hw/ppc/mac.h | 2 -- hw/ppc/mac_newworld.c | 22 ++++++++------ hw/ppc/mac_oldworld.c | 67 ++++++++++++++++++++++++------------------- 3 files changed, 52 insertions(+), 39 deletions(-)
This is the cut down version of the earlier series omitting unfinished patches that I plan to rework later and rebased to Mark's qemu-macppc branch. Compared to v7 the only change is the cast to (target_ulong) from (uint32_t) as requested by Mark in patch 1. Regards, BALATON Zoltan BALATON Zoltan (5): mac_oldworld: Allow loading binary ROM image mac_newworld: Allow loading binary ROM image mac_oldworld: Drop a variable, use get_system_memory() directly mac_oldworld: Drop some variables mac_oldworld: Change PCI address of macio to match real hardware hw/ppc/mac.h | 2 -- hw/ppc/mac_newworld.c | 22 ++++++++------ hw/ppc/mac_oldworld.c | 67 ++++++++++++++++++++++++------------------- 3 files changed, 52 insertions(+), 39 deletions(-) -- 2.21.3
On 16/10/2020 00:47, BALATON Zoltan via wrote: > This is the cut down version of the earlier series omitting unfinished > patches that I plan to rework later and rebased to Mark's qemu-macppc > branch. Compared to v7 the only change is the cast to (target_ulong) > from (uint32_t) as requested by Mark in patch 1. FWIW the reason for suggesting the cast to target_ulong is so that the same code works for both qemu-system-ppc and qemu-system-ppc64. For qemu-system-ppc that should correctly drop the sign extension from 32-bit, whilst still allowing someone to load a 64-bit ELF into qemu-system-ppc64 if requested. Can you confirm that the sign extension behaviour is still correct for both qemu-system-ppc and qemu-system-ppc64? If so I'm happy to give it a R-B tag. ATB, Mark.
On Fri, 16 Oct 2020, Mark Cave-Ayland wrote: > On 16/10/2020 00:47, BALATON Zoltan via wrote: >> This is the cut down version of the earlier series omitting unfinished >> patches that I plan to rework later and rebased to Mark's qemu-macppc >> branch. Compared to v7 the only change is the cast to (target_ulong) >> from (uint32_t) as requested by Mark in patch 1. > > FWIW the reason for suggesting the cast to target_ulong is so that the same > code works for both qemu-system-ppc and qemu-system-ppc64. For > qemu-system-ppc that should correctly drop the sign extension from 32-bit, > whilst still allowing someone to load a 64-bit ELF into qemu-system-ppc64 if > requested. > > Can you confirm that the sign extension behaviour is still correct for both > qemu-system-ppc and qemu-system-ppc64? If so I'm happy to give it a R-B tag. I've tried it now again with both ppc and ppc64: both OpenBIOS and a G3 beige ROM can be loaded with qemu-system-ppc but qemu-system-ppc64 fails with OpenBIOS when casting to target_ulong (i think because target_ulong is 64 bit there but g3beige is still 32 bit but I haven't throughly debugged it). But everything works with my original uint32_t cast, so ditch it and use my original version. Should I resubmit or you can fix up? (I think I wait until it's clear if this will be taken by David or you and send a fixed version cc-ing David if this is decided to go through the PPC queue.) Regards, BALATON Zoltan
On 16/10/2020 13:19, BALATON Zoltan via wrote: > On Fri, 16 Oct 2020, Mark Cave-Ayland wrote: >> On 16/10/2020 00:47, BALATON Zoltan via wrote: >>> This is the cut down version of the earlier series omitting unfinished >>> patches that I plan to rework later and rebased to Mark's qemu-macppc >>> branch. Compared to v7 the only change is the cast to (target_ulong) >>> from (uint32_t) as requested by Mark in patch 1. >> >> FWIW the reason for suggesting the cast to target_ulong is so that the same code >> works for both qemu-system-ppc and qemu-system-ppc64. For qemu-system-ppc that >> should correctly drop the sign extension from 32-bit, whilst still allowing someone >> to load a 64-bit ELF into qemu-system-ppc64 if requested. >> >> Can you confirm that the sign extension behaviour is still correct for both >> qemu-system-ppc and qemu-system-ppc64? If so I'm happy to give it a R-B tag. > > I've tried it now again with both ppc and ppc64: both OpenBIOS and a G3 beige ROM can > be loaded with qemu-system-ppc but qemu-system-ppc64 fails with OpenBIOS when casting > to target_ulong (i think because target_ulong is 64 bit there but g3beige is still 32 > bit but I haven't throughly debugged it). But everything works with my original > uint32_t cast, so ditch it and use my original version. Should I resubmit or you can > fix up? (I think I wait until it's clear if this will be taken by David or you and > send a fixed version cc-ing David if this is decided to go through the PPC queue.) Hmmm yes I see that qemu-system-ppc64 fails because target_ulong will always represent a 64-bit quantity, even if you request a 32-bit CPU. Rather than add some CPU-specific code let's keep the uint32_t cast for now as I would hope it is unlikely someone would load an ELF in high memory, and perhaps the sign-extended address bug will get fixed later. With the cast reverted to uint32_t then for patches 1 and 2: Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> If you can send a v9 with the cast fixed I'll apply this to my qemu-macppc branch right away. ATB, Mark.
