Re: 82fef0ad811f "x86/mm: unencrypted non-blocking DMA allocations use coherent pools" was Re: next-0519 on thinkpad x60: sound related? window manager crash
Excerpts from David Rientjes's message of June 7, 2020 8:57 pm:
> Thanks for trying it out, Alex. Would you mind sending your .config and
> command line? I assume either mem_encrypt=on or
> CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT is enabled.
>
> Could you also give this a try?
>
> diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
> --- a/kernel/dma/direct.c
> +++ b/kernel/dma/direct.c
> @@ -99,10 +99,11 @@ static inline bool dma_should_alloc_from_pool(struct
> device *dev, gfp_t gfp,
> static inline bool dma_should_free_from_pool(struct device *dev,
>unsigned long attrs)
> {
> - if (IS_ENABLED(CONFIG_DMA_COHERENT_POOL))
> + if (!IS_ENABLED(CONFIG_DMA_COHERENT_POOL))
> + return false;
> + if (force_dma_unencrypted(dev))
> return true;
> - if ((attrs & DMA_ATTR_NO_KERNEL_MAPPING) &&
> - !force_dma_unencrypted(dev))
> + if (attrs & DMA_ATTR_NO_KERNEL_MAPPING)
> return false;
> if (IS_ENABLED(CONFIG_DMA_DIRECT_REMAP))
> return true;
>
This patch doesn't work for me either. It has since occurred to me that
while I do have CONFIG_AMD_MEM_ENCYRPT=y, I have
CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT=n, because it was broken with
amdgpu (unfortunately a downgrade from radeon in this respect). Tried it
again just now and it looks like it's now able to enable KMS, but all it
displays is serious-looking errors.
Sorry for not mentioning that earlier. I'll send you my .config and
command line off-list.
Thanks,
Alex.
Re: 82fef0ad811f "x86/mm: unencrypted non-blocking DMA allocations use coherent pools" was Re: next-0519 on thinkpad x60: sound related? window manager crash
On Sun, 7 Jun 2020, Alex Xu (Hello71) wrote: > > On Sun, 7 Jun 2020, Pavel Machek wrote: > > > >> > I have a similar issue, caused between aaa2faab4ed8 and b170290c2836. > >> > > >> > [ 20.263098] BUG: unable to handle page fault for address: > >> > b2b582cc2000 > >> > [ 20.263104] #PF: supervisor write access in kernel mode > >> > [ 20.263105] #PF: error_code(0x000b) - reserved bit violation > >> > [ 20.263107] PGD 3fd03b067 P4D 3fd03b067 PUD 3fd03c067 PMD 3f8822067 > >> > PTE 8000273942ab2163 > >> > [ 20.263113] Oops: 000b [#1] PREEMPT SMP > >> > [ 20.263117] CPU: 3 PID: 691 Comm: mpv Not tainted > >> > 5.7.0-11262-gb170290c2836 #1 > >> > [ 20.263119] Hardware name: To Be Filled By O.E.M. To Be Filled By > >> > O.E.M./B450 Pro4, BIOS P4.10 03/05/2020 > >> > [ 20.263125] RIP: 0010:__memset+0x24/0x30 > >> > [ 20.263128] Code: cc cc cc cc cc cc 0f 1f 44 00 00 49 89 f9 48 89 d1 > >> > 83 e2 07 48 c1 e9 03 40 0f b6 f6 48 b8 01 01 01 01 01 01 01 01 48 0f af > >> > c6 48 ab 89 d1 f3 aa 4c 89 c8 c3 90 49 89 f9 40 88 f0 48 89 d1 f3 > >> > [ 20.263131] RSP: 0018:b2b583d07e10 EFLAGS: 00010216 > >> > [ 20.263133] RAX: RBX: 8b8000102c00 RCX: > >> > 4000 > >> > [ 20.263134] RDX: RSI: RDI: > >> > b2b582cc2000 > >> > [ 20.263136] RBP: 8b8000101000 R08: R09: > >> > b2b582cc2000 > >> > [ 20.263137] R10: 5356 R11: 8b8000102c18 R12: > >> > > >> > [ 20.263139] R13: R14: 8b8039944200 R15: > >> > 9794daa0 > >> > [ 