DMA from the Kernel[1], as it was
originally meant to be used only by old PCs, where the DMA controllers
used only  on the bottom 16 MB memory address (24 bits). IMHO, it is 
very unlikely that any ARM SoC have such limitation.

[1] https://lwn.net/Articles/753273/ (article will be freely available
on May, 17)

Anyway, before the removal of GFP_DMA happens, I'd like to better 
understand why we're using it at media, and if we can, instead,
set the DMA bit mask, just like almost all other media drivers
that require to confine DMA into a certain range do. In the case
of ARM, this is what we currently have:

drivers/media/platform/exynos-gsc/gsc-core.c:   vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
drivers/media/platform/exynos4-is/fimc-core.c:  vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
drivers/media/platform/exynos4-is/fimc-is.c:    vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
drivers/media/platform/exynos4-is/fimc-lite.c:  vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
drivers/media/platform/mtk-mdp/mtk_mdp_core.c:  vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
drivers/media/platform/omap3isp/isp.c:  ret = dma_coerce_mask_and_coherent(isp->dev, DMA_BIT_MASK(32));
drivers/media/platform/s5p-g2d/g2d.c:   vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
drivers/media/platform/s5p-jpeg/jpeg-core.c:    vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
drivers/media/platform/s5p-mfc/s5p_mfc.c:                                       DMA_BIT_MASK(32));
drivers/media/platform/s5p-mfc/s5p_mfc.c:                                       DMA_BIT_MASK(32));
drivers/media/platform/s5p-mfc/s5p_mfc.c:       vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));

> 
> BR
> 
> Fabien
> 
> >>> 	drivers/media/spi/cxd2880-spi.c
> >>>
> >>> Could you please check if GFP_DMA is really needed there, or if it is
> >>> just because of some cut-and-paste from some other place?  
> >> I started looking at that for the omap3isp driver but Sakari beat me at
> >> submitting a patch. GFP_DMA isn't needed for omap3isp.
> >>  
> > Thank you both for looking into it.
> >
> > Regards,
> > Mauro
> >
> >
> >
> > Thanks,
> > Mauro  



Thanks,
Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Mauro Carvalho Chehab]

Em Mon, 14 May 2018 07:35:03 -0300
Mauro Carvalho Chehab <mchehab+samsung@kernel.org> escreveu:

> Hi Fabien,
> 
> Em Mon, 14 May 2018 08:00:37 +0000
> Fabien DESSENNE <fabien.dessenne@st.com> escreveu:
> 
> > On 07/05/18 17:19, Mauro Carvalho Chehab wrote:
> > > Em Mon, 07 May 2018 16:26:08 +0300
> > > Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:
> > >  
> > >> Hi Mauro,
> > >>
> > >> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:  
> > >>> There was a recent discussion about the use/abuse of GFP_DMA flag when
> > >>> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> > >>>
> > >>> The idea seems to be to remove it, using CMA instead. Before doing that,
> > >>> better to check if what we have on media is are valid use cases for it, or
> > >>> if it is there just due to some misunderstanding (or because it was
> > >>> copied from some other code).
> > >>>
> > >>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> > >>> also posting today two other patches meant to stop abuse of it on USB
> > >>> drivers. Still, there are 4 platform drivers using it:
> > >>>
> > >>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> > >>> 	drivers/media/platform/omap3isp/ispstat.c
> > >>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> > >>> 	drivers/media/platform/sti/hva/hva-mem.c  
> > 
> > Hi Mauro,
> > 
> > The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run 
> > on ARM platforms, not on x86.
> > Since this thread deals with x86 & DMA trouble, I am not sure that we 
> > actually have a problem for the sti drivers.
> > 
> > There are some other sti drivers that make use of this GFP_DMA flag 
> > (drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.
> > 
> > Nevertheless I can see that the media sti drivers depend on COMPILE_TEST 
> > (which is not the case for the DRM ones).
> > Would it be an acceptable solution to remove the COMPILE_TEST dependency?
> 
> This has nothing to do with either x86 or COMPILE_TEST. The thing is
> that there's a plan for removing GFP_DMA from the Kernel[1], as it was
> originally meant to be used only by old PCs, where the DMA controllers
> used only  on the bottom 16 MB memory address (24 bits). IMHO, it is 
> very unlikely that any ARM SoC have such limitation.
> 
> [1] https://lwn.net/Articles/753273/ (article will be freely available
> on May, 17)

