| Message ID | 20220617125750.728590-4-arnd@kernel.org (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | phase out CONFIG_VIRT_TO_BUS | expand |
Arnd, Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: > From: Arnd Bergmann <arnd@arndb.de> > > All architecture-independent users of virt_to_bus() and bus_to_virt() > have been fixed to use the dma mapping interfaces or have been > removed now. This means the definitions on most architectures, and the > CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. > > The only exceptions to this are a few network and scsi drivers for m68k > Amiga and VME machines and ppc32 Macintosh. These drivers work correctly > with the old interfaces and are probably not worth changing. The Amiga SCSI drivers are all old WD33C93 ones, and replacing virt_to_bus by virt_to_phys in the dma_setup() function there would cause no functional change at all. drivers/vme/bridges/vme_ca91cx42.c hasn't been used at all on m68k (it is a PCI-to-VME bridge chipset driver that would be needed on architectures that natively use a PCI bus). I haven't found anything that selects that driver, so not sure it is even still in use?? That would allow you to drop the remaining virt_to_bus define from arch/m68k/include/asm/virtconvert.h. I could submit a patch to convert the Amiga SCSI drivers to use virt_to_phys if Geert and the SCSI maintainers think it's worth the churn. 32bit powerpc is a different matter though. Cheers, Michael
On Sat, Jun 18, 2022 at 3:06 AM Michael Schmitz <schmitzmic@gmail.com> wrote: > Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: > > > > All architecture-independent users of virt_to_bus() and bus_to_virt() > > have been fixed to use the dma mapping interfaces or have been > > removed now. This means the definitions on most architectures, and the > > CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. > > > > The only exceptions to this are a few network and scsi drivers for m68k > > Amiga and VME machines and ppc32 Macintosh. These drivers work correctly > > with the old interfaces and are probably not worth changing. > > The Amiga SCSI drivers are all old WD33C93 ones, and replacing > virt_to_bus by virt_to_phys in the dma_setup() function there would > cause no functional change at all. Ok, thanks for taking a look here. > drivers/vme/bridges/vme_ca91cx42.c hasn't been used at all on m68k (it > is a PCI-to-VME bridge chipset driver that would be needed on > architectures that natively use a PCI bus). I haven't found anything > that selects that driver, so not sure it is even still in use?? It's gone now, Greg has already taken my patches for this through the staging tree. > That would allow you to drop the remaining virt_to_bus define from > arch/m68k/include/asm/virtconvert.h. > > I could submit a patch to convert the Amiga SCSI drivers to use > virt_to_phys if Geert and the SCSI maintainers think it's worth the churn. I don't think using virt_to_phys() is an improvement here, as virt_to_bus() was originally meant as a better abstraction to replace the use of virt_to_phys() to make drivers portable, before it got replaced by the dma-mapping interface in turn. It looks like the Amiga SCSI drivers have an open-coded version of what dma_map_single() does, to do bounce buffering and cache management. The ideal solution would be to convert the drivers actually use the appropriate dma-mapping interfaces and remove this custom code. The same could be done for the two vme drivers (scsi/mvme147.c and ethernet/82596.c), which do the cache management but apparently don't need swiotlb bounce buffering. Rewriting the drivers to modern APIs is of course non-trivial, and if you want a shortcut here, I would suggest introducing platform specific helpers similar to isa_virt_to_bus() and call them amiga_virt_to_bus() and vme_virt_to_bus, respectively. Putting these into a platform specific header file at least helps clarify that both the helper functions and the drivers using them are non-portable. > 32bit powerpc is a different matter though. It's similar, but unrelated. The two apple ethernet drivers (bmac and mace) can again either get changed to use the dma-mapping interfaces, or get a custom pmac_virt_to_bus()/ pmac_bus_to_virt() helper. There is also drivers/tty/serial/cpm_uart/cpm_uart_cpm2.c, which I think just needs a trivial change, but I'm not sure how to do it correctly. Arnd
Arnd, Am 24.06.2022 um 21:10 schrieb Arnd Bergmann: > On Sat, Jun 18, 2022 at 3:06 AM Michael Schmitz <schmitzmic@gmail.com> wrote: >> Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: >>> >>> All architecture-independent users of virt_to_bus() and bus_to_virt() >>> have been fixed to use the dma mapping interfaces or have been >>> removed now. This means the definitions on most architectures, and the >>> CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. >>> >>> The only exceptions to this are a few network and scsi drivers for m68k >>> Amiga and VME machines and ppc32 Macintosh. These drivers work correctly >>> with the old interfaces and are probably not worth changing. >> >> The Amiga SCSI drivers are all old WD33C93 ones, and replacing >> virt_to_bus by virt_to_phys in the dma_setup() function there would >> cause no functional change at all. > > Ok, thanks for taking a look here. > >> drivers/vme/bridges/vme_ca91cx42.c hasn't been used at all on m68k (it >> is a PCI-to-VME bridge chipset driver that would be needed on >> architectures that natively use a PCI bus). I haven't found anything >> that selects that driver, so not sure it is even still in use?? > > It's gone now, Greg has already taken my patches for this through > the staging tree. One less to worry about, thanks. >> That would allow you to drop the remaining virt_to_bus define from >> arch/m68k/include/asm/virtconvert.h. >> >> I could submit a patch to convert the Amiga SCSI drivers to use >> virt_to_phys if Geert and the SCSI maintainers think it's worth the churn. > > I don't think using virt_to_phys() is an improvement here, as > virt_to_bus() was originally meant as a better abstraction to > replace the use of virt_to_phys() to make drivers portable, before > it got replaced by the dma-mapping interface in turn. > > It looks like the Amiga SCSI drivers have an open-coded version of > what dma_map_single() does, to do bounce buffering and cache > management. The ideal solution would be to convert the drivers > actually use the appropriate dma-mapping interfaces and remove > this custom code. I've taken another look at these drivers' dma_setup() functions and they all look much more complex than the Amiga ESP drivers (which do use the dma-mapping interface for parts of the DMA setup). From my limited understanding, the difference between the ESP and WD33C93 drivers is that the former are used on 040/060 accelerator boards only (where the processor does do bus snooping and DMA can access all of RAM). The latter ones would need cache management, could only use non-coherent mappings and would require special case handling for DMA-inaccessible RAM inside a device-specific dma ops' map_page() function. That's several bridges too far for me ... I have no Amiga hardware whatsoever, and know no one who could test changes to WD33C93 drivers for me. What I have is a NCR5380 with the proverbial 'pathological DMA' integration example (and its driver was never changed to even use virt_to_bus()!). I might learn enough about using the dma-mapping API on that one eventually (though the requirement for at least 1 MB swiotlb bounce buffers looks hard to meet), and use that to convert the WD33C93 drivers, but it would still remain untested. > The same could be done for the two vme drivers (scsi/mvme147.c > and ethernet/82596.c), which do the cache management but > apparently don't need swiotlb bounce buffering. > > Rewriting the drivers to modern APIs is of course non-trivial, > and if you want a shortcut here, I would suggest introducing > platform specific helpers similar to isa_virt_to_bus() and call > them amiga_virt_to_bus() and vme_virt_to_bus, respectively. I don't think Amiga and m68k VME differ at all in that respect, so might just call it m68k_virt_to_bus() for now. > Putting these into a platform specific header file at least helps > clarify that both the helper functions and the drivers using them > are non-portable. There are no platform specific header files other than asm/amigahw.h and asm/mvme147hw.h, currently only holding register address definitions. Would it be OK to add m68k_virt_to_bus() in there if it can't remain in asm/virtconvert.h, Geert? > >> 32bit powerpc is a different matter though. > > It's similar, but unrelated. The two apple ethernet drivers > (bmac and mace) can again either get changed to use the > dma-mapping interfaces, or get a custom pmac_virt_to_bus()/ > pmac_bus_to_virt() helper. Hmmm - I see Finn had done the DMA API conversion on macmace.c which might give some hints on what to do about mace.c ... no idea about bmac.c though. And again, haven't got hardware to test, so custom helpers is it, then. Cheers, Michael > There is also drivers/tty/serial/cpm_uart/cpm_uart_cpm2.c, > which I think just needs a trivial change, but I'm not sure > how to do it correctly. > > Arnd >
