diff options
| author | Christophe Leroy <christophe.leroy@csgroup.eu> | 2021-06-03 08:41:44 +0000 |
|---|---|---|
| committer | Michael Ellerman <mpe@ellerman.id.au> | 2021-06-17 00:09:08 +1000 |
| commit | 16132529cee586ee9a058bb33cfbdcb5d884f6b3 (patch) | |
| tree | 8f0cbcdf446efa599ec886f65150040223f28a6f /arch/powerpc/mm/book3s32 | |
| parent | 6b4d630068b0c5cdd6d8e599182b131448e0cb06 (diff) | |
powerpc/32s: Rework Kernel Userspace Access Protection
On book3s/32, KUAP is provided by toggling Ks bit in segment registers.
One segment register addresses 256M of virtual memory.
At the time being, KUAP implements a complex logic to apply the
unlock/lock on the exact number of segments covering the user range
to access, with saving the boundaries of the range of segments in
a member of thread struct.
But most if not all user accesses are within a single segment.
Rework KUAP with a different approach:
- Open only one segment, the one corresponding to the starting
address of the range to be accessed.
- If a second segment is involved, it will generate a page fault. The
segment will then be open by the page fault handler.
The kuap member of thread struct will now contain:
- The start address of the current on going user access, that will be
used to know which segment to lock at the end of the user access.
- ~0 when no user access is open
- ~1 when additionnal segments are opened by a page fault.
Then, at lock time
- When only one segment is open, close it.
- When several segments are open, close all user segments.
Almost 100% of the time, only one segment will be involved.
In interrupts, inline the function that unlock/lock all segments,
because not inlining them implies a lot of register save/restore.
With the patch, writing value 128 in userspace in perf_copy_attr() is
done with 16 instructions:
3890: 93 82 04 dc stw r28,1244(r2)
3894: 7d 20 e5 26 mfsrin r9,r28
3898: 55 29 00 80 rlwinm r9,r9,0,2,0
389c: 7d 20 e1 e4 mtsrin r9,r28
38a0: 4c 00 01 2c isync
38a4: 39 20 00 80 li r9,128
38a8: 91 3c 00 00 stw r9,0(r28)
38ac: 81 42 04 dc lwz r10,1244(r2)
38b0: 39 00 ff ff li r8,-1
38b4: 91 02 04 dc stw r8,1244(r2)
38b8: 2c 0a ff fe cmpwi r10,-2
38bc: 41 82 00 88 beq 3944 <perf_copy_attr+0x36c>
38c0: 7d 20 55 26 mfsrin r9,r10
38c4: 65 29 40 00 oris r9,r9,16384
38c8: 7d 20 51 e4 mtsrin r9,r10
38cc: 4c 00 01 2c isync
...
3944: 48 00 00 01 bl 3944 <perf_copy_attr+0x36c>
3944: R_PPC_REL24 kuap_lock_all_ool
Before the patch it was 118 instructions. In reality only 42 are
executed in most cases, but GCC is not able to see that a properly
aligned user access cannot involve more than one segment.
