[4/4] powerpc: Book3S 64-bit "heavyweight" KASAN support

KASAN support on powerpc64 is interesting:

 - We want to be able to support inline instrumentation so as to be
   able to catch global and stack issues.

 - We run a lot of code at boot in real mode. This includes stuff like
   printk(), so it's not feasible to just disable instrumentation
   around it.

   [For those not immersed in ppc64, in real mode, the top nibble or
   2 bits (depending on radix/hash mmu) of the address is ignored. To
   make things work, we put the linear mapping at
   0xc000000000000000. This means that a pointer to part of the linear
   mapping will work both in real mode, where it will be interpreted
   as a physical address of the form 0x000..., and out of real mode,
   where it will go via the linear mapping.]

 - Inline instrumentation requires a fixed offset.

 - Because of our running things in real mode, the offset has to
   point to valid memory both in and out of real mode.

This makes finding somewhere to put the KASAN shadow region a bit fun.

One approach is just to give up on inline instrumentation and override
the address->shadow calculation. This way we can delay all checking
until after we get everything set up to our satisfaction. However,
we'd really like to do better.

What we can do - if we know _at compile time_ how much contiguous
physical memory we have - is to set aside the top 1/8th of the memory
and use that. This is a big hammer (hence the "heavyweight" name) and
comes with 3 big consequences:

 - kernels will simply fail to boot on machines with less memory than
   specified when compiling.

 - kernels running on machines with more memory than specified when
   compiling will simply ignore the extra memory.

 - there's no nice way to handle physically discontiguous memory, so
   you are restricted to the first physical memory block.

If you can bear all this, you get pretty full support for KASAN.

Despite the limitations, it can still find bugs,
e.g. http://patchwork.ozlabs.org/patch/1103775/

The current implementation is Radix only. I am open to extending
it to hash at some point but I don't think it should hold up v1.

Massive thanks to mpe, who had the idea for the initial design.

Signed-off-by: Daniel Axtens <dja@axtens.net>

---
Changes since the rfc:

 - Boots real and virtual hardware, kvm works.

 - disabled reporting when we're checking the stack for exception
   frames. The behaviour isn't wrong, just incompatible with KASAN.

 - Documentation!

 - Dropped old module stuff in favour of KASAN_VMALLOC.

The bugs with ftrace and kuap were due to kernel bloat pushing
prom_init calls to be done via the plt. Because we did not have
a relocatable kernel, and they are done very early, this caused
everything to explode. Compile with CONFIG_RELOCATABLE!

---
 Documentation/dev-tools/kasan.rst            |   7 +-
 Documentation/powerpc/kasan.txt              | 111 +++++++++++++++++++
 arch/powerpc/Kconfig                         |   4 +
 arch/powerpc/Kconfig.debug                   |  21 ++++
 arch/powerpc/Makefile                        |   7 ++
 arch/powerpc/include/asm/book3s/64/radix.h   |   5 +
 arch/powerpc/include/asm/kasan.h             |  35 +++++-
 arch/powerpc/kernel/process.c                |   8 ++
 arch/powerpc/kernel/prom.c                   |  57 +++++++++-
 arch/powerpc/mm/kasan/Makefile               |   1 +
 arch/powerpc/mm/kasan/kasan_init_book3s_64.c |  76 +++++++++++++
 11 files changed, 326 insertions(+), 6 deletions(-)
 create mode 100644 Documentation/powerpc/kasan.txt
 create mode 100644 arch/powerpc/mm/kasan/kasan_init_book3s_64.c

Le 07/08/2019 à 01:38, Daniel Axtens a écrit :
> KASAN support on powerpc64 is interesting:
> 
>   - We want to be able to support inline instrumentation so as to be
>     able to catch global and stack issues.
> 
>   - We run a lot of code at boot in real mode. This includes stuff like
>     printk(), so it's not feasible to just disable instrumentation
>     around it.

Have you definitely given up the idea of doing a standard implementation 
of KASAN like other 64 bits arches have done ?

Isn't it possible to setup an early 1:1 mapping and go in virtual mode 
earlier ? What is so different between book3s64 and book3e64 ?
On book3e64, we've been able to setup KASAN before printing anything 
(except when using EARLY_DEBUG). Isn't it feasible on book3s64 too ?

> 
>     [For those not immersed in ppc64, in real mode, the top nibble or
>     2 bits (depending on radix/hash mmu) of the address is ignored. To
>     make things work, we put the linear mapping at
>     0xc000000000000000. This means that a pointer to part of the linear
>     mapping will work both in real mode, where it will be interpreted
>     as a physical address of the form 0x000..., and out of real mode,
>     where it will go via the linear mapping.]
> 
>   - Inline instrumentation requires a fixed offset.
> 
>   - Because of our running things in real mode, the offset has to
>     point to valid memory both in and out of real mode.
> 
> This makes finding somewhere to put the KASAN shadow region a bit fun.
> 
> One approach is just to give up on inline instrumentation and override
> the address->shadow calculation. This way we can delay all checking
> until after we get everything set up to our satisfaction. However,
> we'd really like to do better.
> 
> What we can do - if we know _at compile time_ how much contiguous
> physical memory we have - is to set aside the top 1/8th of the memory
> and use that. This is a big hammer (hence the "heavyweight" name) and
> comes with 3 big consequences:
> 
>   - kernels will simply fail to boot on machines with less memory than
>     specified when compiling.
> 
>   - kernels running on machines with more memory than specified when
>     compiling will simply ignore the extra memory.
> 
>   - there's no nice way to handle physically discontiguous memory, so
>     you are restricted to the first physical memory block.
> 
> If you can bear all this, you get pretty full support for KASAN.
> 
> Despite the limitations, it can still find bugs,
> e.g. http://patchwork.ozlabs.org/patch/1103775/
> 
> The current implementation is Radix only. I am open to extending
> it to hash at some point but I don't think it should hold up v1.
> 
> Massive thanks to mpe, who had the idea for the initial design.
> 
> Signed-off-by: Daniel Axtens <dja@axtens.net>
> 
> ---
> Changes since the rfc:
> 
>   - Boots real and virtual hardware, kvm works.
> 
>   - disabled reporting when we're checking the stack for exception
>     frames. The behaviour isn't wrong, just incompatible with KASAN.

Does this applies to / impacts PPC32 at all ?

> 
>   - Documentation!
> 
>   - Dropped old module stuff in favour of KASAN_VMALLOC.

You said in the cover that this is done to avoid having to split modules 
out of VMALLOC area. Would it be an issue to perform that split ?
I can understand it is not easy on 32 bits because vmalloc space is 
rather small, but on 64 bits don't we have enough virtual space to 
confortably split modules out of vmalloc ? The 64 bits already splits 
ioremap away from vmalloc whereas 32 bits have them merged too.

> 
> The bugs with ftrace and kuap were due to kernel bloat pushing
> prom_init calls to be done via the plt. Because we did not have
> a relocatable kernel, and they are done very early, this caused
> everything to explode. Compile with CONFIG_RELOCATABLE!
> 
> ---
>   Documentation/dev-tools/kasan.rst            |   7 +-
>   Documentation/powerpc/kasan.txt              | 111 +++++++++++++++++++
>   arch/powerpc/Kconfig                         |   4 +
>   arch/powerpc/Kconfig.debug                   |  21 ++++
>   arch/powerpc/Makefile                        |   7 ++
>   arch/powerpc/include/asm/book3s/64/radix.h   |   5 +
>   arch/powerpc/include/asm/kasan.h             |  35 +++++-
>   arch/powerpc/kernel/process.c                |   8 ++
>   arch/powerpc/kernel/prom.c                   |  57 +++++++++-
>   arch/powerpc/mm/kasan/Makefile               |   1 +
>   arch/powerpc/mm/kasan/kasan_init_book3s_64.c |  76 +++++++++++++
>   11 files changed, 326 insertions(+), 6 deletions(-)
>   create mode 100644 Documentation/powerpc/kasan.txt
>   create mode 100644 arch/powerpc/mm/kasan/kasan_init_book3s_64.c
> 
> diff --git a/Documentation/dev-tools/kasan.rst b/Documentation/dev-tools/kasan.rst
> index 35fda484a672..48d3b669e577 100644
> --- a/Documentation/dev-tools/kasan.rst
> +++ b/Documentation/dev-tools/kasan.rst
> @@ -22,7 +22,9 @@ global variables yet.
>   Tag-based KASAN is only supported in Clang and requires version 7.0.0 or later.
>   
>   Currently generic KASAN is supported for the x86_64, arm64, xtensa and s390
> -architectures, and tag-based KASAN is supported only for arm64.
> +architectures. It is also supported on powerpc for 32-bit kernels, and for
> +64-bit kernels running under the radix MMU. Tag-based KASAN is supported only
> +for arm64.

Could the 32 bits documentation stuff be a separate patch ?