(On Sun, Jun 26, 2022 at 7:21 AM Michael Schmitz <schmitzmic@gmail.com> wrote: > > The same could be done for the two vme drivers (scsi/mvme147.c > > and ethernet/82596.c), which do the cache management but > > apparently don't need swiotlb bounce buffering. > > > > Rewriting the drivers to modern APIs is of course non-trivial, > > and if you want a shortcut here, I would suggest introducing > > platform specific helpers similar to isa_virt_to_bus() and call > > them amiga_virt_to_bus() and vme_virt_to_bus, respectively. > > I don't think Amiga and m68k VME differ at all in that respect, so might > just call it m68k_virt_to_bus() for now. > > > Putting these into a platform specific header file at least helps > > clarify that both the helper functions and the drivers using them > > are non-portable. > > There are no platform specific header files other than asm/amigahw.h and > asm/mvme147hw.h, currently only holding register address definitions. > Would it be OK to add m68k_virt_to_bus() in there if it can't remain in > asm/virtconvert.h, Geert? In that case, I would just leave it under the current name and not change m68k at all. I don't like the m68k_virt_to_bus() name because there is not anything CPU specific in what it does, and keeping it in a common header does nothing to prevent it from being used on other platforms either. > >> 32bit powerpc is a different matter though. > > > > It's similar, but unrelated. The two apple ethernet drivers > > (bmac and mace) can again either get changed to use the > > dma-mapping interfaces, or get a custom pmac_virt_to_bus()/ > > pmac_bus_to_virt() helper. > > Hmmm - I see Finn had done the DMA API conversion on macmace.c which > might give some hints on what to do about mace.c ... no idea about > bmac.c though. And again, haven't got hardware to test, so custom > helpers is it, then. Ok. Arnd
Arnd, Am 26.06.2022 um 20:36 schrieb Arnd Bergmann: >> There are no platform specific header files other than asm/amigahw.h and >> asm/mvme147hw.h, currently only holding register address definitions. >> Would it be OK to add m68k_virt_to_bus() in there if it can't remain in >> asm/virtconvert.h, Geert? > > In that case, I would just leave it under the current name and not change > m68k at all. I don't like the m68k_virt_to_bus() name because there is > not anything CPU specific in what it does, and keeping it in a common > header does nothing to prevent it from being used on other platforms > either. Fair enough. >>>> 32bit powerpc is a different matter though. >>> >>> It's similar, but unrelated. The two apple ethernet drivers >>> (bmac and mace) can again either get changed to use the >>> dma-mapping interfaces, or get a custom pmac_virt_to_bus()/ >>> pmac_bus_to_virt() helper. >> >> Hmmm - I see Finn had done the DMA API conversion on macmace.c which >> might give some hints on what to do about mace.c ... no idea about >> bmac.c though. And again, haven't got hardware to test, so custom >> helpers is it, then. > > Ok. Again, no platform specific headers to shift renamed helpers to, so may as well keep this as-is. Cheers, Michael > > Arnd >
Hi Michael, On Sat, Jun 18, 2022 at 3:06 AM Michael Schmitz <schmitzmic@gmail.com> wrote: > Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: > > From: Arnd Bergmann <arnd@arndb.de> > > > > All architecture-independent users of virt_to_bus() and bus_to_virt() > > have been fixed to use the dma mapping interfaces or have been > > removed now. This means the definitions on most architectures, and the > > CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. > > > > The only exceptions to this are a few network and scsi drivers for m68k > > Amiga and VME machines and ppc32 Macintosh. These drivers work correctly > > with the old interfaces and are probably not worth changing. > > The Amiga SCSI drivers are all old WD33C93 ones, and replacing > virt_to_bus by virt_to_phys in the dma_setup() function there would > cause no functional change at all. FTR, the sgiwd93 driver use dma_map_single(). Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Hi Geert, On 27/06/22 20:26, Geert Uytterhoeven wrote: > Hi Michael, > > On Sat, Jun 18, 2022 at 3:06 AM Michael Schmitz <schmitzmic@gmail.com> wrote: >> Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: >>> From: Arnd Bergmann <arnd@arndb.de> >>> >>> All architecture-independent users of virt_to_bus() and bus_to_virt() >>> have been fixed to use the dma mapping interfaces or have been >>> removed now. This means the definitions on most architectures, and the >>> CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. >>> >>> The only exceptions to this are a few network and scsi drivers for m68k >>> Amiga and VME machines and ppc32 Macintosh. These drivers work correctly >>> with the old interfaces and are probably not worth changing. >> The Amiga SCSI drivers are all old WD33C93 ones, and replacing >> virt_to_bus by virt_to_phys in the dma_setup() function there would >> cause no functional change at all. > FTR, the sgiwd93 driver use dma_map_single(). Thanks! From what I see, it doesn't have to deal with bounce buffers though? Cheers, Michael > > Gr{oetje,eeting}s, > > Geert > > -- > Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org > > In personal conversations with technical people, I call myself a hacker. But > when I'm talking to journalists I just say "programmer" or something like that. > -- Linus Torvalds
Hii Geert Am 28.06.2022 um 09:12 schrieb Michael Schmitz: > Hi Geert, > > On 27/06/22 20:26, Geert Uytterhoeven wrote: >> Hi Michael, >> >> On Sat, Jun 18, 2022 at 3:06 AM Michael Schmitz <schmitzmic@gmail.com> >> wrote: >>> Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: >>>> From: Arnd Bergmann <arnd@arndb.de> >>>> >>>> All architecture-independent users of virt_to_bus() and bus_to_virt() >>>> have been fixed to use the dma mapping interfaces or have been >>>> removed now. This means the definitions on most architectures, and the >>>> CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. >>>> >>>> The only exceptions to this are a few network and scsi drivers for m68k >>>> Amiga and VME machines and ppc32 Macintosh. These drivers work >>>> correctly >>>> with the old interfaces and are probably not worth changing. >>> The Amiga SCSI drivers are all old WD33C93 ones, and replacing >>> virt_to_bus by virt_to_phys in the dma_setup() function there would >>> cause no functional change at all. >> FTR, the sgiwd93 driver use dma_map_single(). > > Thanks! From what I see, it doesn't have to deal with bounce buffers > though? Leaving the bounce buffer handling in place, and taking a few other liberties - this is what converting the easiest case (a3000 SCSI) might look like. Any obvious mistakes? The mvme147 driver would be very similar to handle (after conversion to a platform device). The driver allocates bounce buffers using kmalloc if it hits an unaligned data buffer - can such buffers still even happen these days? If I understand dma_map_single() correctly, the resulting dma handle would be equally misaligned? To allocate a bounce buffer, would it be OK to use dma_alloc_coherent() even though AFAIU memory used for DMA buffers generally isn't consistent on m68k? Thinking ahead to the other two Amiga drivers - I wonder whether allocating a static bounce buffer or a DMA pool at driver init is likely to succeed if the kernel runs from the low 16 MB RAM chunk? It certainly won't succeed if the kernel runs from a higher memory address, so the present bounce buffer logic around amiga_chip_alloc() might still need to be used here. Leaves the question whether converting the gvp11 and a2091 drivers is actually worth it, if bounce buffers still have to be handled explicitly. Untested (except for compile testing), un-checkpatched, don't try this on any disk with valuable data ... Cheers, Michael
