On Tue Nov 5, 2024 at 6:45 PM CET, Claudio Imbrenda wrote:
> Add a new test to check that the host can use 1M large pages to back
> protected guests when the corresponding feature is present.
>
> Signed-off-by: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
> ---
> s390x/Makefile | 2 +
> lib/s390x/asm/arch_def.h | 1 +
> lib/s390x/asm/uv.h | 18 ++
> s390x/pv-edat1.c | 472 +++++++++++++++++++++++++++++++++++
> s390x/snippets/c/pv-memhog.c | 59 +++++
> 5 files changed, 552 insertions(+)
> create mode 100644 s390x/pv-edat1.c
> create mode 100644 s390x/snippets/c/pv-memhog.c
>
> diff --git a/s390x/Makefile b/s390x/Makefile
> index 23342bd6..c5c6f92c 100644
> --- a/s390x/Makefile
> +++ b/s390x/Makefile
> @@ -48,6 +48,7 @@ tests += $(TEST_DIR)/sie-dat.elf
> pv-tests += $(TEST_DIR)/pv-diags.elf
> pv-tests += $(TEST_DIR)/pv-icptcode.elf
> pv-tests += $(TEST_DIR)/pv-ipl.elf
> +pv-tests += $(TEST_DIR)/pv-edat1.elf
>
> ifneq ($(HOST_KEY_DOCUMENT),)
> ifneq ($(GEN_SE_HEADER),)
> @@ -137,6 +138,7 @@ $(TEST_DIR)/pv-icptcode.elf: pv-snippets += $(SNIPPET_DIR)/asm/icpt-loop.gbin
> $(TEST_DIR)/pv-icptcode.elf: pv-snippets += $(SNIPPET_DIR)/asm/loop.gbin
> $(TEST_DIR)/pv-icptcode.elf: pv-snippets += $(SNIPPET_DIR)/asm/pv-icpt-vir-timing.gbin
> $(TEST_DIR)/pv-ipl.elf: pv-snippets += $(SNIPPET_DIR)/asm/pv-diag-308.gbin
> +$(TEST_DIR)/pv-edat1.elf: pv-snippets += $(SNIPPET_DIR)/c/pv-memhog.gbin
>
> ifneq ($(GEN_SE_HEADER),)
> snippets += $(pv-snippets)
> diff --git a/lib/s390x/asm/arch_def.h b/lib/s390x/asm/arch_def.h
> index 745a3387..481ede8f 100644
> --- a/lib/s390x/asm/arch_def.h
> +++ b/lib/s390x/asm/arch_def.h
> @@ -249,6 +249,7 @@ extern struct lowcore lowcore;
> #define PGM_INT_CODE_REGION_FIRST_TRANS 0x39
> #define PGM_INT_CODE_REGION_SECOND_TRANS 0x3a
> #define PGM_INT_CODE_REGION_THIRD_TRANS 0x3b
> +#define PGM_INT_CODE_SECURE_PAGE_SIZE 0x3c
> #define PGM_INT_CODE_SECURE_STOR_ACCESS 0x3d
> #define PGM_INT_CODE_NON_SECURE_STOR_ACCESS 0x3e
> #define PGM_INT_CODE_SECURE_STOR_VIOLATION 0x3f
> diff --git a/lib/s390x/asm/uv.h b/lib/s390x/asm/uv.h
> index 611dcd3f..7527be48 100644
> --- a/lib/s390x/asm/uv.h
> +++ b/lib/s390x/asm/uv.h
> @@ -35,6 +35,7 @@
> #define UVC_CMD_CONV_TO_SEC_STOR 0x0200
> #define UVC_CMD_CONV_FROM_SEC_STOR 0x0201
> #define UVC_CMD_DESTR_SEC_STOR 0x0202
> +#define UVC_CMD_VERIFY_LARGE_FRAME 0x0203
> #define UVC_CMD_SET_SEC_CONF_PARAMS 0x0300
> #define UVC_CMD_UNPACK_IMG 0x0301
> #define UVC_CMD_VERIFY_IMG 0x0302
> @@ -74,6 +75,11 @@ enum uv_cmds_inst {
> BIT_UVC_CMD_PIN_PAGE_SHARED = 21,
> BIT_UVC_CMD_UNPIN_PAGE_SHARED = 22,