>   
>   Usage
>   -----
> @@ -252,7 +254,8 @@ CONFIG_KASAN_VMALLOC
>   ~~~~~~~~~~~~~~~~~~~~
>   
>   With ``CONFIG_KASAN_VMALLOC``, KASAN can cover vmalloc space at the
> -cost of greater memory usage. Currently this is only supported on x86.
> +cost of greater memory usage. Currently this is optional on x86, and
> +required on 64-bit powerpc.
>   
>   This works by hooking into vmalloc and vmap, and dynamically
>   allocating real shadow memory to back the mappings.
> diff --git a/Documentation/powerpc/kasan.txt b/Documentation/powerpc/kasan.txt
> new file mode 100644
> index 000000000000..a5592454353b
> --- /dev/null
> +++ b/Documentation/powerpc/kasan.txt
> @@ -0,0 +1,111 @@
> +KASAN is supported on powerpc on 32-bit and 64-bit Radix only.
> +
> +32 bit support
> +==============
> +
> +KASAN is supported on both hash and nohash MMUs on 32-bit.
> +
> +The shadow area sits at the top of the kernel virtual memory space above the
> +fixmap area and occupies one eighth of the total kernel virtual memory space.
> +
> +Instrumentation of the vmalloc area is not currently supported, but modules are.
> +
> +64 bit support
> +==============
> +
> +Currently, only the radix MMU is supported. There have been versions for Book3E
> +processors floating around on the mailing list, but nothing has been merged.
> +
> +KASAN support on Book3S is interesting:

And support for others is not interesting ? :)

> +
> + - We want to be able to support inline instrumentation so as to be able to
> +   catch global and stack issues.
> +
> + - Inline instrumentation requires a fixed offset.
> +
> + - We run a lot of code at boot in real mode. This includes stuff like printk(),
> +   so it's not feasible to just disable instrumentation around it.
> +
> + - Because of our running things in real mode, the offset has to point to valid
> +   memory both in and out of real mode.
> +
> +This makes finding somewhere to put the KASAN shadow region a bit fun.
> +
> +One approach is just to give up on inline instrumentation. This way we can delay
> +all checks until after we get everything set up to our satisfaction. However,
> +we'd really like to do better.
> +
> +If we know _at compile time_ how much contiguous physical memory we have, we can
> +set aside the top 1/8th of physical memory and use that. This is a big hammer
> +and comes with 2 big consequences:
> +
> + - kernels will simply fail to boot on machines with less memory than specified
> +   when compiling.
> +
> + - kernels running on machines with more memory than specified when compiling
> +   will simply ignore the extra memory.
> +
> +If you can live with this, you get full support for KASAN.
> +
> +Tips
> +----
> +
> + - Compile with CONFIG_RELOCATABLE.
> +
> +   In development, we found boot hangs when building with ftrace and KUAP. These
> +   ended up being due to kernel bloat pushing prom_init calls to be done via the
> +   PLT. Because we did not have a relocatable kernel, and they are done very
> +   early, this caused us to jump off into somewhere invalid. Enabling relocation
> +   fixes this.
> +
> +NUMA/discontiguous physical memory
> +----------------------------------
> +
> +We currently cannot really deal with discontiguous physical memory. You are
> +restricted to the physical memory that is contiguous from physical address zero,
> +and must specify the size of that memory, not total memory, when configuring
> +your kernel.
> +
> +Discontiguous memory can occur when you have a machine with memory spread across
> +multiple nodes. For example, on a Talos II with 64GB of RAM:
> +
> + - 32GB runs from 0x0 to 0x0000_0008_0000_0000,
> + - then there's a gap,
> + - then the final 32GB runs from 0x0000_2000_0000_0000 to 0x0000_2008_0000_0000.
> +
> +This can create _significant_ issues:
> +
> + - If we try to treat the machine as having 64GB of _contiguous_ RAM, we would
> +   assume that ran from 0x0 to 0x0000_0010_0000_0000. We'd then reserve the last
> +   1/8th - 0x0000_000e_0000_0000 to 0x0000_0010_0000_0000 as the shadow
> +   region. But when we try to access any of that, we'll try to access pages that
> +   are not physically present.
> +
> + - If we try to base the shadow region size on the top address, we'll need to
> +   reserve 0x2008_0000_0000 / 8 = 0x0401_0000_0000 bytes = 4100 GB of memory,
> +   which will clearly not work on a system with 64GB of RAM.
> +
> +Therefore, you are restricted to the memory in the node starting at 0x0. For
> +this system, that's 32GB. If you specify a contiguous physical memory size
> +greater than the size of the first contiguous region of memory, the system will
> +be unable to boot or even print an error message warning you.

Can't we restrict ourselve to the first block at startup while we are in 
real mode, but then support the entire mem once we have switched on the 
MMU ?

> +
> +You can determine the layout of your system's memory by observing the messages
> +that the Radix MMU prints on boot. The Talos II discussed earlier has:
> +
> +radix-mmu: Mapped 0x0000000000000000-0x0000000040000000 with 1.00 GiB pages (exec)
> +radix-mmu: Mapped 0x0000000040000000-0x0000000800000000 with 1.00 GiB pages
> +radix-mmu: Mapped 0x0000200000000000-0x0000200800000000 with 1.00 GiB pages
> +
> +As discussed, you'd configure this system for 32768 MB.
> +
> +Another system prints:
> +
> +radix-mmu: Mapped 0x0000000000000000-0x0000000040000000 with 1.00 GiB pages (exec)
> +radix-mmu: Mapped 0x0000000040000000-0x0000002000000000 with 1.00 GiB pages
> +radix-mmu: Mapped 0x0000200000000000-0x0000202000000000 with 1.00 GiB pages
> +
> +This machine has more memory: 0x0000_0040_0000_0000 total, but only
> +0x0000_0020_0000_0000 is physically contiguous from zero, so we'd configure the
> +kernel for 131072 MB of physically contiguous memory.
> +
> diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
> index d8dcd8820369..3d6deee100e2 100644
> --- a/arch/powerpc/Kconfig
> +++ b/arch/powerpc/Kconfig
> @@ -171,6 +171,10 @@ config PPC
>   	select HAVE_ARCH_HUGE_VMAP		if PPC_BOOK3S_64 && PPC_RADIX_MMU
>   	select HAVE_ARCH_JUMP_LABEL
>   	select HAVE_ARCH_KASAN			if PPC32
> +	select HAVE_ARCH_KASAN			if PPC_BOOK3S_64 && PPC_RADIX_MMU
> +	select ARCH_HAS_KASAN_EARLY_SHADOW	if PPC_BOOK3S_64

See comment on patch 1, would be better to avoid that.

> +	select HAVE_ARCH_KASAN_VMALLOC		if PPC_BOOK3S_64
> +	select KASAN_VMALLOC			if KASAN
>   	select HAVE_ARCH_KGDB
>   	select HAVE_ARCH_MMAP_RND_BITS
>   	select HAVE_ARCH_MMAP_RND_COMPAT_BITS	if COMPAT
> diff --git a/arch/powerpc/Kconfig.debug b/arch/powerpc/Kconfig.debug
> index c59920920ddc..2d6fb7b1ba59 100644
> --- a/arch/powerpc/Kconfig.debug
> +++ b/arch/powerpc/Kconfig.debug
> @@ -394,6 +394,27 @@ config PPC_FAST_ENDIAN_SWITCH
>           help
>   	  If you're unsure what this is, say N.
>   
> +config PHYS_MEM_SIZE_FOR_KASAN
> +	int "Contiguous physical memory size for KASAN (MB)"
> +	depends on 

Drop the depend and maybe do:
	int "Contiguous physical memory size for KASAN (MB)" if KASAN && 
PPC_BOOK3S_64
	default 0

