s390x: add tests for diag258
v2 cover s390x: add tests for diag258
v2:
---
* do not run test under TCG
Add tests for diag258 handling on s390x.
There recently was a bugfix in the kernel:
https://lore.kernel.org/r/[email protected]
This adds tests for it.
Nico Boehr (2):
s390x: add test for diag258
s390x: edat: move LC_SIZE to arch_def.h
lib/s390x/asm/arch_def.h | 1 + s390x/Makefile | 1 + s390x/diag258.c | 258 +++++++++++++++++++++++++++++++++++++++ s390x/edat.c | 1 - s390x/unittests.cfg | 3 + 5 files changed, 263 insertions(+), 1 deletion(-) create mode 100644 s390x/diag258.c
Signature · 1 line
2.46.0
PATCH v2 1/2 s390x: add test for diag258
This adds a test for diag258 (page ref service/async page fault).
There recently was a virtual-real address confusion bug, so we should test:
- diag258 parameter Rx is a real adress
- crossing the end of RAM with the parameter list yields an addressing exception
- invalid diagcode in the parameter block yields an specification exception
- diag258 correctly applies prefixing.
Note that we're just testing error cases as of now.
--- s390x/Makefile | 1 + s390x/diag258.c | 258 ++++++++++++++++++++++++++++++++++++++++++++ s390x/unittests.cfg | 3 + 3 files changed, 262 insertions(+) create mode 100644 s390x/diag258.c
s390x/Makefile +1 -0
@@ -44,6 +44,7 @@ tests += $(TEST_DIR)/exittime.elf tests += $(TEST_DIR)/ex.elf tests += $(TEST_DIR)/topology.elf tests += $(TEST_DIR)/sie-dat.elf +tests += $(TEST_DIR)/diag258.elf pv-tests += $(TEST_DIR)/pv-diags.elf pv-tests += $(TEST_DIR)/pv-icptcode.elf
s390x/diag258.c new file +258 -0
258 lines not on this page · raw message
s390x/unittests.cfg +3 -0
@@ -392,3 +392,6 @@ file = sie-dat.elf [pv-attest] file = pv-attest.elf + +[diag258] +file = diag258.elf
Signature · 1 line
2.46.0
Re: [kvm-unit-tests PATCH v2 1/2] s390x: add test for diag258
Quoted 50 lines · On Mon, 23 Sep 2024 08:26:03 +0200 Nico Boehr <[email protected]> wrote:
This adds a test for diag258 (page ref service/async page fault).
There recently was a virtual-real address confusion bug, so we should test:
- diag258 parameter Rx is a real adress
- crossing the end of RAM with the parameter list yields an addressing exception
- invalid diagcode in the parameter block yields an specification exception
- diag258 correctly applies prefixing.Note that we're just testing error cases as of now.
Signed-off-by Nico Boehr <[email protected]>--- s390x/Makefile | 1 + s390x/diag258.c | 258 ++++++++++++++++++++++++++++++++++++++++++++ s390x/unittests.cfg | 3 + 3 files changed, 262 insertions(+) create mode 100644 s390x/diag258.cs390x/Makefile +1
-0@@ -44,6 +44,7 @@ tests += $(TEST_DIR)/exittime.elf tests += $(TEST_DIR)/ex.elf tests += $(TEST_DIR)/topology.elf tests += $(TEST_DIR)/sie-dat.elf +tests += $(TEST_DIR)/diag258.elf pv-tests += $(TEST_DIR)/pv-diags.elf pv-tests += $(TEST_DIR)/pv-icptcode.elfs390x/diag258.c new file +11
-0@@ -0,0 +1,258 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* +* Diag 258: Async Page Fault Handler +* +* Copyright (c) 2024 IBM Corp +* +* Authors: +* Nico Boehr <[email protected]> +*/ + +#include <libcflat.h>
s390x/diag258.c line 21
Quoted 10 lines
s390x/diag258.c new file +10
-0+#include <asm-generic/barrier.h> +#include <asm/asm-offsets.h> +#include <asm/interrupt.h> +#include <asm/mem.h> +#include <asm/pgtable.h> +#include <mmu.h> +#include <sclp.h> +#include <vmalloc.h> + +static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));
wait, you are using LC_SIZE in this file... even though it's only defined in the next patch.