> BIT_UVC_CMD_ATTESTATION = 28,
> + BIT_UVC_CMD_VERIFY_LARGE_FRAME = 32,
> +};
> +
> +enum uv_features {
> + BIT_UV_1M_BACKING = 6,
> };
>
> struct uv_cb_header {
> @@ -312,6 +318,18 @@ static inline int uv_import(uint64_t handle, unsigned long gaddr)
> return uv_call(0, (uint64_t)&uvcb);
> }
>
> +static inline int uv_merge(uint64_t handle, unsigned long gaddr)
> +{
> + struct uv_cb_cts uvcb = {
> + .header.cmd = UVC_CMD_VERIFY_LARGE_FRAME,
> + .header.len = sizeof(uvcb),
> + .guest_handle = handle,
> + .gaddr = gaddr,
> + };
> +
> + return uv_call(0, (uint64_t)&uvcb);
> +}
> +
> static inline int uv_export(unsigned long paddr)
> {
> struct uv_cb_cfs uvcb = {
> diff --git a/s390x/pv-edat1.c b/s390x/pv-edat1.c
> new file mode 100644
> index 00000000..a211c6f9
> --- /dev/null
> +++ b/s390x/pv-edat1.c
> @@ -0,0 +1,472 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * PV huge page backing tests
> + *
> + * Copyright (c) 2024 IBM Corp
> + *
> + * Authors:
> + * Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
> + */
> +#include <libcflat.h>
> +#include <vmalloc.h>
> +#include <sie.h>
> +#include <sclp.h>
> +#include <snippet.h>
> +#include <pv_icptdata.h>
> +#include <asm/interrupt.h>
> +#include <asm/facility.h>
> +#include <asm/uv.h>
> +#include <asm/page.h>
> +#include <asm/mem.h>
> +#include <mmu.h>
> +#include <asm/time.h>
Maybe sort these for improved maintainability.
> +
> +#define GUEST_ORDER REGION3_SHIFT
> +#define GUEST_SIZE BIT_ULL(REGION3_SHIFT)
> +#define N_ITER 32
> +#define MAGIC_ADDRESS 0x12a000
> +
> +#define FIRST 42
> +#define SECOND 23
> +
> +#define STR(x) ((x) ? "1m" : "4k")
> +#define STR3(x) ((x) != -1 ? STR(x) : "no")
> +#define PARAM(n, step) (((unsigned long)(n) << 32) | (step))
> +
> +static struct vm vm;
> +static void *root;
> +
> +extern const char SNIPPET_NAME_START(c, pv_memhog)[];
> +extern const char SNIPPET_NAME_END(c, pv_memhog)[];
> +extern const char SNIPPET_HDR_START(c, pv_memhog)[];
> +extern const char SNIPPET_HDR_END(c, pv_memhog)[];
> +
> +static void init_snippet(struct vm *vm) {
> + const unsigned long size_hdr = SNIPPET_HDR_LEN(c, pv_memhog);
> + const unsigned long size_gbin = SNIPPET_LEN(c, pv_memhog);
> +
> + snippet_pv_init(vm, SNIPPET_NAME_START(c, pv_memhog),
> + SNIPPET_HDR_START(c, pv_memhog),
> + size_gbin, size_hdr, SNIPPET_UNPACK_OFF);
> +}
> +
> +static uint64_t tod_to_us(uint64_t tod)
> +{
> + return tod >> STCK_SHIFT_US;
> +}
> +/*
> +static uint64_t tod_to_ns(uint64_t tod)
> +{
> + return tod_to_us(tod * 1000);
> +}
> +*/
Do we really want to add commented code?