Will allow you to not enclose it into #ifdef KASAN in the code below

> +	help
> +
> +	  To get inline instrumentation support for KASAN on 64-bit Book3S
> +	  machines, you need to know how much contiguous physical memory your
> +	  system has. A shadow offset will be calculated based on this figure,
> +	  which will be compiled in to the kernel. KASAN will use this offset
> +	  to access its shadow region, which is used to verify memory accesses.
> +
> +	  If you attempt to boot on a system with less memory than you specify
> +	  here, your system will fail to boot very early in the process. If you
> +	  boot on a system with more memory than you specify, the extra memory
> +	  will wasted - it will be reserved and not used.
> +
> +	  For systems with discontiguous blocks of physical memory, specify the
> +	  size of the block starting at 0x0. You can determine this by looking
> +	  at the memory layout info printed to dmesg by the radix MMU code
> +	  early in boot. See Documentation/powerpc/kasan.txt.
> +
>   config KASAN_SHADOW_OFFSET
>   	hex
>   	depends on KASAN
> diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile
> index c345b79414a9..33e7bba4c8db 100644
> --- a/arch/powerpc/Makefile
> +++ b/arch/powerpc/Makefile
> @@ -229,6 +229,13 @@ ifdef CONFIG_476FPE_ERR46
>   		-T $(srctree)/arch/powerpc/platforms/44x/ppc476_modules.lds
>   endif
>   
> +ifdef CONFIG_KASAN
> +ifdef CONFIG_PPC_BOOK3S_64
> +# 0xa800000000000000 = 12105675798371893248
> +KASAN_SHADOW_OFFSET = $(shell echo 7 \* 1024 \* 1024 \* $(CONFIG_PHYS_MEM_SIZE_FOR_KASAN) / 8 + 12105675798371893248 | bc)
> +endif
> +endif
> +
>   # No AltiVec or VSX instructions when building kernel
>   KBUILD_CFLAGS += $(call cc-option,-mno-altivec)
>   KBUILD_CFLAGS += $(call cc-option,-mno-vsx)
> diff --git a/arch/powerpc/include/asm/book3s/64/radix.h b/arch/powerpc/include/asm/book3s/64/radix.h
> index e04a839cb5b9..4c011cc15e05 100644
> --- a/arch/powerpc/include/asm/book3s/64/radix.h
> +++ b/arch/powerpc/include/asm/book3s/64/radix.h
> @@ -35,6 +35,11 @@
>   #define RADIX_PMD_SHIFT		(PAGE_SHIFT + RADIX_PTE_INDEX_SIZE)
>   #define RADIX_PUD_SHIFT		(RADIX_PMD_SHIFT + RADIX_PMD_INDEX_SIZE)
>   #define RADIX_PGD_SHIFT		(RADIX_PUD_SHIFT + RADIX_PUD_INDEX_SIZE)
> +
> +#define R_PTRS_PER_PTE		(1 << RADIX_PTE_INDEX_SIZE)
> +#define R_PTRS_PER_PMD		(1 << RADIX_PMD_INDEX_SIZE)
> +#define R_PTRS_PER_PUD		(1 << RADIX_PUD_INDEX_SIZE)
> +

See my suggestion in patch 1.

>   /*
>    * Size of EA range mapped by our pagetables.
>    */
> diff --git a/arch/powerpc/include/asm/kasan.h b/arch/powerpc/include/asm/kasan.h
> index 296e51c2f066..d6b4028c296b 100644
> --- a/arch/powerpc/include/asm/kasan.h
> +++ b/arch/powerpc/include/asm/kasan.h
> @@ -14,13 +14,20 @@
>   
>   #ifndef __ASSEMBLY__
>   
> -#include <asm/page.h>
> +#ifdef CONFIG_KASAN
> +void kasan_init(void);
> +#else
> +static inline void kasan_init(void) { }
> +#endif
>   
>   #define KASAN_SHADOW_SCALE_SHIFT	3
>   
>   #define KASAN_SHADOW_START	(KASAN_SHADOW_OFFSET + \
>   				 (PAGE_OFFSET >> KASAN_SHADOW_SCALE_SHIFT))
>   
> +#ifdef CONFIG_PPC32
> +#include <asm/page.h>
> +
>   #define KASAN_SHADOW_OFFSET	ASM_CONST(CONFIG_KASAN_SHADOW_OFFSET)
>   
>   #define KASAN_SHADOW_END	0UL
> @@ -30,11 +37,33 @@
>   #ifdef CONFIG_KASAN
>   void kasan_early_init(void);
>   void kasan_mmu_init(void);
> -void kasan_init(void);
>   #else
> -static inline void kasan_init(void) { }
>   static inline void kasan_mmu_init(void) { }
>   #endif
> +#endif
> +
> +#ifdef CONFIG_PPC_BOOK3S_64
> +#include <asm/pgtable.h>
> +
> +/*
> + * The KASAN shadow offset is such that linear map (0xc000...) is shadowed by
> + * the last 8th of linearly mapped physical memory. This way, if the code uses
> + * 0xc addresses throughout, accesses work both in in real mode (where the top
> + * 2 bits are ignored) and outside of real mode.
> + */
> +#define KASAN_SHADOW_OFFSET ((u64)CONFIG_PHYS_MEM_SIZE_FOR_KASAN * \
> +				1024 * 1024 * 7 / 8 + 0xa800000000000000UL)

Already calculated in the Makefile, can't we reuse it ?

'X * 1024 * 1024' is usually better understood is 'X << 20' instead.


> +
> +#define KASAN_SHADOW_SIZE ((u64)CONFIG_PHYS_MEM_SIZE_FOR_KASAN * \
> +				1024 * 1024 * 1 / 8)
> +
> +extern unsigned char kasan_early_shadow_page[PAGE_SIZE];
> +
> +extern pte_t kasan_early_shadow_pte[R_PTRS_PER_PTE];
> +extern pmd_t kasan_early_shadow_pmd[R_PTRS_PER_PMD];
> +extern pud_t kasan_early_shadow_pud[R_PTRS_PER_PUD];
> +extern p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D];
> +#endif /* CONFIG_PPC_BOOK3S_64 */
>   
>   #endif /* __ASSEMBLY */
>   #endif
> diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
> index 8fc4de0d22b4..31602536e72b 100644
> --- a/arch/powerpc/kernel/process.c
> +++ b/arch/powerpc/kernel/process.c
> @@ -2097,7 +2097,14 @@ void show_stack(struct task_struct *tsk, unsigned long *stack)
>   		/*
>   		 * See if this is an exception frame.
>   		 * We look for the "regshere" marker in the current frame.
> +		 *
> +		 * KASAN may complain about this. If it is an exception frame,
> +		 * we won't have unpoisoned the stack in asm when we set the
> +		 * exception marker. If it's not an exception frame, who knows
> +		 * how things are laid out - the shadow could be in any state
> +		 * at all. Just disable KASAN reporting for now.
>   		 */
> +		kasan_disable_current();
>   		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
>   		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
>   			struct pt_regs *regs = (struct pt_regs *)
> @@ -2107,6 +2114,7 @@ void show_stack(struct task_struct *tsk, unsigned long *stack)
>   			       regs->trap, (void *)regs->nip, (void *)lr);
>   			firstframe = 1;
>   		}
> +		kasan_enable_current();
>   
>   		sp = newsp;
>   	} while (count++ < kstack_depth_to_print);
> diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c
> index 7159e791a70d..dde5f2896ab6 100644
> --- a/arch/powerpc/kernel/prom.c
> +++ b/arch/powerpc/kernel/prom.c
> @@ -71,6 +71,7 @@ unsigned long tce_alloc_start, tce_alloc_end;
>   u64 ppc64_rma_size;
>   #endif
>   static phys_addr_t first_memblock_size;
> +static phys_addr_t top_phys_addr;
>   static int __initdata boot_cpu_count;
>   
>   static int __init early_parse_mem(char *p)
> @@ -448,6 +449,21 @@ static bool validate_mem_limit(u64 base, u64 *size)
>   {
>   	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
>   
> +#ifdef CONFIG_KASAN
> +	/*
> +	 * To handle the NUMA/discontiguous memory case, don't allow a block
> +	 * to be added if it falls completely beyond the configured physical
> +	 * memory.
> +	 *
> +	 * See Documentation/powerpc/kasan.txt
> +	 */
> +	if (base >= (u64)CONFIG_PHYS_MEM_SIZE_FOR_KASAN * 1024 * 1024) {
> +		pr_warn("KASAN: not adding mem block at %llx (size %llx)",
> +			base, *size);
> +		return false;
> +	}
> +#endif
> +
>   	if (base >= max_mem)
>   		return false;
>   	if ((base + *size) > max_mem)
> @@ -571,8 +587,11 @@ void __init early_init_dt_add_memory_arch(u64 base, u64 size)
>   
>   	/* Add the chunk to the MEMBLOCK list */
>   	if (add_mem_to_memblock) {
> -		if (validate_mem_limit(base, &size))
> +		if (validate_mem_limit(base, &size)) {
>   			memblock_add(base, size);
> +			if (base + size > top_phys_addr)
> +				top_phys_addr = base + size;
> +		}
>   	}
>   }
>   
> @@ -612,6 +631,8 @@ static void __init early_reserve_mem_dt(void)
>   static void __init early_reserve_mem(void)
>   {
>   	__be64 *reserve_map;
> +	phys_addr_t kasan_shadow_start __maybe_unused;
> +	phys_addr_t kasan_memory_size __maybe_unused;

Could we avoid those uggly __maybe_unused ?

>   
>   	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
>   			fdt_off_mem_rsvmap(initial_boot_params));
> @@ -650,6 +671,40 @@ static void __init early_reserve_mem(void)
>   		return;
>   	}
>   #endif
> +
> +#if defined(CONFIG_KASAN) && defined(CONFIG_PPC_BOOK3S_64)

Would be better to do following instead of the #ifdef

if (IS_ENABLED(CONFIG_KASAN) && IS_ENABLED(CONFIG_PPC_BOOK3S_64)) {
}

This would avoid the __maybe_unused above, and would allow to valide the 
code even when CONFIG_KASAN is not selected.

