| Message ID | cover.1668958803.git.baskov@ispras.ru (mailing list archive) |
|---|---|
| Headers | show |
| Series | x86_64: Improvements at compressed kernel stage | expand |
On Tue, Nov 22, 2022 at 02:12:09PM +0300, Evgeniy Baskov wrote: > This patchset is aimed > * to improve UEFI compatibility of compressed kernel code for x86_64 > * to setup proper memory access attributes for code and rodata sections > * to implement W^X protection policy throughout the whole execution > of compressed kernel for EFISTUB code path. Hi Evgeniy, I've tested this patch set on hardware and QEMU+MU firmware, and it works for me with a couple of minor issues: - on one machine that has the DXE protocol but not the EFI one, we get an error because the firmware doesn't support EFI_MEMORY_RP - on QEMU I'm seeing the size of "(unsigned long)_head - image_base" wind up as 0, which leads to an EFI_INVALID_PARAMETER on the clear_memory_attributes() call. In both cases the system winds up working, but with unnecessary console output. I'll send you patches as a follow-up to this mail. In the mean time: Tested-by: Peter Jones <pjones@redhat.com>
On Tue, Dec 13, 2022 at 01:03:17PM -0500, Peter Jones wrote: > On Tue, Nov 22, 2022 at 02:12:09PM +0300, Evgeniy Baskov wrote: > > This patchset is aimed > > * to improve UEFI compatibility of compressed kernel code for x86_64 > > * to setup proper memory access attributes for code and rodata sections > > * to implement W^X protection policy throughout the whole execution > > of compressed kernel for EFISTUB code path. > > Hi Evgeniy, > > I've tested this patch set on hardware and QEMU+MU firmware, and it > works for me with a couple of minor issues: > > - on one machine that has the DXE protocol but not the EFI one, we get > an error because the firmware doesn't support EFI_MEMORY_RP > - on QEMU I'm seeing the size of "(unsigned long)_head - image_base" > wind up as 0, which leads to an EFI_INVALID_PARAMETER on the > clear_memory_attributes() call. > > In both cases the system winds up working, but with unnecessary console > output. I just realized I've overstated here - I haven't actually hit the first problem on x86, only on ARM, where we don't currently use this code. I discovered it in grub, and checked your patch set to see if you had the same issue I did. That said, "in both cases the system winds up working" is probably still true - in that the edk2 code supports EFI_MEMORY_RP on x86 but not ARM, so x86 won't hit the issue when using DXE unless someone cooks up another implementation. Nevertheless I believe the patch to fix it is correct and should be applied. Thanks!
On 2022-12-13 21:13, Peter Jones wrote: > On Tue, Dec 13, 2022 at 01:03:17PM -0500, Peter Jones wrote: >> On Tue, Nov 22, 2022 at 02:12:09PM +0300, Evgeniy Baskov wrote: >> > This patchset is aimed >> > * to improve UEFI compatibility of compressed kernel code for x86_64 >> > * to setup proper memory access attributes for code and rodata sections >> > * to implement W^X protection policy throughout the whole execution >> > of compressed kernel for EFISTUB code path. >> >> Hi Evgeniy, >> >> I've tested this patch set on hardware and QEMU+MU firmware, and it >> works for me with a couple of minor issues: >> >> - on one machine that has the DXE protocol but not the EFI one, we get >> an error because the firmware doesn't support EFI_MEMORY_RP >> - on QEMU I'm seeing the size of "(unsigned long)_head - image_base" >> wind up as 0, which leads to an EFI_INVALID_PARAMETER on the >> clear_memory_attributes() call. >> >> In both cases the system winds up working, but with unnecessary >> console >> output. > > I just realized I've overstated here - I haven't actually hit the first > problem on x86, only on ARM, where we don't currently use this code. I > discovered it in grub, and checked your patch set to see if you had the > same issue I did. That said, "in both cases the system winds up > working" is probably still true - in that the edk2 code supports > EFI_MEMORY_RP on x86 but not ARM, so x86 won't hit the issue when using > DXE unless someone cooks up another implementation. Nevertheless I > believe the patch to fix it is correct and should be applied. > > Thanks! Hi, Thank you for testing and fixes! I have also discovered one issue with v3, that can only be hit when booting with grub -- there's one kernel_add_identity_map() missing in the get_acpi_srat_table() function, before header->length is read. So I'll prepare the v4 soon and include your new patches there. Thanks, Evgeniy Baskov
