Skip to contentSkip to conversation

riscv: fix kprobes on minimal kernel configs

PATCH Superseded Xiaofeng Yuan 5 participants 9 messages Jul 20 – Aug 14 RISC-V

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover riscv: fix kprobes on minimal kernel configs

When building with an allnoconfig-derived minimal configuration (without CONFIG_STRICT_MODULE_RWX), kprobes fails because instruction slot pages from EXECMEM_KPROBES are read-only and patch_map() bypasses the fixmap writable alias.

These two patches fix the infrastructure.

PATCH 1/2 changelog:
v2: use CONFIG_STRICT_MODULE_RWX instead of ARCH_HAS_EXECMEM_ROX
v3: add commit description
v4: fix indent alignment

PATCH 2/2 changelog:
v2: add commit description
v3: early return when !CONFIG_STRICT_MODULE_RWX

Xiaofeng Yuan (2):
riscv: mm: make EXECMEM_KPROBES writable without
CONFIG_STRICT_MODULE_RWX
riscv: patch: skip fixmap mapping when kernel text is already
writable

 arch/riscv/kernel/patch.c | 4 ++--
 arch/riscv/mm/init.c      | 4 +++-
 2 files changed, 6 insertions(+), 2 deletions(-)

--
2.43.0

#1 lore raw

PATCH v4 1/2 riscv: mm: make EXECMEM_KPROBES writable without CONFIG_STRICT_MODULE_RWX

When CONFIG_STRICT_MODULE_RWX is not set, execmem cannot create temporary writable mappings for read-only executable pages. In this case, the execmem ranges must already have writable permissions.

Currently EXECMEM_KPROBES unconditionally uses PAGE_KERNEL_READ_EXEC, which causes kprobe instruction slot writes to trigger page faults on systems where CONFIG_STRICT_MODULE_RWX is not enabled.

Fix this by using PAGE_KERNEL_EXEC when CONFIG_STRICT_MODULE_RWX is not available.

---
v2: use CONFIG_STRICT_MODULE_RWX instead of CONFIG_ARCH_HAS_EXECMEM_ROX (per Nam Cao's review)
v3: add commit description
v4: fix indent alignment (per Nam Cao's review)

arch/riscv/mm/init.c +3 -1
index 5b1b3c88b4..7951d72a12 100644
@@ -1457,7 +1457,9 @@ struct execmem_info __init *execmem_arch_setup(void)
 			[EXECMEM_KPROBES] = {
 				.start	= VMALLOC_START,
 				.end	= VMALLOC_END,
-				.pgprot	= PAGE_KERNEL_READ_EXEC,
+				.pgprot	= IS_ENABLED(CONFIG_STRICT_MODULE_RWX) ?
+					  PAGE_KERNEL_READ_EXEC :
+					  PAGE_KERNEL_EXEC,
 				.alignment = 1,
 			},
 			[EXECMEM_BPF] = {
Signature · 1 line
2.43.0
#2 lore raw

PATCH v4 2/2 riscv: patch: skip fixmap mapping when kernel text is already writable

In reply to Xiaofeng Yuan

Currently patch_map() always creates a temporary writable mapping via fixmap for kernel text addresses, even when CONFIG_STRICT_MODULE_RWX is disabled and the kernel text is already mapped with _PAGE_WRITE.

This is unnecessary overhead at best, and on minimal configurations it can cause page faults.

Skip the fixmap path for kernel text when CONFIG_STRICT_MODULE_RWX is not enabled, since the text pages are already writable in this case.