Btw, you can also read about that at:
	https://lwn.net/Articles/753274/

> 
> Anyway, before the removal of GFP_DMA happens, I'd like to better 
> understand why we're using it at media, and if we can, instead,
> set the DMA bit mask, just like almost all other media drivers
> that require to confine DMA into a certain range do. In the case
> of ARM, this is what we currently have:
> 
> drivers/media/platform/exynos-gsc/gsc-core.c:   vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> drivers/media/platform/exynos4-is/fimc-core.c:  vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> drivers/media/platform/exynos4-is/fimc-is.c:    vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> drivers/media/platform/exynos4-is/fimc-lite.c:  vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> drivers/media/platform/mtk-mdp/mtk_mdp_core.c:  vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> drivers/media/platform/omap3isp/isp.c:  ret = dma_coerce_mask_and_coherent(isp->dev, DMA_BIT_MASK(32));
> drivers/media/platform/s5p-g2d/g2d.c:   vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> drivers/media/platform/s5p-jpeg/jpeg-core.c:    vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> drivers/media/platform/s5p-mfc/s5p_mfc.c:                                       DMA_BIT_MASK(32));
> drivers/media/platform/s5p-mfc/s5p_mfc.c:                                       DMA_BIT_MASK(32));
> drivers/media/platform/s5p-mfc/s5p_mfc.c:       vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> 
> > 
> > BR
> > 
> > Fabien
> > 
> > >>> 	drivers/media/spi/cxd2880-spi.c
> > >>>
> > >>> Could you please check if GFP_DMA is really needed there, or if it is
> > >>> just because of some cut-and-paste from some other place?  
> > >> I started looking at that for the omap3isp driver but Sakari beat me at
> > >> submitting a patch. GFP_DMA isn't needed for omap3isp.
> > >>  
> > > Thank you both for looking into it.
> > >
> > > Regards,
> > > Mauro
> > >
> > >
> > >
> > > Thanks,
> > > Mauro  
> 
> 
> 
> Thanks,
> Mauro



Thanks,
Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Fabien DESSENNE]

On 14/05/18 12:39, Mauro Carvalho Chehab wrote:
> Em Mon, 14 May 2018 07:35:03 -0300
> Mauro Carvalho Chehab <mchehab+samsung@kernel.org> escreveu:
>
>> Hi Fabien,
>>
>> Em Mon, 14 May 2018 08:00:37 +0000
>> Fabien DESSENNE <fabien.dessenne@st.com> escreveu:
>>
>>> On 07/05/18 17:19, Mauro Carvalho Chehab wrote:
>>>> Em Mon, 07 May 2018 16:26:08 +0300
>>>> Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:
>>>>   
>>>>> Hi Mauro,
>>>>>
>>>>> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:
>>>>>> There was a recent discussion about the use/abuse of GFP_DMA flag when
>>>>>> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
>>>>>>
>>>>>> The idea seems to be to remove it, using CMA instead. Before doing that,
>>>>>> better to check if what we have on media is are valid use cases for it, or
>>>>>> if it is there just due to some misunderstanding (or because it was
>>>>>> copied from some other code).
>>>>>>
>>>>>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
>>>>>> also posting today two other patches meant to stop abuse of it on USB
>>>>>> drivers. Still, there are 4 platform drivers using it:
>>>>>>
>>>>>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
>>>>>> 	drivers/media/platform/omap3isp/ispstat.c
>>>>>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
>>>>>> 	drivers/media/platform/sti/hva/hva-mem.c
>>> Hi Mauro,
>>>
>>> The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run
>>> on ARM platforms, not on x86.
>>> Since this thread deals with x86 & DMA trouble, I am not sure that we
>>> actually have a problem for the sti drivers.
>>>
>>> There are some other sti drivers that make use of this GFP_DMA flag
>>> (drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.
>>>
>>> Nevertheless I can see that the media sti drivers depend on COMPILE_TEST
>>> (which is not the case for the DRM ones).
>>> Would it be an acceptable solution to remove the COMPILE_TEST dependency?
>> This has nothing to do with either x86 or COMPILE_TEST. The thing is
>> that there's a plan for removing GFP_DMA from the Kernel[1], as it was
>> originally meant to be used only by old PCs, where the DMA controllers
>> used only  on the bottom 16 MB memory address (24 bits). IMHO, it is
>> very unlikely that any ARM SoC have such limitation.
>>
>> [1] https://lwn.net/Articles/753273/ (article will be freely available
>> on May, 17)
> Btw, you can also read about that at:
> 	https://lwn.net/Articles/753274/
>
>> Anyway, before the removal of GFP_DMA happens, I'd like to better
>> understand why we're using it at media, and if we can, instead,
>> set the DMA bit mask, just like almost all other media drivers
>> that require to confine DMA into a certain range do. In the case
>> of ARM, this is what we currently have:
>>
>> drivers/media/platform/exynos-gsc/gsc-core.c:   vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
>> drivers/media/platform/exynos4-is/fimc-core.c:  vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
>> drivers/media/platform/exynos4-is/fimc-is.c:    vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
>> drivers/media/platform/exynos4-is/fimc-lite.c:  vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
>> drivers/media/platform/mtk-mdp/mtk_mdp_core.c:  vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
>> drivers/media/platform/omap3isp/isp.c:  ret = dma_coerce_mask_and_coherent(isp->dev, DMA_BIT_MASK(32));
>> drivers/media/platform/s5p-g2d/g2d.c:   vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
>> drivers/media/platform/s5p-jpeg/jpeg-core.c:    vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
>> drivers/media/platform/s5p-mfc/s5p_mfc.c:                                       DMA_BIT_MASK(32));
>> drivers/media/platform/s5p-mfc/s5p_mfc.c:                                       DMA_BIT_MASK(32));
>> drivers/media/platform/s5p-mfc/s5p_mfc.c:       vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
>>

