From c74a3041f019d6492ff0cd4d46d694fc0262e65f Mon Sep 17 00:00:00 2001
From: Conrad Meyer <cem@FreeBSD.org>
Date: Fri, 3 Jul 2020 14:54:46 +0000
Subject: [PATCH] Add domain policy allocation for amd64 fpu_kern_ctx

Like other types of allocation, fpu_kern_ctx are frequently allocated per-cpu.
Provide the API and sketch some example consumers.

fpu_kern_alloc_ctx_domain() preferentially allocates memory from the
provided domain, and falls back to other domains if that one is empty
(DOMAINSET_PREF(domain) policy).

Maybe it makes more sense to just shove one of these in the DPCPU area
sooner or later -- left for future work.

Reviewed by:	markj
Differential Revision:	https://reviews.freebsd.org/D22053
---
 sys/amd64/amd64/fpu.c                | 31 +++++++++++++++++++++-------
 sys/amd64/include/fpu.h              |  1 +
 sys/crypto/aesni/aesni.c             |  7 ++++++-
 sys/crypto/blake2/blake2_cryptodev.c |  7 ++++++-
 4 files changed, 36 insertions(+), 10 deletions(-)

diff --git a/sys/amd64/amd64/fpu.c b/sys/amd64/amd64/fpu.c
index b554394a17b..53819131e5b 100644
--- a/sys/amd64/amd64/fpu.c
+++ b/sys/amd64/amd64/fpu.c
@@ -38,6 +38,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/bus.h>
+#include <sys/domainset.h>
 #include <sys/kernel.h>
 #include <sys/lock.h>
 #include <sys/malloc.h>
@@ -1030,17 +1031,31 @@ struct fpu_kern_ctx {
 	char hwstate1[];
 };
 
+static inline size_t __pure2
+fpu_kern_alloc_sz(u_int max_est)
+{
+	return (sizeof(struct fpu_kern_ctx) + XSAVE_AREA_ALIGN + max_est);
+}
+
+static inline int __pure2
+fpu_kern_malloc_flags(u_int fpflags)
+{
+	return (((fpflags & FPU_KERN_NOWAIT) ? M_NOWAIT : M_WAITOK) | M_ZERO);
+}
+
+struct fpu_kern_ctx *
+fpu_kern_alloc_ctx_domain(int domain, u_int flags)
+{
+	return (malloc_domainset(fpu_kern_alloc_sz(cpu_max_ext_state_size),
+	    M_FPUKERN_CTX, DOMAINSET_PREF(domain),
+	    fpu_kern_malloc_flags(flags)));
+}
+
 struct fpu_kern_ctx *
 fpu_kern_alloc_ctx(u_int flags)
 {
-	struct fpu_kern_ctx *res;
-	size_t sz;
-
-	sz = sizeof(struct fpu_kern_ctx) + XSAVE_AREA_ALIGN +
-	    cpu_max_ext_state_size;
-	res = malloc(sz, M_FPUKERN_CTX, ((flags & FPU_KERN_NOWAIT) ?
-	    M_NOWAIT : M_WAITOK) | M_ZERO);
-	return (res);
+	return (malloc(fpu_kern_alloc_sz(cpu_max_ext_state_size),
+	    M_FPUKERN_CTX, fpu_kern_malloc_flags(flags)));
 }
 
 void
diff --git a/sys/amd64/include/fpu.h b/sys/amd64/include/fpu.h
index 6b9782e7207..8f957eede39 100644
--- a/sys/amd64/include/fpu.h
+++ b/sys/amd64/include/fpu.h
@@ -71,6 +71,7 @@ int	fputrap_sse(void);
 int	fputrap_x87(void);
 void	fpuuserinited(struct thread *td);
 struct fpu_kern_ctx *fpu_kern_alloc_ctx(u_int flags);
+struct fpu_kern_ctx *fpu_kern_alloc_ctx_domain(int domain, u_int flags);
 void	fpu_kern_free_ctx(struct fpu_kern_ctx *ctx);
 void	fpu_kern_enter(struct thread *td, struct fpu_kern_ctx *ctx,
 	    u_int flags);
diff --git a/sys/crypto/aesni/aesni.c b/sys/crypto/aesni/aesni.c
index 42973b50718..3a5a9e0fb1f 100644
--- a/sys/crypto/aesni/aesni.c
+++ b/sys/crypto/aesni/aesni.c
@@ -180,7 +180,12 @@ aesni_attach(device_t dev)
 	    M_WAITOK|M_ZERO);
 
 	CPU_FOREACH(i) {
-		ctx_fpu[i] = fpu_kern_alloc_ctx(0);
+#ifdef __amd64__
+		ctx_fpu[i] = fpu_kern_alloc_ctx_domain(
+		    pcpu_find(i)->pc_domain, FPU_KERN_NORMAL);
+#else
+		ctx_fpu[i] = fpu_kern_alloc_ctx(FPU_KERN_NORMAL);
+#endif
 		mtx_init(&ctx_mtx[i], "anifpumtx", NULL, MTX_DEF|MTX_NEW);
 	}
 
diff --git a/sys/crypto/blake2/blake2_cryptodev.c b/sys/crypto/blake2/blake2_cryptodev.c
index c374be28ea1..fcb3ec56059 100644
--- a/sys/crypto/blake2/blake2_cryptodev.c
+++ b/sys/crypto/blake2/blake2_cryptodev.c
@@ -142,7 +142,12 @@ blake2_attach(device_t dev)
 	    M_WAITOK | M_ZERO);
 
 	CPU_FOREACH(i) {
-		ctx_fpu[i] = fpu_kern_alloc_ctx(0);
+#ifdef __amd64__
+		ctx_fpu[i] = fpu_kern_alloc_ctx_domain(
+		    pcpu_find(i)->pc_domain, FPU_KERN_NORMAL);
+#else
+		ctx_fpu[i] = fpu_kern_alloc_ctx(FPU_KERN_NORMAL);
+#endif
 		mtx_init(&ctx_mtx[i], "bl2fpumtx", NULL, MTX_DEF | MTX_NEW);
 	}