Re: [BUG][PATCH] arm64: bti: fix BTI to handle local indirect branches

From: Jeremy Linton
Date: Mon Oct 05 2020 - 16:10:46 EST


Hi,


On 10/5/20 2:59 PM, Mark Brown wrote:
On Mon, Oct 05, 2020 at 01:18:04PM -0500, Jeremy Linton wrote:

The AES code uses a 'br x7' as part of a function called by
a macro, that ends up needing a BTI_J as a target. Lets
define SYN_CODE_START_LOCAL() for this and replace the
SYM_FUNC_START_LOCAL with a SYM_FUNC_CODE_LOCAL in the AES block.

Really what the subject here should say is that this code is not a
standard function and therefore should not be annotated as such - it's
wrong with or without BTI, BTI just makes it very apparent. It'd also
be better to split the change in linkage.h out into a separate patch,
that'd make things clearer for review.

CPU: 1 PID: 265 Comm: cryptomgr_test Not tainted 5.8.11-300.fc33.aarch64 #1
pstate: 20400c05 (nzCv daif +PAN -UAO BTYPE=j-)
pc : aesbs_encrypt8+0x0/0x5f0 [aes_neon_bs]
lr : aesbs_xts_encrypt+0x48/0xe0 [aes_neon_bs]
sp : ffff80001052b730
x29: ffff80001052b730 x28: 0000000000000001
x27: ffff0001ec8f4000 x26: ffff0001ec5d27b0

Please think hard before including complete backtraces in upstream
reports, they are very large and contain almost no useful information
relative to their size so often obscure the relevant content in your
message. If part of the backtrace is usefully illustrative (it often is
for search engines if nothing else) then it's usually better to pull out
the relevant sections.

-SYM_FUNC_START_LOCAL(aesbs_encrypt8)
+SYM_CODE_START_LOCAL(aesbs_encrypt8)
ldr q9, [bskey], #16 // round 0 key
ldr q8, M0SR
ldr q24, SR
@@ -488,10 +488,10 @@ SYM_FUNC_START_LOCAL(aesbs_encrypt8)
eor v2.16b, v2.16b, v12.16b
eor v5.16b, v5.16b, v12.16b
ret
-SYM_FUNC_END(aesbs_encrypt8)
+SYM_END(aesbs_encrypt8)

This should be SYM_CODE_END() to match the opening. However...

* When using in-kernel BTI we need to ensure that PCS-conformant assembly
@@ -42,6 +43,9 @@
SYM_START(name, SYM_L_WEAK, SYM_A_NONE) \
BTI_C
+#define SYM_CODE_START_LOCAL(name) \
+ SYM_START(name, SYM_L_LOCAL, SYM_A_ALIGN) \
+ BTI_JC

...this is going to cause problems, SYM_CODE means that we should
assemble *exactly* what was written since it's some non-standard thing -
we use it for the vectors table for example. Looking at the code it's
not 100% clear that the best approach here isn't just to change the call
to a regular function call, this isn't a fast path or anything as far as
I can see so it's unclear to me why we need to tail call.

Well for some workloads its could be AFAIK. OTOH, Ard mentioned dumping the tail call too, and I think that is pretty reasonable. So it looks like that is a better plan since it also avoids all this SYM_ flailing.



Failing that I think we need an annotation for tail called functions,
that'd need to be a new thing as I am not seeing anything appropriate in
the current generic annotations.