Skip to contentSkip to conversation

dump: Add arch section and s390x PV dump

PATCH Superseded Janosch Frank 3 participants 32 messages Jul 21, 2022 – Jul 22, 2022

v3 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v3 cover dump: Add arch section and s390x PV dump

Previously this series was two separate series:
* Arch section support
Adds the possibility for arch code to add custom section data.

* s390 PV dump support
Adds PV dump data to the custom arch sections.

I've chosen to merge them so it's easier to understand why the arch section support has been implement the way it is.

Additionally I've added a cleanup patch beforehand which cleans up the GuestPhysBlock usage.

v3:
* I forgot to reserve the new ELF note so I'm currently discussing its name and over which tree it will be pulled with the kernel devs
* Split code into "dump: Rename write_elf_loads to write_elf_phdr_loads"
* Refined a lot of the commit messages
* Split the string table patch into two: the swap of the section/segment and the string table support
* Renamed write_elf_section_hdr_zero() to prepare_elf_section_hdr_zero()
* Removed rogue code from "dump/dump: Add arch section support"

v2:
* Added "dump: Cleanup memblock usage"
* Fixed whitespace problems and review comments
* Added missing *errp check in dump_end

Janosch Frank (14):
dump: Introduce GuestPhysBlock offset and length filter functions
dump: Rename write_elf_loads to write_elf_phdr_loads
dump: Convert GuestPhysBlock iterators and use the filter functions
dump: Allocate header
dump: Split write of section headers and data and add a prepare step
dump: Reorder struct DumpState
dump: Swap segment and section header locations dump/dump: Add section string table support
dump/dump: Add arch section support
linux header sync
s390x: Add protected dump cap
s390x: Introduce PV query interface
s390x: Add KVM PV dump interface
s390x: pv: Add dump support

 dump/dump.c                  | 432 ++++++++++++++++++++++-------------
 hw/s390x/pv.c                | 112 +++++++++
 hw/s390x/s390-virtio-ccw.c   |   5 +
 include/elf.h                |   1 +
 include/hw/s390x/pv.h        |  18 ++
 include/sysemu/dump-arch.h   |  27 +++
 include/sysemu/dump.h        |  69 +++++-
 linux-headers/linux/kvm.h    |  54 +++++
 target/s390x/arch_dump.c     | 248 +++++++++++++++++---
 target/s390x/kvm/kvm.c       |   7 +
 target/s390x/kvm/kvm_s390x.h |   1 +
 11 files changed, 767 insertions(+), 207 deletions(-)
Signature · 1 line
2.34.1
#1 lore raw

PATCH v3 1/14 dump: Introduce GuestPhysBlock offset and length filter functions

As the code is used in multiple places in dump/dump.c we should introduce functions that calculate the filtered length and offset of a GuestPhysBlock so we can use them in the next patch.

---
 include/sysemu/dump.h | 37 +++++++++++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)
include/sysemu/dump.h +37 -0
index ffc2ea1072..2b39abeeae 100644
@@ -203,4 +203,41 @@ typedef struct DumpState {
 uint16_t cpu_to_dump16(DumpState *s, uint16_t val);
 uint32_t cpu_to_dump32(DumpState *s, uint32_t val);
 uint64_t cpu_to_dump64(DumpState *s, uint64_t val);
+
+static inline int64_t dump_get_memblock_size(GuestPhysBlock *block, int64_t filter_area_start,
+                                             int64_t filter_area_length)
+{
+    int64_t size, left, right;
+
+    /* No filter, return full size */
+    if (!filter_area_length) {
+        return block->target_end - block->target_start;
+    }
+
+    /* calculate the overlapped region. */
+    left = MAX(filter_area_start, block->target_start);
+    right = MIN(filter_area_start + filter_area_length, block->target_end);
+    size = right - left;
+    size = size > 0 ? size : 0;
+
+    return size;
+}
+
+static inline int64_t dump_get_memblock_start(GuestPhysBlock *block, int64_t filter_area_start,
+                                  int64_t filter_area_length)
+{
+    if (filter_area_length) {
+        /* return -1 if the block is not within filter area */
+        if (block->target_start >= filter_area_start + filter_area_length ||
+            block->target_end <= filter_area_start) {
+            return -1;
+        }
+
+        if (filter_area_start > block->target_start) {
+            return filter_area_start - block->target_start;
+        }
+    }
+
+    return 0;
+}
 #endif
Signature · 1 line
2.34.1
#3 lore raw

PATCH v3 2/14 dump: Rename write_elf_loads to write_elf_phdr_loads

In reply to Janosch Frank

Let's make it a bit clearer that we write the program headers of the PT_LOAD type.

---
 dump/dump.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
dump/dump.c +3 -3
index 4d9658ffa2..0ed7cf9c7b 100644
@@ -490,7 +490,7 @@ static void get_offset_range(hwaddr phys_addr,
     }
 }
 
-static void write_elf_loads(DumpState *s, Error **errp)
+static void write_elf_phdr_loads(DumpState *s, Error **errp)
 {
     ERRP_GUARD();
     hwaddr offset, filesz;
@@ -573,8 +573,8 @@ static void dump_begin(DumpState *s, Error **errp)
         return;
     }
 
-    /* write all PT_LOAD to vmcore */
-    write_elf_loads(s, errp);
+    /* write all PT_LOADs to vmcore */
+    write_elf_phdr_loads(s, errp);
     if (*errp) {
         return;
     }
Signature · 1 line
2.34.1
#10 lore raw

PATCH v3 5/14 dump: Split write of section headers and data and add a prepare step

In reply to Janosch Frank

By splitting the writing of the section headers and (future) section data we prepare for the addition of a string table section and architecture sections.

At the same time we move the writing of the section to the end of the dump process. This allows the upcoming architecture section code to add data after all of the common dump data has been written.