On Sat, 17 Oct 2020, Mark Cave-Ayland wrote: > On 16/10/2020 13:19, BALATON Zoltan via wrote: >> On Fri, 16 Oct 2020, Mark Cave-Ayland wrote: >>> On 16/10/2020 00:47, BALATON Zoltan via wrote: >>>> This is the cut down version of the earlier series omitting unfinished >>>> patches that I plan to rework later and rebased to Mark's qemu-macppc >>>> branch. Compared to v7 the only change is the cast to (target_ulong) >>>> from (uint32_t) as requested by Mark in patch 1. >>> >>> FWIW the reason for suggesting the cast to target_ulong is so that the >>> same code works for both qemu-system-ppc and qemu-system-ppc64. For >>> qemu-system-ppc that should correctly drop the sign extension from 32-bit, >>> whilst still allowing someone to load a 64-bit ELF into qemu-system-ppc64 >>> if requested. >>> >>> Can you confirm that the sign extension behaviour is still correct for >>> both qemu-system-ppc and qemu-system-ppc64? If so I'm happy to give it a >>> R-B tag. >> >> I've tried it now again with both ppc and ppc64: both OpenBIOS and a G3 >> beige ROM can be loaded with qemu-system-ppc but qemu-system-ppc64 fails >> with OpenBIOS when casting to target_ulong (i think because target_ulong is >> 64 bit there but g3beige is still 32 bit but I haven't throughly debugged >> it). But everything works with my original uint32_t cast, so ditch it and >> use my original version. Should I resubmit or you can fix up? (I think I >> wait until it's clear if this will be taken by David or you and send a >> fixed version cc-ing David if this is decided to go through the PPC queue.) > > Hmmm yes I see that qemu-system-ppc64 fails because target_ulong will always > represent a 64-bit quantity, even if you request a 32-bit CPU. Rather than > add some CPU-specific code let's keep the uint32_t cast for now as I would > hope it is unlikely someone would load an ELF in high memory, and perhaps the > sign-extended address bug will get fixed later. > > With the cast reverted to uint32_t then for patches 1 and 2: > Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> > > If you can send a v9 with the cast fixed I'll apply this to my qemu-macppc > branch right away. You could've really just edit the single cast in patch 1 before applying to change the it back but I've resent the changed patch 1 as v9 also adding your R-b for your convenience. Other patches are unchanged so you can take the v8 for those, I haven't resent those, let me know if you want the whole series but this is really getting much more work that it should be for such a simple change. (There is no cast in patch 2 as I've already stated several times.) Regards, BALATON Zoltan
On 17/10/2020 16:56, BALATON Zoltan via wrote: >> If you can send a v9 with the cast fixed I'll apply this to my qemu-macppc branch >> right away. > > You could've really just edit the single cast in patch 1 before applying to change > the it back but I've resent the changed patch 1 as v9 also adding your R-b for your > convenience. Other patches are unchanged so you can take the v8 for those, I haven't > resent those, let me know if you want the whole series but this is really getting > much more work that it should be for such a simple change. (There is no cast in patch > 2 as I've already stated several times.) Thanks - this has been included in the PR I just sent. ATB, Mark.