That's clearer now, thank you for the clarification
I am about to send patches for the sti drivers (set the DMA bit mask)

BR,
Fabien

>>> BR
>>>
>>> Fabien
>>>
>>>>>> 	drivers/media/spi/cxd2880-spi.c
>>>>>>
>>>>>> Could you please check if GFP_DMA is really needed there, or if it is
>>>>>> just because of some cut-and-paste from some other place?
>>>>> I started looking at that for the omap3isp driver but Sakari beat me at
>>>>> submitting a patch. GFP_DMA isn't needed for omap3isp.
>>>>>   
>>>> Thank you both for looking into it.
>>>>
>>>> Regards,
>>>> Mauro
>>>>
>>>>
>>>>
>>>> Thanks,
>>>> Mauro
>>
>>
>> Thanks,
>> Mauro
>
>
> Thanks,
> Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Laurent Pinchart]

Hello,

On Tuesday, 15 May 2018 10:30:28 EEST Fabien DESSENNE wrote:
> On 14/05/18 12:39, Mauro Carvalho Chehab wrote:
> > Em Mon, 14 May 2018 07:35:03 -0300 Mauro Carvalho Chehab escreveu:
> >> Em Mon, 14 May 2018 08:00:37 +0000 Fabien DESSENNE escreveu:
> >>> On 07/05/18 17:19, Mauro Carvalho Chehab wrote:
> >>>> Em Mon, 07 May 2018 16:26:08 +0300 Laurent Pinchart escreveu:
> >>>>> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:
> >>>>> 
> >>>>>> There was a recent discussion about the use/abuse of GFP_DMA flag
> >>>>>> when allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> >>>>>>
> >>>>>> The idea seems to be to remove it, using CMA instead. Before doing
> >>>>>> that, better to check if what we have on media is are valid use cases
> >>>>>> for it, or if it is there just due to some misunderstanding (or
> >>>>>> because it was copied from some other code).
> >>>>>>
> >>>>>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> >>>>>> also posting today two other patches meant to stop abuse of it on
> >>>>>> USB drivers. Still, there are 4 platform drivers using it:
> >>>>>>
> >>>>>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> >>>>>> 	drivers/media/platform/omap3isp/ispstat.c
> >>>>>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> >>>>>> 	drivers/media/platform/sti/hva/hva-mem.c
> >>>
> >>> The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run
> >>> on ARM platforms, not on x86. Since this thread deals with x86 & DMA
> >>> trouble, I am not sure that we actually have a problem for the sti
> >>> drivers.
> >>>
> >>> There are some other sti drivers that make use of this GFP_DMA flag
> >>> (drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.
> >>>
> >>> Nevertheless I can see that the media sti drivers depend on COMPILE_TEST
> >>> (which is not the case for the DRM ones).
> >>> Would it be an acceptable solution to remove the COMPILE_TEST
> >>> dependency?
> >> 
> >> This has nothing to do with either x86 or COMPILE_TEST. The thing is
> >> that there's a plan for removing GFP_DMA from the Kernel[1], as it was
> >> originally meant to be used only by old PCs, where the DMA controllers
> >> used only  on the bottom 16 MB memory address (24 bits). IMHO, it is
> >> very unlikely that any ARM SoC have such limitation.