---
 dump/dump.c           | 114 ++++++++++++++++++++++++++++++++----------
 include/sysemu/dump.h |   4 ++
 2 files changed, 92 insertions(+), 26 deletions(-)
dump/dump.c +88 -26
index 2d04e06815..980702d476 100644
@@ -342,30 +342,71 @@ static void write_elf_phdr_note(DumpState *s, Error **errp)
     }
 }
 
-static void write_elf_section(DumpState *s, int type, Error **errp)
+static size_t prepare_elf_section_hdr_zero(DumpState *s, void *buff)
 {
-    Elf32_Shdr shdr32;
-    Elf64_Shdr shdr64;
-    int shdr_size;
-    void *shdr;
-    int ret;
+    if (dump_is_64bit(s)) {
+        Elf64_Shdr *shdr64 = buff;
 
-    if (type == 0) {
-        shdr_size = sizeof(Elf32_Shdr);
-        memset(&shdr32, 0, shdr_size);
-        shdr32.sh_info = cpu_to_dump32(s, s->phdr_num);
-        shdr = &shdr32;
+        shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
     } else {
-        shdr_size = sizeof(Elf64_Shdr);
-        memset(&shdr64, 0, shdr_size);
-        shdr64.sh_info = cpu_to_dump32(s, s->phdr_num);
-        shdr = &shdr64;
+        Elf32_Shdr *shdr32 = buff;
+
+        shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
     }
 
-    ret = fd_write_vmcore(shdr, shdr_size, s);
+    return dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+}
+
+static void prepare_elf_section_hdrs(DumpState *s)
+{
+    uint8_t *buff_hdr;
+    size_t len, sizeof_shdr;
+
+    /*
+     * Section ordering:
+     * - HDR zero (if needed)
+     */
+    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+    len = sizeof_shdr * s->shdr_num;
+    s->elf_section_hdrs = g_malloc0(len);
+    buff_hdr = s->elf_section_hdrs;
+
+    /* Write special section first */
+    if (s->phdr_num == PN_XNUM) {
+            prepare_elf_section_hdr_zero(s, buff_hdr);
+    }
+}
+
+static void prepare_elf_sections(DumpState *s, Error **errp)
+{
+    if (!s->shdr_num) {
+        return;
+    }
+
+    prepare_elf_section_hdrs(s);
+}
+
+static void write_elf_section_headers(DumpState *s, Error **errp)
+{
+    size_t sizeof_shdr;
+    int ret;
+
+    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+
+    ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
     if (ret < 0) {
-        error_setg_errno(errp, -ret,
-                         "dump: failed to write section header table");
+        error_setg_errno(errp, -ret, "dump: failed to write section headers");
+    }
+}
+
+static void write_elf_sections(DumpState *s, Error **errp)
+{
+    int ret;
+
+    /* Write section zero */
+    ret = fd_write_vmcore(s->elf_section_data, s->elf_section_data_size, s);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "dump: failed to write section data");
     }
 }
 
@@ -557,12 +598,22 @@ static void dump_begin(DumpState *s, Error **errp)
     /* Write elf header to buffer */
     prepare_elf_header(s);
 
+    prepare_elf_sections(s, errp);
+    if (*errp) {
+        return;
+    }
+
     /* Start to write stuff into file descriptor */
     write_elf_header(s, errp);
     if (*errp) {
         return;
     }
 
+    write_elf_section_headers(s, errp);
+    if (*errp) {
+        return;
+    }
+
     /* write PT_NOTE to vmcore */
     write_elf_phdr_note(s, errp);
     if (*errp) {
@@ -575,14 +626,6 @@ static void dump_begin(DumpState *s, Error **errp)
         return;
     }
 
-    /* write section to vmcore */
-    if (s->shdr_num) {
-        write_elf_section(s, 1, errp);
-        if (*errp) {
-            return;
-        }
-    }
-
     /* write notes to vmcore */
     write_elf_notes(s, errp);
 }
@@ -610,6 +653,19 @@ static void dump_iterate(DumpState *s, Error **errp)
     }
 }
 
+static void dump_end(DumpState *s, Error **errp)
+{
+    ERRP_GUARD();
+
+    if (!s->elf_section_data_size) {
+        return;
+    }
+    s->elf_section_data = g_malloc0(s->elf_section_data_size);
+
+    /* write sections to vmcore */
+    write_elf_sections(s, errp);
+}
+
 static void create_vmcore(DumpState *s, Error **errp)
 {
     ERRP_GUARD();
@@ -620,6 +676,12 @@ static void create_vmcore(DumpState *s, Error **errp)
     }
 
     dump_iterate(s, errp);
+    if (*errp) {
+        return;
+    }
+
+    /* Write section data after memory has been dumped */
+    dump_end(s, errp);
 }
 
 static int write_start_flat_header(int fd)
include/sysemu/dump.h +4 -0
index 58f41bbf45..dad10dee0b 100644
@@ -172,6 +172,10 @@ typedef struct DumpState {
     int64_t length;            /* Length of the dump we want to dump */
 
     void *elf_header;
+    void *elf_section_hdrs;
+    uint64_t elf_section_data_size;
+    void *elf_section_data;
+
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
     uint32_t nr_cpus;           /* number of guest's cpu */
Signature · 1 line
2.34.1
#5 lore raw

PATCH v3 3/14 dump: Convert GuestPhysBlock iterators and use the filter functions

In reply to Janosch Frank

The iteration over the memblocks is hard to understand so it's about time to clean it up. Instead of manually grabbing the next memblock we can use QTAILQ_FOREACH to iterate over all memblocks.

Additionally we move the calculation of the offset and length out by using the dump_get_memblock_*() functions.