> +
> +static inline uint64_t guest_start(struct vm *vm)
> +{
> + return vm->sblk->mso;
> +}
> +
> +static inline uint64_t guest_end(struct vm *vm)
> +{
> + return vm->sblk->msl + SZ_1M;
> +}
> +
> +static inline uint64_t guest_size(struct vm *vm)
> +{
> + return guest_end(vm) - guest_start(vm);
> +}
> +
> +static uint16_t uvc_merge(struct vm *vm, uint64_t addr)
> +{
> + struct uv_cb_cts uvcb = {
> + .header.cmd = UVC_CMD_VERIFY_LARGE_FRAME,
> + .header.len = sizeof(uvcb),
> + .guest_handle = vm->uv.vm_handle,
> + .gaddr = addr,
> + };
> + int cc;
> +
> + cc = uv_call(0, (uint64_t)&uvcb);
> + assert(cc ? uvcb.header.rc != 1 : uvcb.header.rc == 1);
> + return uvcb.header.rc;
> +}
> +
> +static void map_identity_range(struct vm *vm, uint64_t start, uint64_t end, bool map1m)
> +{
> + uint64_t mem = guest_start(vm);
> +
> + if (map1m) {
> + for (start &= HPAGE_MASK; start < end; start += HPAGE_SIZE)
> + install_large_page(root, mem + start, (void *)(mem + start));
> + } else {
> + for (start &= PAGE_SIZE; start < end; start += PAGE_SIZE)
> + install_page(root, mem + start, (void *)(mem + start));
> + }
> +}
> +
> +static inline void map_identity_all(struct vm *vm, bool map1m)
> +{
> + map_identity_range(vm, 0, guest_size(vm), map1m);
> +}
> +
> +static void import_range(struct vm *vm, uint64_t start, uint64_t end)
> +{
> + for (start &= PAGE_MASK; start < end; start += PAGE_SIZE)
> + assert(uv_import(vm->uv.vm_handle, guest_start(vm) + start) == 0);
> +}
> +
> +static inline void import_all(struct vm *vm)
> +{
> + import_range(vm, 0, guest_size(vm));
> +}
> +
> +static void export_range(struct vm *vm, uint64_t start, uint64_t end)
> +{
> + for (start &= PAGE_MASK; start < end; start += PAGE_SIZE)
> + assert(uv_export(guest_start(vm) + start) == 0);
> +}
> +
> +static void merge_range(struct vm *vm, uint64_t start, uint64_t end)
> +{
> + for (start &= HPAGE_MASK; start < end; start += HPAGE_SIZE)
> + assert(uvc_merge(vm, guest_start(vm) + start) == 1);
> +}
> +
> +static inline void merge_all(struct vm *vm)
> +{
> + merge_range(vm, 0, guest_size(vm));
> +}
> +
> +static inline bool pgm_3c_addr_is(struct vm *vm, uint64_t gaddr)
> +{
> + union teid teid = { .val = lowcore.trans_exc_id };
> + bool res;
> +
> + res = (lowcore.pgm_int_code == PGM_INT_CODE_SECURE_PAGE_SIZE) &&
> + (teid.addr == (guest_start(vm) + gaddr) >> PAGE_SHIFT);
> + clear_pgm_int();
> + return res;
> +}
> +
> +static inline void assert_diag500_val(struct vm *vm, uint64_t val)
> +{
> + assert(pv_icptdata_check_diag(vm, 0x500));
> + assert(vm->save_area.guest.grs[2] == val);
> +}
> +
> +static void run_iterations(struct vm *vm, uint64_t param, bool ptlb_always, bool merge)
> +{
> + uint64_t total, elapsed, best, worst;
> + uint64_t before, after;
> + int i;
> +
> + map_identity_all(vm, 0);
> + init_snippet(vm);
> + import_all(vm);
> + map_identity_all(vm, merge);
> + if (merge)
> + merge_all(vm);
> +
> + sie(vm);
> + assert_diag500_val(vm, FIRST);
> + vm->save_area.guest.grs[2] = PARAM(0, param);
This is just param, is it not?
So we can directly set "= param;" for clarity.