> >>
> >> [1] https://lwn.net/Articles/753273/ (article will be freely available
> >> on May, 17)
> > 
> > Btw, you can also read about that at:
> > 
> > 	https://lwn.net/Articles/753274/
> >
> >> Anyway, before the removal of GFP_DMA happens, I'd like to better
> >> understand why we're using it at media, and if we can, instead,
> >> set the DMA bit mask, just like almost all other media drivers
> >> that require to confine DMA into a certain range do. In the case
> >> of ARM, this is what we currently have:
> >>
> >> drivers/media/platform/exynos-gsc/gsc-core.c:  
> >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/exynos4-is/fimc-core.c: 
> >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/exynos4-is/fimc-is.c:   
> >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/exynos4-is/fimc-lite.c: 
> >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/mtk-mdp/mtk_mdp_core.c: 
> >> vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/omap3isp/isp.c:  ret =
> >> dma_coerce_mask_and_coherent(isp->dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/s5p-g2d/g2d.c:  
> >> vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/s5p-jpeg/jpeg-core.c:   
> >> vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> >> drivers/media/platform/s5p-mfc/s5p_mfc.c:                               
> >>        DMA_BIT_MASK(32));
> >> drivers/media/platform/s5p-mfc/s5p_mfc.c:     
> >>                                  DMA_BIT_MASK(32));
> >> drivers/media/platform/s5p-mfc/s5p_mfc.c:      
> >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> 
> That's clearer now, thank you for the clarification
> I am about to send patches for the sti drivers (set the DMA bit mask)

Some drivers call vb2_dma_contig_set_max_seg_size() and some call 
dma_coerce_mask_and_coherent(). Both are likely needed, the former telling the 
DMA mapping API about the maximum size of a scatter-gather chunk that the 
device supports (when using vb2-dma-contig that size should really be the full 
address space supported by the device as we want DMA-contiguous buffers), and 
the latter telling the DMA mapping API about the address space that is 
accessible through DMA (and thus in which address range buffers must be 
placed).

I wonder why the omap3isp driver works without a 
vb2_dma_contig_set_max_seg_size() call. Sakari, any insight ?

> >>>>>> 	drivers/media/spi/cxd2880-spi.c
> >>>>>>
> >>>>>> Could you please check if GFP_DMA is really needed there, or if it
> >>>>>> is just because of some cut-and-paste from some other place?
> >>>>> 
> >>>>> I started looking at that for the omap3isp driver but Sakari beat me
> >>>>> at submitting a patch. GFP_DMA isn't needed for omap3isp.
> >>>> 
> >>>> Thank you both for looking into it.

-- 
Regards,

Laurent Pinchart

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Mauro Carvalho Chehab]

Hi Laurent,

Em Tue, 15 May 2018 11:27:44 +0300
Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:

> Hello,
> 
> On Tuesday, 15 May 2018 10:30:28 EEST Fabien DESSENNE wrote:
> > On 14/05/18 12:39, Mauro Carvalho Chehab wrote:  
> > > Em Mon, 14 May 2018 07:35:03 -0300 Mauro Carvalho Chehab escreveu:  
> > >> Em Mon, 14 May 2018 08:00:37 +0000 Fabien DESSENNE escreveu:  
> > >>> On 07/05/18 17:19, Mauro Carvalho Chehab wrote:  
> > >>>> Em Mon, 07 May 2018 16:26:08 +0300 Laurent Pinchart escreveu:  
> > >>>>> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:
> > >>>>>   
> > >>>>>> There was a recent discussion about the use/abuse of GFP_DMA flag
> > >>>>>> when allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> > >>>>>>
> > >>>>>> The idea seems to be to remove it, using CMA instead. Before doing
> > >>>>>> that, better to check if what we have on media is are valid use cases
> > >>>>>> for it, or if it is there just due to some misunderstanding (or
> > >>>>>> because it was copied from some other code).
> > >>>>>>
> > >>>>>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> > >>>>>> also posting today two other patches meant to stop abuse of it on
> > >>>>>> USB drivers. Still, there are 4 platform drivers using it:
> > >>>>>>
> > >>>>>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> > >>>>>> 	drivers/media/platform/omap3isp/ispstat.c
> > >>>>>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> > >>>>>> 	drivers/media/platform/sti/hva/hva-mem.c  
> > >>>
> > >>> The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run
> > >>> on ARM platforms, not on x86. Since this thread deals with x86 & DMA
> > >>> trouble, I am not sure that we actually have a problem for the sti
> > >>> drivers.
> > >>>
> > >>> There are some other sti drivers that make use of this GFP_DMA flag
> > >>> (drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.
> > >>>
> > >>> Nevertheless I can see that the media sti drivers depend on COMPILE_TEST
> > >>> (which is not the case for the DRM ones).
> > >>> Would it be an acceptable solution to remove the COMPILE_TEST
> > >>> dependency?  
> > >> 
> > >> This has nothing to do with either x86 or COMPILE_TEST. The thing is
> > >> that there's a plan for removing GFP_DMA from the Kernel[1], as it was
> > >> originally meant to be used only by old PCs, where the DMA controllers
> > >> used only  on the bottom 16 MB memory address (24 bits). IMHO, it is
> > >> very unlikely that any ARM SoC have such limitation.
> > >>
> > >> [1] https://lwn.net/Articles/753273/ (article will be freely available
> > >> on May, 17)  
> > > 
> > > Btw, you can also read about that at:
> > > 
> > > 	https://lwn.net/Articles/753274/
> > >  
> > >> Anyway, before the removal of GFP_DMA happens, I'd like to better
> > >> understand why we're using it at media, and if we can, instead,
> > >> set the DMA bit mask, just like almost all other media drivers
> > >> that require to confine DMA into a certain range do. In the case
> > >> of ARM, this is what we currently have:
> > >>
> > >> drivers/media/platform/exynos-gsc/gsc-core.c:  
> > >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/exynos4-is/fimc-core.c: 
> > >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/exynos4-is/fimc-is.c:   
> > >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/exynos4-is/fimc-lite.c: 
> > >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/mtk-mdp/mtk_mdp_core.c: 
> > >> vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/omap3isp/isp.c:  ret =
> > >> dma_coerce_mask_and_coherent(isp->dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/s5p-g2d/g2d.c:  
> > >> vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/s5p-jpeg/jpeg-core.c:   
> > >> vb2_dma_contig_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
> > >> drivers/media/platform/s5p-mfc/s5p_mfc.c:                               
> > >>        DMA_BIT_MASK(32));
> > >> drivers/media/platform/s5p-mfc/s5p_mfc.c:     
> > >>                                  DMA_BIT_MASK(32));
> > >> drivers/media/platform/s5p-mfc/s5p_mfc.c:      
> > >> vb2_dma_contig_set_max_seg_size(dev, DMA_BIT_MASK(32));  
> > 
> > That's clearer now, thank you for the clarification
> > I am about to send patches for the sti drivers (set the DMA bit mask)  
> 
> Some drivers call vb2_dma_contig_set_max_seg_size() and some call 
> dma_coerce_mask_and_coherent(). Both are likely needed, the former telling the 
> DMA mapping API about the maximum size of a scatter-gather chunk that the 
> device supports (when using vb2-dma-contig that size should really be the full 
> address space supported by the device as we want DMA-contiguous buffers), and 
> the latter telling the DMA mapping API about the address space that is 
> accessible through DMA (and thus in which address range buffers must be 
> placed).
> 
> I wonder why the omap3isp driver works without a 
> vb2_dma_contig_set_max_seg_size() call. Sakari, any insight ?