---
v2: add commit description
v3: early return when !CONFIG_STRICT_MODULE_RWX (per Nam Cao's suggestion)

arch/riscv/kernel/patch.c +4 -3
index 16b243376f..caef41d5ef 100644
@@ -44,15 +44,16 @@ static __always_inline void *patch_map(void *addr, const unsigned int fixmap)
 	uintptr_t uintaddr = (uintptr_t) addr;
 	phys_addr_t phys;
 
+	if (!IS_ENABLED(CONFIG_STRICT_MODULE_RWX))
+		return addr;
+
 	if (core_kernel_text(uintaddr) || is_kernel_exittext(uintaddr)) {
 		phys = __pa_symbol(addr);
-	} else if (IS_ENABLED(CONFIG_STRICT_MODULE_RWX)) {
+	} else {
 		struct page *page = vmalloc_to_page(addr);
 
 		BUG_ON(!page);
 		phys = page_to_phys(page) + offset_in_page(addr);
-	} else {
-		return addr;
 	}
 
 	return (void *)set_fixmap_offset(fixmap, phys);
Signature · 1 line
2.43.0
#3 lore raw

Re: [PATCH v4 2/2] riscv: patch: skip fixmap mapping when kernel text is already writable

Hi,

Quoted 19 lines · Xiaofeng Yuan

Currently patch_map() always creates a temporary writable mapping via fixmap for kernel text addresses, even when CONFIG_STRICT_MODULE_RWX is disabled and the kernel text is already mapped with _PAGE_WRITE.

This is unnecessary overhead at best, and on minimal configurations it can cause page faults.

Skip the fixmap path for kernel text when CONFIG_STRICT_MODULE_RWX is not enabled, since the text pages are already writable in this case.

---
v2: add commit description
v3: early return when !CONFIG_STRICT_MODULE_RWX (per Nam Cao's suggestion)

arch/riscv/kernel/patch.c +0 -0
index 16b243376f..caef41d5ef 100644

arch/riscv/kernel/patch.c static __always_inline void *patch_map(void *addr, const unsigned int fixmap) · line 48

Quoted 5 lines
arch/riscv/kernel/patch.c +2 -0
@@ -44,15 +44,16 @@ static __always_inline void *patch_map(void *addr, const unsigned int fixmap)
 	uintptr_t uintaddr = (uintptr_t) addr;
 	phys_addr_t phys;
 
+	if (!IS_ENABLED(CONFIG_STRICT_MODULE_RWX))
+		return addr;

Is this correct when STRICT_KERNEL_RWX is set and not STRICT_MODULE_RWX (e.g. when modules are not enabled)?

Quoted 17 lines
arch/riscv/kernel/patch.c +2 -3
+
 	if (core_kernel_text(uintaddr) || is_kernel_exittext(uintaddr)) {
 		phys = __pa_symbol(addr);
-	} else if (IS_ENABLED(CONFIG_STRICT_MODULE_RWX)) {
+	} else {
 		struct page *page = vmalloc_to_page(addr);
 
 		BUG_ON(!page);
 		phys = page_to_phys(page) + offset_in_page(addr);
-	} else {
-		return addr;
 	}
 
 	return (void *)set_fixmap_offset(fixmap, phys);
Signature · 1 line
2.43.0

This patch causes the following oops on my BPI-F3:

 Unable to handle kernel paging request at virtual address 0000006789abce0b
 Current swapper/0 pgtable: 4K pagesize, 39-bit VAs, pgdp=0x000000000269d000
 [0000006789abce0b] pgd=0000000040031c01, p4d=0000000040031c01, pud=0000000040031c01, pmd=0000000000000000
 Oops [#1]
 Tainted: [W]=WARN
 Hardware name: Banana Pi BPI-F3 (DT)
 epc : kmem_cache_alloc_lru_noprof (/home/klara/git/linux/trees/bisect/mm/slub.c:4941)
  ra : __d_alloc (/home/klara/git/linux/trees/bisect/fs/dcache.c:1902)
 epc : ffffffff8038379e ra : ffffffff804053b2 sp : ffffffff82203ab0
  gp : ffffffff8248da18 tp : ffffffff822200c0 t0 : ffffffd7010660e8
  t1 : ffffffff921904d0 t2 : 000000005d6ccc9e s0 : ffffffff82203b30
  s1 : 0000000000000000 a0 : 0123456789abcdef a1 : ffffffd700b3f510
  a2 : 0000000000000cc0 a3 : 0000000000000002 a4 : 0000000000000000
  a5 : 0123456700000000 a6 : 00ffffffff899275 a7 : ffffffff82203918
  s2 : 0000000000000000 s3 : ffffffd700b3f000 s4 : ffffffff81a0d848
  s5 : ffffffff82492018 s6 : 0000000000000000 s7 : 00000000000003e8
  s8 : ffffffff814010e0 s9 : 0000000000000000 s10: 0000000000200000
  s11: 00000000024910d8 t3 : 0000000000000014 t4 : 0000000000000026
  t5 : 000000003137ae71 t6 : ffffffff82203b18 ssp : 0000000000000000
 status: 0000000200000120 badaddr: 0000006789abce0b cause: 000000000000000d
 kmem_cache_alloc_lru_noprof (/home/klara/git/linux/trees/bisect/mm/slub.c:4941)
 __d_alloc (/home/klara/git/linux/trees/bisect/fs/dcache.c:1902)
 d_make_root (/home/klara/git/linux/trees/bisect/fs/dcache.c:1999 /home/klara/git/linux/trees/bisect/fs/dcache.c:2213)
 shmem_fill_super (/home/klara/git/linux/trees/bisect/mm/shmem.c:5050)
 get_tree_nodev (/home/klara/git/linux/trees/bisect/fs/super.c:1273 /home/klara/git/linux/trees/bisect/fs/super.c:1292)
 shmem_get_tree (/home/klara/git/linux/trees/bisect/mm/shmem.c:5062)
 vfs_get_tree (/home/klara/git/linux/trees/bisect/fs/super.c:1700)
 fc_mount (/home/klara/git/linux/trees/bisect/fs/namespace.c:1198)
 vfs_kern_mount.part.0 (/home/klara/git/linux/trees/bisect/fs/namespace.c:1236)
 kern_mount (/home/klara/git/linux/trees/bisect/fs/namespace.c:6290 /home/klara/git/linux/trees/bisect/fs/namespace.c:6292)
 shmem_init (/home/klara/git/linux/trees/bisect/mm/shmem.c:5367)
 mnt_init (/home/klara/git/linux/trees/bisect/fs/namespace.c:6274)
 vfs_caches_init (/home/klara/git/linux/trees/bisect/fs/dcache.c:3517)
 start_kernel (/home/klara/git/linux/trees/bisect/init/main.c:1157)
 Code: 0001 0001 0001 7119 f8a2 f4a6 0100 f0ca ecce fc86 (6703) 01c5
 All code
 ========
    0:	0001                	.insn	2, 0x0001
    2:	0001                	.insn	2, 0x0001
    4:	0001                	.insn	2, 0x0001
    6:	7119                	.insn	2, 0x7119
    8:	f8a2                	.insn	2, 0xf8a2
    a:	f4a6                	.insn	2, 0xf4a6
    c:	0100                	.insn	2, 0x0100
    e:	f0ca                	.insn	2, 0xf0ca
   10:	ecce                	.insn	2, 0xecce
   12:	fc86                	.insn	2, 0xfc86
   14:*	01c56703          	lwu	a4,28(a0)		<-- trapping instruction

Code starting with the faulting instruction
===========================================

    0:	01c56703          	lwu	a4,28(a0)
 ---[ end trace 0000000000000000 ]---
 Kernel panic - not syncing: Attempted to kill the idle task!

I have MODULES disabled and thus also STRICT_MODULE_RWX disabled, but still STRICT_KERNEL_RWX enabled. I saw that an earlier version of this patch[1] instead gated the first branch behind STRICT_KERNEL_RWX. That version works fine for me.