I think you should swap patch 1 and 2
Quoted 65 lines
s390x/diag258.c new file +65
-0+ +#define __PF_RES_FIELD 0x8000000000000000UL + +/* copied from Linux arch/s390/mm/pfault.c */ +struct pfault_refbk { +u16 refdiagc; +u16 reffcode; +u16 refdwlen; +u16 refversn; +u64 refgaddr; +u64 refselmk; +u64 refcmpmk; +u64 reserved; +}; + +uint64_t pfault_token =0x0123fadec0fe3210UL ; + +static struct pfault_refbk pfault_init_refbk __attribute__((aligned(8))) = { + .refdiagc = 0x258, + .reffcode = 0, /* TOKEN */ + .refdwlen = sizeof(struct pfault_refbk) / sizeof(uint64_t), + .refversn = 2, + .refgaddr = (u64)&pfault_token, + .refselmk = 1UL << 48, + .refcmpmk = 1UL << 48, + .reserved = __PF_RES_FIELD +}; + +static struct pfault_refbk pfault_cancel_refbk __attribute((aligned(8))) = { + .refdiagc = 0x258, + .reffcode = 1, /* CANCEL */ + .refdwlen = sizeof(struct pfault_refbk) / sizeof(uint64_t), + .refversn = 2, + .refgaddr = 0, + .refselmk = 0, + .refcmpmk = 0, + .reserved = 0 +}; + +static inline int diag258 (struct pfault_refbk *refbk) +{ +int rc =-1 ; + + asmvolatile ( +" diag %[refbk],%[rc],0x258\n" + : [rc]"+d" (rc) + : [refbk]"a" (refbk),"m" (*(refbk)) + :"cc" ); +return rc; +} + +static void test_priv (void ) +{ +report_prefix_push ("privileged" ); +expect_pgm_int (); +enter_pstate (); +diag258 (&pfault_init_refbk); +check_pgm_int_code (PGM_INT_CODE_PRIVILEGED_OPERATION ); +report_prefix_pop (); +} + +static void *page_map_outside_real_space (phys_addr_t page_real) +{ +pgd_t *root = (pgd_t *)(stctg (1 ) &PAGE_MASK ); +void * vaddr =alloc_vpage ();
s390x/diag258.c line 96
Quoted 10 lines
s390x/diag258.c new file +10
-0+ +install_page (root, page_real, vaddr); + +return vaddr; +} + +/* +* Verify that the refbk pointer is a real address and not a virtual +* address. This is tested by enabling DAT and establishing a mapping +* for the refbk that is outside of the bounds of our (guest-)physical
s/physical/real/ (or absolute)
s390x/diag258.c line 103
Quoted 7 lines
s390x/diag258.c new file +7
-0+* address space. +*/ +static void test_refbk_real (void ) +{ +pgd_t *root; +struct pfault_refbk *refbk; +void *refbk_page;
please reverse Christmas tree
Quoted 31 lines
s390x/diag258.c new file +31
-0+ +report_prefix_push ("refbk is real" ); + +/* Set up virtual memory and allocate a physical page for storing the refbk */ +setup_vm (); + refbk_page =alloc_page (); + +/* Map refblk page outside of physical memory identity mapping */ + root = (pgd_t *)(stctg (1 ) &PAGE_MASK ); + refbk =page_map_outside_real_space (virt_to_pte_phys (root, refbk_page)); + +/* Assert the mapping really is outside identity mapping */ +report_info ("refbk is at 0x%lx" , (u64 )refbk); +report_info ("ram size is 0x%lx" ,get_ram_size ()); +assert ((u64 )refbk >get_ram_size ()); + +/* Copy the init refbk to the page */ +memcpy (refbk, &pfault_init_refbk,sizeof (struct pfault_refbk )); + +/* Protect the virtual mapping to avoid diag258 actually doing something */ +protect_page (refbk,PAGE_ENTRY_I ); + +expect_pgm_int (); +diag258 (refbk); +check_pgm_int_code (PGM_INT_CODE_ADDRESSING ); +report_prefix_pop (); + +free_page (refbk_page); +disable_dat (); +irq_set_dat_mode (false,0 ); +}
s390x/diag258.c line 144
Quoted 10 lines
s390x/diag258.c new file +10
-0+ +/* +* Verify diag258 correctly applies prefixing. +*/ +static void test_refbk_prefixing (void ) +{ +uint64_t ry; +uint32_t old_prefix; +struct pfault_refbk *refbk_in_prefix, *refbk_in_reverse_prefix; +const size_t lowcore_offset_for_refbk = offsetof(struct lowcore , pad_0x03a0);