This would probably require to set a default value for 
CONFIG_PHYS_MEM_SIZE_FOR_KASAN when KASAN is not set, or maybe define an 
intermediate const somewhere in some .h which takes 
CONFIG_PHYS_MEM_SIZE_FOR_KASAN when KASAN is there and 0 when KASAN is 
not there

> +	kasan_memory_size = ((phys_addr_t)CONFIG_PHYS_MEM_SIZE_FOR_KASAN
> +		* 1024 * 1024);
> +
> +	if (top_phys_addr < kasan_memory_size) {
> +		/*
> +		 * We are doomed. Attempts to call e.g. panic() are likely to
> +		 * fail because they call out into instrumented code, which
> +		 * will almost certainly access memory beyond the end of
> +		 * physical memory. Hang here so that at least the NIP points
> +		 * somewhere that will help you debug it if you look at it in
> +		 * qemu.
> +		 */
> +		while (true)
> +			;

A bit gross ?


> +	} else if (top_phys_addr > kasan_memory_size) {
> +		/* print a biiiig warning in hopes people notice */
> +		pr_err("===========================================\n"
> +			"Physical memory exceeds compiled-in maximum!\n"
> +			"This kernel was compiled for KASAN with %u MB physical memory.\n"
> +			"The actual physical memory detected is %llu MB.\n"
> +			"Memory above the compiled limit will not be used!\n"
> +			"===========================================\n",
> +			CONFIG_PHYS_MEM_SIZE_FOR_KASAN,
> +			top_phys_addr / (1024 * 1024));
> +	}
> +
> +	kasan_shadow_start = _ALIGN_DOWN(kasan_memory_size * 7 / 8, PAGE_SIZE);
> +	DBG("reserving %llx -> %llx for KASAN",
> +	    kasan_shadow_start, top_phys_addr);
> +	memblock_reserve(kasan_shadow_start,
> +			 top_phys_addr - kasan_shadow_start);
> +#endif
>   }
>   
>   #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
> diff --git a/arch/powerpc/mm/kasan/Makefile b/arch/powerpc/mm/kasan/Makefile
> index 6577897673dd..ff8143ba1e4d 100644
> --- a/arch/powerpc/mm/kasan/Makefile
> +++ b/arch/powerpc/mm/kasan/Makefile
> @@ -3,3 +3,4 @@
>   KASAN_SANITIZE := n
>   
>   obj-$(CONFIG_PPC32)           += kasan_init_32.o
> +obj-$(CONFIG_PPC_BOOK3S_64)   += kasan_init_book3s_64.o
> diff --git a/arch/powerpc/mm/kasan/kasan_init_book3s_64.c b/arch/powerpc/mm/kasan/kasan_init_book3s_64.c

In the same spirit as what was done in other mm subdirs, could we rename 
those files to:
init_32.o
init_book3s64.o


> new file mode 100644
> index 000000000000..fafda3d5e9a3
> --- /dev/null
> +++ b/arch/powerpc/mm/kasan/kasan_init_book3s_64.c
> @@ -0,0 +1,76 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * KASAN for 64-bit Book3S powerpc
> + *
> + * Copyright (C) 2019 IBM Corporation
> + * Author: Daniel Axtens <dja@axtens.net>
> + */
> +
> +#define DISABLE_BRANCH_PROFILING
> +
> +#include <linux/kasan.h>
> +#include <linux/printk.h>
> +#include <linux/sched/task.h>
> +#include <asm/pgalloc.h>
> +
> +unsigned char kasan_early_shadow_page[PAGE_SIZE] __page_aligned_bss;

What's the difference with what's defined in the kasan generic part and 
that you have opted out in first patch ?

> +
> +pte_t kasan_early_shadow_pte[R_PTRS_PER_PTE] __page_aligned_bss;
> +pmd_t kasan_early_shadow_pmd[R_PTRS_PER_PMD] __page_aligned_bss;
> +pud_t kasan_early_shadow_pud[R_PTRS_PER_PUD] __page_aligned_bss;
> +p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D] __page_aligned_bss;

See my suggestion for those in patch 1.

> +
> +void __init kasan_init(void)
> +{
> +	int i;
> +	void *k_start = kasan_mem_to_shadow((void *)RADIX_KERN_VIRT_START);
> +	void *k_end = kasan_mem_to_shadow((void *)RADIX_VMEMMAP_END);
> +
> +	unsigned long pte_val = __pa(kasan_early_shadow_page)
> +					| pgprot_val(PAGE_KERNEL) | _PAGE_PTE;
> +
> +	if (!early_radix_enabled())
> +		panic("KASAN requires radix!");
> +
> +	for (i = 0; i < PTRS_PER_PTE; i++)
> +		kasan_early_shadow_pte[i] = __pte(pte_val);

Shouldn't you use __set_pte_at() here ?

> +
> +	for (i = 0; i < PTRS_PER_PMD; i++)
> +		pmd_populate_kernel(&init_mm, &kasan_early_shadow_pmd[i],
> +				    kasan_early_shadow_pte);
> +
> +	for (i = 0; i < PTRS_PER_PUD; i++)
> +		pud_populate(&init_mm, &kasan_early_shadow_pud[i],
> +			     kasan_early_shadow_pmd);
> +
> +
> +	memset(kasan_mem_to_shadow((void *)PAGE_OFFSET), KASAN_SHADOW_INIT,
> +		KASAN_SHADOW_SIZE);
> +
> +	kasan_populate_early_shadow(
> +		kasan_mem_to_shadow((void *)RADIX_KERN_VIRT_START),
> +		kasan_mem_to_shadow((void *)RADIX_VMALLOC_START));
> +
> +	/* leave a hole here for vmalloc */
> +
> +	kasan_populate_early_shadow(
> +		kasan_mem_to_shadow((void *)RADIX_VMALLOC_END),
> +		kasan_mem_to_shadow((void *)RADIX_VMEMMAP_END));
> +
> +	flush_tlb_kernel_range((unsigned long)k_start, (unsigned long)k_end);
> +
> +	/* mark early shadow region as RO and wipe */
> +	for (i = 0; i < PTRS_PER_PTE; i++)
> +		__set_pte_at(&init_mm, (unsigned long)kasan_early_shadow_page,
> +			&kasan_early_shadow_pte[i],
> +			pfn_pte(virt_to_pfn(kasan_early_shadow_page),
> +			__pgprot(_PAGE_PTE | _PAGE_KERNEL_RO | _PAGE_BASE)),

Isn't there an already existing val for the above set of flags ?

> +			0);
> +	memset(kasan_early_shadow_page, 0, PAGE_SIZE);
> +
> +	kasan_init_tags();

You said in the doc that only arm supports that ...

> +
> +	/* Enable error messages */
> +	init_task.kasan_depth = 0;
> +	pr_info("KASAN init done (64-bit Book3S heavyweight mode)\n");
> +}
> 

Christophe

Hi Daniel,

Le 07/08/2019 à 18:34, Christophe Leroy a écrit :
> 
> 
> Le 07/08/2019 à 01:38, Daniel Axtens a écrit :
>> KASAN support on powerpc64 is interesting:
>>
>>   - We want to be able to support inline instrumentation so as to be
>>     able to catch global and stack issues.
>>
>>   - We run a lot of code at boot in real mode. This includes stuff like
>>     printk(), so it's not feasible to just disable instrumentation
>>     around it.
> 
> Have you definitely given up the idea of doing a standard implementation 
> of KASAN like other 64 bits arches have done ?
> 
> Isn't it possible to setup an early 1:1 mapping and go in virtual mode 
> earlier ? What is so different between book3s64 and book3e64 ?
> On book3e64, we've been able to setup KASAN before printing anything 
> (except when using EARLY_DEBUG). Isn't it feasible on book3s64 too ?
> 

I looked at it once more, and cannot find that "We run a lot of code at 
boot in real mode. This includes stuff like printk()".

Can you provide exemples ?

AFAICS, there are two things which are run in real mode at boot:
1/ prom_init() in kernel/prom_init.c
2/ early_setup() in kernel/setup_64.c

1/ KASAN is already inhibited for prom_init(), and prom_init() only uses 
prom_printf() to display stuff.
2/ early_setup() only call a subset of simple functions. By regrouping 
things in a new file called early_64.c as done for PPC32 with 
early_32.c, we can easily inhibit kasan for those few stuff. printk() is 
not used there either, there is even a comment at the startup of 
early_setup() telling /* -------- printk is _NOT_ safe to use here ! 
------- */. The only things that perform display is the function 
udbg_printf(), which is called only when DEBUG is set and which is 
linked to CONFIG_PPC_EARLY_DEBUG. We already discussed that and agreed 
that CONFIG_PPC_EARLY_DEBUG could be made exclusive of CONFIG_KASAN.

Once early_setup() has run, BOOK3S64 goes in virtual mode, just like 
BOOK3E does.

What am I missing ?