Hi Michael, On Tue, Jun 28, 2022 at 5:26 AM Michael Schmitz <schmitzmic@gmail.com> wrote: > Am 28.06.2022 um 09:12 schrieb Michael Schmitz: > > On 27/06/22 20:26, Geert Uytterhoeven wrote: > >> On Sat, Jun 18, 2022 at 3:06 AM Michael Schmitz <schmitzmic@gmail.com> > >> wrote: > >>> Am 18.06.2022 um 00:57 schrieb Arnd Bergmann: > >>>> From: Arnd Bergmann <arnd@arndb.de> > >>>> > >>>> All architecture-independent users of virt_to_bus() and bus_to_virt() > >>>> have been fixed to use the dma mapping interfaces or have been > >>>> removed now. This means the definitions on most architectures, and the > >>>> CONFIG_VIRT_TO_BUS symbol are now obsolete and can be removed. > >>>> > >>>> The only exceptions to this are a few network and scsi drivers for m68k > >>>> Amiga and VME machines and ppc32 Macintosh. These drivers work > >>>> correctly > >>>> with the old interfaces and are probably not worth changing. > >>> The Amiga SCSI drivers are all old WD33C93 ones, and replacing > >>> virt_to_bus by virt_to_phys in the dma_setup() function there would > >>> cause no functional change at all. > >> FTR, the sgiwd93 driver use dma_map_single(). > > > > Thanks! From what I see, it doesn't have to deal with bounce buffers > > though? > > Leaving the bounce buffer handling in place, and taking a few other > liberties - this is what converting the easiest case (a3000 SCSI) might > look like. Any obvious mistakes? The mvme147 driver would be very > similar to handle (after conversion to a platform device). Thanks, looks reasonable. > The driver allocates bounce buffers using kmalloc if it hits an > unaligned data buffer - can such buffers still even happen these days? No idea. > If I understand dma_map_single() correctly, the resulting dma handle > would be equally misaligned? > > To allocate a bounce buffer, would it be OK to use dma_alloc_coherent() > even though AFAIU memory used for DMA buffers generally isn't consistent > on m68k? > > Thinking ahead to the other two Amiga drivers - I wonder whether > allocating a static bounce buffer or a DMA pool at driver init is likely > to succeed if the kernel runs from the low 16 MB RAM chunk? It certainly > won't succeed if the kernel runs from a higher memory address, so the > present bounce buffer logic around amiga_chip_alloc() might still need > to be used here. > > Leaves the question whether converting the gvp11 and a2091 drivers is > actually worth it, if bounce buffers still have to be handled explicitly. A2091 should be straight-forward, as A3000 is basically A2091 on the motherboard (comparing the two drivers, looks like someone's been sprinkling mb()s over the A3000 driver). I don't have any of these SCSI host adapters (not counting the A590 (~A2091) expansion of the old A500, which is not Linux-capable, and hasn't been powered on for 20 years). Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
On Tue, Jun 28, 2022 at 5:25 AM Michael Schmitz <schmitzmic@gmail.com> wrote: > Am 28.06.2022 um 09:12 schrieb Michael Schmitz: > > Leaving the bounce buffer handling in place, and taking a few other > liberties - this is what converting the easiest case (a3000 SCSI) might > look like. Any obvious mistakes? The mvme147 driver would be very > similar to handle (after conversion to a platform device). > > The driver allocates bounce buffers using kmalloc if it hits an > unaligned data buffer - can such buffers still even happen these days? > If I understand dma_map_single() correctly, the resulting dma handle > would be equally misaligned? > > To allocate a bounce buffer, would it be OK to use dma_alloc_coherent() > even though AFAIU memory used for DMA buffers generally isn't consistent > on m68k? I think it makes sense to skip the bounce buffering as you do here: the only standardized way we have for integrating that part is to use the swiotlb infrastructure, but as you mentioned earlier that part is probably too resource-heavy here for Amiga. I see two other problems with your patch though: a) you still duplicate the cache handling: the cache_clear()/cache_push() is supposed to already be done by dma_map_single() when the device is not cache-coherent. b) The bounce buffer is never mapped here, instead you have the virt_to_phys() here, which is not the same. I think you need to map the pointer that actually gets passed down to the device after deciding to use a bouce buffer or not. Arnd
Hi Geert, On 28/06/22 19:03, Geert Uytterhoeven wrote: > >> Leaving the bounce buffer handling in place, and taking a few other >> liberties - this is what converting the easiest case (a3000 SCSI) might >> look like. Any obvious mistakes? The mvme147 driver would be very >> similar to handle (after conversion to a platform device). > Thanks, looks reasonable. Thanks, I'll take care of Arnd's comments and post a corrected version later. >> The driver allocates bounce buffers using kmalloc if it hits an >> unaligned data buffer - can such buffers still even happen these days? > No idea. Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this code path is still being used. > >> If I understand dma_map_single() correctly, the resulting dma handle >> would be equally misaligned? >> >> To allocate a bounce buffer, would it be OK to use dma_alloc_coherent() >> even though AFAIU memory used for DMA buffers generally isn't consistent >> on m68k? >> >> Thinking ahead to the other two Amiga drivers - I wonder whether >> allocating a static bounce buffer or a DMA pool at driver init is likely >> to succeed if the kernel runs from the low 16 MB RAM chunk? It certainly >> won't succeed if the kernel runs from a higher memory address, so the >> present bounce buffer logic around amiga_chip_alloc() might still need >> to be used here. >> >> Leaves the question whether converting the gvp11 and a2091 drivers is >> actually worth it, if bounce buffers still have to be handled explicitly. > A2091 should be straight-forward, as A3000 is basically A2091 on the > motherboard (comparing the two drivers, looks like someone's been > sprinkling mb()s over the A3000 driver). Yep, and at least the ones in the dma_setup() function are there for no reason (the compiler won't reorder stores around the cache flush calls, I hope?). Just leaves the 24 bit DMA mask there (and likely need for bounce buffers). > I don't have any of these SCSI host adapters (not counting the A590 > (~A2091) expansion of the old A500, which is not Linux-capable, and > hasn't been powered on for 20 years). I wonder whether kullervo has survived - that one was an A3000. Should have gone to Adrian a few years ago... Cheers, Michael > > Gr{oetje,eeting}s, > > Geert > > -- > Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org > > In personal conversations with technical people, I call myself a hacker. But > when I'm talking to journalists I just say "programmer" or something like that. > -- Linus Torvalds
Hi Arnd, On 28/06/22 19:08, Arnd Bergmann wrote: > On Tue, Jun 28, 2022 at 5:25 AM Michael Schmitz <schmitzmic@gmail.com> wrote: >> Am 28.06.2022 um 09:12 schrieb Michael Schmitz: >> >> Leaving the bounce buffer handling in place, and taking a few other >> liberties - this is what converting the easiest case (a3000 SCSI) might >> look like. Any obvious mistakes? The mvme147 driver would be very >> similar to handle (after conversion to a platform device). >> >> The driver allocates bounce buffers using kmalloc if it hits an >> unaligned data buffer - can such buffers still even happen these days? >> If I understand dma_map_single() correctly, the resulting dma handle >> would be equally misaligned? >> >> To allocate a bounce buffer, would it be OK to use dma_alloc_coherent() >> even though AFAIU memory used for DMA buffers generally isn't consistent >> on m68k? > I think it makes sense to skip the bounce buffering as you do here: > the only standardized way we have for integrating that part is to > use the swiotlb infrastructure, but as you mentioned earlier that > part is probably too resource-heavy here for Amiga. OK, leaving the old custom logic in place allows to convert the 24 bit DMA drivers more easily. > > I see two other problems with your patch though: > > a) you still duplicate the cache handling: the cache_clear()/cache_push() > is supposed to already be done by dma_map_single() when the device > is not cache-coherent. That's one of the 'liberties' I alluded to. The reason I left these in is that I'm none too certain what device feature the DMA API uses to decide a device isn't cache-coherent. If it's dev->coherent_dma_mask, the way I set up the device in the a3000 driver should leave the coherent mask unchanged. For the Zorro drivers, devices are set up to use the same storage to store normal and coherent masks - something we most likely want to change. I need to think about the ramifications of that. Note that zorro_esp.c uses dma_sync_single_for_device() and uses a 32 bit coherent DMA mask which does work OK. I might ask Adrian to test a change to only set dev->dma_mask, and drop the dma_sync_single_for_device() calls if there's any doubt on this aspect. > b) The bounce buffer is never mapped here, instead you have the > virt_to_phys() here, which is not the same. I think you need to map > the pointer that actually gets passed down to the device after deciding > to use a bouce buffer or not. I hadn't realized that I can map the bounce buffer just as it's done for the SCp data buffer. Should have been obvious, but I'm still learning about the DMA API. I've updated the patch now, will re-send as part of a complete series once done. Cheers, Michael > > Arnd
