[v11, 00/10] riscv: support kernel-mode Vector

[Björn Töpel]

Hi Andy,

On Mon, 15 Jan 2024 at 06:59, Andy Chiu <andy.chiu at sifive.com> wrote:
>
> This series provides support running Vector in kernel mode.
> Additionally, kernel-mode Vector can be configured to run without
> turnning off preemption on a CONFIG_PREEMPT kernel. Along with the
> suport, we add Vector optimized copy_{to,from}_user. And provide a
> simple threshold to decide when to run the vectorized functions.

[...]

> Changelog v11:
>  - This is a quick respin to address boot failing on ubuntu and alpine.
>  - Pass the updated copy size when calling scalar fallback.
>  - Skip some bytes for scalar fallback when fauting at vse8.v with a
>    non-zero $vstart. (Guo)
>  - Guard riscv_v_setup_ctx_cache() with  has_vector() check.

For convenience, here's the complete diff from v10:

--8<--
diff --git a/arch/riscv/kernel/vector.c b/arch/riscv/kernel/vector.c
index f9769703fd39..6727d1d3b8f2 100644
--- a/arch/riscv/kernel/vector.c
+++ b/arch/riscv/kernel/vector.c
@@ -53,6 +53,9 @@ int riscv_v_setup_vsize(void)

 void __init riscv_v_setup_ctx_cache(void)
 {
+    if (!has_vector())
+        return;
+
     riscv_v_user_cachep = kmem_cache_create_usercopy("riscv_vector_ctx",
                              riscv_v_vsize, 16, SLAB_PANIC,
                              0, riscv_v_vsize, NULL);
diff --git a/arch/riscv/lib/riscv_v_helpers.c b/arch/riscv/lib/riscv_v_helpers.c
index 6cac8f4e69e9..be38a93cedae 100644
--- a/arch/riscv/lib/riscv_v_helpers.c
+++ b/arch/riscv/lib/riscv_v_helpers.c
@@ -33,6 +33,7 @@ asmlinkage int enter_vector_usercopy(void *dst, void
*src, size_t n)
         copied = n - remain;
         dst += copied;
         src += copied;
+        n = remain;
         goto fallback;
     }

diff --git a/arch/riscv/lib/uaccess_vector.S b/arch/riscv/lib/uaccess_vector.S
index 566739f6331a..51ab5588e9ff 100644
--- a/arch/riscv/lib/uaccess_vector.S
+++ b/arch/riscv/lib/uaccess_vector.S
@@ -23,22 +23,31 @@

 SYM_FUNC_START(__asm_vector_usercopy)
     /* Enable access to user memory */
-    li t6, SR_SUM
-    csrs CSR_STATUS, t6
+    li    t6, SR_SUM
+    csrs    CSR_STATUS, t6

 loop:
     vsetvli iVL, iNum, e8, ELEM_LMUL_SETTING, ta, ma
     fixup vle8.v vData, (pSrc), 10f
-    fixup vse8.v vData, (pDst), 10f
     sub iNum, iNum, iVL
     add pSrc, pSrc, iVL
+    fixup vse8.v vData, (pDst), 11f
     add pDst, pDst, iVL
     bnez iNum, loop

-    /* Exception fixup code. It's the same as normal exit */
+    /* Exception fixup for vector load is shared with normal exit */
 10:
     /* Disable access to user memory */
     csrc    CSR_STATUS, t6
     mv    a0, iNum
     ret
+
+    /* Exception fixup code for vector store. */
+11:
+    /* Undo the subtraction after vle8.v */
+    add    iNum, iNum, iVL
+    /* Make sure the scalar fallback skip already processed bytes */
+    csrr    t2, CSR_VSTART
+    sub    iNum, iNum, t2
+    j    10b
 SYM_FUNC_END(__asm_vector_usercopy)
--8<--

One nit that sticks out is the assembly alignment for
__asm_vector_usercopy, where tab/notab layouts are mixed.

I'll kick off a full build/boot with the v11!


Cheers,
Björn