Re: [RFC 03/16] kgr: initial code

[Andi Kleen <andi@firstfloor.org>]

Jiri Slaby <jslaby@suse.cz> writes:
>  	OFFSET(crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
> diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
> index 1e96c3628bf2..a03b1e9d2de3 100644
> --- a/arch/x86/kernel/entry_64.S
> +++ b/arch/x86/kernel/entry_64.S
> @@ -615,6 +615,7 @@ GLOBAL(system_call_after_swapgs)
>  	movq  %rax,ORIG_RAX-ARGOFFSET(%rsp)
>  	movq  %rcx,RIP-ARGOFFSET(%rsp)
>  	CFI_REL_OFFSET rip,RIP-ARGOFFSET
> +	movw $0, TI_kgr_in_progress+THREAD_INFO(%rsp,RIP-ARGOFFSET)

Better use 4 bytes. This has the potential to cause an expensive
Length Changing Prefixes Stall on Intel CPUs.
> +
> +static int kgr_init_ftrace_ops(const struct kgr_patch_fun *patch_fun)
> +{
> +	struct kgr_loc_caches *caches;
> +	unsigned long fentry_loc;
> +
> +	/*
> +	 * Initialize the ftrace_ops->private with pointers to the fentry
> +	 * sites of both old and new functions. This is used as a
> +	 * redirection target in the per-arch stubs.
> +	 *
> +	 * Beware! -- freeing (once unloading will be implemented)
> +	 * will require synchronize_sched() etc.
> +	 */
> +
> +	caches = kmalloc(sizeof(*caches), GFP_KERNEL);
> +	if (!caches) {
> +		kgr_debug("kgr: unable to allocate fentry caches\n");
> +		return -ENOMEM;
> +	}

All the error paths in this function leak memory.


-Andi
-- 
ak@linux.intel.com -- Speaking for myself only