On 10/16/20 11:58 AM, Mark Cave-Ayland wrote: > On 16/10/2020 00:47, BALATON Zoltan via wrote: > >> This is the cut down version of the earlier series omitting unfinished >> patches that I plan to rework later and rebased to Mark's qemu-macppc >> branch. Compared to v7 the only change is the cast to (target_ulong) >> from (uint32_t) as requested by Mark in patch 1. > > FWIW the reason for suggesting the cast to target_ulong is so that the > same code works for both qemu-system-ppc and qemu-system-ppc64. For > qemu-system-ppc that should correctly drop the sign extension from > 32-bit, whilst still allowing someone to load a 64-bit ELF into > qemu-system-ppc64 if requested. IMO this is part of a bigger design problem. Not all machines main bus is 64-bit. I did some experiments but changing that involves a lot of work.
On Fri, 16 Oct 2020, Philippe Mathieu-Daudé wrote: > On 10/16/20 11:58 AM, Mark Cave-Ayland wrote: >> On 16/10/2020 00:47, BALATON Zoltan via wrote: >> >>> This is the cut down version of the earlier series omitting unfinished >>> patches that I plan to rework later and rebased to Mark's qemu-macppc >>> branch. Compared to v7 the only change is the cast to (target_ulong) >>> from (uint32_t) as requested by Mark in patch 1. >> >> FWIW the reason for suggesting the cast to target_ulong is so that the same >> code works for both qemu-system-ppc and qemu-system-ppc64. For >> qemu-system-ppc that should correctly drop the sign extension from 32-bit, >> whilst still allowing someone to load a 64-bit ELF into qemu-system-ppc64 >> if requested. > > IMO this is part of a bigger design problem. Not all > machines main bus is 64-bit. I did some experiments > but changing that involves a lot of work. Did not want to reply to this to not bring it to your attention before patch gets in finally but it's too late... Not sure what you refer to but in this particular case the problem only seems to be load_elf loading 32 bit ELF files returning sign extended 64 bit address which looks bogus but since this function is widely used I did not feel confident enough to propose a patch to load_elf. By the way, also the parameters of load_elf could take a clean up to remove all the mostly NULL values as I've pointed out before: https://lists.nongnu.org/archive/html/qemu-devel/2019-01/msg03427.html but all this could wait until later, these don't seem to be urgent problems to prevent moving mac machines forward now and could all be addressen in separate elf loading series. So just note the problem and move on for now please. Reagards. BALATON Zoltan
On 10/17/20 6:39 PM, BALATON Zoltan via wrote: > On Fri, 16 Oct 2020, Philippe Mathieu-Daudé wrote: >> On 10/16/20 11:58 AM, Mark Cave-Ayland wrote: >>> On 16/10/2020 00:47, BALATON Zoltan via wrote: >>> >>>> This is the cut down version of the earlier series omitting unfinished >>>> patches that I plan to rework later and rebased to Mark's qemu-macppc >>>> branch. Compared to v7 the only change is the cast to (target_ulong) >>>> from (uint32_t) as requested by Mark in patch 1. >>> >>> FWIW the reason for suggesting the cast to target_ulong is so that >>> the same code works for both qemu-system-ppc and qemu-system-ppc64. >>> For qemu-system-ppc that should correctly drop the sign extension >>> from 32-bit, whilst still allowing someone to load a 64-bit ELF into >>> qemu-system-ppc64 if requested. >> >> IMO this is part of a bigger design problem. Not all >> machines main bus is 64-bit. I did some experiments >> but changing that involves a lot of work. > > Did not want to reply to this to not bring it to your attention before > patch gets in finally but it's too late... > > Not sure what you refer to I refer to having machines with a N-bit main bus using a N-bit main bus (currently all main busses are 64-bit wide). Some 32-bit machines have access to 64-bit busses, some don't. I have been wondering about it because of the AVR CPU which uses a pair of busses, each less than 32-bit. If one day I can finish my work there, the Old World mac might benefit from it. > but in this particular case the problem only > seems to be load_elf loading 32 bit ELF files returning sign extended 64 > bit address which looks bogus but since this function is widely used I > did not feel confident enough to propose a patch to load_elf. > > By the way, also the parameters of load_elf could take a clean up to > remove all the mostly NULL values as I've pointed out before: > > https://lists.nongnu.org/archive/html/qemu-devel/2019-01/msg03427.html > > but all this could wait until later, these don't seem to be urgent > problems to prevent moving mac machines forward now and could all be > addressen in separate elf loading series. So just note the problem and > move on for now please. > > Reagards. > BALATON Zoltan
© 2016 - 2024 Red Hat, Inc.