[PATCH v2 3/3] RISC-V: Use Zkr to seed KASLR base address

Jesse Taube jesse at rivosinc.com
Thu Jun 27 10:55:19 PDT 2024



On 6/27/24 07:45, Conor Dooley wrote:
> Hey Jesse,
> 
> On Wed, Jun 26, 2024 at 01:16:52PM -0400, Jesse Taube wrote:
>> Parse the device tree for Zkr in isa string.
>> If Zkr is present, use it to seed the kernel base address.
>>
>> Signed-off-by: Jesse Taube <jesse at rivosinc.com>
>> ---
>>   arch/riscv/kernel/pi/Makefile           |  2 +-
>>   arch/riscv/kernel/pi/archrandom_early.c | 30 ++++++++
>>   arch/riscv/kernel/pi/fdt_early.c        | 94 +++++++++++++++++++++++++
>>   arch/riscv/kernel/pi/pi.h               |  3 +
>>   arch/riscv/mm/init.c                    |  5 +-
>>   5 files changed, 132 insertions(+), 2 deletions(-)
>>   create mode 100644 arch/riscv/kernel/pi/archrandom_early.c
>>
>> diff --git a/arch/riscv/kernel/pi/Makefile b/arch/riscv/kernel/pi/Makefile
>> index 1ef7584be0c3..dba902f2a538 100644
>> --- a/arch/riscv/kernel/pi/Makefile
>> +++ b/arch/riscv/kernel/pi/Makefile
>> @@ -33,5 +33,5 @@ $(obj)/string.o: $(srctree)/lib/string.c FORCE
>>   $(obj)/ctype.o: $(srctree)/lib/ctype.c FORCE
>>   	$(call if_changed_rule,cc_o_c)
>>   
>> -obj-y		:= cmdline_early.pi.o fdt_early.pi.o string.pi.o ctype.pi.o lib-fdt.pi.o lib-fdt_ro.pi.o
>> +obj-y		:= cmdline_early.pi.o fdt_early.pi.o string.pi.o ctype.pi.o lib-fdt.pi.o lib-fdt_ro.pi.o archrandom_early.pi.o
>>   extra-y		:= $(patsubst %.pi.o,%.o,$(obj-y))
>> diff --git a/arch/riscv/kernel/pi/archrandom_early.c b/arch/riscv/kernel/pi/archrandom_early.c
>> new file mode 100644
>> index 000000000000..c6261165e8a6
>> --- /dev/null
>> +++ b/arch/riscv/kernel/pi/archrandom_early.c
>> @@ -0,0 +1,30 @@
>> +// SPDX-License-Identifier: GPL-2.0-only
>> +
>> +#include <asm/csr.h>
>> +#include <linux/processor.h>
>> +
>> +#include "pi.h"
>> +
>> +/*
>> + * To avoid rewriting code include asm/archrandom.h and create macros
>> + * for the functions that won't be included.
>> + */
>> +#undef riscv_has_extension_unlikely
>> +#define riscv_has_extension_likely(...) false
>> +#undef pr_err_once
>> +#define pr_err_once(...)
>> +
>> +#include <asm/archrandom.h>
>> +
>> +u64 get_kaslr_seed_zkr(const uintptr_t dtb_pa)
>> +{
>> +	unsigned long seed = 0;
>> +
>> +	if (!early_isa_str((const void *)dtb_pa, "zkr"))
>> +		return 0;
>> +
>> +	if (!csr_seed_long(&seed))
>> +		return 0;
>> +
>> +	return seed;
>> +}
>> diff --git a/arch/riscv/kernel/pi/fdt_early.c b/arch/riscv/kernel/pi/fdt_early.c
>> index 40ee299702bf..ba76197b44d1 100644
>> --- a/arch/riscv/kernel/pi/fdt_early.c
>> +++ b/arch/riscv/kernel/pi/fdt_early.c
>> @@ -23,3 +23,97 @@ u64 get_kaslr_seed(uintptr_t dtb_pa)
>>   	*prop = 0;
>>   	return ret;
>>   }
>> +
>> +/* Based off of fdt_stringlist_contains */
>> +static int isa_string_contains(const char *strlist, int listlen, const char *str)
> 
> The variable names here are needlessly confusing IMO. The function also
> returns a bool, not an int.
> 
>> +{
>> +	int len = strlen(str);
>> +	const char *p;
>> +
>> +	while (listlen >= len) {
>> +		if (strncasecmp(str, strlist, len) == 0)
>> +			return 1;
> 
> How does this handle searching a devicetree containing "rv64ima_zksed_zkr"
> for the extension zks? Hint: https://godbolt.org/z/YfhTqe54e
> I think this works for fdt_stringlist_contains() because it also
> compares the null chars - which you're not doing so I think this also
> brakes for something like riscv,isa-extensions = "rv64ima\0zksed\0zkr"
> while searching for zks.
> 
>> +		p = memchr(strlist, '_', listlen);
> 
> Or how does this handle searching "rv64imafdczkr" for zkr? It's gonna
> run right off the end of the string without finding anything, right?