20.263141] FS: 7f41aa4b4200() GS:8b803ecc() > >> > knlGS: > >> > [ 20.263143] CS: 0010 DS: ES: CR0: 80050033 > >> > [ 20.263144] CR2: b2b582cc2000 CR3: 0003b6731000 CR4: > >> > 003406e0 > >> > [ 20.263146] Call Trace: > >> > [ 20.263151] ? snd_pcm_hw_params+0x3f3/0x47a > >> > [ 20.263154] ? snd_pcm_common_ioctl+0xf2/0xf73 > >> > [ 20.263158] ? snd_pcm_ioctl+0x1e/0x29 > >> > [ 20.263161] ? ksys_ioctl+0x77/0x91 > >> > [ 20.263163] ? __x64_sys_ioctl+0x11/0x14 > >> > [ 20.263166] ? do_syscall_64+0x3d/0xf5 > >> > [ 20.263170] ? entry_SYSCALL_64_after_hwframe+0x44/0xa9 > >> > [ 20.263173] Modules linked in: uvcvideo videobuf2_vmalloc > >> > videobuf2_memops videobuf2_v4l2 videodev snd_usb_audio videobuf2_common > >> > snd_hwdep snd_usbmidi_lib input_leds snd_rawmidi led_class > >> > [ 20.263182] CR2: b2b582cc2000 > >> > [ 20.263184] ---[ end trace c6b47a774b91f0a0 ]--- > >> > [ 20.263187] RIP: 0010:__memset+0x24/0x30 > >> > [ 20.263190] Code: cc cc cc cc cc cc 0f 1f 44 00 00 49 89 f9 48 89 d1 > >> > 83 e2 07 48 c1 e9 03 40 0f b6 f6 48 b8 01 01 01 01 01 01 01 01 48 0f af > >> > c6 48 ab 89 d1 f3 aa 4c 89 c8 c3 90 49 89 f9 40 88 f0 48 89 d1 f3 > >> > [ 20.263192] RSP: 0018:b2b583d07e10 EFLAGS: 00010216 > >> > [ 20.263193] RAX: RBX: 8b8000102c00 RCX: > >> > 4000 > >> > [ 20.263195] RDX: RSI: RDI: > >> > b2b582cc2000 > >> > [ 20.263196] RBP: 8b8000101000 R08: R09: > >> > b2b582cc2000 > >> > [ 20.263197] R10: 5356 R11: 8b8000102c18 R12: > >> > > >> > [ 20.263199] R13: R14: 8b8039944200 R15: > >> > 9794daa0 > >> > [ 20.263201] FS: 7f41aa4b4200() GS:8b803ecc() > >> > knlGS: > >> > [ 20.263202] CS: 0010 DS: ES: CR0: 80050033 > >> > [ 20.263204] CR2: b2b582cc2000 CR3: 0003b6731000 CR4: > >> > 003406e0 > >> > > >> > I bisected this to 82fef0ad811f "x86/mm: unencrypted non-blocking DMA > >> > allocations use coherent pools". Reverting 1ee18de92927 resolves the > >> > issue. > >> > > >> > Looks like Thinkpad X60 doesn't have VT-d, but could still be DMA > >> > related. > >> > >> Note that newer -next releases seem to behave okay for me. The commit > >> pointed out by siection is really simple: > >> > >> AFAIK you could verify it is responsible by turning off > >> CONFIG_AMD_MEM_ENCRYPT on latest kernel... > >> > >> Best regards, > >>Pavel > >> > >> index 1d6104ea8af0..2bf819d3 100644 > >> --- a/arch/x86/Kconfig > >> +++ b/arch/x86/Kconfig > >> @@ -1520,6 +1520,7 @@ config X86_CPA_STATISTICS > >> config AMD_MEM_ENCRYPT > >> bool "AMD Secure Memory Encryption (SME) support" > >> depends on X86_64 && CPU_SUP_AMD > >> + select DMA_COHERENT_POOL > >> select DYNAMIC_PHYSICAL_MASK > >> select ARCH_USE_MEMREMAP_PROT > >> select ARCH_HAS_FORCE_DMA_UNENCRYPTED > > > > Thanks for the report! > > > > Besides CONFIG_AMD_MEM_ENCRYPT, do you have CONFIG_DMA_DIRECT_REMAP > > enabled? If so, it may be caused by the virtual address passed to the > >
Re: 82fef0ad811f "x86/mm: unencrypted non-blocking DMA allocations use coherent pools" was Re: next-0519 on thinkpad x60: sound related? window manager crash
Excerpts from David Rientjes's message of June 7, 2020 3:41 pm:
> On Sun, 7 Jun 2020, Pavel Machek wrote:
>
>> > I have a similar issue, caused between aaa2faab4ed8 and b170290c2836.