On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz <schmitzmic@gmail.com> wrote: > On 28/06/22 19:03, Geert Uytterhoeven wrote: > >> The driver allocates bounce buffers using kmalloc if it hits an > >> unaligned data buffer - can such buffers still even happen these days? > > No idea. > Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this > code path is still being used. kmalloc() guarantees alignment to the next power-of-two size or KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it is cacheline aligned. Arnd
On Tue, Jun 28, 2022 at 11:38 PM Michael Schmitz <schmitzmic@gmail.com> wrote: > On 28/06/22 19:08, Arnd Bergmann wrote: > > I see two other problems with your patch though: > > > > a) you still duplicate the cache handling: the cache_clear()/cache_push() > > is supposed to already be done by dma_map_single() when the device > > is not cache-coherent. > > That's one of the 'liberties' I alluded to. The reason I left these in > is that I'm none too certain what device feature the DMA API uses to > decide a device isn't cache-coherent. If it's dev->coherent_dma_mask, > the way I set up the device in the a3000 driver should leave the > coherent mask unchanged. For the Zorro drivers, devices are set up to > use the same storage to store normal and coherent masks - something we > most likely want to change. I need to think about the ramifications of > that. > > Note that zorro_esp.c uses dma_sync_single_for_device() and uses a 32 > bit coherent DMA mask which does work OK. I might ask Adrian to test a > change to only set dev->dma_mask, and drop the > dma_sync_single_for_device() calls if there's any doubt on this aspect. The "coherent_mask" is independent of the cache flushing. On some architectures, a device can indicate whether it needs cache management or not to guarantee coherency, but on m68k it appears that we always assume it does, see arch/m68k/kernel/dma.c > > b) The bounce buffer is never mapped here, instead you have the > > virt_to_phys() here, which is not the same. I think you need to map > > the pointer that actually gets passed down to the device after deciding > > to use a bouce buffer or not. > > I hadn't realized that I can map the bounce buffer just as it's done for > the SCp data buffer. Should have been obvious, but I'm still learning > about the DMA API. > > I've updated the patch now, will re-send as part of a complete series > once done. I suppose you can just drop the bounce buffer if this just comes from kmalloc(). Arnd
Hi Arnd, On 29/06/22 09:50, Arnd Bergmann wrote: > On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz <schmitzmic@gmail.com> wrote: >> On 28/06/22 19:03, Geert Uytterhoeven wrote: >>>> The driver allocates bounce buffers using kmalloc if it hits an >>>> unaligned data buffer - can such buffers still even happen these days? >>> No idea. >> Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this >> code path is still being used. > kmalloc() guarantees alignment to the next power-of-two size or > KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it > is cacheline aligned. And all SCSI buffers are allocated using kmalloc? No way at all for user space to pass unaligned data? (SCSI is a weird beast - I have used a SCSI DAT tape driver many many years ago, which broke all sorts of assumptions about transfer block sizes ... but that might actually have been in the v0.99 days, many rewrites of SCSI midlevel ago). Just being cautious, as getting any of this tested will be a stretch. Cheers, Michael > > Arnd
Hi Arnd, On 29/06/22 09:55, Arnd Bergmann wrote: > On Tue, Jun 28, 2022 at 11:38 PM Michael Schmitz <schmitzmic@gmail.com> wrote: >> On 28/06/22 19:08, Arnd Bergmann wrote: >>> I see two other problems with your patch though: >>> >>> a) you still duplicate the cache handling: the cache_clear()/cache_push() >>> is supposed to already be done by dma_map_single() when the device >>> is not cache-coherent. >> That's one of the 'liberties' I alluded to. The reason I left these in >> is that I'm none too certain what device feature the DMA API uses to >> decide a device isn't cache-coherent. If it's dev->coherent_dma_mask, >> the way I set up the device in the a3000 driver should leave the >> coherent mask unchanged. For the Zorro drivers, devices are set up to >> use the same storage to store normal and coherent masks - something we >> most likely want to change. I need to think about the ramifications of >> that. >> >> Note that zorro_esp.c uses dma_sync_single_for_device() and uses a 32 >> bit coherent DMA mask which does work OK. I might ask Adrian to test a >> change to only set dev->dma_mask, and drop the >> dma_sync_single_for_device() calls if there's any doubt on this aspect. > The "coherent_mask" is independent of the cache flushing. On some > architectures, a device can indicate whether it needs cache management > or not to guarantee coherency, but on m68k it appears that we always > assume it does, see arch/m68k/kernel/dma.c Thanks - what I see there indicates that on the relevant platforms, pages mapped for DMA have their page table cache bits modified to make them non-cacheable (and I suppose unmapping restores the default cache bits). That means I should use dma_set_mask_and_coherent() here to take advantage of this, and no need to mess around with dma_sync_single_for_device() in the drivers' dma_setup() functions. >>> b) The bounce buffer is never mapped here, instead you have the >>> virt_to_phys() here, which is not the same. I think you need to map >>> the pointer that actually gets passed down to the device after deciding >>> to use a bouce buffer or not. >> I hadn't realized that I can map the bounce buffer just as it's done for >> the SCp data buffer. Should have been obvious, but I'm still learning >> about the DMA API. >> >> I've updated the patch now, will re-send as part of a complete series >> once done. > I suppose you can just drop the bounce buffer if this just comes > from kmalloc(). That's only true for a3000 and mvme147 though. Cheers, Michael > > Arnd
On 6/28/22 16:09, Michael Schmitz wrote: > On 29/06/22 09:50, Arnd Bergmann wrote: >> On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz >> <schmitzmic@gmail.com> wrote: >>> On 28/06/22 19:03, Geert Uytterhoeven wrote: >>>>> The driver allocates bounce buffers using kmalloc if it hits an >>>>> unaligned data buffer - can such buffers still even happen these days? >>>> No idea. >>> Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this >>> code path is still being used. >> kmalloc() guarantees alignment to the next power-of-two size or >> KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it >> is cacheline aligned. > > And all SCSI buffers are allocated using kmalloc? No way at all for user > space to pass unaligned data? > > (SCSI is a weird beast - I have used a SCSI DAT tape driver many many > years ago, which broke all sorts of assumptions about transfer block > sizes ... but that might actually have been in the v0.99 days, many > rewrites of SCSI midlevel ago). > > Just being cautious, as getting any of this tested will be a stretch. An example of a user space application that passes an SG I/O data buffer to the kernel that is aligned to a four byte boundary but not to an eight byte boundary if the -s (scattered) command line option is used: https://github.com/osandov/blktests/blob/master/src/discontiguous-io.cpp Bart.
Hi Bart, On 29/06/22 11:50, Bart Van Assche wrote: > On 6/28/22 16:09, Michael Schmitz wrote: >> On 29/06/22 09:50, Arnd Bergmann wrote: >>> On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz >>> <schmitzmic@gmail.com> wrote: >>>> On 28/06/22 19:03, Geert Uytterhoeven wrote: >>>>>> The driver allocates bounce buffers using kmalloc if it hits an >>>>>> unaligned data buffer - can such buffers still even happen these >>>>>> days? >>>>> No idea. >>>> Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether >>>> this >>>> code path is still being used. >>> kmalloc() guarantees alignment to the next power-of-two size or >>> KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it >>> is cacheline aligned. >> >> And all SCSI buffers are allocated using kmalloc? No way at all for >> user space to pass unaligned data? >> >> (SCSI is a weird beast - I have used a SCSI DAT tape driver many many >> years ago, which broke all sorts of assumptions about transfer block >> sizes ... but that might actually have been in the v0.99 days, many >> rewrites of SCSI midlevel ago). >> >> Just being cautious, as getting any of this tested will be a stretch. > > An example of a user space application that passes an SG I/O data > buffer to the kernel that is aligned to a four byte boundary but not > to an eight byte boundary if the -s (scattered) command line option is > used: > https://github.com/osandov/blktests/blob/master/src/discontiguous-io.cpp Thanks - four byte alignment actually wouldn't be an issue for me. It's two byte or smaller that would trip up the SCSI DMA. While I'm sure such an even more pathological test case could be written, I was rather worried about st.c and sr.c input ... Cheers, Michael > > Bart.