---
 dump/dump.c           | 91 +++++++++++--------------------------------
 include/sysemu/dump.h |  9 ++---
 2 files changed, 27 insertions(+), 73 deletions(-)
dump/dump.c +23 -68
index 0ed7cf9c7b..5c9ed25c5a 100644
@@ -591,56 +591,27 @@ static void dump_begin(DumpState *s, Error **errp)
     write_elf_notes(s, errp);
 }
 
-static int get_next_block(DumpState *s, GuestPhysBlock *block)
-{
-    while (1) {
-        block = QTAILQ_NEXT(block, next);
-        if (!block) {
-            /* no more block */
-            return 1;
-        }
-
-        s->start = 0;
-        s->next_block = block;
-        if (s->has_filter) {
-            if (block->target_start >= s->begin + s->length ||
-                block->target_end <= s->begin) {
-                /* This block is out of the range */
-                continue;
-            }
-
-            if (s->begin > block->target_start) {
-                s->start = s->begin - block->target_start;
-            }
-        }
-
-        return 0;
-    }
-}
-
 /* write all memory to vmcore */
 static void dump_iterate(DumpState *s, Error **errp)
 {
     ERRP_GUARD();
     GuestPhysBlock *block;
-    int64_t size;
+    int64_t memblock_size, memblock_start;
 
-    do {
-        block = s->next_block;
-
-        size = block->target_end - block->target_start;
-        if (s->has_filter) {
-            size -= s->start;
-            if (s->begin + s->length < block->target_end) {
-                size -= block->target_end - (s->begin + s->length);
-            }
+    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
+        memblock_start = dump_get_memblock_start(block, s->begin, s->length);
+        if (memblock_start == -1) {
+            continue;
         }
-        write_memory(s, block, s->start, size, errp);
+
+        memblock_size = dump_get_memblock_size(block, s->begin, s->length);
+
+        /* Write the memory to file */
+        write_memory(s, block, memblock_start, memblock_size, errp);
         if (*errp) {
             return;
         }
-
-    } while (!get_next_block(s, block));
+    }
 }
 
 static void create_vmcore(DumpState *s, Error **errp)
@@ -1490,30 +1461,22 @@ static void create_kdump_vmcore(DumpState *s, Error **errp)
     }
 }
 
-static ram_addr_t get_start_block(DumpState *s)
+static int validate_start_block(DumpState *s)
 {
     GuestPhysBlock *block;
 
     if (!s->has_filter) {
-        s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
         return 0;
     }
 
     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
+        /* This block is out of the range */
         if (block->target_start >= s->begin + s->length ||
             block->target_end <= s->begin) {
-            /* This block is out of the range */
             continue;
         }
-
-        s->next_block = block;
-        if (s->begin > block->target_start) {
-            s->start = s->begin - block->target_start;
-        } else {
-            s->start = 0;
-        }
-        return s->start;
-    }
+        return 0;
+   }
 
     return -1;
 }
@@ -1540,25 +1503,17 @@ bool qemu_system_dump_in_progress(void)
     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
 }
 
-/* calculate total size of memory to be dumped (taking filter into
- * acoount.) */
+/*
+ * calculate total size of memory to be dumped (taking filter into
+ * account.)
+ */
 static int64_t dump_calculate_size(DumpState *s)
 {
     GuestPhysBlock *block;
-    int64_t size = 0, total = 0, left = 0, right = 0;
+    int64_t total = 0;
 
     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
-        if (s->has_filter) {
-            /* calculate the overlapped region. */
-            left = MAX(s->begin, block->target_start);
-            right = MIN(s->begin + s->length, block->target_end);
-            size = right - left;
-            size = size > 0 ? size : 0;
-        } else {
-            /* count the whole region in */
-            size = (block->target_end - block->target_start);
-        }
-        total += size;
+        total += dump_get_memblock_size(block, s->begin, s->length);
     }
 
     return total;
@@ -1660,8 +1615,8 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         goto cleanup;
     }
 
-    s->start = get_start_block(s);
-    if (s->start == -1) {
+    /* Is the filter filtering everything? */
+    if (validate_start_block(s) == -1) {
         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
         goto cleanup;
     }
include/sysemu/dump.h +4 -5
index 2b39abeeae..7025e50682 100644
@@ -166,11 +166,10 @@ typedef struct DumpState {
     hwaddr memory_offset;
     int fd;
 
-    GuestPhysBlock *next_block;
-    ram_addr_t start;
-    bool has_filter;
-    int64_t begin;
-    int64_t length;
+    /* Guest memory related data */
+    bool has_filter;           /* Are we dumping parts of the memory? */
+    int64_t begin;             /* Start address of the chunk we want to dump */
+    int64_t length;            /* Length of the dump we want to dump */
 
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
Signature · 1 line
2.34.1
#8 lore raw

PATCH v3 4/14 dump: Allocate header

In reply to Janosch Frank

Allocating the header lets us write it at a later time and hence also allows us to change section and segment table offsets until we finally write it.

---
 dump/dump.c           | 127 +++++++++++++++++++++---------------------
 include/sysemu/dump.h |   1 +
 2 files changed, 64 insertions(+), 64 deletions(-)
dump/dump.c +63 -64
index 5c9ed25c5a..2d04e06815 100644
@@ -98,6 +98,7 @@ static int dump_cleanup(DumpState *s)
     memory_mapping_list_free(&s->list);
     close(s->fd);
     g_free(s->guest_note);
+    g_free(s->elf_header);
     s->guest_note = NULL;
     if (s->resume) {
         if (s->detached) {
@@ -126,73 +127,49 @@ static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
     return 0;
 }
 
-static void write_elf64_header(DumpState *s, Error **errp)
+static void prepare_elf64_header(DumpState *s)
 {
-    /*
-     * phnum in the elf header is 16 bit, if we have more segments we
-     * set phnum to PN_XNUM and write the real number of segments to a
-     * special section.
-     */
-    uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
-    Elf64_Ehdr elf_header;
-    int ret;
+    uint16_t phnum = s->phdr_num >= PN_XNUM ? PN_XNUM : s->phdr_num;
+    Elf64_Ehdr *elf_header = s->elf_header;
 