Thanks
Christophe

Christophe Leroy <christophe.leroy@c-s.fr> writes:

> Le 07/08/2019 à 01:38, Daniel Axtens a écrit :
>> KASAN support on powerpc64 is interesting:
>> 
>>   - We want to be able to support inline instrumentation so as to be
>>     able to catch global and stack issues.
>> 
>>   - We run a lot of code at boot in real mode. This includes stuff like
>>     printk(), so it's not feasible to just disable instrumentation
>>     around it.
>
> Have you definitely given up the idea of doing a standard implementation 
> of KASAN like other 64 bits arches have done ?
>
> Isn't it possible to setup an early 1:1 mapping and go in virtual mode 
> earlier ? What is so different between book3s64 and book3e64 ?
> On book3e64, we've been able to setup KASAN before printing anything 
> (except when using EARLY_DEBUG). Isn't it feasible on book3s64 too ?

So I got this pretty wrong when trying to explain it. The problem isn't
that we run the code in boot as I said, it's that a bunch of the KVM
code runs in real mode.

>>   - disabled reporting when we're checking the stack for exception
>>     frames. The behaviour isn't wrong, just incompatible with KASAN.
>
> Does this applies to / impacts PPC32 at all ?

It should. I found that when doing stack walks, the code would touch
memory that KASAN hadn't unpoisioned. I'm a bit surprised you haven't
seen it arise, tbh.

>>   - Dropped old module stuff in favour of KASAN_VMALLOC.
>
> You said in the cover that this is done to avoid having to split modules 
> out of VMALLOC area. Would it be an issue to perform that split ?
> I can understand it is not easy on 32 bits because vmalloc space is 
> rather small, but on 64 bits don't we have enough virtual space to 
> confortably split modules out of vmalloc ? The 64 bits already splits 
> ioremap away from vmalloc whereas 32 bits have them merged too.

I could have done this. Maybe I should have done this. But now I have
done vmalloc space support.

>> The bugs with ftrace and kuap were due to kernel bloat pushing
>> prom_init calls to be done via the plt. Because we did not have
>> a relocatable kernel, and they are done very early, this caused
>> everything to explode. Compile with CONFIG_RELOCATABLE!
>> 
>> ---
>>   Documentation/dev-tools/kasan.rst            |   7 +-
>>   Documentation/powerpc/kasan.txt              | 111 +++++++++++++++++++
>>   arch/powerpc/Kconfig                         |   4 +
>>   arch/powerpc/Kconfig.debug                   |  21 ++++
>>   arch/powerpc/Makefile                        |   7 ++
>>   arch/powerpc/include/asm/book3s/64/radix.h   |   5 +
>>   arch/powerpc/include/asm/kasan.h             |  35 +++++-
>>   arch/powerpc/kernel/process.c                |   8 ++
>>   arch/powerpc/kernel/prom.c                   |  57 +++++++++-
>>   arch/powerpc/mm/kasan/Makefile               |   1 +
>>   arch/powerpc/mm/kasan/kasan_init_book3s_64.c |  76 +++++++++++++
>>   11 files changed, 326 insertions(+), 6 deletions(-)
>>   create mode 100644 Documentation/powerpc/kasan.txt
>>   create mode 100644 arch/powerpc/mm/kasan/kasan_init_book3s_64.c
>> 
>> diff --git a/Documentation/dev-tools/kasan.rst b/Documentation/dev-tools/kasan.rst
>> index 35fda484a672..48d3b669e577 100644
>> --- a/Documentation/dev-tools/kasan.rst
>> +++ b/Documentation/dev-tools/kasan.rst
>> @@ -22,7 +22,9 @@ global variables yet.
>>   Tag-based KASAN is only supported in Clang and requires version 7.0.0 or later.
>>   
>>   Currently generic KASAN is supported for the x86_64, arm64, xtensa and s390
>> -architectures, and tag-based KASAN is supported only for arm64.
>> +architectures. It is also supported on powerpc for 32-bit kernels, and for
>> +64-bit kernels running under the radix MMU. Tag-based KASAN is supported only
>> +for arm64.
>
> Could the 32 bits documentation stuff be a separate patch ?

Done.
>
>>   
>>   Usage
>>   -----
>> @@ -252,7 +254,8 @@ CONFIG_KASAN_VMALLOC
>>   ~~~~~~~~~~~~~~~~~~~~
>>   
>>   With ``CONFIG_KASAN_VMALLOC``, KASAN can cover vmalloc space at the
>> -cost of greater memory usage. Currently this is only supported on x86.
>> +cost of greater memory usage. Currently this is optional on x86, and
>> +required on 64-bit powerpc.
>>   
>>   This works by hooking into vmalloc and vmap, and dynamically
>>   allocating real shadow memory to back the mappings.
>> diff --git a/Documentation/powerpc/kasan.txt b/Documentation/powerpc/kasan.txt
>> new file mode 100644
>> index 000000000000..a5592454353b
>> --- /dev/null
>> +++ b/Documentation/powerpc/kasan.txt
>> @@ -0,0 +1,111 @@
>> +KASAN is supported on powerpc on 32-bit and 64-bit Radix only.
>> +
>> +32 bit support
>> +==============
>> +
>> +KASAN is supported on both hash and nohash MMUs on 32-bit.
>> +
>> +The shadow area sits at the top of the kernel virtual memory space above the
>> +fixmap area and occupies one eighth of the total kernel virtual memory space.
>> +
>> +Instrumentation of the vmalloc area is not currently supported, but modules are.
>> +
>> +64 bit support
>> +==============
>> +
>> +Currently, only the radix MMU is supported. There have been versions for Book3E
>> +processors floating around on the mailing list, but nothing has been merged.
>> +
>> +KASAN support on Book3S is interesting:
>
> And support for others is not interesting ? :)
>
:) I've reworded it to be a bit more professional.

>> +
>> + - We want to be able to support inline instrumentation so as to be able to
>> +   catch global and stack issues.
>> +
>> + - Inline instrumentation requires a fixed offset.
>> +
>> + - We run a lot of code at boot in real mode. This includes stuff like printk(),
>> +   so it's not feasible to just disable instrumentation around it.
>> +
>> + - Because of our running things in real mode, the offset has to point to valid
>> +   memory both in and out of real mode.
>> +
>> +This makes finding somewhere to put the KASAN shadow region a bit fun.
>> +
>> +One approach is just to give up on inline instrumentation. This way we can delay
>> +all checks until after we get everything set up to our satisfaction. However,
>> +we'd really like to do better.
>> +
>> +If we know _at compile time_ how much contiguous physical memory we have, we can
>> +set aside the top 1/8th of physical memory and use that. This is a big hammer
>> +and comes with 2 big consequences:
>> +
>> + - kernels will simply fail to boot on machines with less memory than specified
>> +   when compiling.
>> +
>> + - kernels running on machines with more memory than specified when compiling
>> +   will simply ignore the extra memory.
>> +
>> +If you can live with this, you get full support for KASAN.
>> +
>> +Tips
>> +----
>> +
>> + - Compile with CONFIG_RELOCATABLE.
>> +
>> +   In development, we found boot hangs when building with ftrace and KUAP. These
>> +   ended up being due to kernel bloat pushing prom_init calls to be done via the
>> +   PLT. Because we did not have a relocatable kernel, and they are done very
>> +   early, this caused us to jump off into somewhere invalid. Enabling relocation
>> +   fixes this.
>> +
>> +NUMA/discontiguous physical memory
>> +----------------------------------
>> +
>> +We currently cannot really deal with discontiguous physical memory. You are
>> +restricted to the physical memory that is contiguous from physical address zero,
>> +and must specify the size of that memory, not total memory, when configuring
>> +your kernel.
>> +
>> +Discontiguous memory can occur when you have a machine with memory spread across
>> +multiple nodes. For example, on a Talos II with 64GB of RAM:
>> +
>> + - 32GB runs from 0x0 to 0x0000_0008_0000_0000,
>> + - then there's a gap,
>> + - then the final 32GB runs from 0x0000_2000_0000_0000 to 0x0000_2008_0000_0000.
>> +
>> +This can create _significant_ issues:
>> +
>> + - If we try to treat the machine as having 64GB of _contiguous_ RAM, we would
>> +   assume that ran from 0x0 to 0x0000_0010_0000_0000. We'd then reserve the last
>> +   1/8th - 0x0000_000e_0000_0000 to 0x0000_0010_0000_0000 as the shadow
>> +   region. But when we try to access any of that, we'll try to access pages that
>> +   are not physically present.
>> +
>> + - If we try to base the shadow region size on the top address, we'll need to
>> +   reserve 0x2008_0000_0000 / 8 = 0x0401_0000_0000 bytes = 4100 GB of memory,
>> +   which will clearly not work on a system with 64GB of RAM.
>> +
>> +Therefore, you are restricted to the memory in the node starting at 0x0. For
>> +this system, that's 32GB. If you specify a contiguous physical memory size
>> +greater than the size of the first contiguous region of memory, the system will
>> +be unable to boot or even print an error message warning you.
>
> Can't we restrict ourselve to the first block at startup while we are in 
> real mode, but then support the entire mem once we have switched on the 
> MMU ?