>> >
>> > [ 20.263098] BUG: unable to handle page fault for address:
>> > b2b582cc2000
>> > [ 20.263104] #PF: supervisor write access in kernel mode
>> > [ 20.263105] #PF: error_code(0x000b) - reserved bit violation
>> > [ 20.263107] PGD 3fd03b067 P4D 3fd03b067 PUD 3fd03c067 PMD 3f8822067 PTE
>> > 8000273942ab2163
>> > [ 20.263113] Oops: 000b [#1] PREEMPT SMP
>> > [ 20.263117] CPU: 3 PID: 691 Comm: mpv Not tainted
>> > 5.7.0-11262-gb170290c2836 #1
>> > [ 20.263119] Hardware name: To Be Filled By O.E.M. To Be Filled By
>> > O.E.M./B450 Pro4, BIOS P4.10 03/05/2020
>> > [ 20.263125] RIP: 0010:__memset+0x24/0x30
>> > [ 20.263128] Code: cc cc cc cc cc cc 0f 1f 44 00 00 49 89 f9 48 89 d1 83
>> > e2 07 48 c1 e9 03 40 0f b6 f6 48 b8 01 01 01 01 01 01 01 01 48 0f af c6
>> > 48 ab 89 d1 f3 aa 4c 89 c8 c3 90 49 89 f9 40 88 f0 48 89 d1 f3
>> > [ 20.263131] RSP: 0018:b2b583d07e10 EFLAGS: 00010216
>> > [ 20.263133] RAX: RBX: 8b8000102c00 RCX:
>> > 4000
>> > [ 20.263134] RDX: RSI: RDI:
>> > b2b582cc2000
>> > [ 20.263136] RBP: 8b8000101000 R08: R09:
>> > b2b582cc2000
>> > [ 20.263137] R10: 5356 R11: 8b8000102c18 R12:
>> >
>> > [ 20.263139] R13: R14: 8b8039944200 R15:
>> > 9794daa0
>> > [ 20.263141] FS: 7f41aa4b4200() GS:8b803ecc()
>> > knlGS:
>> > [ 20.263143] CS: 0010 DS: ES: CR0: 80050033
>> > [ 20.263144] CR2: b2b582cc2000 CR3: 0003b6731000 CR4:
>> > 003406e0
>> > [ 20.263146] Call Trace:
>> > [ 20.263151] ? snd_pcm_hw_params+0x3f3/0x47a
>> > [ 20.263154] ? snd_pcm_common_ioctl+0xf2/0xf73
>> > [ 20.263158] ? snd_pcm_ioctl+0x1e/0x29
>> > [ 20.263161] ? ksys_ioctl+0x77/0x91
>> > [ 20.263163] ? __x64_sys_ioctl+0x11/0x14
>> > [ 20.263166] ? do_syscall_64+0x3d/0xf5
>> > [ 20.263170] ? entry_SYSCALL_64_after_hwframe+0x44/0xa9
>> > [ 20.263173] Modules linked in: uvcvideo videobuf2_vmalloc
>> > videobuf2_memops videobuf2_v4l2 videodev snd_usb_audio videobuf2_common
>> > snd_hwdep snd_usbmidi_lib input_leds snd_rawmidi led_class
>> > [ 20.263182] CR2: b2b582cc2000
>> > [ 20.263184] ---[ end trace c6b47a774b91f0a0 ]---
>> > [ 20.263187] RIP: 0010:__memset+0x24/0x30
>> > [ 20.263190] Code: cc cc cc cc cc cc 0f 1f 44 00 00 49 89 f9 48 89 d1 83
>> > e2 07 48 c1 e9 03 40 0f b6 f6 48 b8 01 01 01 01 01 01 01 01 48 0f af c6
>> > 48 ab 89 d1 f3 aa 4c 89 c8 c3 90 49 89 f9 40 88 f0 48 89 d1 f3
>> > [ 20.263192] RSP: 0018:b2b583d07e10 EFLAGS: 00010216
>> > [ 20.263193] RAX: RBX: 8b8000102c00 RCX:
>> > 4000
>> > [ 20.263195] RDX: RSI: RDI:
>> > b2b582cc2000
>> > [ 20.263196] RBP: 8b8000101000 R08: R09:
>> > b2b582cc2000
>> > [ 20.263197] R10: 5356 R11: 8b8000102c18 R12:
>> >
>> > [ 20.263199] R13: R14: 8b8039944200 R15:
>> > 9794daa0
>> > [ 20.263201] FS: 7f41aa4b4200() GS:8b803ecc()
>> > knlGS:
>> > [ 20.263202] CS: 0010 DS: ES: CR0: 80050033
>> > [ 20.263204] CR2: b2b582cc2000 CR3: 0003b6731000 CR4:
>> > 003406e0
>> >
>> > I bisected this to 82fef0ad811f "x86/mm: unencrypted non-blocking DMA
>> > allocations use coherent pools". Reverting 1ee18de92927 resolves the
>> > issue.