-    memset(&elf_header, 0, sizeof(Elf64_Ehdr));
-    memcpy(&elf_header, ELFMAG, SELFMAG);
-    elf_header.e_ident[EI_CLASS] = ELFCLASS64;
-    elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
-    elf_header.e_ident[EI_VERSION] = EV_CURRENT;
-    elf_header.e_type = cpu_to_dump16(s, ET_CORE);
-    elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
-    elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
-    elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
-    elf_header.e_phoff = cpu_to_dump64(s, s->phdr_offset);
-    elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
-    elf_header.e_phnum = cpu_to_dump16(s, phnum);
+    memcpy(elf_header, ELFMAG, SELFMAG);
+    elf_header->e_ident[EI_CLASS] = ELFCLASS64;
+    elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
+    elf_header->e_ident[EI_VERSION] = EV_CURRENT;
+    elf_header->e_type = cpu_to_dump16(s, ET_CORE);
+    elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
+    elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
+    elf_header->e_ehsize = cpu_to_dump16(s, sizeof(*elf_header));
+    elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
+    elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
+    elf_header->e_phnum = cpu_to_dump16(s, phnum);
     if (s->shdr_num) {
-        elf_header.e_shoff = cpu_to_dump64(s, s->shdr_offset);
-        elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
-        elf_header.e_shnum = cpu_to_dump16(s, s->shdr_num);
-    }
-
-    ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
-    if (ret < 0) {
-        error_setg_errno(errp, -ret, "dump: failed to write elf header");
+        elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
+        elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
+        elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
     }
 }
 
-static void write_elf32_header(DumpState *s, Error **errp)
+static void prepare_elf32_header(DumpState *s)
 {
-    /*
-     * phnum in the elf header is 16 bit, if we have more segments we
-     * set phnum to PN_XNUM and write the real number of segments to a
-     * special section.
-     */
-    uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
-    Elf32_Ehdr elf_header;
-    int ret;
+    uint16_t phnum = s->phdr_num >= PN_XNUM ? PN_XNUM : s->phdr_num;
+    Elf32_Ehdr *elf_header = s->elf_header;
 
-    memset(&elf_header, 0, sizeof(Elf32_Ehdr));
-    memcpy(&elf_header, ELFMAG, SELFMAG);
-    elf_header.e_ident[EI_CLASS] = ELFCLASS32;
-    elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
-    elf_header.e_ident[EI_VERSION] = EV_CURRENT;
-    elf_header.e_type = cpu_to_dump16(s, ET_CORE);
-    elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
-    elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
-    elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
-    elf_header.e_phoff = cpu_to_dump32(s, s->phdr_offset);
-    elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
-    elf_header.e_phnum = cpu_to_dump16(s, phnum);
+    memcpy(elf_header, ELFMAG, SELFMAG);
+    elf_header->e_ident[EI_CLASS] = ELFCLASS32;
+    elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
+    elf_header->e_ident[EI_VERSION] = EV_CURRENT;
+    elf_header->e_type = cpu_to_dump16(s, ET_CORE);
+    elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
+    elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
+    elf_header->e_ehsize = cpu_to_dump16(s, sizeof(*elf_header));
+    elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
+    elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
+    elf_header->e_phnum = cpu_to_dump16(s, phnum);
     if (s->shdr_num) {
-        elf_header.e_shoff = cpu_to_dump32(s, s->shdr_offset);
-        elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
-        elf_header.e_shnum = cpu_to_dump16(s, s->shdr_num);
-    }
-
-    ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
-    if (ret < 0) {
-        error_setg_errno(errp, -ret, "dump: failed to write elf header");
+        elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
+        elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
+        elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
     }
 }
 
@@ -528,6 +505,26 @@ static void write_elf_notes(DumpState *s, Error **errp)
     }
 }
 
+static void prepare_elf_header(DumpState *s)
+{
+    if (dump_is_64bit(s)) {
+        prepare_elf64_header(s);
+    } else {
+        prepare_elf32_header(s);
+    }
+}
+
+static void write_elf_header(DumpState *s, Error **errp)
+{
+    size_t size = dump_is_64bit(s) ? sizeof(Elf64_Ehdr) : sizeof(Elf32_Ehdr);
+    int ret;
+
+    ret = fd_write_vmcore(s->elf_header, size, s);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "dump: failed to write elf header");
+    }
+}
+
 /* write elf header, PT_NOTE and elf note to vmcore. */
 static void dump_begin(DumpState *s, Error **errp)
 {
@@ -557,12 +554,11 @@ static void dump_begin(DumpState *s, Error **errp)
      * vmcore.
      */
 
-    /* write elf header to vmcore */
-    if (dump_is_64bit(s)) {
-        write_elf64_header(s, errp);
-    } else {
-        write_elf32_header(s, errp);
-    }
+    /* Write elf header to buffer */
+    prepare_elf_header(s);
+
+    /* Start to write stuff into file descriptor */
+    write_elf_header(s, errp);
     if (*errp) {
         return;
     }
@@ -1642,6 +1638,9 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         goto cleanup;
     }
 
+    s->elf_header = g_malloc0(dump_is_64bit(s) ?
+                              sizeof(Elf64_Ehdr) : sizeof(Elf32_Ehdr));
+
     /*
      * The goal of this block is to (a) update the previously guessed
      * phys_base, (b) copy the guest note out of the guest.
include/sysemu/dump.h +1 -0
index 7025e50682..58f41bbf45 100644
@@ -171,6 +171,7 @@ typedef struct DumpState {
     int64_t begin;             /* Start address of the chunk we want to dump */
     int64_t length;            /* Length of the dump we want to dump */
 
+    void *elf_header;
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
     uint32_t nr_cpus;           /* number of guest's cpu */
Signature · 1 line
2.34.1
#13 lore raw

PATCH v3 6/14 dump: Reorder struct DumpState

In reply to Janosch Frank

Let's move ELF related members into one block and guest memory related ones into another to improve readability.