We could only do this if we could guarantee that all memory accessed by
KVM real mode handlers was in the the first block. I don't think I can
guarantee tihs.

>> +
>> +You can determine the layout of your system's memory by observing the messages
>> +that the Radix MMU prints on boot. The Talos II discussed earlier has:
>> +
>> +radix-mmu: Mapped 0x0000000000000000-0x0000000040000000 with 1.00 GiB pages (exec)
>> +radix-mmu: Mapped 0x0000000040000000-0x0000000800000000 with 1.00 GiB pages
>> +radix-mmu: Mapped 0x0000200000000000-0x0000200800000000 with 1.00 GiB pages
>> +
>> +As discussed, you'd configure this system for 32768 MB.
>> +
>> +Another system prints:
>> +
>> +radix-mmu: Mapped 0x0000000000000000-0x0000000040000000 with 1.00 GiB pages (exec)
>> +radix-mmu: Mapped 0x0000000040000000-0x0000002000000000 with 1.00 GiB pages
>> +radix-mmu: Mapped 0x0000200000000000-0x0000202000000000 with 1.00 GiB pages
>> +
>> +This machine has more memory: 0x0000_0040_0000_0000 total, but only
>> +0x0000_0020_0000_0000 is physically contiguous from zero, so we'd configure the
>> +kernel for 131072 MB of physically contiguous memory.
>> +
>> diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
>> index d8dcd8820369..3d6deee100e2 100644
>> --- a/arch/powerpc/Kconfig
>> +++ b/arch/powerpc/Kconfig
>> @@ -171,6 +171,10 @@ config PPC
>>   	select HAVE_ARCH_HUGE_VMAP		if PPC_BOOK3S_64 && PPC_RADIX_MMU
>>   	select HAVE_ARCH_JUMP_LABEL
>>   	select HAVE_ARCH_KASAN			if PPC32
>> +	select HAVE_ARCH_KASAN			if PPC_BOOK3S_64 && PPC_RADIX_MMU
>> +	select ARCH_HAS_KASAN_EARLY_SHADOW	if PPC_BOOK3S_64
>
> See comment on patch 1, would be better to avoid that.
>
>> +	select HAVE_ARCH_KASAN_VMALLOC		if PPC_BOOK3S_64
>> +	select KASAN_VMALLOC			if KASAN
>>   	select HAVE_ARCH_KGDB
>>   	select HAVE_ARCH_MMAP_RND_BITS
>>   	select HAVE_ARCH_MMAP_RND_COMPAT_BITS	if COMPAT
>> diff --git a/arch/powerpc/Kconfig.debug b/arch/powerpc/Kconfig.debug
>> index c59920920ddc..2d6fb7b1ba59 100644
>> --- a/arch/powerpc/Kconfig.debug
>> +++ b/arch/powerpc/Kconfig.debug
>> @@ -394,6 +394,27 @@ config PPC_FAST_ENDIAN_SWITCH
>>           help
>>   	  If you're unsure what this is, say N.
>>   
>> +config PHYS_MEM_SIZE_FOR_KASAN
>> +	int "Contiguous physical memory size for KASAN (MB)"
>> +	depends on 
>
> Drop the depend and maybe do:
> 	int "Contiguous physical memory size for KASAN (MB)" if KASAN && 
> PPC_BOOK3S_64
> 	default 0
>

Done.

> Will allow you to not enclose it into #ifdef KASAN in the code below
>
>> +	help
>> +
>> +	  To get inline instrumentation support for KASAN on 64-bit Book3S
>> +	  machines, you need to know how much contiguous physical memory your
>> +	  system has. A shadow offset will be calculated based on this figure,
>> +	  which will be compiled in to the kernel. KASAN will use this offset
>> +	  to access its shadow region, which is used to verify memory accesses.
>> +
>> +	  If you attempt to boot on a system with less memory than you specify
>> +	  here, your system will fail to boot very early in the process. If you
>> +	  boot on a system with more memory than you specify, the extra memory
>> +	  will wasted - it will be reserved and not used.
>> +
>> +	  For systems with discontiguous blocks of physical memory, specify the
>> +	  size of the block starting at 0x0. You can determine this by looking
>> +	  at the memory layout info printed to dmesg by the radix MMU code
>> +	  early in boot. See Documentation/powerpc/kasan.txt.
>> +
>>   config KASAN_SHADOW_OFFSET
>>   	hex
>>   	depends on KASAN
>> diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile
>> index c345b79414a9..33e7bba4c8db 100644
>> --- a/arch/powerpc/Makefile
>> +++ b/arch/powerpc/Makefile
>> @@ -229,6 +229,13 @@ ifdef CONFIG_476FPE_ERR46
>>   		-T $(srctree)/arch/powerpc/platforms/44x/ppc476_modules.lds
>>   endif
>>   
>> +ifdef CONFIG_KASAN
>> +ifdef CONFIG_PPC_BOOK3S_64
>> +# 0xa800000000000000 = 12105675798371893248
>> +KASAN_SHADOW_OFFSET = $(shell echo 7 \* 1024 \* 1024 \* $(CONFIG_PHYS_MEM_SIZE_FOR_KASAN) / 8 + 12105675798371893248 | bc)
>> +endif
>> +endif
>> +
>>   # No AltiVec or VSX instructions when building kernel
>>   KBUILD_CFLAGS += $(call cc-option,-mno-altivec)
>>   KBUILD_CFLAGS += $(call cc-option,-mno-vsx)
>> diff --git a/arch/powerpc/include/asm/book3s/64/radix.h b/arch/powerpc/include/asm/book3s/64/radix.h
>> index e04a839cb5b9..4c011cc15e05 100644
>> --- a/arch/powerpc/include/asm/book3s/64/radix.h
>> +++ b/arch/powerpc/include/asm/book3s/64/radix.h
>> @@ -35,6 +35,11 @@
>>   #define RADIX_PMD_SHIFT		(PAGE_SHIFT + RADIX_PTE_INDEX_SIZE)
>>   #define RADIX_PUD_SHIFT		(RADIX_PMD_SHIFT + RADIX_PMD_INDEX_SIZE)
>>   #define RADIX_PGD_SHIFT		(RADIX_PUD_SHIFT + RADIX_PUD_INDEX_SIZE)
>> +
>> +#define R_PTRS_PER_PTE		(1 << RADIX_PTE_INDEX_SIZE)
>> +#define R_PTRS_PER_PMD		(1 << RADIX_PMD_INDEX_SIZE)
>> +#define R_PTRS_PER_PUD		(1 << RADIX_PUD_INDEX_SIZE)
>> +
>
> See my suggestion in patch 1.
>
>>   /*
>>    * Size of EA range mapped by our pagetables.
>>    */
>> diff --git a/arch/powerpc/include/asm/kasan.h b/arch/powerpc/include/asm/kasan.h
>> index 296e51c2f066..d6b4028c296b 100644
>> --- a/arch/powerpc/include/asm/kasan.h
>> +++ b/arch/powerpc/include/asm/kasan.h
>> @@ -14,13 +14,20 @@
>>   
>>   #ifndef __ASSEMBLY__
>>   
>> -#include <asm/page.h>
>> +#ifdef CONFIG_KASAN
>> +void kasan_init(void);
>> +#else
>> +static inline void kasan_init(void) { }
>> +#endif
>>   
>>   #define KASAN_SHADOW_SCALE_SHIFT	3
>>   
>>   #define KASAN_SHADOW_START	(KASAN_SHADOW_OFFSET + \
>>   				 (PAGE_OFFSET >> KASAN_SHADOW_SCALE_SHIFT))
>>   
>> +#ifdef CONFIG_PPC32
>> +#include <asm/page.h>
>> +
>>   #define KASAN_SHADOW_OFFSET	ASM_CONST(CONFIG_KASAN_SHADOW_OFFSET)
>>   
>>   #define KASAN_SHADOW_END	0UL
>> @@ -30,11 +37,33 @@
>>   #ifdef CONFIG_KASAN
>>   void kasan_early_init(void);
>>   void kasan_mmu_init(void);
>> -void kasan_init(void);
>>   #else
>> -static inline void kasan_init(void) { }
>>   static inline void kasan_mmu_init(void) { }
>>   #endif
>> +#endif
>> +
>> +#ifdef CONFIG_PPC_BOOK3S_64
>> +#include <asm/pgtable.h>
>> +
>> +/*
>> + * The KASAN shadow offset is such that linear map (0xc000...) is shadowed by
>> + * the last 8th of linearly mapped physical memory. This way, if the code uses
>> + * 0xc addresses throughout, accesses work both in in real mode (where the top
>> + * 2 bits are ignored) and outside of real mode.
>> + */
>> +#define KASAN_SHADOW_OFFSET ((u64)CONFIG_PHYS_MEM_SIZE_FOR_KASAN * \
>> +				1024 * 1024 * 7 / 8 + 0xa800000000000000UL)
>
> Already calculated in the Makefile, can't we reuse it ?
>
> 'X * 1024 * 1024' is usually better understood is 'X << 20' instead.
>
>
>> +
>> +#define KASAN_SHADOW_SIZE ((u64)CONFIG_PHYS_MEM_SIZE_FOR_KASAN * \
>> +				1024 * 1024 * 1 / 8)
>> +
>> +extern unsigned char kasan_early_shadow_page[PAGE_SIZE];
>> +
>> +extern pte_t kasan_early_shadow_pte[R_PTRS_PER_PTE];
>> +extern pmd_t kasan_early_shadow_pmd[R_PTRS_PER_PMD];
>> +extern pud_t kasan_early_shadow_pud[R_PTRS_PER_PUD];
>> +extern p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D];
>> +#endif /* CONFIG_PPC_BOOK3S_64 */
>>   
>>   #endif /* __ASSEMBLY */
>>   #endif
>> diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
>> index 8fc4de0d22b4..31602536e72b 100644
>> --- a/arch/powerpc/kernel/process.c
>> +++ b/arch/powerpc/kernel/process.c
>> @@ -2097,7 +2097,14 @@ void show_stack(struct task_struct *tsk, unsigned long *stack)
>>   		/*
>>   		 * See if this is an exception frame.
>>   		 * We look for the "regshere" marker in the current frame.
>> +		 *
>> +		 * KASAN may complain about this. If it is an exception frame,
>> +		 * we won't have unpoisoned the stack in asm when we set the
>> +		 * exception marker. If it's not an exception frame, who knows
>> +		 * how things are laid out - the shadow could be in any state
>> +		 * at all. Just disable KASAN reporting for now.
>>   		 */
>> +		kasan_disable_current();
>>   		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
>>   		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
>>   			struct pt_regs *regs = (struct pt_regs *)
>> @@ -2107,6 +2114,7 @@ void show_stack(struct task_struct *tsk, unsigned long *stack)
>>   			       regs->trap, (void *)regs->nip, (void *)lr);
>>   			firstframe = 1;
>>   		}
>> +		kasan_enable_current();
>>   
>>   		sp = newsp;
>>   	} while (count++ < kstack_depth_to_print);
>> diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c
>> index 7159e791a70d..dde5f2896ab6 100644
>> --- a/arch/powerpc/kernel/prom.c
>> +++ b/arch/powerpc/kernel/prom.c
>> @@ -71,6 +71,7 @@ unsigned long tce_alloc_start, tce_alloc_end;
>>   u64 ppc64_rma_size;
>>   #endif
>>   static phys_addr_t first_memblock_size;
>> +static phys_addr_t top_phys_addr;
>>   static int __initdata boot_cpu_count;
>>   
>>   static int __init early_parse_mem(char *p)
>> @@ -448,6 +449,21 @@ static bool validate_mem_limit(u64 base, u64 *size)
>>   {
>>   	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
>>   
>> +#ifdef CONFIG_KASAN
>> +	/*
>> +	 * To handle the NUMA/discontiguous memory case, don't allow a block
>> +	 * to be added if it falls completely beyond the configured physical
>> +	 * memory.
>> +	 *
>> +	 * See Documentation/powerpc/kasan.txt
>> +	 */
>> +	if (base >= (u64)CONFIG_PHYS_MEM_SIZE_FOR_KASAN * 1024 * 1024) {
>> +		pr_warn("KASAN: not adding mem block at %llx (size %llx)",
>> +			base, *size);
>> +		return false;
>> +	}
>> +#endif
>> +
>>   	if (base >= max_mem)
>>   		return false;
>>   	if ((base + *size) > max_mem)
>> @@ -571,8 +587,11 @@ void __init early_init_dt_add_memory_arch(u64 base, u64 size)
>>   
>>   	/* Add the chunk to the MEMBLOCK list */
>>   	if (add_mem_to_memblock) {
>> -		if (validate_mem_limit(base, &size))
>> +		if (validate_mem_limit(base, &size)) {
>>   			memblock_add(base, size);
>> +			if (base + size > top_phys_addr)
>> +				top_phys_addr = base + size;
>> +		}
>>   	}
>>   }
>>   
>> @@ -612,6 +631,8 @@ static void __init early_reserve_mem_dt(void)
>>   static void __init early_reserve_mem(void)
>>   {
>>   	__be64 *reserve_map;
>> +	phys_addr_t kasan_shadow_start __maybe_unused;
>> +	phys_addr_t kasan_memory_size __maybe_unused;
>
> Could we avoid those uggly __maybe_unused ?
>
>>   
>>   	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
>>   			fdt_off_mem_rsvmap(initial_boot_params));
>> @@ -650,6 +671,40 @@ static void __init early_reserve_mem(void)
>>   		return;
>>   	}
>>   #endif
>> +
>> +#if defined(CONFIG_KASAN) && defined(CONFIG_PPC_BOOK3S_64)
>
> Would be better to do following instead of the #ifdef
>
> if (IS_ENABLED(CONFIG_KASAN) && IS_ENABLED(CONFIG_PPC_BOOK3S_64)) {
> }
>
> This would avoid the __maybe_unused above, and would allow to valide the 
> code even when CONFIG_KASAN is not selected.
>
> This would probably require to set a default value for 
> CONFIG_PHYS_MEM_SIZE_FOR_KASAN when KASAN is not set, or maybe define an 
> intermediate const somewhere in some .h which takes 
> CONFIG_PHYS_MEM_SIZE_FOR_KASAN when KASAN is there and 0 when KASAN is 
> not there
>
Done.