Regards,
Klara Modin

# bad: [66566bdc5a42d707d4c3f54587333db00e955268] Merge branch 'unstable/spacemit-k1-wdt' into unstable/next-local
git bisect start 'HEAD'
# status: waiting for 'good' commit(s), 'bad' commit known
# good: [f5bbbfec59b4e2fb7520a91de3df8a6174325d6a] Merge tag 'probes-fixes-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace
git bisect good f5bbbfec59b4e2fb7520a91de3df8a6174325d6a
# bad: [9eba0000697167ff6960428f7401e9f65f7abba3] Merge branch 'libcrypto-next' of https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux.git
git bisect bad 9eba0000697167ff6960428f7401e9f65f7abba3
# bad: [6a66dbb2b9b0a0a5b1225052267c0eaedf34492d] Merge branch 'xtensa-for-next' of https://github.com/jcmvbkbc/linux-xtensa.git
git bisect bad 6a66dbb2b9b0a0a5b1225052267c0eaedf34492d
# good: [3787df98a48a0e699f6a06d4a42fc20597368db1] Merge branch 'for-next/core' of https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
git bisect good 3787df98a48a0e699f6a06d4a42fc20597368db1
# good: [aeba6e4f8ace1a3f016feb64c0a417c7ddcbcf6b] Merge tag 'qcom-arm64-for-7.3-2' of https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux into soc/dt git bisect good aeba6e4f8ace1a3f016feb64c0a417c7ddcbcf6b
# good: [15213090630d0a4417dbcdef0de776cd2fac0fac] Merge branch 'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/khilman/linux-omap.git
git bisect good 15213090630d0a4417dbcdef0de776cd2fac0fac
# good: [e347a696fd95bc9b47e23e3d8ac7f6263ee502b2] Merge branch 'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/abelvesa/linux.git
git bisect good e347a696fd95bc9b47e23e3d8ac7f6263ee502b2
# good: [b7307ccce8238c792fb3a6412ca5cac9ba8784bc] Merge branch 'next' of https://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux.git
git bisect good b7307ccce8238c792fb3a6412ca5cac9ba8784bc
# good: [b5d4268affa543793ccaf838a10915c2c9f0db1c] Merge branch 'features' into for-next
git bisect good b5d4268affa543793ccaf838a10915c2c9f0db1c
# good: [ebdec8d2c156b8e662cb350ce05b0f92275f6ffd] RISC-V: Add Ssccfg/Smcdeleg ISA extension definition and parsing
git bisect good ebdec8d2c156b8e662cb350ce05b0f92275f6ffd
# good: [f61959a3a8b5522eb43cf71f293e091417bbf11c] riscv: Add Ziccamoa, Ziccif, Ziccrse, and Za64rs to cpufeature and hwprobe
git bisect good f61959a3a8b5522eb43cf71f293e091417bbf11c
# bad: [54fefa110db4a407d96625b66a76d874713ddae2] riscv: patch: skip fixmap mapping when kernel text is already writable
git bisect bad 54fefa110db4a407d96625b66a76d874713ddae2
# good: [42cd1e1fc8995ec7ed5cc61957fc503eb065e0eb] riscv: alternative: Also patch the compat vDSO
git bisect good 42cd1e1fc8995ec7ed5cc61957fc503eb065e0eb
# good: [83ba459c9b5893590fd47177363b08b453a86168] riscv: mm: make EXECMEM_KPROBES writable without CONFIG_STRICT_MODULE_RWX
git bisect good 83ba459c9b5893590fd47177363b08b453a86168

 # first 'bad' commit: [54fefa110db4a407d96625b66a76d874713ddae2] riscv: patch: skip fixmap mapping when kernel text is already writable

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

#4 lore raw

Re: [PATCH v4 2/2] riscv: patch: skip fixmap mapping when kernel text is already writable

Quoted 4 lines · Klara Modin

I have MODULES disabled and thus also STRICT_MODULE_RWX disabled, but still STRICT_KERNEL_RWX enabled. I saw that an earlier version of this patch[1] instead gated the first branch behind STRICT_KERNEL_RWX. That version works fine for me.

Urgh, I got confused between STRICT_MODULE_RWX and STRICT_KERNEL_RWX, I suggested the final version.

Xiaofeng, can you send a patch?

Nam

maintainer
#5 lore raw

Re: [PATCH v4 2/2] riscv: patch: skip fixmap mapping when kernel text is already writable

Quoted 10 lines · Nam Cao
Quoted 4 lines · Klara Modin

I have MODULES disabled and thus also STRICT_MODULE_RWX disabled, but still STRICT_KERNEL_RWX enabled. I saw that an earlier version of this patch[1] instead gated the first branch behind STRICT_KERNEL_RWX. That version works fine for me.