reverse Christmas tree here too
Quoted 44 lines
s390x/diag258.c new file +44
-0+ +report_prefix_push ("refbk prefixing" ); + +report_info ("refbk at lowcore offset 0x%lx" , lowcore_offset_for_refbk); + +assert ((unsigned long )&prefix_buf <SZ_2G ); + +memcpy (prefix_buf,0 ,LC_SIZE ); + +/* +* After the call to set_prefix() below, this will refer to absolute +* address lowcore_offset_for_refbk (reverse prefixing). +*/ + refbk_in_reverse_prefix = (struct pfault_refbk *)(&prefix_buf[0 ] + lowcore_offset_for_refbk); + +/* +* After the call to set_prefix() below, this will refer to absolute +* address &prefix_buf[0] + lowcore_offset_for_refbk (forward prefixing). +*/ + refbk_in_prefix = (struct pfault_refbk *)OPAQUE_PTR (lowcore_offset_for_refbk); + + old_prefix =get_prefix (); +set_prefix ((uint32_t )(uintptr_t )prefix_buf); + +/* +* If diag258 would not be applying prefixing on access to +* refbk_in_reverse_prefix correctly, it would access absolute address +* refbk_in_reverse_prefix (which to us is accessible at real address +* refbk_in_prefix). +* Make sure it really fails by putting invalid function code +* at refbk_in_prefix. +*/ + refbk_in_prefix->refdiagc =0xc0fe ; + +/* +* Put a valid refbk at refbk_in_reverse_prefix. +*/ +memcpy (refbk_in_reverse_prefix, &pfault_init_refbk,sizeof (pfault_init_refbk)); + + ry =diag258 (refbk_in_reverse_prefix); +report (!ry,"real address refbk accessed" ); + +/* +* Activating should have worked. Cancel the activation and expect
s390x/diag258.c line 198
Quoted 10 lines
s390x/diag258.c new file +10
-0+* return 0. If activation would not have worked, this should return with +* 4 (pfault handshaking not active). +*/ + ry =diag258 (&pfault_cancel_refbk); +report (!ry,"handshaking canceled" ); + +set_prefix (old_prefix); + +report_prefix_pop (); +}
it seems like you are only testing the first page of lowcore; can you expand the test to also test the second page?
s390x/diag258.c line 206
Quoted 8 lines
s390x/diag258.c new file +8
-0+ +/* +* Verify that a refbk exceeding physical memory is not accepted, even +* when crossing a frame boundary. +*/ +static void test_refbk_crossing (void ) +{ +const size_t bytes_in_last_page =8 ;
are there any alignment requirements for the buffer?
if so, that should also be tested (either that a fault is triggered or that the lowest bytes are ignored, depending on how it is defined to work)
Quoted 16 lines
s390x/diag258.c new file +16
-0+struct pfault_refbk *refbk = (struct pfault_refbk *)(get_ram_size () - bytes_in_last_page); + +report_prefix_push ("refbk crossing" ); + +report_info ("refbk is at 0x%lx" , (u64 )refbk); +report_info ("ram size is 0x%lx" ,get_ram_size ()); +assert (sizeof (struct pfault_refbk ) > bytes_in_last_page); + +/* Copy bytes_in_last_page bytes of the init refbk to the page */ +memcpy (refbk, &pfault_init_refbk, bytes_in_last_page); + +expect_pgm_int (); +diag258 (refbk); +check_pgm_int_code (PGM_INT_CODE_ADDRESSING ); +report_prefix_pop (); +}
s390x/diag258.c line 232
Quoted 10 lines
s390x/diag258.c new file +10
-0+ +/* +* Verify that a refbk with an invalid refdiagc is not accepted. +*/ +static void test_refbk_invalid_diagcode (void ) +{ +struct pfault_refbk refbk = pfault_init_refbk; + +report_prefix_push ("invalid refdiagc" ); + refbk.refdiagc =0xc0fe ;
other testcases in this file depend on invalid codes failing; maybe move this test up?