> +
> + total = worst = 0;
s/ / /
> + best = -1;
> +
> + ptlb();
> + for (i = 0; i < N_ITER; i++) {
> + if (ptlb_always)
> + ptlb();
> +
> + stckf(&before);
> + sie(vm);
> + stckf(&after);
> +
> + assert_diag500_val(vm, SECOND);
> + vm->save_area.guest.grs[2] = 42;
> +
> + elapsed = tod_to_us(after - before);
> +
> + total += elapsed;
> +
> + best = best > elapsed ? elapsed : best;
> + worst = worst < elapsed ? elapsed : worst;
> + }
> + uv_destroy_guest(vm);
> +
> + report_info("%5lu %5lu %5lu", best, total / N_ITER, worst);
> +}
> +
> +static void timings(void)
> +{
> + const uint64_t seqs[] = { 0, 1, 3, 7, 17, 27, 63, 127};
> + unsigned int purge, map1m, seq;
> +
> + report_prefix_push("timings");
> +
> + report_info("Averages over %u iterations, in us (best/average/worst)", N_ITER);
> + for (purge = 0; purge < 2; purge++) {
> + report_prefix_pushf("ptlb %s", purge ? "always" : " once");
> + for (seq = 0; seq < ARRAY_SIZE(seqs); seq++) {
> + if (seqs[seq])
> + report_prefix_pushf("seq step %3lu", seqs[seq]);
> + else
> + report_prefix_push("pseudorandom");
> + for (map1m = 0; map1m < 2; map1m++) {
> + report_prefix_pushf("%s", STR(map1m));
> + run_iterations(&vm, seqs[seq], purge, map1m);
> + report_prefix_pop();
> + }
> + report_prefix_pop();
> + }
> + report_prefix_pop();
> + }
> +
> + report_pass("Timing tests successful");
> + report_prefix_pop();
> +}
> +
> +
> +static void do_one_test(struct vm *vm, bool init1m, bool import1m, int merge, bool run1m)
> +{
> + map_identity_all(vm, init1m);
> + init_snippet(vm);
> +
> + map_identity_all(vm, import1m);
> + import_all(vm);
> +
> + if (merge >= 0) {
> + map_identity_all(vm, merge);
> + merge_all(vm);
> + }
> +
> + map_identity_all(vm, run1m);
> + if ((merge == -1) && run1m) {
> + sie(vm);
> + report(vm->sblk->icptcode == ICPT_PV_PREF, "ICPT 112");
> + merge_range(vm, 0, SZ_1M);
> + expect_pgm_int();
> + sie(vm);
> + report(pgm_3c_addr_is(vm, MAGIC_ADDRESS), "PGM 3C at address %#x", MAGIC_ADDRESS);
> + map_identity_all(vm, false);
> + }
> +
> + sie(vm);
> + assert_diag500_val(vm, FIRST);
> + vm->save_area.guest.grs[2] = PARAM(4096, 1);
s/4096/PAGE_SIZE/ ?
> + sie(vm);
> + assert_diag500_val(vm, 23);
> +
> + uv_destroy_guest(vm);
> +}
> +
> +
> +static void test_run(void)
> +{
> + int init1m, import1m, merge, run1m;
> +
> + report_prefix_push("test run");
> +
> + for (init1m = 0; init1m < 1; init1m++) {
> + for (import1m = 0; import1m < 2; import1m++) {
> + for (merge = -1; merge < 2; merge++) {
> + for (run1m = 0; run1m < 2; run1m++) {
> + report_prefix_pushf("init %s, import %s, merge %s, run %s",
> + STR(init1m), STR(import1m), STR3(merge), STR(run1m));
> +
> + do_one_test(&vm, init1m, import1m, merge, run1m);
> + report_pass("Execution successful");
> +
> + report_prefix_pop();
> + }
> + }
> + }
> + }
> +
> + report_prefix_pop();
> +}
> +
> +static void test_merge(void)
> +{
> + uint64_t tmp, mem;
> + int cc;
> +
> + report_prefix_push("merge");
> + init_snippet(&vm);
> +
> + mem = guest_start(&vm);
> +
> + map_identity_all(&vm, false);
> + install_page(root, mem + 0x101000, (void *)(mem + 0x102000));
> + install_page(root, mem + 0x102000, (void *)(mem + 0x101000));
> + install_page(root, mem + 0x205000, (void *)(mem + 0x305000));
