lkml.org 
[lkml]   [2020]   [Jun]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: 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: ffffb2b582cc2000
> > [ 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 <f3> 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:ffffb2b583d07e10 EFLAGS: 00010216
> > [ 20.263133] RAX: 0000000000000000 RBX: ffff8b8000102c00 RCX: 0000000000004000
> > [ 20.263134] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffb2b582cc2000
> > [ 20.263136] RBP: ffff8b8000101000 R08: 0000000000000000 R09: ffffb2b582cc2000
> > [ 20.263137] R10: 0000000000005356 R11: ffff8b8000102c18 R12: 0000000000000000
> > [ 20.263139] R13: 0000000000000000 R14: ffff8b8039944200 R15: ffffffff9794daa0
> > [ 20.263141] FS: 00007f41aa4b4200(0000) GS:ffff8b803ecc0000(0000) knlGS:0000000000000000
> > [ 20.263143] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [ 20.263144] CR2: ffffb2b582cc2000 CR3: 00000003b6731000 CR4: 00000000003406e0
> > [ 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: ffffb2b582cc2000
> > [ 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 <f3> 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:ffffb2b583d07e10 EFLAGS: 00010216
> > [ 20.263193] RAX: 0000000000000000 RBX: ffff8b8000102c00 RCX: 0000000000004000
> > [ 20.263195] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffb2b582cc2000
> > [ 20.263196] RBP: ffff8b8000101000 R08: 0000000000000000 R09: ffffb2b582cc2000
> > [ 20.263197] R10: 0000000000005356 R11: ffff8b8000102c18 R12: 0000000000000000
> > [ 20.263199] R13: 0000000000000000 R14: ffff8b8039944200 R15: ffffffff9794daa0
> > [ 20.263201] FS: 00007f41aa4b4200(0000) GS:ffff8b803ecc0000(0000) knlGS:0000000000000000
> > [ 20.263202] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [ 20.263204] CR2: ffffb2b582cc2000 CR3: 00000003b6731000 CR4: 00000000003406e0
> >
> > 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..2bf2222819d3 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 it helps?
---
diff --git a/kernel/dma/pool.c b/kernel/dma/pool.c
--- a/kernel/dma/pool.c
+++ b/kernel/dma/pool.c
@@ -13,6 +13,8 @@
#include <linux/slab.h>
#include <linux/workqueue.h>

+static struct gen_pool *atomic_pool __ro_after_init;
+static unsigned long pool_size;
static struct gen_pool *atomic_pool_dma __ro_after_init;
static unsigned long pool_size_dma;
static struct gen_pool *atomic_pool_dma32 __ro_after_init;
@@ -41,24 +43,37 @@ static void __init dma_atomic_pool_debugfs_init(void)
if (IS_ERR_OR_NULL(root))
return;

+ debugfs_create_ulong("pool_size", 0400, root, &pool_size);
debugfs_create_ulong("pool_size_dma", 0400, root, &pool_size_dma);
debugfs_create_ulong("pool_size_dma32", 0400, root, &pool_size_dma32);
debugfs_create_ulong("pool_size_kernel", 0400, root, &pool_size_kernel);
}

-static void dma_atomic_pool_size_add(gfp_t gfp, size_t size)
+static gfp_t dma_atomic_pool_gfp(void)
{
- if (gfp & __GFP_DMA)
+ if (IS_ENABLED(CONFIG_ZONE_DMA))
+ return GFP_KERNEL | GFP_DMA;
+ if (IS_ENABLED(CONFIG_ZONE_DMA32))
+ return GFP_KERNEL | GFP_DMA32;
+ return GFP_KERNEL;
+}
+
+static void dma_atomic_pool_size_add(struct gen_pool *pool, size_t size)
+{
+ if (pool == atomic_pool)
+ pool_size += size;
+ else if (pool == atomic_pool_dma)
pool_size_dma += size;
- else if (gfp & __GFP_DMA32)
+ else if (pool == atomic_pool_dma32)
pool_size_dma32 += size;
- else
+ else if (pool == atomic_pool_kernel)
pool_size_kernel += size;
}

static int atomic_pool_expand(struct gen_pool *pool, size_t pool_size,
gfp_t gfp)
{
+ bool decrypt = pool != atomic_pool;
unsigned int order;
struct page *page;
void *addr;
@@ -94,8 +109,9 @@ static int atomic_pool_expand(struct gen_pool *pool, size_t pool_size,
* Memory in the atomic DMA pools must be unencrypted, the pools do not
* shrink so no re-encryption occurs in dma_direct_free_pages().
*/
- ret = set_memory_decrypted((unsigned long)page_to_virt(page),
- 1 << order);
+ if (decrypt)
+ ret = set_memory_decrypted((unsigned long)page_to_virt(page),
+ 1 << order);
if (ret)
goto remove_mapping;
ret = gen_pool_add_virt(pool, (unsigned long)addr, page_to_phys(page),
@@ -103,12 +119,13 @@ static int atomic_pool_expand(struct gen_pool *pool, size_t pool_size,
if (ret)
goto encrypt_mapping;

- dma_atomic_pool_size_add(gfp, pool_size);
+ dma_atomic_pool_size_add(pool, pool_size);
return 0;

encrypt_mapping:
- ret = set_memory_encrypted((unsigned long)page_to_virt(page),
- 1 << order);
+ if (decrypt)
+ ret = set_memory_encrypted((unsigned long)page_to_virt(page),
+ 1 << order);
if (WARN_ON_ONCE(ret)) {
/* Decrypt succeeded but encrypt failed, purposely leak */
goto out;
@@ -132,6 +149,7 @@ static void atomic_pool_resize(struct gen_pool *pool, gfp_t gfp)

static void atomic_pool_work_fn(struct work_struct *work)
{
+ atomic_pool_resize(atomic_pool, dma_atomic_pool_gfp());
if (IS_ENABLED(CONFIG_ZONE_DMA))
atomic_pool_resize(atomic_pool_dma,
GFP_KERNEL | GFP_DMA);
@@ -182,6 +200,10 @@ static int __init dma_atomic_pool_init(void)
}
INIT_WORK(&atomic_pool_work, atomic_pool_work_fn);

+ atomic_pool = __dma_atomic_pool_init(atomic_pool_size,
+ dma_atomic_pool_gfp());
+ if (!atomic_pool)
+ ret = -ENOMEM;
atomic_pool_kernel = __dma_atomic_pool_init(atomic_pool_size,
GFP_KERNEL);
if (!atomic_pool_kernel)
@@ -209,6 +231,9 @@ static inline struct gen_pool *dev_to_pool(struct device *dev)
u64 phys_mask;
gfp_t gfp;

+ if (!force_dma_unencrypted(dev))
+ return atomic_pool;
+
gfp = dma_direct_optimal_gfp_mask(dev, dev->coherent_dma_mask,
&phys_mask);
if (IS_ENABLED(CONFIG_ZONE_DMA) && gfp == GFP_DMA)
\
 
 \ /
  Last update: 2020-06-07 21:42    [W:0.156 / U:0.032 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site