---
 include/sysemu/dump.h | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
include/sysemu/dump.h +10 -7
index dad10dee0b..a94d81303d 100644
@@ -154,15 +154,8 @@ typedef struct DumpState {
     GuestPhysBlockList guest_phys_blocks;
     ArchDumpInfo dump_info;
     MemoryMappingList list;
-    uint32_t phdr_num;
-    uint32_t shdr_num;
     bool resume;
     bool detached;
-    ssize_t note_size;
-    hwaddr shdr_offset;
-    hwaddr phdr_offset;
-    hwaddr section_offset;
-    hwaddr note_offset;
     hwaddr memory_offset;
     int fd;
 
@@ -171,6 +164,16 @@ typedef struct DumpState {
     int64_t begin;             /* Start address of the chunk we want to dump */
     int64_t length;            /* Length of the dump we want to dump */
 
+    /* Elf dump related data */
+    uint32_t phdr_num;
+    uint32_t shdr_num;
+    uint32_t sh_info;
+    ssize_t note_size;
+    hwaddr shdr_offset;
+    hwaddr phdr_offset;
+    hwaddr note_offset;
+    hwaddr section_offset;
+
     void *elf_header;
     void *elf_section_hdrs;
     uint64_t elf_section_data_size;
Signature · 1 line
2.34.1
#18 lore raw

PATCH v3 8/14 dump/dump: Add section string table support

In reply to Janosch Frank

As sections don't have a type like the notes do we need another way to determine their contents. The string table allows us to assign each section an identification string which architectures can then use to tag their sections with.

There will be no string table if the architecture doesn't add custom sections which are introduced in a following patch.

---
 dump/dump.c           | 81 +++++++++++++++++++++++++++++++++++++++++--
 include/sysemu/dump.h |  1 +
 2 files changed, 80 insertions(+), 2 deletions(-)
dump/dump.c +79 -2
index 6f3274c5af..944217349a 100644
@@ -99,6 +99,7 @@ static int dump_cleanup(DumpState *s)
     close(s->fd);
     g_free(s->guest_note);
     g_free(s->elf_header);
+    g_array_unref(s->string_table_buf);
     s->guest_note = NULL;
     if (s->resume) {
         if (s->detached) {
@@ -357,14 +358,47 @@ static size_t prepare_elf_section_hdr_zero(DumpState *s, void *buff)
     return dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
 }
 
+static void write_elf_section_hdr_string(DumpState *s, void *buff)
+{
+    Elf32_Shdr shdr32;
+    Elf64_Shdr shdr64;
+    int shdr_size;
+    void *shdr = buff;
+
+    if (dump_is_64bit(s)) {
+        shdr_size = sizeof(Elf64_Shdr);
+        memset(&shdr64, 0, shdr_size);
+        shdr64.sh_type = SHT_STRTAB;
+        shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
+        shdr64.sh_name = s->string_table_buf->len;
+        g_array_append_vals(s->string_table_buf, ".strtab", sizeof(".strtab"));
+        shdr64.sh_size = s->string_table_buf->len;
+        shdr = &shdr64;
+    } else {
+        shdr_size = sizeof(Elf32_Shdr);
+        memset(&shdr32, 0, shdr_size);
+        shdr32.sh_type = SHT_STRTAB;
+        shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
+        shdr32.sh_name = s->string_table_buf->len;
+        g_array_append_vals(s->string_table_buf, ".strtab", sizeof(".strtab"));
+        shdr32.sh_size = s->string_table_buf->len;
+        shdr = &shdr32;
+    }
+
+    memcpy(buff, shdr, shdr_size);
+}
+
 static void prepare_elf_section_hdrs(DumpState *s)
 {
     uint8_t *buff_hdr;
-    size_t len, sizeof_shdr;
+    size_t len, size = 0, sizeof_shdr;
+    Elf64_Ehdr *hdr64 = s->elf_header;
+    Elf32_Ehdr *hdr32 = s->elf_header;
 
     /*
      * Section ordering:
      * - HDR zero (if needed)
+     * - String table hdr
      */
     sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
     len = sizeof_shdr * s->shdr_num;
@@ -375,6 +409,22 @@ static void prepare_elf_section_hdrs(DumpState *s)
     if (s->phdr_num == PN_XNUM) {
             prepare_elf_section_hdr_zero(s, buff_hdr);
     }
+    buff_hdr += size;
+
+    if (s->shdr_num < 2) {
+        return;
+    }
+
+    /*
+     * String table needs to be last section since strings are added
+     * via arch_sections_write_hdr().
+     */
+    write_elf_section_hdr_string(s, buff_hdr);
+    if (dump_is_64bit(s)) {
+        hdr64->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
+    } else {
+        hdr32->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
+    }
 }
 
 static void prepare_elf_sections(DumpState *s, Error **errp)
@@ -403,11 +453,18 @@ static void write_elf_sections(DumpState *s, Error **errp)
 {
     int ret;
 
-    /* Write section zero */
+    /* Write section zero and arch sections */
     ret = fd_write_vmcore(s->elf_section_data, s->elf_section_data_size, s);
     if (ret < 0) {
         error_setg_errno(errp, -ret, "dump: failed to write section data");
     }
+
+    /* Write string table data */
+    ret = fd_write_vmcore(s->string_table_buf->data,
+                          s->string_table_buf->len, s);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "dump: failed to write string table data");
+    }
 }
 
 static void write_data(DumpState *s, void *buf, int length, Error **errp)
@@ -678,6 +735,7 @@ static void create_vmcore(DumpState *s, Error **errp)
         return;
     }
 