>> +	kasan_memory_size = ((phys_addr_t)CONFIG_PHYS_MEM_SIZE_FOR_KASAN
>> +		* 1024 * 1024);
>> +
>> +	if (top_phys_addr < kasan_memory_size) {
>> +		/*
>> +		 * We are doomed. Attempts to call e.g. panic() are likely to
>> +		 * fail because they call out into instrumented code, which
>> +		 * will almost certainly access memory beyond the end of
>> +		 * physical memory. Hang here so that at least the NIP points
>> +		 * somewhere that will help you debug it if you look at it in
>> +		 * qemu.
>> +		 */
>> +		while (true)
>> +			;
>
> A bit gross ?
>

It is, but I don't know how to do it better. I don't want to panic(), as
explained... What did you have in mind?

>
>> +	} else if (top_phys_addr > kasan_memory_size) {
>> +		/* print a biiiig warning in hopes people notice */
>> +		pr_err("===========================================\n"
>> +			"Physical memory exceeds compiled-in maximum!\n"
>> +			"This kernel was compiled for KASAN with %u MB physical memory.\n"
>> +			"The actual physical memory detected is %llu MB.\n"
>> +			"Memory above the compiled limit will not be used!\n"
>> +			"===========================================\n",
>> +			CONFIG_PHYS_MEM_SIZE_FOR_KASAN,
>> +			top_phys_addr / (1024 * 1024));
>> +	}
>> +
>> +	kasan_shadow_start = _ALIGN_DOWN(kasan_memory_size * 7 / 8, PAGE_SIZE);
>> +	DBG("reserving %llx -> %llx for KASAN",
>> +	    kasan_shadow_start, top_phys_addr);
>> +	memblock_reserve(kasan_shadow_start,
>> +			 top_phys_addr - kasan_shadow_start);
>> +#endif
>>   }
>>   
>>   #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
>> diff --git a/arch/powerpc/mm/kasan/Makefile b/arch/powerpc/mm/kasan/Makefile
>> index 6577897673dd..ff8143ba1e4d 100644
>> --- a/arch/powerpc/mm/kasan/Makefile
>> +++ b/arch/powerpc/mm/kasan/Makefile
>> @@ -3,3 +3,4 @@
>>   KASAN_SANITIZE := n
>>   
>>   obj-$(CONFIG_PPC32)           += kasan_init_32.o
>> +obj-$(CONFIG_PPC_BOOK3S_64)   += kasan_init_book3s_64.o
>> diff --git a/arch/powerpc/mm/kasan/kasan_init_book3s_64.c b/arch/powerpc/mm/kasan/kasan_init_book3s_64.c
>
> In the same spirit as what was done in other mm subdirs, could we rename 
> those files to:
> init_32.o
> init_book3s64.o

Done.