Hi Bart, On 29/06/22 12:01, Michael Schmitz wrote: > >> An example of a user space application that passes an SG I/O data >> buffer to the kernel that is aligned to a four byte boundary but not >> to an eight byte boundary if the -s (scattered) command line option >> is used: >> https://github.com/osandov/blktests/blob/master/src/discontiguous-io.cpp > > Thanks - four byte alignment actually wouldn't be an issue for me. > It's two byte or smaller that would trip up the SCSI DMA. > > While I'm sure such an even more pathological test case could be > written, I was rather worried about st.c and sr.c input ... Nevermind - I just see m68k defines ARCH_DMA_MINALIGN to be four bytes. Should be safe for all that matters, then. Cheers, Michael
On Wed, Jun 29, 2022 at 11:09:00AM +1200, Michael Schmitz wrote: > And all SCSI buffers are allocated using kmalloc? No way at all for user > space to pass unaligned data? Most that you will see actually comes from the page allocator. But the block layer has a dma_alignment limit, and when userspace sends I/O that is not properly aligned it will be bounce buffered before it it sent to the driver.
On Wed, Jun 29, 2022 at 09:38:00AM +1200, Michael Schmitz wrote: > That's one of the 'liberties' I alluded to. The reason I left these in is > that I'm none too certain what device feature the DMA API uses to decide a > device isn't cache-coherent. The DMA API does not look at device features at all. It needs to be told so by the platform code. Once an architecture implements the hooks to support non-coherent DMA all devices are treated as non-coherent by default unless overriden by the architecture either globally (using the global dma_default_coherent variable) or per-device (using the dev->dma_coherent field, usually set by arch_setup_dma_ops). > If it's dev->coherent_dma_mask, the way I set > up the device in the a3000 driver should leave the coherent mask unchanged. > For the Zorro drivers, devices are set up to use the same storage to store > normal and coherent masks - something we most likely want to change. I need > to think about the ramifications of that. No, the coherent mask is slightly misnamed amd not actually related.
From: Michael Schmitz > Sent: 29 June 2022 00:09 > > Hi Arnd, > > On 29/06/22 09:50, Arnd Bergmann wrote: > > On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz <schmitzmic@gmail.com> wrote: > >> On 28/06/22 19:03, Geert Uytterhoeven wrote: > >>>> The driver allocates bounce buffers using kmalloc if it hits an > >>>> unaligned data buffer - can such buffers still even happen these days? > >>> No idea. > >> Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this > >> code path is still being used. > > kmalloc() guarantees alignment to the next power-of-two size or > > KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it > > is cacheline aligned. > > And all SCSI buffers are allocated using kmalloc? No way at all for user > space to pass unaligned data? I didn't think kmalloc() gave any such guarantee about alignment. There are cache-line alignment requirements on systems with non-coherent dma, but otherwise the alignment can be much smaller. One of the allocators adds a header to each item, IIRC that can lead to 'unexpected' alignments - especially on m68k. dma_alloc_coherent() does align to next 'power of 2'. And sometimes you need (eg) 16k allocates that are 16k aligned. David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Le 30/06/2022 à 10:04, David Laight a écrit : > From: Michael Schmitz >> Sent: 29 June 2022 00:09 >> >> Hi Arnd, >> >> On 29/06/22 09:50, Arnd Bergmann wrote: >>> On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz <schmitzmic@gmail.com> wrote: >>>> On 28/06/22 19:03, Geert Uytterhoeven wrote: >>>>>> The driver allocates bounce buffers using kmalloc if it hits an >>>>>> unaligned data buffer - can such buffers still even happen these days? >>>>> No idea. >>>> Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this >>>> code path is still being used. >>> kmalloc() guarantees alignment to the next power-of-two size or >>> KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it >>> is cacheline aligned. >> >> And all SCSI buffers are allocated using kmalloc? No way at all for user >> space to pass unaligned data? > > I didn't think kmalloc() gave any such guarantee about alignment. I does since commit 59bb47985c1d ("mm, sl[aou]b: guarantee natural alignment for kmalloc(power-of-two)") Christophe > There are cache-line alignment requirements on systems with non-coherent > dma, but otherwise the alignment can be much smaller. > > One of the allocators adds a header to each item, IIRC that can > lead to 'unexpected' alignments - especially on m68k. > > dma_alloc_coherent() does align to next 'power of 2'. > And sometimes you need (eg) 16k allocates that are 16k aligned. > > David > > - > Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK > Registration No: 1397386 (Wales)
From: Christophe Leroy > Sent: 30 June 2022 10:40 > > Le 30/06/2022 à 10:04, David Laight a écrit : > > From: Michael Schmitz > >> Sent: 29 June 2022 00:09 > >> > >> Hi Arnd, > >> > >> On 29/06/22 09:50, Arnd Bergmann wrote: > >>> On Tue, Jun 28, 2022 at 11:03 PM Michael Schmitz <schmitzmic@gmail.com> wrote: > >>>> On 28/06/22 19:03, Geert Uytterhoeven wrote: > >>>>>> The driver allocates bounce buffers using kmalloc if it hits an > >>>>>> unaligned data buffer - can such buffers still even happen these days? > >>>>> No idea. > >>>> Hmmm - I think I'll stick a WARN_ONCE() in there so we know whether this > >>>> code path is still being used. > >>> kmalloc() guarantees alignment to the next power-of-two size or > >>> KMALLOC_MIN_ALIGN, whichever is bigger. On m68k this means it > >>> is cacheline aligned. > >> > >> And all SCSI buffers are allocated using kmalloc? No way at all for user > >> space to pass unaligned data? > > > > I didn't think kmalloc() gave any such guarantee about alignment. > > I does since commit 59bb47985c1d ("mm, sl[aou]b: guarantee natural > alignment for kmalloc(power-of-two)") Looks like it is done for 'power-of-two' less than PAGE_SIZE. This may not help scsi tape writes which could easily be (say) 47 bytes. I think that only guarantees 2 byte alignment on m68k. (Although increasing the min-alignment on m68k to 4 (or even 8) will probably make no measurable difference.) What happens above PAGE_SIZE? Any structure with a trailing [] field could easily request '64k + a_bit' bytes. You don't really want to extend this to 128k - but I suspect that is what happens. David > > Christophe > > > There are cache-line alignment requirements on systems with non-coherent > > dma, but otherwise the alignment can be much smaller. > > > > One of the allocators adds a header to each item, IIRC that can > > lead to 'unexpected' alignments - especially on m68k. > > > > dma_alloc_coherent() does align to next 'power of 2'. > > And sometimes you need (eg) 16k allocates that are 16k aligned. > > > > David > > > > - > > Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK > > Registration No: 1397386 (Wales) - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Hi Christoph, On 29/06/22 18:21, Christoph Hellwig wrote: > On Wed, Jun 29, 2022 at 11:09:00AM +1200, Michael Schmitz wrote: >> And all SCSI buffers are allocated using kmalloc? No way at all for user >> space to pass unaligned data? > Most that you will see actually comes from the page allocator. But > the block layer has a dma_alignment limit, and when userspace sends > I/O that is not properly aligned it will be bounce buffered before > it it sent to the driver. That limit is set to L1_CACHE_BYTES on m68k so we're good here. Thanks, Michael