Urgh, I got confused between STRICT_MODULE_RWX and STRICT_KERNEL_RWX, I suggested the final version.

Xiaofeng, can you send a patch?

Just dropped it from riscv/for-next, so, no need to send a revert. Xiaofeng, could you send an updated patch?

Thanks to Klara and Prabhakar for reporting the failures promptly!

- Paul

#6 lore raw

Re: [PATCH v4 2/2] riscv: patch: skip fixmap mapping when kernel text is already writable

In reply to Xiaofeng Yuan

Hi Xiaofeng,

Thank you for the patch.

Quoted 28 lines · Xiaofeng Yuan

Currently patch_map() always creates a temporary writable mapping via fixmap for kernel text addresses, even when CONFIG_STRICT_MODULE_RWX is disabled and the kernel text is already mapped with _PAGE_WRITE.

This is unnecessary overhead at best, and on minimal configurations it can cause page faults.

Skip the fixmap path for kernel text when CONFIG_STRICT_MODULE_RWX is not enabled, since the text pages are already writable in this case.

---
v2: add commit description
v3: early return when !CONFIG_STRICT_MODULE_RWX (per Nam Cao's suggestion)

arch/riscv/kernel/patch.c +3 -0
index 16b243376f..caef41d5ef 100644
@@ -44,15 +44,16 @@ static __always_inline void *patch_map(void *addr, const unsigned int fixmap)
        uintptr_t uintaddr = (uintptr_t) addr;
        phys_addr_t phys;
 
+       if (!IS_ENABLED(CONFIG_STRICT_MODULE_RWX))
+               return addr;
+
        if (core_kernel_text(uintaddr) || is_kernel_exittext(uintaddr)) {
                phys = __pa_symbol(addr);

arch/riscv/kernel/patch.c static __always_inline void *patch_map(void *addr, const unsigned int fixmap) · line 58

Quoted 10 lines
arch/riscv/kernel/patch.c +1 -3
-       } else if (IS_ENABLED(CONFIG_STRICT_MODULE_RWX)) {
+       } else {
                struct page *page = vmalloc_to_page(addr);
 
                BUG_ON(!page);
                phys = page_to_phys(page) + offset_in_page(addr);
-       } else {
-               return addr;
        }
 

With this patch applied and using renesas_defconfig I see below Oops on SMARC RZ/Five.