On Tue, 13 Dec 2022 at 23:16, Evgeniy Baskov <baskov@ispras.ru> wrote: > > On 2022-12-13 21:13, Peter Jones wrote: > > On Tue, Dec 13, 2022 at 01:03:17PM -0500, Peter Jones wrote: > >> On Tue, Nov 22, 2022 at 02:12:09PM +0300, Evgeniy Baskov wrote: > >> > This patchset is aimed > >> > * to improve UEFI compatibility of compressed kernel code for x86_64 > >> > * to setup proper memory access attributes for code and rodata sections > >> > * to implement W^X protection policy throughout the whole execution > >> > of compressed kernel for EFISTUB code path. > >> > >> Hi Evgeniy, > >> > >> I've tested this patch set on hardware and QEMU+MU firmware, and it > >> works for me with a couple of minor issues: > >> > >> - on one machine that has the DXE protocol but not the EFI one, we get > >> an error because the firmware doesn't support EFI_MEMORY_RP > >> - on QEMU I'm seeing the size of "(unsigned long)_head - image_base" > >> wind up as 0, which leads to an EFI_INVALID_PARAMETER on the > >> clear_memory_attributes() call. > >> > >> In both cases the system winds up working, but with unnecessary > >> console > >> output. > > > > I just realized I've overstated here - I haven't actually hit the first > > problem on x86, only on ARM, where we don't currently use this code. I > > discovered it in grub, and checked your patch set to see if you had the > > same issue I did. That said, "in both cases the system winds up > > working" is probably still true - in that the edk2 code supports > > EFI_MEMORY_RP on x86 but not ARM, so x86 won't hit the issue when using > > DXE unless someone cooks up another implementation. Nevertheless I > > believe the patch to fix it is correct and should be applied. > > > > Thanks! > > Hi, > > Thank you for testing and fixes! > > I have also discovered one issue with v3, that can only be hit when > booting with grub -- there's one kernel_add_identity_map() missing in > the get_acpi_srat_table() function, before header->length is read. > So I'll prepare the v4 soon and include your new patches there. > Happy to see this is converging. Please rebase onto latest mainline as well - some cleanups to the early boot code have landed there yesterday.
On 2022-12-14 13:09, Ard Biesheuvel wrote: > On Tue, 13 Dec 2022 at 23:16, Evgeniy Baskov <baskov@ispras.ru> wrote: >> >> On 2022-12-13 21:13, Peter Jones wrote: >> > On Tue, Dec 13, 2022 at 01:03:17PM -0500, Peter Jones wrote: >> >> On Tue, Nov 22, 2022 at 02:12:09PM +0300, Evgeniy Baskov wrote: >> >> > This patchset is aimed >> >> > * to improve UEFI compatibility of compressed kernel code for x86_64 >> >> > * to setup proper memory access attributes for code and rodata sections >> >> > * to implement W^X protection policy throughout the whole execution >> >> > of compressed kernel for EFISTUB code path. >> >> >> >> Hi Evgeniy, >> >> >> >> I've tested this patch set on hardware and QEMU+MU firmware, and it >> >> works for me with a couple of minor issues: >> >> >> >> - on one machine that has the DXE protocol but not the EFI one, we get >> >> an error because the firmware doesn't support EFI_MEMORY_RP >> >> - on QEMU I'm seeing the size of "(unsigned long)_head - image_base" >> >> wind up as 0, which leads to an EFI_INVALID_PARAMETER on the >> >> clear_memory_attributes() call. >> >> >> >> In both cases the system winds up working, but with unnecessary >> >> console >> >> output. >> > >> > I just realized I've overstated here - I haven't actually hit the first >> > problem on x86, only on ARM, where we don't currently use this code. I >> > discovered it in grub, and checked your patch set to see if you had the >> > same issue I did. That said, "in both cases the system winds up >> > working" is probably still true - in that the edk2 code supports >> > EFI_MEMORY_RP on x86 but not ARM, so x86 won't hit the issue when using >> > DXE unless someone cooks up another implementation. Nevertheless I >> > believe the patch to fix it is correct and should be applied. >> > >> > Thanks! >> >> Hi, >> >> Thank you for testing and fixes! >> >> I have also discovered one issue with v3, that can only be hit when >> booting with grub -- there's one kernel_add_identity_map() missing in >> the get_acpi_srat_table() function, before header->length is read. >> So I'll prepare the v4 soon and include your new patches there. >> > > Happy to see this is converging. Please rebase onto latest mainline as > well - some cleanups to the early boot code have landed there > yesterday. Will do, thanks!