>
>> new file mode 100644
>> index 000000000000..fafda3d5e9a3
>> --- /dev/null
>> +++ b/arch/powerpc/mm/kasan/kasan_init_book3s_64.c
>> @@ -0,0 +1,76 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * KASAN for 64-bit Book3S powerpc
>> + *
>> + * Copyright (C) 2019 IBM Corporation
>> + * Author: Daniel Axtens <dja@axtens.net>
>> + */
>> +
>> +#define DISABLE_BRANCH_PROFILING
>> +
>> +#include <linux/kasan.h>
>> +#include <linux/printk.h>
>> +#include <linux/sched/task.h>
>> +#include <asm/pgalloc.h>
>> +
>> +unsigned char kasan_early_shadow_page[PAGE_SIZE] __page_aligned_bss;
>
> What's the difference with what's defined in the kasan generic part and 
> that you have opted out in first patch ?
>
>> +
>> +pte_t kasan_early_shadow_pte[R_PTRS_PER_PTE] __page_aligned_bss;
>> +pmd_t kasan_early_shadow_pmd[R_PTRS_PER_PMD] __page_aligned_bss;
>> +pud_t kasan_early_shadow_pud[R_PTRS_PER_PUD] __page_aligned_bss;
>> +p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D] __page_aligned_bss;
>
> See my suggestion for those in patch 1.
>
>> +
>> +void __init kasan_init(void)
>> +{
>> +	int i;
>> +	void *k_start = kasan_mem_to_shadow((void *)RADIX_KERN_VIRT_START);
>> +	void *k_end = kasan_mem_to_shadow((void *)RADIX_VMEMMAP_END);
>> +
>> +	unsigned long pte_val = __pa(kasan_early_shadow_page)
>> +					| pgprot_val(PAGE_KERNEL) | _PAGE_PTE;
>> +
>> +	if (!early_radix_enabled())
>> +		panic("KASAN requires radix!");
>> +
>> +	for (i = 0; i < PTRS_PER_PTE; i++)
>> +		kasan_early_shadow_pte[i] = __pte(pte_val);
>
> Shouldn't you use __set_pte_at() here ?
>
Yes, done.

>> +
>> +	for (i = 0; i < PTRS_PER_PMD; i++)
>> +		pmd_populate_kernel(&init_mm, &kasan_early_shadow_pmd[i],
>> +				    kasan_early_shadow_pte);
>> +
>> +	for (i = 0; i < PTRS_PER_PUD; i++)
>> +		pud_populate(&init_mm, &kasan_early_shadow_pud[i],
>> +			     kasan_early_shadow_pmd);
>> +
>> +
>> +	memset(kasan_mem_to_shadow((void *)PAGE_OFFSET), KASAN_SHADOW_INIT,
>> +		KASAN_SHADOW_SIZE);
>> +
>> +	kasan_populate_early_shadow(
>> +		kasan_mem_to_shadow((void *)RADIX_KERN_VIRT_START),
>> +		kasan_mem_to_shadow((void *)RADIX_VMALLOC_START));
>> +
>> +	/* leave a hole here for vmalloc */
>> +
>> +	kasan_populate_early_shadow(
>> +		kasan_mem_to_shadow((void *)RADIX_VMALLOC_END),
>> +		kasan_mem_to_shadow((void *)RADIX_VMEMMAP_END));
>> +
>> +	flush_tlb_kernel_range((unsigned long)k_start, (unsigned long)k_end);
>> +
>> +	/* mark early shadow region as RO and wipe */
>> +	for (i = 0; i < PTRS_PER_PTE; i++)
>> +		__set_pte_at(&init_mm, (unsigned long)kasan_early_shadow_page,
>> +			&kasan_early_shadow_pte[i],
>> +			pfn_pte(virt_to_pfn(kasan_early_shadow_page),
>> +			__pgprot(_PAGE_PTE | _PAGE_KERNEL_RO | _PAGE_BASE)),
>
> Isn't there an already existing val for the above set of flags ?
>
See if you like the simplified version in v2.

>> +			0);
>> +	memset(kasan_early_shadow_page, 0, PAGE_SIZE);
>> +
>> +	kasan_init_tags();
>
> You said in the doc that only arm supports that ...
>
Indeed - serves me right for copying arm code!

Regards,
Daniel

>> +
>> +	/* Enable error messages */
>> +	init_task.kasan_depth = 0;
>> +	pr_info("KASAN init done (64-bit Book3S heavyweight mode)\n");
>> +}
>> 
>
> Christophe

Le 10/12/2019 à 06:10, Daniel Axtens a écrit :
> Christophe Leroy <christophe.leroy@c-s.fr> writes:
> 
>> Le 07/08/2019 à 01:38, Daniel Axtens a écrit :
>>> KASAN support on powerpc64 is interesting:
>>>
>>>    - We want to be able to support inline instrumentation so as to be
>>>      able to catch global and stack issues.
>>>
>>>    - We run a lot of code at boot in real mode. This includes stuff like
>>>      printk(), so it's not feasible to just disable instrumentation
>>>      around it.
>>
>> Have you definitely given up the idea of doing a standard implementation
>> of KASAN like other 64 bits arches have done ?
>>
>> Isn't it possible to setup an early 1:1 mapping and go in virtual mode
>> earlier ? What is so different between book3s64 and book3e64 ?
>> On book3e64, we've been able to setup KASAN before printing anything
>> (except when using EARLY_DEBUG). Isn't it feasible on book3s64 too ?
> 
> So I got this pretty wrong when trying to explain it. The problem isn't
> that we run the code in boot as I said, it's that a bunch of the KVM
> code runs in real mode.

Ok.

Does it mean we would be able to implement it the standard way when 
CONFIG_KVM is not selected ?

> 
>>>    - disabled reporting when we're checking the stack for exception
>>>      frames. The behaviour isn't wrong, just incompatible with KASAN.
>>
>> Does this applies to / impacts PPC32 at all ?
> 
> It should. I found that when doing stack walks, the code would touch
> memory that KASAN hadn't unpoisioned. I'm a bit surprised you haven't
> seen it arise, tbh.

How do you trigger that ?

I've tried to provoke some faults with LKDTM that provoke BUG dumps, but 
it doesn't trip.
I also performed task state listing via sysrq, and I don't get anything 
wrong either.

> 
>>>    - Dropped old module stuff in favour of KASAN_VMALLOC.
>>
>> You said in the cover that this is done to avoid having to split modules
>> out of VMALLOC area. Would it be an issue to perform that split ?
>> I can understand it is not easy on 32 bits because vmalloc space is
>> rather small, but on 64 bits don't we have enough virtual space to
>> confortably split modules out of vmalloc ? The 64 bits already splits
>> ioremap away from vmalloc whereas 32 bits have them merged too.
> 
> I could have done this. Maybe I should have done this. But now I have
> done vmalloc space support.

So you force the use of KASAN_VMALLOC ? Doesn't it have a performance 
impact ?


Christophe

Hi Christophe,

>>>>    - We run a lot of code at boot in real mode. This includes stuff like
>>>>      printk(), so it's not feasible to just disable instrumentation
>>>>      around it.
>>>
>>> Have you definitely given up the idea of doing a standard implementation
>>> of KASAN like other 64 bits arches have done ?
>>>
>>> Isn't it possible to setup an early 1:1 mapping and go in virtual mode
>>> earlier ? What is so different between book3s64 and book3e64 ?
>>> On book3e64, we've been able to setup KASAN before printing anything
>>> (except when using EARLY_DEBUG). Isn't it feasible on book3s64 too ?
>> 
>> So I got this pretty wrong when trying to explain it. The problem isn't
>> that we run the code in boot as I said, it's that a bunch of the KVM
>> code runs in real mode.
>
> Ok.
>
> Does it mean we would be able to implement it the standard way when 
> CONFIG_KVM is not selected ?

I suppose, but KVM is pretty important to me!

>>>>    - disabled reporting when we're checking the stack for exception
>>>>      frames. The behaviour isn't wrong, just incompatible with KASAN.
>>>
>>> Does this applies to / impacts PPC32 at all ?
>> 
>> It should. I found that when doing stack walks, the code would touch
>> memory that KASAN hadn't unpoisioned. I'm a bit surprised you haven't
>> seen it arise, tbh.
>
> How do you trigger that ?
>
> I've tried to provoke some faults with LKDTM that provoke BUG dumps, but 
> it doesn't trip.
> I also performed task state listing via sysrq, and I don't get anything 
> wrong either.

I'll try to disable this and see if I can trigger it again.

>>>>    - Dropped old module stuff in favour of KASAN_VMALLOC.
>>>
>>> You said in the cover that this is done to avoid having to split modules
>>> out of VMALLOC area. Would it be an issue to perform that split ?
>>> I can understand it is not easy on 32 bits because vmalloc space is
>>> rather small, but on 64 bits don't we have enough virtual space to
>>> confortably split modules out of vmalloc ? The 64 bits already splits
>>> ioremap away from vmalloc whereas 32 bits have them merged too.
>> 
>> I could have done this. Maybe I should have done this. But now I have
>> done vmalloc space support.
>
> So you force the use of KASAN_VMALLOC ? Doesn't it have a performance 
> impact ?

It has a perfomance impact when allocating and freeing virtual address
space in the vmalloc region, yes. There should be no discernable impact
when using vmalloc space.

My team is actively working on vmap-stack support for ppc64, with the
end goal of running syzkaller with vmap-stack and kasan. vmap-stack plus
kasan requires kasan-vmalloc, so for my purposes doing things in this
order makes sense.

I'd be happy to have a later series introduce the split and then make
KASAN_VMALLOC optional. I would need to understand the implications of
splitting the address space from a KASLR point of view: I don't want to
accidentally overly restrict the available randomness.

Regards,
Daniel

>
> Christophe