>> >
>> > Looks like Thinkpad X60 doesn't have VT-d, but could still be DMA
>> > related.
>>
>> Note that newer -next releases seem to behave okay for me. The commit
>> pointed out by siection is really simple:
>>
>> AFAIK you could verify it is responsible by turning off
>> CONFIG_AMD_MEM_ENCRYPT on latest kernel...
>>
>> Best regards,
>> Pavel
>>
>> index 1d6104ea8af0..2bf819d3 100644
>> --- a/arch/x86/Kconfig
>> +++ b/arch/x86/Kconfig
>> @@ -1520,6 +1520,7 @@ config X86_CPA_STATISTICS
>> config AMD_MEM_ENCRYPT
>> bool "AMD Secure Memory Encryption (SME) support"
>> depends on X86_64 && CPU_SUP_AMD
>> + select DMA_COHERENT_POOL
>> select DYNAMIC_PHYSICAL_MASK
>> select ARCH_USE_MEMREMAP_PROT
>> select ARCH_HAS_FORCE_DMA_UNENCRYPTED
>
> Thanks for the report!
>
> Besides CONFIG_AMD_MEM_ENCRYPT, do you have CONFIG_DMA_DIRECT_REMAP
> enabled? If so, it may be caused by the virtual address passed to the
> set_memory_{decrypted,encrypted}() functions.
>
> And I assume you are enabling SME by using mem_encrypt=on on the kernel
> command line or CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT
Re: 82fef0ad811f "x86/mm: unencrypted non-blocking DMA allocations use coherent pools" was Re: next-0519 on thinkpad x60: sound related? window manager crash
On Sun, 7 Jun 2020, Pavel Machek wrote:
> > I have a similar issue, caused between aaa2faab4ed8 and b170290c2836.
> >
> > [ 20.263098] BUG: unable to handle page fault for address:
> > b2b582cc2000
> > [ 20.263104] #PF: supervisor write access in kernel mode
> > [ 20.263105] #PF: error_code(0x000b) - reserved bit violation
> > [ 20.263107] PGD 3fd03b067 P4D 3fd03b067 PUD 3fd03c067 PMD 3f8822067 PTE
> > 8000273942ab2163
> > [ 20.263113] Oops: 000b [#1] PREEMPT SMP
> > [ 20.263117] CPU: 3 PID: 691 Comm: mpv Not tainted
> > 5.7.0-11262-gb170290c2836 #1
> > [ 20.263119] Hardware name: To Be Filled By O.E.M. To Be Filled By
> > O.E.M./B450 Pro4, BIOS P4.10 03/05/2020
> > [ 20.263125] RIP: 0010:__memset+0x24/0x30
> > [ 20.263128] Code: cc cc cc cc cc cc 0f 1f 44 00 00 49 89 f9 48 89 d1 83
> > e2 07 48 c1 e9 03 40 0f b6 f6 48 b8 01 01 01 01 01 01 01 01 48 0f af c6
> > 48 ab 89 d1 f3 aa 4c 89 c8 c3 90 49 89 f9 40 88 f0 48 89 d1 f3
> > [ 20.263131] RSP: 0018:b2b583d07e10 EFLAGS: 00010216
> > [ 20.263133] RAX: RBX: 8b8000102c00 RCX:
> > 4000
> > [ 20.263134] RDX: RSI: RDI:
> > b2b582cc2000
> > [ 20.263136] RBP: 8b8000101000 R08: R09:
> > b2b582cc2000
> > [ 20.263137] R10: 5356 R11: 8b8000102c18 R12:
> >
> > [ 20.263139] R13: R14: 8b8039944200 R15:
> > 9794daa0
> > [ 20.263141] FS: 7f41aa4b4200() GS:8b803ecc()
> > knlGS:
> > [ 20.263143] CS: 0010 DS: ES: CR0: 80050033
> > [ 20.263144] CR2: b2b582cc2000 CR3: 0003b6731000 CR4:
> > 003406e0
> > [ 20.263146] Call Trace:
> > [ 20.263151] ? snd_pcm_hw_params+0x3f3/0x47a