+    /* Iterate over memory and dump it to file */
     dump_iterate(s, errp);
     if (*errp) {
         return;
@@ -1660,6 +1718,13 @@ static void dump_init(DumpState *s, int fd, bool has_format,
     s->has_filter = has_filter;
     s->begin = begin;
     s->length = length;
+    /* First index is 0, it's the special null name */
+    s->string_table_buf = g_array_new(FALSE, TRUE, 1);
+    /*
+     * Allocate the null name, due to the clearing option set to true
+     * it will be 0.
+     */
+    g_array_set_size(s->string_table_buf, 1);
 
     memory_mapping_list_init(&s->list);
 
@@ -1820,6 +1885,18 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         }
     }
 
+    /*
+     * calculate shdr_num and elf_section_data_size so we know the offsets and
+     * sizes of all parts.
+     *
+     * If phdr_num overflowed we have at least one section header
+     * More sections/hdrs can be added by the architectures
+     */
+    if (s->shdr_num > 1) {
+        /* Reserve the string table */
+        s->shdr_num += 1;
+    }
+
     tmp = (s->phdr_num == PN_XNUM) ? s->sh_info : s->phdr_num;
     if (dump_is_64bit(s)) {
         s->shdr_offset = sizeof(Elf64_Ehdr);
include/sysemu/dump.h +1 -0
index a94d81303d..af2dd1f090 100644
@@ -178,6 +178,7 @@ typedef struct DumpState {
     void *elf_section_hdrs;
     uint64_t elf_section_data_size;
     void *elf_section_data;
+    GArray *string_table_buf;  /* String table section */
 
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
Signature · 1 line
2.34.1
#15 lore raw

PATCH v3 7/14 dump: Swap segment and section header locations

In reply to Janosch Frank

For the upcoming string table and arch section support we need to modify the elf layout a bit. Instead of the segments, i.e. the guest's memory contents, beeing the last area the section data will live at the end of the file. This will allow us to write the section data after all guest memory has been dumped which is important for the s390 PV dump support.

---
 dump/dump.c | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)
dump/dump.c +12 -9
index 980702d476..6f3274c5af 100644
@@ -590,6 +590,9 @@ static void dump_begin(DumpState *s, Error **errp)
      *   --------------
      *   |  memory     |
      *   --------------
+     *   |  sectn data |
+     *   --------------
+
      *
      * we only know where the memory is saved after we write elf note into
      * vmcore.
@@ -1817,18 +1820,18 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         }
     }
 
+    tmp = (s->phdr_num == PN_XNUM) ? s->sh_info : s->phdr_num;
     if (dump_is_64bit(s)) {
-        s->phdr_offset = sizeof(Elf64_Ehdr);
-        s->shdr_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
-        s->note_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
-        s->memory_offset = s->note_offset + s->note_size;
+        s->shdr_offset = sizeof(Elf64_Ehdr);
+        s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
+        s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * tmp;
     } else {
-
-        s->phdr_offset = sizeof(Elf32_Ehdr);
-        s->shdr_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
-        s->note_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
-        s->memory_offset = s->note_offset + s->note_size;
+        s->shdr_offset = sizeof(Elf32_Ehdr);
+        s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
+        s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * tmp;
     }
+    s->memory_offset = s->note_offset + s->note_size;
+    s->section_offset = s->memory_offset + s->total_size;
 
     return;
 
Signature · 1 line
2.34.1
#26 lore raw

PATCH v3 12/14 s390x: Introduce PV query interface

In reply to Janosch Frank

Introduce an interface over which we can get information about UV data.

---
 hw/s390x/pv.c              | 61 ++++++++++++++++++++++++++++++++++++++
 hw/s390x/s390-virtio-ccw.c |  5 ++++
 include/hw/s390x/pv.h      | 10 +++++++
 3 files changed, 76 insertions(+)
hw/s390x/pv.c +61 -0
index 401b63d6cb..a5af4ddf46 100644
@@ -20,6 +20,11 @@
 #include "exec/confidential-guest-support.h"
 #include "hw/s390x/ipl.h"
 #include "hw/s390x/pv.h"
+#include "target/s390x/kvm/kvm_s390x.h"
+
+static bool info_valid;
+static struct kvm_s390_pv_info_vm info_vm;
+static struct kvm_s390_pv_info_dump info_dump;
 
 static int __s390_pv_cmd(uint32_t cmd, const char *cmdname, void *data)
 {
@@ -56,6 +61,42 @@ static int __s390_pv_cmd(uint32_t cmd, const char *cmdname, void *data)
     }                                  \
 }
 
+int s390_pv_query_info(void)
+{
+    struct kvm_s390_pv_info info = {
+        .header.id = KVM_PV_INFO_VM,
+        .header.len_max = sizeof(info.header) + sizeof(info.vm),
+    };
+    int rc;
+
+    /* Info API's first user is dump so they are bundled */
+    if (!kvm_s390_get_protected_dump()) {
+        return 0;
+    }
+
+    rc = s390_pv_cmd(KVM_PV_INFO, &info);
+    if (rc) {
+        error_report("KVM PV INFO cmd %x failed: %s",
+                     info.header.id, strerror(rc));
+        return rc;
+    }
+    memcpy(&info_vm, &info.vm, sizeof(info.vm));
+
+    info.header.id = KVM_PV_INFO_DUMP;
+    info.header.len_max = sizeof(info.header) + sizeof(info.dump);
+    rc = s390_pv_cmd(KVM_PV_INFO, &info);
+    if (rc) {
+        error_report("KVM PV INFO cmd %x failed: %s",
+                     info.header.id, strerror(rc));
+        return rc;
+    }
+
+    memcpy(&info_dump, &info.dump, sizeof(info.dump));
+    info_valid = true;
+
+    return rc;
+}
+
 int s390_pv_vm_enable(void)
 {
     return s390_pv_cmd(KVM_PV_ENABLE, NULL);
@@ -114,6 +155,26 @@ void s390_pv_inject_reset_error(CPUState *cs)
     env->regs[r1 + 1] = DIAG_308_RC_INVAL_FOR_PV;
 }
 