Quoted 8 lines
s390x/diag258.c new file +8
-0+ +expect_pgm_int (); +diag258 (&refbk); +check_pgm_int_code (PGM_INT_CODE_SPECIFICATION ); +report_prefix_pop (); +} + +int main (void )
s390x/diag258.c line 250
Quoted 10 lines
s390x/diag258.c new file +10
-0+{ +report_prefix_push ("diag258" ); + +expect_pgm_int (); +diag258 (&pfault_init_refbk); +if (clear_pgm_int () ==PGM_INT_CODE_SPECIFICATION ) { +report_skip ("diag258 not supported" ); + }else { +test_priv (); +test_refbk_real ();
should probably go here....
s390x/diag258.c line 253
Quoted 3 lines
s390x/diag258.c new file +3
-0+test_refbk_prefixing (); +test_refbk_crossing (); +test_refbk_invalid_diagcode ();
.... this one ^
Quoted 16 lines
s390x/diag258.c new file +5
-0+ } + +report_prefix_pop (); +return report_summary (); +}s390x/unittests.cfg +3
-0@@ -392,3 +392,6 @@ file = sie-dat.elf [pv-attest] file = pv-attest.elf + +[diag258] +file = diag258.elf
Re: [kvm-unit-tests PATCH v2 1/2] s390x: add test for diag258
Quoting Claudio Imbrenda (2024-09-25 17:34:52)
Quoted 6 lines
Quoted 1 line
+static
uint8_tprefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));wait, you are using
LC_SIZEin this file... even though it's only defined in the next patch.I think you should swap patch 1 and 2
will do
[...]
Quoted 6 lines
Quoted 4 lines
+/*
+ * Verify that the refbk pointer is a real address and not a virtual
+ * address. This is tested by enabling DAT and establishing a mapping
+ * for the refbk that is outside of the bounds of our (guest-)physicals/physical/real/ (or absolute)
Huh? Why? Talking about the size here, so absolute, real and physical are all equivalent here.
[...]
Quoted 15 lines
Quoted 15 lines
s390x/diag258.c new file +15
-0+/* +* Put a valid refbk at refbk_in_reverse_prefix. +*/ +memcpy (refbk_in_reverse_prefix, &pfault_init_refbk,sizeof (pfault_init_refbk)); + + ry =diag258 (refbk_in_reverse_prefix); +report (!ry,"real address refbk accessed" ); + +/* +* Activating should have worked. Cancel the activation and expect +* return 0. If activation would not have worked, this should return with +* 4 (pfault handshaking not active). +*/ + ry =diag258 (&pfault_cancel_refbk); +report (!ry,"handshaking canceled" );
Quoted 8 lines
s390x/diag258.c line 198
Quoted 5 lines
s390x/diag258.c new file +5
-0+ +set_prefix (old_prefix); + +report_prefix_pop (); +}it seems like you are only testing the first page of lowcore; can you expand the test to also test the second page?
Would you mind leaving this for a future extension?
Quoted 5 lines
Quoted 5 lines
s390x/diag258.c new file +5
-0+ +/* +* Verify that a refbk exceeding physical memory is not accepted, even +* when crossing a frame boundary. +*/
Quoted 8 lines
s390x/diag258.c line 206
Quoted 3 lines
s390x/diag258.c new file +3
-0+static void test_refbk_crossing (void ) +{ +const size_t bytes_in_last_page =8 ;are there any alignment requirements for the buffer?
if so, that should also be tested (either that a fault is triggered or that the lowest bytes are ignored, depending on how it is defined to work)
There are alignment requirements. Would it be OK to do this in a future extension?