> > [ 20.263154] ? snd_pcm_common_ioctl+0xf2/0xf73
> > [ 20.263158] ? snd_pcm_ioctl+0x1e/0x29
> > [ 20.263161] ? ksys_ioctl+0x77/0x91
> > [ 20.263163] ? __x64_sys_ioctl+0x11/0x14
> > [ 20.263166] ? do_syscall_64+0x3d/0xf5
> > [ 20.263170] ? entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > [ 20.263173] Modules linked in: uvcvideo videobuf2_vmalloc
> > videobuf2_memops videobuf2_v4l2 videodev snd_usb_audio videobuf2_common
> > snd_hwdep snd_usbmidi_lib input_leds snd_rawmidi led_class
> > [ 20.263182] CR2: b2b582cc2000
> > [ 20.263184] ---[ end trace c6b47a774b91f0a0 ]---
> > [ 20.263187] RIP: 0010:__memset+0x24/0x30
> > [ 20.263190] Code: cc cc cc cc cc cc 0f 1f 44 00 00 49 89 f9 48 89 d1 83
> > e2 07 48 c1 e9 03 40 0f b6 f6 48 b8 01 01 01 01 01 01 01 01 48 0f af c6
> > 48 ab 89 d1 f3 aa 4c 89 c8 c3 90 49 89 f9 40 88 f0 48 89 d1 f3
> > [ 20.263192] RSP: 0018:b2b583d07e10 EFLAGS: 00010216
> > [ 20.263193] RAX: RBX: 8b8000102c00 RCX:
> > 4000
> > [ 20.263195] RDX: RSI: RDI:
> > b2b582cc2000
> > [ 20.263196] RBP: 8b8000101000 R08: R09:
> > b2b582cc2000
> > [ 20.263197] R10: 5356 R11: 8b8000102c18 R12:
> >
> > [ 20.263199] R13: R14: 8b8039944200 R15:
> > 9794daa0
> > [ 20.263201] FS: 7f41aa4b4200() GS:8b803ecc()
> > knlGS:
> > [ 20.263202] CS: 0010 DS: ES: CR0: 80050033
> > [ 20.263204] CR2: b2b582cc2000 CR3: 0003b6731000 CR4:
> > 003406e0
> >
> > I bisected this to 82fef0ad811f "x86/mm: unencrypted non-blocking DMA
> > allocations use coherent pools". Reverting 1ee18de92927 resolves the
> > issue.
> >
> > Looks like Thinkpad X60 doesn't have VT-d, but could still be DMA
> > related.
>
> Note that newer -next releases seem to behave okay for me. The commit
> pointed out by siection is really simple:
>
> AFAIK you could verify it is responsible by turning off
> CONFIG_AMD_MEM_ENCRYPT on latest kernel...
>
> Best regards,
> Pavel
>
> index 1d6104ea8af0..2bf819d3 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -1520,6 +1520,7 @@ config X86_CPA_STATISTICS
> config AMD_MEM_ENCRYPT
> bool "AMD Secure Memory Encryption (SME) support"
> depends on X86_64 && CPU_SUP_AMD
> + select DMA_COHERENT_POOL
> select DYNAMIC_PHYSICAL_MASK
> select ARCH_USE_MEMREMAP_PROT
> select ARCH_HAS_FORCE_DMA_UNENCRYPTED
Thanks for the report!
Besides CONFIG_AMD_MEM_ENCRYPT, do you have CONFIG_DMA_DIRECT_REMAP
enabled? If so, it may be caused by the virtual address passed to the
set_memory_{decrypted,encrypted}() functions.
And I assume you are enabling SME by using mem_encrypt=on on the kernel
command line or CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT is enabled.
We likely need an atomic pool for devices that support DMA to addresses in
sme_me_mask as well. I can test this tomorrow, but wanted to get it out
early to see if