Hi Christoph, On 29/06/22 18:25, Christoph Hellwig wrote: > On Wed, Jun 29, 2022 at 09:38:00AM +1200, Michael Schmitz wrote: >> That's one of the 'liberties' I alluded to. The reason I left these in is >> that I'm none too certain what device feature the DMA API uses to decide a >> device isn't cache-coherent. > The DMA API does not look at device features at all. It needs to be > told so by the platform code. Once an architecture implements the > hooks to support non-coherent DMA all devices are treated as > non-coherent by default unless overriden by the architecture either > globally (using the global dma_default_coherent variable) or per-device > (using the dev->dma_coherent field, usually set by arch_setup_dma_ops). Haven't got any of that, so non-coherent DMA is all we can use (even though some of the RAM used for bounce buffers may actually be coherent due to the page table cache bits). > >> If it's dev->coherent_dma_mask, the way I set >> up the device in the a3000 driver should leave the coherent mask unchanged. >> For the Zorro drivers, devices are set up to use the same storage to store >> normal and coherent masks - something we most likely want to change. I need >> to think about the ramifications of that. > No, the coherent mask is slightly misnamed amd not actually related. Thanks, that had me confused. Cheers, Michael
diff --git a/Documentation/core-api/bus-virt-phys-mapping.rst b/Documentation/core-api/bus-virt-phys-mapping.rst deleted file mode 100644 index c72b24a7d52c..000000000000 --- a/Documentation/core-api/bus-virt-phys-mapping.rst +++ /dev/null @@ -1,220 +0,0 @@ -========================================================== -How to access I/O mapped memory from within device drivers -========================================================== - -:Author: Linus - -.. warning:: - - The virt_to_bus() and bus_to_virt() functions have been - superseded by the functionality provided by the PCI DMA interface - (see Documentation/core-api/dma-api-howto.rst). They continue - to be documented below for historical purposes, but new code - must not use them. --davidm 00/12/12 - -:: - - [ This is a mail message in response to a query on IO mapping, thus the - strange format for a "document" ] - -The AHA-1542 is a bus-master device, and your patch makes the driver give the -controller the physical address of the buffers, which is correct on x86 -(because all bus master devices see the physical memory mappings directly). - -However, on many setups, there are actually **three** different ways of looking -at memory addresses, and in this case we actually want the third, the -so-called "bus address". - -Essentially, the three ways of addressing memory are (this is "real memory", -that is, normal RAM--see later about other details): - - - CPU untranslated. This is the "physical" address. Physical address - 0 is what the CPU sees when it drives zeroes on the memory bus. - - - CPU translated address. This is the "virtual" address, and is - completely internal to the CPU itself with the CPU doing the appropriate - translations into "CPU untranslated". - - - bus address. This is the address of memory as seen by OTHER devices, - not the CPU. Now, in theory there could be many different bus - addresses, with each device seeing memory in some device-specific way, but - happily most hardware designers aren't actually actively trying to make - things any more complex than necessary, so you can assume that all - external hardware sees the memory the same way. - -Now, on normal PCs the bus address is exactly the same as the physical -address, and things are very simple indeed. However, they are that simple -because the memory and the devices share the same address space, and that is -not generally necessarily true on other PCI/ISA setups. - -Now, just as an example, on the PReP (PowerPC Reference Platform), the -CPU sees a memory map something like this (this is from memory):: - - 0-2 GB "real memory" - 2 GB-3 GB "system IO" (inb/out and similar accesses on x86) - 3 GB-4 GB "IO memory" (shared memory over the IO bus) - -Now, that looks simple enough. However, when you look at the same thing from -the viewpoint of the devices, you have the reverse, and the physical memory -address 0 actually shows up as address 2 GB for any IO master. - -So when the CPU wants any bus master to write to physical memory 0, it -has to give the master address 0x80000000 as the memory address. - -So, for example, depending on how the kernel is actually mapped on the -PPC, you can end up with a setup like this:: - - physical address: 0 - virtual address: 0xC0000000 - bus address: 0x80000000 - -where all the addresses actually point to the same thing. It's just seen -through different translations.. - -Similarly, on the Alpha, the normal translation is:: - - physical address: 0 - virtual address: 0xfffffc0000000000 - bus address: 0x40000000 - -(but there are also Alphas where the physical address and the bus address -are the same). - -Anyway, the way to look up all these translations, you do:: - - #include <asm/io.h> - - phys_addr = virt_to_phys(virt_addr); - virt_addr = phys_to_virt(phys_addr); - bus_addr = virt_to_bus(virt_addr); - virt_addr = bus_to_virt(bus_addr); - -Now, when do you need these? - -You want the **virtual** address when you are actually going to access that -pointer from the kernel. So you can have something like this:: - - /* - * this is the hardware "mailbox" we use to communicate with - * the controller. The controller sees this directly. - */ - struct mailbox { - __u32 status; - __u32 bufstart; - __u32 buflen; - .. - } mbox; - - unsigned char * retbuffer; - - /* get the address from the controller */ - retbuffer = bus_to_virt(mbox.bufstart); - switch (retbuffer[0]) { - case STATUS_OK: - ... - -on the other hand, you want the bus address when you have a buffer that -you want to give to the controller:: - - /* ask the controller to read the sense status into "sense_buffer" */ - mbox.bufstart = virt_to_bus(&sense_buffer); - mbox.buflen = sizeof(sense_buffer); - mbox.status = 0; - notify_controller(&mbox); - -And you generally **never** want to use the physical address, because you can't -use that from the CPU (the CPU only uses translated virtual addresses), and -you can't use it from the bus master. - -So why do we care about the physical address at all? We do need the physical -address in some cases, it's just not very often in normal code. The physical -address is needed if you use memory mappings, for example, because the -"remap_pfn_range()" mm function wants the physical address of the memory to -be remapped as measured in units of pages, a.k.a. the pfn (the memory -management layer doesn't know about devices outside the CPU, so it -shouldn't need to know about "bus addresses" etc). - -.. note:: - - The above is only one part of the whole equation. The above - only talks about "real memory", that is, CPU memory (RAM). - -There is a completely different type of memory too, and that's the "shared -memory" on the PCI or ISA bus. That's generally not RAM (although in the case -of a video graphics card it can be normal DRAM that is just used for a frame -buffer), but can be things like a packet buffer in a network card etc. - -This memory is called "PCI memory" or "shared memory" or "IO memory" or -whatever, and there is only one way to access it: the readb/writeb and -related functions. You should never take the address of such memory, because -there is really nothing you can do with such an address: it's not -conceptually in the same memory space as "real memory" at all, so you cannot -just dereference a pointer. (Sadly, on x86 it **is** in the same memory space, -so on x86 it actually works to just deference a pointer, but it's not -portable). - -For such memory, you can do things like: - - - reading:: - - /* - * read first 32 bits from ISA memory at 0xC0000, aka - * C000:0000 in DOS terms - */ - unsigned int signature = isa_readl(0xC0000); - - - remapping