5060: 39 1d 00 04 addi r8,r29,4
5064: 3d 20 b0 00 lis r9,-20480
5068: 7c 08 48 40 cmplw r8,r9
506c: 40 81 00 08 ble 5074 <perf_copy_attr+0x2cc>
5070: 3d 00 b0 00 lis r8,-20480
5074: 39 28 ff ff addi r9,r8,-1
5078: 57 aa 00 06 rlwinm r10,r29,0,0,3
507c: 55 29 27 3e rlwinm r9,r9,4,28,31
5080: 39 29 00 01 addi r9,r9,1
5084: 7d 29 53 78 or r9,r9,r10
5088: 91 22 04 dc stw r9,1244(r2)
508c: 7d 20 ed 26 mfsrin r9,r29
5090: 55 29 00 80 rlwinm r9,r9,0,2,0
5094: 7c 08 50 40 cmplw r8,r10
5098: 40 81 00 c0 ble 5158 <perf_copy_attr+0x3b0>
509c: 7d 46 50 f8 not r6,r10
50a0: 7c c6 42 14 add r6,r6,r8
50a4: 54 c6 27 be rlwinm r6,r6,4,30,31
50a8: 7d 20 51 e4 mtsrin r9,r10
50ac: 3c ea 10 00 addis r7,r10,4096
50b0: 39 29 01 11 addi r9,r9,273
50b4: 7f 88 38 40 cmplw cr7,r8,r7
50b8: 55 29 02 06 rlwinm r9,r9,0,8,3
50bc: 40 9d 00 9c ble cr7,5158 <perf_copy_attr+0x3b0>
50c0: 2f 86 00 00 cmpwi cr7,r6,0
50c4: 41 9e 00 4c beq cr7,5110 <perf_copy_attr+0x368>
50c8: 2f 86 00 01 cmpwi cr7,r6,1
50cc: 41 9e 00 2c beq cr7,50f8 <perf_copy_attr+0x350>
50d0: 2f 86 00 02 cmpwi cr7,r6,2
50d4: 41 9e 00 14 beq cr7,50e8 <perf_copy_attr+0x340>
50d8: 7d 20 39 e4 mtsrin r9,r7
50dc: 39 29 01 11 addi r9,r9,273
50e0: 3c e7 10 00 addis r7,r7,4096
50e4: 55 29 02 06 rlwinm r9,r9,0,8,3
50e8: 7d 20 39 e4 mtsrin r9,r7
50ec: 39 29 01 11 addi r9,r9,273
50f0: 3c e7 10 00 addis r7,r7,4096
50f4: 55 29 02 06 rlwinm r9,r9,0,8,3
50f8: 7d 20 39 e4 mtsrin r9,r7
50fc: 3c e7 10 00 addis r7,r7,4096
5100: 39 29 01 11 addi r9,r9,273
5104: 7f 88 38 40 cmplw cr7,r8,r7
5108: 55 29 02 06 rlwinm r9,r9,0,8,3
510c: 40 9d 00 4c ble cr7,5158 <perf_copy_attr+0x3b0>
5110: 7d 20 39 e4 mtsrin r9,r7
5114: 39 29 01 11 addi r9,r9,273
5118: 3c c7 10 00 addis r6,r7,4096
511c: 55 29 02 06 rlwinm r9,r9,0,8,3
5120: 7d 20 31 e4 mtsrin r9,r6
5124: 39 29 01 11 addi r9,r9,273
5128: 3c c6 10 00 addis r6,r6,4096
512c: 55 29 02 06 rlwinm r9,r9,0,8,3
5130: 7d 20 31 e4 mtsrin r9,r6
5134: 39 29 01 11 addi r9,r9,273
5138: 3c c7 30 00 addis r6,r7,12288
513c: 55 29 02 06 rlwinm r9,r9,0,8,3
5140: 7d 20 31 e4 mtsrin r9,r6
5144: 3c e7 40 00 addis r7,r7,16384
5148: 39 29 01 11 addi r9,r9,273
514c: 7f 88 38 40 cmplw cr7,r8,r7
5150: 55 29 02 06 rlwinm r9,r9,0,8,3
5154: 41 9d ff bc bgt cr7,5110 <perf_copy_attr+0x368>
5158: 4c 00 01 2c isync
515c: 39 20 00 80 li r9,128
5160: 91 3d 00 00 stw r9,0(r29)
5164: 38 e0 00 00 li r7,0
5168: 90 e2 04 dc stw r7,1244(r2)
516c: 7d 20 ed 26 mfsrin r9,r29
5170: 65 29 40 00 oris r9,r9,16384
5174: 40 81 00 c0 ble 5234 <perf_copy_attr+0x48c>
5178: 7d 47 50 f8 not r7,r10
517c: 7c e7 42 14 add r7,r7,r8
5180: 54 e7 27 be rlwinm r7,r7,4,30,31
5184: 7d 20 51 e4 mtsrin r9,r10