> + install_page(root, mem + 0x305000, (void *)(mem + 0x205000));
> + import_range(&vm, 0, 0x400000);
> + import_range(&vm, 0x401000, 0x501000);
> + import_range(&vm, 0x600000, 0x700000);
> +
> + /* Address lower than MSO */
> + report(uvc_merge(&vm, 0) == 0x103, "Address below MSO");
> + report(uvc_merge(&vm, mem - SZ_1M) == 0x103, "Address below MSO");
> +
> + /* Address higher than MSL */
> + install_large_page(root, guest_end(&vm), (void *)guest_end(&vm));
> + report(uvc_merge(&vm, guest_end(&vm)) == 0x103, "Address above MSL");
> +
> + /* Not all pages are imported */
> + report(uvc_merge(&vm, mem + 0x400000) == 0x106, "First page not imported");
> + report(uvc_merge(&vm, mem + 0x500000) == 0x106, "Only first page imported");
> +
> + /* Large 2G page used */
> + tmp = mem & REGION3_ENTRY_RFAA;
> + install_huge_page(root, tmp, (void *)tmp);
> + report(uvc_merge(&vm, mem) == 0x109, "Large region-3 table entry (2G page)");
> + map_identity_range(&vm, 0, SZ_2G, false);
> +
> + /* Not all pages are aligned correctly */
> + report(uvc_merge(&vm, mem + 0x100000) == 0x104, "Pages not consecutive");
> + report(uvc_merge(&vm, mem + 0x200000) == 0x104, "Pages not in the same 1M frame");
> +
> + /* Invalid host virtual to host aboslute mapping */
> + install_large_page(root, get_ram_size(), (void *)(mem + 0x600000));
> + report(uvc_merge(&vm, mem + 0x600000) == 0x107, "Invalid mapping");
> + map_identity_range(&vm, 0x600000, 0x700000, false);
> +
> + /* Success */
> + report(uvc_merge(&vm, mem) == 1, "Successful merge");
> +
> + cc = uv_export(mem + 0xff000);
> + report(cc == 0, "Successful export of merged page");
> +
> + uv_destroy_guest(&vm);
> + report_prefix_pop();
> +}
> +
> +static void test_one_export(struct vm *vm, int pre1m, bool exportfirst, int page, bool post1m)
> +{
> + uint64_t addr = SZ_1M + page * PAGE_SIZE;
> + int expected = FIRST;
> +
> + report_prefix_pushf("page %d", page);
> +
> + map_identity_all(vm, pre1m == 1);
> + init_snippet(vm);
> + import_all(vm);
> + merge_all(vm);
> +
> + if (pre1m != -1) {
> + sie(vm);
> + assert_diag500_val(vm, expected);
> + vm->save_area.guest.grs[2] = PARAM(4, 256 + 42);
> + expected = SECOND;
> + }
> +
> + if (exportfirst) {
> + export_range(vm, addr, addr + PAGE_SIZE);
> + if (pre1m != 1)
> + map_identity_all(vm, true);
> + } else {
> + if (pre1m != 1)
> + map_identity_all(vm, true);
> + export_range(vm, addr, addr + PAGE_SIZE);
> + }
> + expect_pgm_int();
> + sie(vm);
> + assert(pgm_3c_addr_is(vm, MAGIC_ADDRESS));
> +
> + import_range(vm, addr, addr + PAGE_SIZE);
> + if (post1m) {
> + merge_range(vm, SZ_1M, 2 * SZ_1M);
> + } else {
> + map_identity_all(vm, false);
> + }
> + sie(vm);
> + assert_diag500_val(vm, expected);
> + report_pass("Successful");
> +
> + uv_destroy_guest(vm);
> + report_prefix_pop();
> +}
> +
> +static void test_export(void)
> +{
> + int pre1m, post1m, exportfirst;
> +
> + report_prefix_push("export");
> +
> + for (pre1m = -1; pre1m < 1; pre1m++) {
> + for (post1m = 0; post1m < 2; post1m++) {
> + for (exportfirst = 0; exportfirst < 2; exportfirst++) {
> + report_prefix_pushf("%s pre-run, %s post-run, export %s remap",
> + STR3(pre1m), STR(post1m), exportfirst ? "before" : "after");
> +
> + test_one_export(&vm, pre1m, exportfirst, 0, post1m);