+uint64_t kvm_s390_pv_dmp_get_size_cpu(void)
+{
+    return info_dump.dump_cpu_buffer_len;
+}
+
+uint64_t kvm_s390_pv_dmp_get_size_complete(void)
+{
+    return info_dump.dump_config_finalize_len;
+}
+
+uint64_t kvm_s390_pv_dmp_get_size_mem(void)
+{
+    return info_dump.dump_config_mem_buffer_per_1m;
+}
+
+bool kvm_s390_pv_info_basic_valid(void)
+{
+    return info_valid;
+}
+
 #define TYPE_S390_PV_GUEST "s390-pv-guest"
 OBJECT_DECLARE_SIMPLE_TYPE(S390PVGuest, S390_PV_GUEST)
 
hw/s390x/s390-virtio-ccw.c +5 -0
index cc3097bfee..f9401e392b 100644
@@ -366,6 +366,11 @@ static int s390_machine_protect(S390CcwMachineState *ms)
 
     ms->pv = true;
 
+    rc = s390_pv_query_info();
+    if (rc) {
+        goto out_err;
+    }
+
     /* Set SE header and unpack */
     rc = s390_ipl_prepare_pv_header();
     if (rc) {
include/hw/s390x/pv.h +10 -0
index 1f1f545bfc..6fa55bf70e 100644
@@ -38,6 +38,7 @@ static inline bool s390_is_pv(void)
     return ccw->pv;
 }
 
+int s390_pv_query_info(void);
 int s390_pv_vm_enable(void);
 void s390_pv_vm_disable(void);
 int s390_pv_set_sec_parms(uint64_t origin, uint64_t length);
@@ -46,8 +47,13 @@ void s390_pv_prep_reset(void);
 int s390_pv_verify(void);
 void s390_pv_unshare(void);
 void s390_pv_inject_reset_error(CPUState *cs);
+uint64_t kvm_s390_pv_dmp_get_size_cpu(void);
+uint64_t kvm_s390_pv_dmp_get_size_mem(void);
+uint64_t kvm_s390_pv_dmp_get_size_complete(void);
+bool kvm_s390_pv_info_basic_valid(void);
 #else /* CONFIG_KVM */
 static inline bool s390_is_pv(void) { return false; }
+static inline int s390_pv_query_info(void) { return 0; }
 static inline int s390_pv_vm_enable(void) { return 0; }
 static inline void s390_pv_vm_disable(void) {}
 static inline int s390_pv_set_sec_parms(uint64_t origin, uint64_t length) { return 0; }
@@ -56,6 +62,10 @@ static inline void s390_pv_prep_reset(void) {}
 static inline int s390_pv_verify(void) { return 0; }
 static inline void s390_pv_unshare(void) {}
 static inline void s390_pv_inject_reset_error(CPUState *cs) {};
+static inline uint64_t kvm_s390_pv_dmp_get_size_cpu(void) { return 0; }
+static inline uint64_t kvm_s390_pv_dmp_get_size_mem(void) { return 0; }
+static inline uint64_t kvm_s390_pv_dmp_get_size_complete(void) { return 0; }
+static inline bool kvm_s390_pv_info_basic_valid(void) { return false; }
 #endif /* CONFIG_KVM */
 
 int s390_pv_kvm_init(ConfidentialGuestSupport *cgs, Error **errp);
Signature · 1 line
2.34.1
#20 lore raw

PATCH v3 9/14 dump/dump: Add arch section support

In reply to Janosch Frank

Add hooks which architectures can use to add arbitrary data to custom sections.

---
 dump/dump.c                |  6 ++++++
 include/sysemu/dump-arch.h | 27 +++++++++++++++++++++++++++
 2 files changed, 33 insertions(+)
dump/dump.c +6 -0
index 944217349a..60a164afd5 100644
@@ -398,6 +398,7 @@ static void prepare_elf_section_hdrs(DumpState *s)
     /*
      * Section ordering:
      * - HDR zero (if needed)
+     * - Arch section hdrs
      * - String table hdr
      */
     sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
@@ -415,6 +416,9 @@ static void prepare_elf_section_hdrs(DumpState *s)
         return;
     }
 
+    size = dump_arch_sections_write_hdr(&s->dump_info, s, buff_hdr);
+    buff_hdr += size;
+
     /*
      * String table needs to be last section since strings are added
      * via arch_sections_write_hdr().
@@ -721,6 +725,7 @@ static void dump_end(DumpState *s, Error **errp)
         return;
     }
     s->elf_section_data = g_malloc0(s->elf_section_data_size);
+    dump_arch_sections_write(&s->dump_info, s, s->elf_section_data);
 
     /* write sections to vmcore */
     write_elf_sections(s, errp);
@@ -1892,6 +1897,7 @@ static void dump_init(DumpState *s, int fd, bool has_format,
      * If phdr_num overflowed we have at least one section header
      * More sections/hdrs can be added by the architectures
      */
+    dump_arch_sections_add(&s->dump_info, (void *)s);
     if (s->shdr_num > 1) {
         /* Reserve the string table */
         s->shdr_num += 1;
include/sysemu/dump-arch.h +27 -0
index e25b02e990..de77908424 100644
@@ -21,6 +21,9 @@ typedef struct ArchDumpInfo {
     uint32_t page_size;      /* The target's page size. If it's variable and
                               * unknown, then this should be the maximum. */
     uint64_t phys_base;      /* The target's physmem base. */
+    void (*arch_sections_add_fn)(void *opaque);
+    uint64_t (*arch_sections_write_hdr_fn)(void *opaque, uint8_t *buff);
+    void (*arch_sections_write_fn)(void *opaque, uint8_t *buff);
 } ArchDumpInfo;
 
 struct GuestPhysBlockList; /* memory_mapping.h */
@@ -28,4 +31,28 @@ int cpu_get_dump_info(ArchDumpInfo *info,
                       const struct GuestPhysBlockList *guest_phys_blocks);
 ssize_t cpu_get_note_size(int class, int machine, int nr_cpus);
 