and writing:: - - /* - * remap framebuffer PCI memory area at 0xFC000000, - * size 1MB, so that we can access it: We can directly - * access only the 640k-1MB area, so anything else - * has to be remapped. - */ - void __iomem *baseptr = ioremap(0xFC000000, 1024*1024); - - /* write a 'A' to the offset 10 of the area */ - writeb('A',baseptr+10); - - /* unmap when we unload the driver */ - iounmap(baseptr); - - - copying and clearing:: - - /* get the 6-byte Ethernet address at ISA address E000:0040 */ - memcpy_fromio(kernel_buffer, 0xE0040, 6); - /* write a packet to the driver */ - memcpy_toio(0xE1000, skb->data, skb->len); - /* clear the frame buffer */ - memset_io(0xA0000, 0, 0x10000); - -OK, that just about covers the basics of accessing IO portably. Questions? -Comments? You may think that all the above is overly complex, but one day you -might find yourself with a 500 MHz Alpha in front of you, and then you'll be -happy that your driver works ;) - -Note that kernel versions 2.0.x (and earlier) mistakenly called the -ioremap() function "vremap()". ioremap() is the proper name, but I -didn't think straight when I wrote it originally. People who have to -support both can do something like:: - - /* support old naming silliness */ - #if LINUX_VERSION_CODE < 0x020100 - #define ioremap vremap - #define iounmap vfree - #endif - -at the top of their source files, and then they can use the right names -even on 2.0.x systems. - -And the above sounds worse than it really is. Most real drivers really -don't do all that complex things (or rather: the complexity is not so -much in the actual IO accesses as in error handling and timeouts etc). -It's generally not hard to fix drivers, and in many cases the code -actually looks better afterwards:: - - unsigned long signature = *(unsigned int *) 0xC0000; - vs - unsigned long signature = readl(0xC0000); - -I think the second version actually is more readable, no? diff --git a/Documentation/core-api/dma-api-howto.rst b/Documentation/core-api/dma-api-howto.rst index 358d495456d1..828846804e25 100644 --- a/Documentation/core-api/dma-api-howto.rst +++ b/Documentation/core-api/dma-api-howto.rst @@ -707,20 +707,6 @@ to use the dma_sync_*() interfaces:: } } -Drivers converted fully to this interface should not use virt_to_bus() any -longer, nor should they use bus_to_virt(). Some drivers have to be changed a -little bit, because there is no longer an equivalent to bus_to_virt() in the -dynamic DMA mapping scheme - you have to always store the DMA addresses -returned by the dma_alloc_coherent(), dma_pool_alloc(), and dma_map_single() -calls (dma_map_sg() stores them in the scatterlist itself if the platform -supports dynamic DMA mapping in hardware) in your driver structures and/or -in the card registers. - -All drivers should be using these interfaces with no exceptions. It -is planned to completely remove virt_to_bus() and bus_to_virt() as -they are entirely deprecated. Some ports already do not provide these -as it is impossible to correctly support them. - Handling Errors =============== diff --git a/Documentation/core-api/index.rst b/Documentation/core-api/index.rst index 9ef37e985a40..0550282cfd6f 100644 --- a/Documentation/core-api/index.rst +++ b/Documentation/core-api/index.rst @@ -42,7 +42,6 @@ Library functionality that is used throughout the kernel. rbtree generic-radix-tree packing - bus-virt-phys-mapping this_cpu_ops timekeeping errseq diff --git a/Documentation/translations/zh_CN/core-api/index.rst b/Documentation/translations/zh_CN/core-api/index.rst index 26d9913fc8b6..c52175fc1b61 100644 --- a/Documentation/translations/zh_CN/core-api/index.rst +++ b/Documentation/translations/zh_CN/core-api/index.rst @@ -52,7 +52,6 @@ Todolist: circular-buffers generic-radix-tree packing - bus-virt-phys-mapping this_cpu_ops timekeeping errseq diff --git a/arch/alpha/Kconfig b/arch/alpha/Kconfig index 7d0d26b5b3f5..97fce7386b00 100644 --- a/arch/alpha/Kconfig +++ b/arch/alpha/Kconfig @@ -17,7 +17,6 @@ config ALPHA select HAVE_PERF_EVENTS select NEED_DMA_MAP_STATE select NEED_SG_DMA_LENGTH - select VIRT_TO_BUS select GENERIC_IRQ_PROBE select GENERIC_PCI_IOMAP select AUTO_IRQ_AFFINITY if SMP diff --git a/arch/alpha/include/asm/floppy.h b/arch/alpha/include/asm/floppy.h index 588758685439..64b42d9591fc 100644 --- a/arch/alpha/include/asm/floppy.h +++ b/arch/alpha/include/asm/floppy.h @@ -20,7 +20,7 @@ #define fd_free_dma() free_dma(FLOPPY_DMA) #define fd_clear_dma_ff() clear_dma_ff(FLOPPY_DMA) #define fd_set_dma_mode(mode) set_dma_mode(FLOPPY_DMA,mode) -#define fd_set_dma_addr(addr) set_dma_addr(FLOPPY_DMA,virt_to_bus(addr)) +#define fd_set_dma_addr(addr) set_dma_addr(FLOPPY_DMA,isa_virt_to_bus(addr)) #define fd_set_dma_count(count) set_dma_count(FLOPPY_DMA,count) #define fd_enable_irq() enable_irq(FLOPPY_IRQ) #define fd_disable_irq() disable_irq(FLOPPY_IRQ) diff --git a/arch/alpha/include/asm/io.h b/arch/alpha/include/asm/io.h index c9cb554fbe54..d277189b2677 100644 --- a/arch/alpha/include/asm/io.h +++ b/arch/alpha/include/asm/io.h @@ -106,15 +106,15 @@ static inline void * phys_to_virt(unsigned long address) extern unsigned long __direct_map_base; extern unsigned long __direct_map_size; -static inline unsigned long __deprecated virt_to_bus(volatile void *address) +static inline unsigned long __deprecated isa_virt_to_bus(volatile void *address) { unsigned long phys = virt_to_phys(address); unsigned long bus = phys + __direct_map_base; return phys <= __direct_map_size ? bus : 0; } -#define isa_virt_to_bus virt_to_bus +#define isa_virt_to_bus isa_virt_to_bus -static inline void * __deprecated bus_to_virt(unsigned long address) +static inline void * __deprecated isa_bus_to_virt(unsigned long address) { void *virt; @@ -125,7 +125,7 @@ static inline void * __deprecated bus_to_virt(unsigned long address) virt = phys_to_virt(address); return (long)address <= 0 ? NULL : virt; } -#define isa_bus_to_virt bus_to_virt +#define isa_bus_to_virt isa_bus_to_virt /* * There are different chipsets to interface the Alpha CPUs to the world. diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig index cb93769a9f2a..26ac8ea15a9e 100644 --- a/arch/ia64/Kconfig +++ b/arch/ia64/Kconfig @@ -39,7 +39,6 @@ config IA64 select HAVE_FUNCTION_DESCRIPTORS select HAVE_VIRT_CPU_ACCOUNTING select HUGETLB_PAGE_SIZE_VARIABLE if HUGETLB_PAGE - select VIRT_TO_BUS select GENERIC_IRQ_PROBE select GENERIC_PENDING_IRQ if SMP select GENERIC_IRQ_SHOW diff --git a/arch/ia64/include/asm/io.h b/arch/ia64/include/asm/io.h index 6d93b923b379..ce66dfc0e719 100644 --- a/arch/ia64/include/asm/io.h +++ b/arch/ia64/include/asm/io.h @@ -96,14 +96,6 @@ extern u64 kern_mem_attribute (unsigned long phys_addr, unsigned long size); extern int valid_phys_addr_range (phys_addr_t addr, size_t count); /* efi.c */ extern int valid_mmap_phys_addr_range (unsigned long pfn, size_t count); -/* - * The following two macros are deprecated and scheduled for removal. - * Please use the PCI-DMA interface defined in <asm/pci.h> instead. - */ -#define bus_to_virt phys_to_virt -#define virt_to_bus virt_to_phys -#define page_to_bus page_to_phys - # endif /* KERNEL */ /* diff --git a/arch/m68k/Kconfig b/arch/m68k/Kconfig index 936cce42ae9a..b06faf6c0b27 100644 --- a/arch/m68k/Kconfig +++ b/arch/m68k/Kconfig @@ -30,7 +30,6 @@ config M68K select OLD_SIGACTION select OLD_SIGSUSPEND3 select UACCESS_MEMCPY if !MMU - select VIRT_TO_BUS select ZONE_DMA config CPU_BIG_ENDIAN diff --git a/arch/m68k/include/asm/virtconvert.h b/arch/m68k/include/asm/virtconvert.h index ca91b32dc6ef..0a27905b0036 100644 --- a/arch/m68k/include/asm/virtconvert.h +++ b/arch/m68k/include/asm/virtconvert.h @@ -33,9 +33,11 @@ static inline void *phys_to_virt(unsigned long address) /* * IO bus memory addresses are 1:1 with the physical address, + * deprecated globally