Quoted 4 lines
Quoted 4 lines
s390x/diag258.c new file +4
-0+/* +* Verify that a refbk with an invalid refdiagc is not accepted. +*/ +static void test_refbk_invalid_diagcode (void )
Quoted 8 lines
s390x/diag258.c line 232
Quoted 5 lines
s390x/diag258.c new file +5
-0+{ +struct pfault_refbk refbk = pfault_init_refbk; + +report_prefix_push ("invalid refdiagc" ); + refbk.refdiagc =0xc0fe ;other testcases in this file depend on invalid codes failing; maybe move this test up?
Yes, thanks, done.
Quoted 5 lines
Quoted 5 lines
s390x/diag258.c new file +5
-0+int main (void ) +{ +report_prefix_push ("diag258" ); + +expect_pgm_int ();
Quoted 8 lines
s390x/diag258.c line 250
Quoted 6 lines
s390x/diag258.c new file +6
-0+diag258 (&pfault_init_refbk); +if (clear_pgm_int () ==PGM_INT_CODE_SPECIFICATION ) { +report_skip ("diag258 not supported" ); + }else { +test_priv (); +test_refbk_real ();should probably go here....
test_refbk_real() relies on the invalid diagcode doing nothing, so it should go *before* that one.
PATCH v2 2/2 s390x: edat: move LC_SIZE to arch_def.h
In reply to Nico Boehrstruct lowcore is defined in arch_def.h and LC_SIZE is useful to other tests as well, therefore move it to arch_def.h.
--- lib/s390x/asm/arch_def.h | 1 + s390x/edat.c | 1 - 2 files changed, 1 insertion(+), 1 deletion(-)
lib/s390x/asm/arch_def.h +1 -0
@@ -119,6 +119,7 @@ enum address_space {#define CTL2_GUARDED_STORAGE (63 - 59) +#define LC_SIZE (2 * PAGE_SIZE)struct lowcore {uint8_t pad_0x0000[0x0080 -0x0000 ];/* 0x0000 */ uint32_t ext_int_param;/* 0x0080 */
s390x/edat.c +0 -1
@@ -17,7 +17,6 @@#define PGD_PAGE_SHIFT (REGION1_SHIFT - PAGE_SHIFT) -#define LC_SIZE (2 * PAGE_SIZE)#define VIRT (x) ((void *)((unsigned long)(x) + (unsigned long)mem)) static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));
Signature · 1 line
2.46.0
Re: [kvm-unit-tests PATCH v2 2/2] s390x: edat: move LC_SIZE to arch_def.h
Quoted 4 lines · On Mon, 23 Sep 2024 08:26:04 +0200 Nico Boehr <[email protected]> wrote:
struct lowcore is defined in
arch_def.handLC_SIZEis useful to other tests as well, therefore move it toarch_def.h.Signed-off-by Nico Boehr <[email protected]>
I find this patch a little weird here, since it's not used for the previous patch, and doesn't seem to have anything to do with it either.
the patch itself is otherwise good, I'm just a little puzzled.
nonetheless:
Quoted 29 lines
--- lib/s390x/asm/arch_def.h | 1 + s390x/edat.c | 1 - 2 files changed, 1 insertion(+), 1 deletion(-)lib/s390x/asm/arch_def.h +1
-0@@ -119,6 +119,7 @@ enum address_space {#define CTL2_GUARDED_STORAGE (63 - 59) +#define LC_SIZE (2 * PAGE_SIZE)struct lowcore {uint8_t pad_0x0000[0x0080 -0x0000 ];/* 0x0000 */ uint32_t ext_int_param;/* 0x0080 */ s390x/edat.c +0
-1@@ -17,7 +17,6 @@#define PGD_PAGE_SHIFT (REGION1_SHIFT - PAGE_SHIFT) -#define LC_SIZE (2 * PAGE_SIZE)#define VIRT (x) ((void *)((unsigned long)(x) + (unsigned long)mem)) static uint8_t prefix_buf[LC_SIZE] __attribute__((aligned(LC_SIZE)));