Yes...

Is that a valid isa,string? I will try to copy how cpufeature.c as close 
as posible.

> 
> Handling "riscv,isa" is not trivial, but at least the search for extension
> approach here skips dealing with some of what has to be done in the "real"
> parser with the version numbers...
> 
> Maybe we just say screw "riscv,isa", as it's deprecated anyway, and only.
I think it's important to have.

> add this new feature for "riscv,isa-extensions" which is far simpler to
> parse and can be done using off-the-shelf fdt functions?
> 
> If not, then I think we should use fdt_stringlist_contains verbatim for
> "riscv,isa-extensions".

Ok I had a notion that riscv,isa-extensions could be upercase they 
cant/wont. I will use fdt_stringlist_contains.

> and introduce a custom function for "riscv,isa"
> only.

That was my original thought I will do that.

> 
>> +		if (!p)
>> +			p = memchr(strlist, '\0', listlen);
>> +		if (!p)
>> +			return 0; /* malformed strlist.. */
>> +		listlen -= (p - strlist) + 1;
>> +		strlist = p + 1;
>> +	}
>> +
>> +	return 0;
>> +}
>> +
>> +/* Based off of fdt_nodename_eq_ */
> 
> Why can't we just use fdt_nodename_eq?

Because fdt_nodename_eq_ is static.
I will change the comment to "copy of fdt_nodename_eq_".
Oddly there is `of_node_name_eq` but not `fdt_nodename_eq`

> 
>> +static int fdt_node_name_eq(const void *fdt, int offset,
>> +			    const char *s)
>> +{
>> +	int olen;
>> +	int len = strlen(s);
>> +	const char *p = fdt_get_name(fdt, offset, &olen);
>> +
>> +	if (!p || olen < len)
>> +		/* short match */
>> +		return 0;
>> +
>> +	if (memcmp(p, s, len) != 0)
>> +		return 0;
>> +
>> +	if (p[len] == '\0')
>> +		return 1;
>> +	else if (!memchr(s, '@', len) && (p[len] == '@'))
>> +		return 1;
>> +	else
>> +		return 0;
>> +}
>> +
>> +/*
>> + * Returns true if the extension is in the isa string
>> + * Returns false if the extension is not found
>> + */
>> +static bool get_ext_named(const void *fdt, int node, const char *name)
> 
> Could you rename this function please?

Yes.

> Having something named "get" that
> returns a bool, and not an "ext_named" is odd - and it'd be self
> explanatory in that case. Maybe it can just be moved into the sole caller
> and isn't needed?

I have it as a seperate function if in the future there needed to be a 
per cpu check.

>> +{
>> +	const void *prop;
>> +	int len;
>> +
>> +	prop = fdt_getprop(fdt, node, "riscv,isa-base", &len);
>> +	if (prop && isa_string_contains(prop, len, name))
>> +		return true;
> 
> This shouldn't be here, there'll not be an extension in this property.
Will remove.

> 
>> +	prop = fdt_getprop(fdt, node, "riscv,isa-extensions", &len);
>> +	if (prop && isa_string_contains(prop, len, name))
>> +		return true;
>> +
>> +	prop = fdt_getprop(fdt, node, "riscv,isa", &len);
>> +	if (prop && isa_string_contains(prop, len, name))
>> +		return true;
>> +
>> +	return false;
>> +}
>> +
>> +/*
>> + * Returns true if the extension is in the isa string on all cpus
> 
> Shouldn't we only be checking CPUs that are not disabled or reserved,
> rather than all CPUs?