but still used on two machines. */ +#if defined(CONFIG_AMIGA) || defined(CONFIG_VME) #define virt_to_bus virt_to_phys -#define bus_to_virt phys_to_virt +#endif #endif #endif diff --git a/arch/microblaze/Kconfig b/arch/microblaze/Kconfig index 8cf429ad1c84..415182eeb082 100644 --- a/arch/microblaze/Kconfig +++ b/arch/microblaze/Kconfig @@ -38,7 +38,6 @@ config MICROBLAZE select OF_EARLY_FLATTREE select PCI_DOMAINS_GENERIC if PCI select PCI_SYSCALL if PCI - select VIRT_TO_BUS select CPU_NO_EFFICIENT_FFS select MMU_GATHER_NO_RANGE select SPARSE_IRQ diff --git a/arch/microblaze/include/asm/io.h b/arch/microblaze/include/asm/io.h index b6a57f8468f0..c1d78b8977a6 100644 --- a/arch/microblaze/include/asm/io.h +++ b/arch/microblaze/include/asm/io.h @@ -30,8 +30,6 @@ extern resource_size_t isa_mem_base; #define PCI_IOBASE ((void __iomem *)_IO_BASE) #define IO_SPACE_LIMIT (0xFFFFFFFF) -#define page_to_bus(page) (page_to_phys(page)) - extern void iounmap(volatile void __iomem *addr); extern void __iomem *ioremap(phys_addr_t address, unsigned long size); diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig index 9457894db237..69f3b9b4677b 100644 --- a/arch/mips/Kconfig +++ b/arch/mips/Kconfig @@ -100,7 +100,6 @@ config MIPS select RTC_LIB select SYSCTL_EXCEPTION_TRACE select TRACE_IRQFLAGS_SUPPORT - select VIRT_TO_BUS select ARCH_HAS_ELFCORE_COMPAT select HAVE_ARCH_KCSAN if 64BIT diff --git a/arch/mips/include/asm/io.h b/arch/mips/include/asm/io.h index 6f5c86d2bab4..cd9168f34fb7 100644 --- a/arch/mips/include/asm/io.h +++ b/arch/mips/include/asm/io.h @@ -147,15 +147,6 @@ static inline void *isa_bus_to_virt(unsigned long address) return phys_to_virt(address); } -/* - * However PCI ones are not necessarily 1:1 and therefore these interfaces - * are forbidden in portable PCI drivers. - * - * Allow them for x86 for legacy drivers, though. - */ -#define virt_to_bus virt_to_phys -#define bus_to_virt phys_to_virt - /* * Change "struct page" to physical address. */ diff --git a/arch/parisc/Kconfig b/arch/parisc/Kconfig index 5f2448dc5a2b..b0d68e9e2df0 100644 --- a/arch/parisc/Kconfig +++ b/arch/parisc/Kconfig @@ -43,7 +43,6 @@ config PARISC select SYSCTL_ARCH_UNALIGN_ALLOW select SYSCTL_EXCEPTION_TRACE select HAVE_MOD_ARCH_SPECIFIC - select VIRT_TO_BUS select MODULES_USE_ELF_RELA select CLONE_BACKWARDS select TTY # Needed for pdc_cons.c diff --git a/arch/parisc/include/asm/floppy.h b/arch/parisc/include/asm/floppy.h index 762cfe7778c0..b318a7df52f6 100644 --- a/arch/parisc/include/asm/floppy.h +++ b/arch/parisc/include/asm/floppy.h @@ -179,7 +179,7 @@ static void _fd_chose_dma_mode(char *addr, unsigned long size) { if(can_use_virtual_dma == 2) { if((unsigned int) addr >= (unsigned int) high_memory || - virt_to_bus(addr) >= 0x1000000 || + virt_to_phys(addr) >= 0x1000000 || _CROSS_64KB(addr, size, 0)) use_virtual_dma = 1; else @@ -215,7 +215,7 @@ static int hard_dma_setup(char *addr, unsigned long size, int mode, int io) doing_pdma = 0; clear_dma_ff(FLOPPY_DMA); set_dma_mode(FLOPPY_DMA,mode); - set_dma_addr(FLOPPY_DMA,virt_to_bus(addr)); + set_dma_addr(FLOPPY_DMA,virt_to_phys(addr)); set_dma_count(FLOPPY_DMA,size); enable_dma(FLOPPY_DMA); return 0; diff --git a/arch/parisc/include/asm/io.h b/arch/parisc/include/asm/io.h index 837ddddbac6a..42ffb60a6ea9 100644 --- a/arch/parisc/include/asm/io.h +++ b/arch/parisc/include/asm/io.h @@ -7,8 +7,6 @@ #define virt_to_phys(a) ((unsigned long)__pa(a)) #define phys_to_virt(a) __va(a) -#define virt_to_bus virt_to_phys -#define bus_to_virt phys_to_virt static inline unsigned long isa_bus_to_virt(unsigned long addr) { BUG(); diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index 8f76499919fe..66418dfeb771 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig @@ -277,7 +277,6 @@ config PPC select SYSCTL_EXCEPTION_TRACE select THREAD_INFO_IN_TASK select TRACE_IRQFLAGS_SUPPORT - select VIRT_TO_BUS if !PPC64 # # Please keep this list sorted alphabetically. # diff --git a/arch/powerpc/include/asm/io.h b/arch/powerpc/include/asm/io.h index c5a5f7c9b231..73fcd5cdb662 100644 --- a/arch/powerpc/include/asm/io.h +++ b/arch/powerpc/include/asm/io.h @@ -985,8 +985,6 @@ static inline void * bus_to_virt(unsigned long address) } #define bus_to_virt bus_to_virt -#define page_to_bus(page) (page_to_phys(page) + PCI_DRAM_OFFSET) - #endif /* CONFIG_PPC32 */ /* access ports */ diff --git a/arch/riscv/include/asm/page.h b/arch/riscv/include/asm/page.h index 1526e410e802..ac70b0fd9a9a 100644 --- a/arch/riscv/include/asm/page.h +++ b/arch/riscv/include/asm/page.h @@ -167,7 +167,6 @@ extern phys_addr_t __phys_addr_symbol(unsigned long x); #define page_to_virt(page) (pfn_to_virt(page_to_pfn(page))) #define page_to_phys(page) (pfn_to_phys(page_to_pfn(page))) -#define page_to_bus(page) (page_to_phys(page)) #define phys_to_page(paddr) (pfn_to_page(phys_to_pfn(paddr))) #define sym_to_pfn(x) __phys_to_pfn(__pa_symbol(x)) diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index abe751626a35..9c50d53fa6e7 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -279,7 +279,6 @@ config X86 select THREAD_INFO_IN_TASK select TRACE_IRQFLAGS_SUPPORT select USER_STACKTRACE_SUPPORT - select VIRT_TO_BUS select HAVE_ARCH_KCSAN if X86_64 select X86_FEATURE_NAMES if PROC_FS select PROC_PID_ARCH_STATUS if PROC_FS diff --git a/arch/x86/include/asm/io.h b/arch/x86/include/asm/io.h index 1870b99c3356..e9025640f634 100644 --- a/arch/x86/include/asm/io.h +++ b/arch/x86/include/asm/io.h @@ -169,15 +169,6 @@ static inline unsigned int isa_virt_to_bus(volatile void *address) } #define isa_bus_to_virt phys_to_virt -/* - * However PCI ones are not necessarily 1:1 and therefore these interfaces - * are forbidden in portable PCI drivers. - * - * Allow them on x86 for legacy drivers, though. - */ -#define virt_to_bus virt_to_phys -#define bus_to_virt phys_to_virt - /* * The default ioremap() behavior is non-cached; if you need something * else, you probably want one of the following. diff --git a/arch/xtensa/Kconfig b/arch/xtensa/Kconfig index 0b0f0172cced..92a24ed738a5 100644 --- a/arch/xtensa/Kconfig +++ b/arch/xtensa/Kconfig @@ -50,7 +50,6 @@ config XTENSA select MODULES_USE_ELF_RELA select PERF_USE_VMALLOC select TRACE_IRQFLAGS_SUPPORT - select VIRT_TO_BUS help Xtensa processors are 32-bit RISC machines designed by Tensilica primarily for embedded systems. These processors are both diff --git a/arch/xtensa/include/asm/io.h b/arch/xtensa/include/asm/io.h index 54188e69b988..a5b707e1c0f4 100644 --- a/arch/xtensa/include/asm/io.h +++ b/arch/xtensa/include/asm/io.h @@ -63,9 +63,6 @@ static inline void iounmap(volatile void __iomem *addr) xtensa_iounmap(addr); } -#define virt_to_bus virt_to_phys -#define bus_to_virt phys_to_virt - #endif /* CONFIG_MMU */ #include <asm-generic/io.h> diff --git a/include/asm-generic/io.h b/include/asm-generic/io.h index 7ce93aaf69f8..f57015eaed73 100644 --- a/include/asm-generic/io.h +++ b/include/asm-generic/io.h @@ -1059,20 +1059,6 @@ static inline void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr) } #endif -#ifdef CONFIG_VIRT_TO_BUS -#ifndef virt_to_bus -static inline unsigned long virt_to_bus(void *address) -{ - return (unsigned long)address; -} - -static inline void *bus_to_virt(unsigned long address) -{ - return (void *)address; -} -#endif -#endif - #ifndef memset_io #define memset_io memset_io /** diff --git a/mm/Kconfig b/mm/Kconfig index 169e64192e48..b7a44b17c79f 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -639,14 +639,6 @@ config BOUNCE memory available to the CPU. Enabled by default when HIGHMEM is selected, but you may say n to override this. -config VIRT_TO_BUS - bool - help - An architecture should select this if it implements the - deprecated interface virt_to_bus(). All new architectures - should probably not select this. - - config MMU_NOTIFIER bool select SRCU