> + test_one_export(&vm, pre1m, exportfirst, 1, post1m);
> + test_one_export(&vm, pre1m, exportfirst, 42, post1m);
> + test_one_export(&vm, pre1m, exportfirst, 128, post1m);
> + test_one_export(&vm, pre1m, exportfirst, 254, post1m);
> + test_one_export(&vm, pre1m, exportfirst, 255, post1m);
> +
> + report_prefix_pop();
> + }
> + }
> + }
> +
> + report_prefix_pop();
> +}
> +
> +static bool check_facilities(void)
> +{
> + if (!uv_host_requirement_checks())
> + return false;
> + if (!test_facility(8) || !test_facility(78)) {
> + report_skip("EDAT1 and EDAT2 not available in the host.");
> + return false;
> + }
> + if (!uv_query_test_call(BIT_UVC_CMD_VERIFY_LARGE_FRAME)) {
> + report_skip("Verify Large Frame UVC not supported.");
> + return false;
> + }
> + if (!uv_query_test_feature(BIT_UV_1M_BACKING)) {
> + report_skip("Large frames not supported for Secure Execution.");
> + return false;
> + }
> + return true;
> +}
> +
> +static void init(void)
> +{
> + uint8_t *guest_memory;
> +
> + setup_vm();
> +
> + root = (void *)(stctg(1) & PAGE_MASK);
> + ctl_set_bit(0, CTL0_EDAT);
> +
> + guest_memory = alloc_pages(GUEST_ORDER - PAGE_SHIFT);
> + sie_guest_create(&vm, (uint64_t)guest_memory, GUEST_SIZE);
> + sie_guest_sca_create(&vm);
> + uv_init();
> + uv_setup_asces();
> +}
> +
> +int main(void)
> +{
> + report_prefix_push("uv-edat1-host");
> +
> + if (check_facilities()) {
> + init();
> +
> + test_merge();
> + test_export();
> + test_run();
> +
> + timings();
> +
> + sie_guest_destroy(&vm);
> + }
> +
> + report_prefix_pop();
> + return report_summary();
> +}
> diff --git a/s390x/snippets/c/pv-memhog.c b/s390x/snippets/c/pv-memhog.c
> new file mode 100644
> index 00000000..43f0c2b1
> --- /dev/null
> +++ b/s390x/snippets/c/pv-memhog.c
> @@ -0,0 +1,59 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * Snippet used by the pv-edat1.c test.
> + * This snippet expects to run with at least 2G or memory.
> + *
> + * Copyright (c) 2024 IBM Corp
> + *
> + * Authors:
> + * Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
> + */
> +#include <libcflat.h>
> +#include <asm-generic/page.h>
> +#include <asm/mem.h>
> +
> +#define N_PAGES (SZ_2G / PAGE_SIZE)
> +#define MID_OF_PAGE (PAGE_SIZE / 2)
> +#define MASK_2G (SZ_2G - 1)
> +#define MIDPAGE_PTR(x) ((uint64_t *)((x) + MID_OF_PAGE))
> +
> +static inline void sie_exit(void)
> +{
> + asm volatile("diag 0,0,0x44\n" : : : "memory");
> +}
> +
> +static inline uint64_t get_value(uint64_t res)
> +{
> + asm volatile("lgr %%r2, %[res]\n"
> + " diag 0, 0, 0x500\n"
> + " lgr %[res], %%r2\n"
> + : [res] "+d"(res)
> + :
> + : "2", "memory"
> + );
> + return res;
> +}
> +
> +int main(void)
> +{
> + uint64_t param, addr, i, n;
> +
> + READ_ONCE(*MIDPAGE_PTR(SZ_1M + 42 * PAGE_SIZE));
> + param = get_value(42);
> +
> + n = (param >> 32) & 0x1fffffff;
> + n = n ? n : N_PAGES;
> + param &= 0x7fffffff;
This is ARCH_LOW_ADDRESS_LIMIT. Should we add that to the local arch_def.h or
mem.h to enable reuse?
> +
> + while (true) {
> + for (i = 0; i < n; i++) {
> + addr = ((param ? i * param : i * i * i) * PAGE_SIZE) & MASK_2G;
> + WRITE_ONCE(*MIDPAGE_PTR(addr), addr);
> + }
> +
> + i = get_value(23);
> + if (i != 42)
> + sie_exit();
> + }
> + return 0;
> +}