[    0.012814] printk: legacy console [tty0] enabled
[    0.017634] Calibrating delay loop (skipped), value calculated
using timer frequency.. 24.00 BogoMIPS (lpj=48000)
[    0.027914] pid_max: default: 32768 minimum: 301
[    0.033876] Mount-cache hash table entries: 2048 (order: 2, 16384
bytes, linear)
[    0.041343] Mountpoint-cache hash table entries: 2048 (order: 2,
16384 bytes, linear)
[    0.049639] Unable to handle kernel paging request at virtual
address 0000006f89abce08
[    0.057617] Current swapper/0 pgtable: 4K pagesize, 39-bit VAs,
pgdp=0x00000000494f9000
[    0.065626] [0000006f89abce08] pgd=0000000000000000,
p4d=0000000000000000, pud=0000000000000000
[    0.074352] Oops [#1]
[    0.076635] CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Not tainted
7.2.0-rc7-next-20260812 #21 PREEMPT
[    0.085494] Hardware name: Renesas SMARC EVK based on r9a07g043f01 (DT)
[    0.092096] epc : kmem_cache_alloc_lru_noprof+0x1a/0x1ba
[    0.097418]  ra : __d_alloc+0x42/0x188
[    0.101177] epc : ffffffff80162e20 ra : ffffffff8019874e sp :
ffffffff81203b30
[    0.108385]  gp : ffffffff812ec750 tp : ffffffff8120c780 t0 :
ffffffd60189db18
[    0.115593]  t1 : 0000000000000014 t2 : 0000000016e00403 s0 :
ffffffff81203b90
[    0.122809]  s1 : ffffffd601895800 a0 : 89abcdef89abcdef a1 :
ffffffd601895d10
[    0.130022]  a2 : 0000000000000cc0 a3 : 0000000000000002 a4 :
dead4ead00000001
[    0.137235]  a5 : 0000000000000000 a6 : 000000001de3521b a7 :
ffffffffac7f4a0c
[    0.144448]  s2 : 0000000000000000 s3 : ffffffd601895800 s4 :
0000000000000000
[    0.151660]  s5 : 0000000000000000 s6 : ffffffff812ed018 s7 :
00000000000003e8
[    0.158872]  s8 : 00000000000003e6 s9 : 0000000000000000 s10:
0000000077f29340
[    0.166084]  s11: 0000000000000000 t3 : ffffffffa0aed022 t4 :
0000000002018efb
[    0.173297]  t5 : 00000000101d5200 t6 : ffffffff81203b68 ssp :
0000000000000000
[    0.180596] status: 0000000200000120 badaddr: 0000006f89abce08
cause: 000000000000000d
[    0.188502] [<ffffffff80162e20>] kmem_cache_alloc_lru_noprof+0x1a/0x1ba
[    0.195116] [<ffffffff8019874e>] __d_alloc+0x42/0x188
[    0.200171] [<ffffffff801988fe>] d_alloc_anon+0xe/0x16
[    0.205314] [<ffffffff80198c90>] d_make_root+0x18/0x3a
[    0.210457] [<ffffffff8012792a>] shmem_fill_super+0x218/0x244
[    0.216210] [<ffffffff801865aa>] vfs_get_super+0x42/0x70
[    0.221525] [<ffffffff801865e8>] get_tree_nodev+0x10/0x18
[    0.226925] [<ffffffff80125e6a>] shmem_get_tree+0x14/0x1c
[    0.232327] [<ffffffff801845ba>] vfs_get_tree+0x1a/0xb0
[    0.237561] [<ffffffff801a1fa2>] fc_mount+0x10/0x32
[    0.242445] [<ffffffff801a2016>] vfs_kern_mount.part.0+0x34/0x5c
[    0.248451] [<ffffffff801a204c>] vfs_kern_mount+0xe/0x1a
[    0.253764] [<ffffffff801a206c>] kern_mount+0x14/0x26
[    0.258818] [<ffffffff8080f70a>] shmem_init+0xbe/0x150
[    0.263963] [<ffffffff8081973e>] mnt_init+0x128/0x314
[    0.269020] [<ffffffff80819316>] vfs_caches_init+0xa0/0xce
[    0.274507] [<ffffffff80800fa2>] start_kernel+0x754/0x7ac
[    0.279923] Code: 1080 e0ca ec86 fc4e f852 3783 5402 3423 fcf4 4781
(6783) 01c5
[    0.287344] ---[ end trace 0000000000000000 ]---
[    0.291986] Kernel panic - not syncing: Attempted to kill the idle task!

Reverting this patch fixed the issue.

Cheers,
Prabhakar

Quoted 9 lines
arch/riscv/kernel/patch.c +0 -1
        return (void *)set_fixmap_offset(fixmap, phys);
--

2.43.0

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

Reviewed-by on the v4 series
#7 lore raw
In reply to Xiaofeng Yuan
Quoted 6 lines · Xiaofeng Yuan

When building with an allnoconfig-derived minimal configuration (without CONFIG_STRICT_MODULE_RWX), kprobes fails because instruction slot pages from EXECMEM_KPROBES are read-only and patch_map() bypasses the fixmap writable alias.

These two patches fix the infrastructure.

maintainer
#8 lore raw
In reply to Xiaofeng Yuan
Quoted 6 lines · Xiaofeng Yuan

When building with an allnoconfig-derived minimal configuration (without CONFIG_STRICT_MODULE_RWX), kprobes fails because instruction slot pages from EXECMEM_KPROBES are read-only and patch_map() bypasses the fixmap writable alias.

These two patches fix the infrastructure.

Thanks, queued for v7.3.

- Paul