5188: 3d 4a 10 00 addis r10,r10,4096
518c: 39 29 01 11 addi r9,r9,273
5190: 7c 08 50 40 cmplw r8,r10
5194: 55 29 02 06 rlwinm r9,r9,0,8,3
5198: 40 81 00 9c ble 5234 <perf_copy_attr+0x48c>
519c: 2c 07 00 00 cmpwi r7,0
51a0: 41 82 00 4c beq 51ec <perf_copy_attr+0x444>
51a4: 2c 07 00 01 cmpwi r7,1
51a8: 41 82 00 2c beq 51d4 <perf_copy_attr+0x42c>
51ac: 2c 07 00 02 cmpwi r7,2
51b0: 41 82 00 14 beq 51c4 <perf_copy_attr+0x41c>
51b4: 7d 20 51 e4 mtsrin r9,r10
51b8: 39 29 01 11 addi r9,r9,273
51bc: 3d 4a 10 00 addis r10,r10,4096
51c0: 55 29 02 06 rlwinm r9,r9,0,8,3
51c4: 7d 20 51 e4 mtsrin r9,r10
51c8: 39 29 01 11 addi r9,r9,273
51cc: 3d 4a 10 00 addis r10,r10,4096
51d0: 55 29 02 06 rlwinm r9,r9,0,8,3
51d4: 7d 20 51 e4 mtsrin r9,r10
51d8: 3d 4a 10 00 addis r10,r10,4096
51dc: 39 29 01 11 addi r9,r9,273
51e0: 7c 08 50 40 cmplw r8,r10
51e4: 55 29 02 06 rlwinm r9,r9,0,8,3
51e8: 40 81 00 4c ble 5234 <perf_copy_attr+0x48c>
51ec: 7d 20 51 e4 mtsrin r9,r10
51f0: 39 29 01 11 addi r9,r9,273
51f4: 3c ea 10 00 addis r7,r10,4096
51f8: 55 29 02 06 rlwinm r9,r9,0,8,3
51fc: 7d 20 39 e4 mtsrin r9,r7
5200: 39 29 01 11 addi r9,r9,273
5204: 3c e7 10 00 addis r7,r7,4096
5208: 55 29 02 06 rlwinm r9,r9,0,8,3
520c: 7d 20 39 e4 mtsrin r9,r7
5210: 39 29 01 11 addi r9,r9,273
5214: 3c ea 30 00 addis r7,r10,12288
5218: 55 29 02 06 rlwinm r9,r9,0,8,3
521c: 7d 20 39 e4 mtsrin r9,r7
5220: 3d 4a 40 00 addis r10,r10,16384
5224: 39 29 01 11 addi r9,r9,273
5228: 7c 08 50 40 cmplw r8,r10
522c: 55 29 02 06 rlwinm r9,r9,0,8,3
5230: 41 81 ff bc bgt 51ec <perf_copy_attr+0x444>
5234: 4c 00 01 2c isync
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
[mpe: Export the ool handlers to fix build errors]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d9121f96a7c4302946839a0771f5d1daeeb6968c.1622708530.git.christophe.leroy@csgroup.eu
Diffstat (limited to 'arch/powerpc/mm/book3s32')
| -rw-r--r-- | arch/powerpc/mm/book3s32/kuap.c | 14 |
1 files changed, 13 insertions, 1 deletions
diff --git a/arch/powerpc/mm/book3s32/kuap.c b/arch/powerpc/mm/book3s32/kuap.c index 0e59c8b32c4f..9df6911b8fde 100644 --- a/arch/powerpc/mm/book3s32/kuap.c +++ b/arch/powerpc/mm/book3s32/kuap.c @@ -6,10 +6,22 @@ struct static_key_false disable_kuap_key; EXPORT_SYMBOL(disable_kuap_key); +void kuap_lock_all_ool(void) +{ + kuap_lock_all(); +} +EXPORT_SYMBOL(kuap_lock_all_ool); + +void kuap_unlock_all_ool(void) +{ + kuap_unlock_all(); +} +EXPORT_SYMBOL(kuap_unlock_all_ool); + void __init setup_kuap(bool disabled) { if (!disabled) - kuap_update_sr(mfsr(0) | SR_KS, 0, TASK_SIZE); + kuap_lock_all_ool(); if (smp_processor_id() != boot_cpuid) return; |