I checked the usage of vb2_dma_contig_set_max_seg_size(). What it
does is to allocate dev->dma_parms and call dma_set_max_seg_size().

Allocating dev->dma_parms change the behavior of 2 function pairs 
(inlined at dma_mapping.h):

1) dma_get_seg_boundary() / dma_set_seg_boundary()

As no media drivers use dev->dma_parms->segment_boundary_mask,
it will keep returning DMA_BIT_MASK(32), either calling or not
the VB2-specific function.

1) dma_get_max_seg_size() / dma_set_max_seg_size()

Checking where dma_get_max_seg_size() is used returns:

$ git grep dma_get_max_seg_size
arch/alpha/kernel/pci_iommu.c:  max_seg_size = dev ? dma_get_max_seg_size(dev) : 0;
arch/arm/mm/dma-mapping.c:      unsigned int max = dma_get_max_seg_size(dev);
arch/ia64/hp/common/sba_iommu.c:        unsigned int max_seg_size = dma_get_max_seg_size(dev);
arch/powerpc/kernel/iommu.c:    max_seg_size = dma_get_max_seg_size(dev);
arch/s390/pci/pci_dma.c:        unsigned int max = dma_get_max_seg_size(dev);
arch/sparc/kernel/iommu.c:      max_seg_size = dma_get_max_seg_size(dev);
arch/sparc/kernel/pci_sun4v.c:  max_seg_size = dma_get_max_seg_size(dev);
arch/x86/kernel/amd_gart_64.c:  max_seg_size    = dma_get_max_seg_size(dev);
drivers/firewire/sbp2.c:        if (dma_get_max_seg_size(device->card->device) > SBP2_MAX_SEG_SIZE)
drivers/iio/buffer/industrialio-buffer-dmaengine.c:     dmaengine_buffer->max_size = dma_get_max_seg_size(chan->device->dev);
drivers/iommu/dma-iommu.c:      unsigned int cur_len = 0, max_len = dma_get_max_seg_size(dev);
drivers/media/common/videobuf2/videobuf2-dma-contig.c:  if (dma_get_max_seg_size(dev) < size)
drivers/mmc/host/mmci.c:                unsigned int max_seg_size = dma_get_max_seg_size(dev);
drivers/mmc/host/mmci.c:                unsigned int max_seg_size = dma_get_max_seg_size(dev);
drivers/mmc/host/mxcmmc.c:              mmc->max_seg_size = dma_get_max_seg_size(
drivers/mmc/host/mxs-mmc.c:     mmc->max_seg_size = dma_get_max_seg_size(ssp->dmach->device->dev);
drivers/parisc/iommu-helpers.h: unsigned int max_seg_size = min(dma_get_max_seg_size(dev),
drivers/scsi/scsi_lib.c:        blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
drivers/spi/spi.c:      unsigned int max_seg_size = dma_get_max_seg_size(dev);
include/linux/dma-mapping.h:static inline unsigned int dma_get_max_seg_size(struct device *dev)
lib/dma-debug.c:        unsigned int max_range = dma_get_max_seg_size(ref->dev);
sound/soc/soc-generic-dmaengine-pcm.c:  hw.period_bytes_max = dma_get_max_seg_size(dma_dev);

As vb2_dma_contig_set_max_seg_size() is called only on ARM drivers
for Exynos and Mediatek, we only need to check where it is used inside
arch/arm/mm/dma-mapping.c. There, only __iommu_map_sg() function
calls it. So, except if mis-named, I would be expecting it to be used
only for Scatter/Gather DMA.

So, it seems that, with the current code, setting it for dma_contig 
probably does nothing.

Please double-check, as things like that can be tricky.

IMO, it is worth removing vb2_dma_contig_set_max_seg_size() and be sure that
the drivers calls it are calling dma_coerce_mask_and_coherent().