Its way easier to just check all the cpus rather then make sure we are 
runing on one thats has the extention. I will add a continue for 
dissabled/reserved cpus.

> To use Zkr for KASLR this is kinda irrelevant
> since really we only care about whether or not the boot CPU has Zkr,
> but in general we only want to consider CPUs that we can actually use.
> For example, if you did this for FPU support with mpfs.dtsi, you'd get
> told that the F/D extensions were not present cos hart 0

Can we assume that the boot hart is always 0?

> doesn't have
> them, even though it's disabled and will not be used by Linux.
> 
>> + * Returns false if the extension is not found
>> + */
>> +bool early_isa_str(const void *fdt, const char *ext_name)
> 
> Could you try to match the naming of the stuff used outside of pi?
> Maybe early_isa_ext_available()?

Yes.

Thanks,
Jesse Taube

> 
> Thanks for the update,
> Conor.
> 
>> +{
>> +	int node, parent;
>> +	bool ret = false;
>> +
>> +	parent = fdt_path_offset(fdt, "/cpus");
>> +	if (parent < 0)
>> +		return false;
>> +
>> +	fdt_for_each_subnode(node, fdt, parent) {
>> +		if (!fdt_node_name_eq(fdt, node, "cpu"))
>> +			continue;
>> +
>> +		if (!get_ext_named(fdt, node, ext_name))
>> +			return false;
>> +
>> +		ret = true;
>> +	}
>> +
>> +	return ret;
>> +}
>> diff --git a/arch/riscv/kernel/pi/pi.h b/arch/riscv/kernel/pi/pi.h
>> index 65da99466baf..26e7e5f84a30 100644
>> --- a/arch/riscv/kernel/pi/pi.h
>> +++ b/arch/riscv/kernel/pi/pi.h
>> @@ -4,6 +4,8 @@
>>   
>>   #include <linux/types.h>
>>   
>> +bool early_isa_str(const void *fdt, const char *ext_name);
>> +
>>   /*
>>    * The folowing functions are exported (but prefixed) declare them here so
>>    * that LLVM does not complain it lacks the 'static' keyword (which, if
>> @@ -11,6 +13,7 @@
>>    */
>>   
>>   u64 get_kaslr_seed(uintptr_t dtb_pa);
>> +u64 get_kaslr_seed_zkr(const uintptr_t dtb_pa);
>>   bool set_nokaslr_from_cmdline(uintptr_t dtb_pa);
>>   u64 set_satp_mode_from_cmdline(uintptr_t dtb_pa);
>>   
>> diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
>> index 9940171c79f0..bfb068dc4a64 100644
>> --- a/arch/riscv/mm/init.c
>> +++ b/arch/riscv/mm/init.c
>> @@ -1025,6 +1025,7 @@ static void __init pt_ops_set_late(void)
>>   #ifdef CONFIG_RANDOMIZE_BASE
>>   extern bool __init __pi_set_nokaslr_from_cmdline(uintptr_t dtb_pa);
>>   extern u64 __init __pi_get_kaslr_seed(uintptr_t dtb_pa);
>> +extern u64 __init __pi_get_kaslr_seed_zkr(const uintptr_t dtb_pa);
>>   
>>   static int __init print_nokaslr(char *p)
>>   {
>> @@ -1045,10 +1046,12 @@ asmlinkage void __init setup_vm(uintptr_t dtb_pa)
>>   
>>   #ifdef CONFIG_RANDOMIZE_BASE
>>   	if (!__pi_set_nokaslr_from_cmdline(dtb_pa)) {
>> -		u64 kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
>> +		u64 kaslr_seed = __pi_get_kaslr_seed_zkr(dtb_pa);
>>   		u32 kernel_size = (uintptr_t)(&_end) - (uintptr_t)(&_start);
>>   		u32 nr_pos;
>>   
>> +		if (kaslr_seed == 0)
>> +			kaslr_seed = __pi_get_kaslr_seed(dtb_pa);
>>   		/*
>>   		 * Compute the number of positions available: we are limited
>>   		 * by the early page table that only has one PUD and we must
>> -- 
>> 2.45.2
>>



More information about the linux-riscv mailing list