Thanks,
Mauro

Re: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Luis R. Rodriguez]

On Mon, May 14, 2018 at 07:35:03AM -0300, Mauro Carvalho Chehab wrote:
> Hi Fabien,
> 
> Em Mon, 14 May 2018 08:00:37 +0000
> Fabien DESSENNE <fabien.dessenne@st.com> escreveu:
> 
> > On 07/05/18 17:19, Mauro Carvalho Chehab wrote:
> > > Em Mon, 07 May 2018 16:26:08 +0300
> > > Laurent Pinchart <laurent.pinchart@ideasonboard.com> escreveu:
> > >  
> > >> Hi Mauro,
> > >>
> > >> On Saturday, 5 May 2018 19:08:15 EEST Mauro Carvalho Chehab wrote:  
> > >>> There was a recent discussion about the use/abuse of GFP_DMA flag when
> > >>> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> > >>>
> > >>> The idea seems to be to remove it, using CMA instead. Before doing that,
> > >>> better to check if what we have on media is are valid use cases for it, or
> > >>> if it is there just due to some misunderstanding (or because it was
> > >>> copied from some other code).
> > >>>
> > >>> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> > >>> also posting today two other patches meant to stop abuse of it on USB
> > >>> drivers. Still, there are 4 platform drivers using it:
> > >>>
> > >>> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> > >>> 	drivers/media/platform/omap3isp/ispstat.c
> > >>> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> > >>> 	drivers/media/platform/sti/hva/hva-mem.c  
> > 
> > Hi Mauro,
> > 
> > The two STI drivers (bdisp-hw.c and hva-mem.c) are only expected to run 
> > on ARM platforms, not on x86.
> > Since this thread deals with x86 & DMA trouble, I am not sure that we 
> > actually have a problem for the sti drivers.
> > 
> > There are some other sti drivers that make use of this GFP_DMA flag 
> > (drivers/gpu/drm/sti/sti_*.c) and it does not seem to be a problem.
> > 
> > Nevertheless I can see that the media sti drivers depend on COMPILE_TEST 
> > (which is not the case for the DRM ones).
> > Would it be an acceptable solution to remove the COMPILE_TEST dependency?
> 
> This has nothing to do with either x86 

Actually it does.

> or COMPILE_TEST. The thing is
> that there's a plan for removing GFP_DMA from the Kernel[1], 

That would not be possible given architectures use GFP_DMA for other
things and there are plenty of legacy x86 drivers which still need to be
around. So the focus from mm folks shifted to letting x86 folks map
GFP_DMA onto the CMA pool. Long term, this is nothing that driver developers
need to care for, but just knowing internally behind the scenes there is some
cleaning up being done in terms of architecture.

> as it was
> originally meant to be used only by old PCs, where the DMA controllers
> used only  on the bottom 16 MB memory address (24 bits).

This is actually the part that is x86 specific.

Each other architecture may use it for some other definition and it seems
some architectures use GFP_DMA all over the place. So the topic really should
be about x86.

> IMHO, it is 
> very unlikely that any ARM SoC have such limitation.

Right, how the flag is used on other architectures varies, so in fact the
focus for cleaning up for now should be an x86 effort. Whether or not
other architectures do something silly with GFP_DMA is beyond the scope
of what was discussed.

  Luis

RE: Are media drivers abusing of GFP_DMA? - was: Re: [LSF/MM TOPIC NOTES] x86 ZONE_DMA love

[Yasunari.Takiguchi]

Dear Mauro

> -----Original Message-----
> 
> There was a recent discussion about the use/abuse of GFP_DMA flag when
> allocating memories at LSF/MM 2018 (see Luis notes enclosed).
> 
> The idea seems to be to remove it, using CMA instead. Before doing that,
> better to check if what we have on media is are valid use cases for it,
> or
> if it is there just due to some misunderstanding (or because it was
> copied from some other code).
> 
> Hans de Goede sent us today a patch stopping abuse at gspca, and I'm
> also posting today two other patches meant to stop abuse of it on USB
> drivers. Still, there are 4 platform drivers using it:
> 
> 	$ git grep -l -E "GFP_DMA\\b" drivers/media/
> 	drivers/media/platform/omap3isp/ispstat.c
> 	drivers/media/platform/sti/bdisp/bdisp-hw.c
> 	drivers/media/platform/sti/hva/hva-mem.c
> 	drivers/media/spi/cxd2880-spi.c
> 
> Could you please check if GFP_DMA is really needed there, or if it is
> just because of some cut-and-paste from some other place?
About drivers/media/spi/cxd2880-spi.c,
we referred to kmalloc of driver/spi/spi.c spi_write_then_read() function and made this code. 

Regards,
Takiguchi