+static inline void dump_arch_sections_add(ArchDumpInfo *info, void *opaque)
+{
+    if (info->arch_sections_add_fn) {
+        info->arch_sections_add_fn(opaque);
+    }
+}
+
+static inline uint64_t dump_arch_sections_write_hdr(ArchDumpInfo *info,
+                                                void *opaque, uint8_t *buff)
+{
+    if (info->arch_sections_write_hdr_fn) {
+        return info->arch_sections_write_hdr_fn(opaque, buff);
+    }
+    return 0;
+}
+
+static inline void dump_arch_sections_write(ArchDumpInfo *info, void *opaque,
+                                            uint8_t *buff)
+{
+    if (info->arch_sections_write_fn) {
+        info->arch_sections_write_fn(opaque, buff);
+    }
+}
+
 #endif
Signature · 1 line
2.34.1
#21 lore raw

PATCH v3 10/14 linux header sync

In reply to Janosch Frank

Add the uapi data for KVM_CAP_S390_PROTECTED_DUMP which I expect to be added with 5.20.

---
 linux-headers/linux/kvm.h | 54 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)
linux-headers/linux/kvm.h +54 -0
index 0d05d02ee4..213216c27c 100644
@@ -1150,6 +1150,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_DISABLE_QUIRKS2 213
 /* #define KVM_CAP_VM_TSC_CONTROL 214 */
 #define KVM_CAP_SYSTEM_EVENT_DATA 215
+#define KVM_CAP_S390_PROTECTED_DUMP 217
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1651,6 +1652,55 @@ struct kvm_s390_pv_unp {
 	__u64 tweak;
 };
 
+enum pv_cmd_info_id {
+	KVM_PV_INFO_VM,
+	KVM_PV_INFO_DUMP,
+};
+
+struct kvm_s390_pv_info_dump {
+	__u64 dump_cpu_buffer_len;
+	__u64 dump_config_mem_buffer_per_1m;
+	__u64 dump_config_finalize_len;
+};
+
+struct kvm_s390_pv_info_vm {
+	__u64 inst_calls_list[4];
+	__u64 max_cpus;
+	__u64 max_guests;
+	__u64 max_guest_addr;
+	__u64 feature_indication;
+};
+
+struct kvm_s390_pv_info_header {
+	__u32 id;
+	__u32 len_max;
+	__u32 len_written;
+	__u32 reserved;
+};
+
+struct kvm_s390_pv_info {
+	struct kvm_s390_pv_info_header header;
+	union {
+		struct kvm_s390_pv_info_dump dump;
+		struct kvm_s390_pv_info_vm vm;
+	};
+};
+
+enum pv_cmd_dmp_id {
+        KVM_PV_DUMP_INIT,
+        KVM_PV_DUMP_CONFIG_STATE,
+        KVM_PV_DUMP_COMPLETE,
+        KVM_PV_DUMP_CPU,
+};
+
+struct kvm_s390_pv_dmp {
+        __u64 subcmd;
+        __u64 buff_addr;
+        __u64 buff_len;
+        __u64 gaddr;
+        __u64 reserved[4];
+};
+
 enum pv_cmd_id {
 	KVM_PV_ENABLE,
 	KVM_PV_DISABLE,
@@ -1659,6 +1709,8 @@ enum pv_cmd_id {
 	KVM_PV_VERIFY,
 	KVM_PV_PREP_RESET,
 	KVM_PV_UNSHARE_ALL,
+        KVM_PV_INFO,
+        KVM_PV_DUMP,
 };
 
 struct kvm_pv_cmd {
@@ -2066,4 +2118,6 @@ struct kvm_stats_desc {
 /* Available with KVM_CAP_XSAVE2 */
 #define KVM_GET_XSAVE2		  _IOR(KVMIO,  0xcf, struct kvm_xsave)
 
+#define KVM_S390_PV_CPU_COMMAND _IOWR(KVMIO, 0xd0, struct kvm_pv_cmd)
+
 #endif /* __LINUX_KVM_H */
Signature · 1 line
2.34.1
#24 lore raw

PATCH v3 11/14 s390x: Add protected dump cap

In reply to Janosch Frank

Add a protected dump capability for later feature checking.

---
 target/s390x/kvm/kvm.c       | 7 +++++++
 target/s390x/kvm/kvm_s390x.h | 1 +
 2 files changed, 8 insertions(+)
target/s390x/kvm/kvm.c +7 -0
index 7bd8db0e7b..cbd8c91424 100644
@@ -157,6 +157,7 @@ static int cap_ri;
 static int cap_hpage_1m;
 static int cap_vcpu_resets;
 static int cap_protected;
+static int cap_protected_dump;
 
 static bool mem_op_storage_key_support;
 
@@ -362,6 +363,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
     cap_vcpu_resets = kvm_check_extension(s, KVM_CAP_S390_VCPU_RESETS);
     cap_protected = kvm_check_extension(s, KVM_CAP_S390_PROTECTED);
+    cap_protected_dump = kvm_check_extension(s, KVM_CAP_S390_PROTECTED_DUMP);
 
     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
@@ -2043,6 +2045,11 @@ int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
 }
 
+int kvm_s390_get_protected_dump(void)
+{
+    return cap_protected_dump;
+}
+
 int kvm_s390_get_ri(void)
 {
     return cap_ri;
target/s390x/kvm/kvm_s390x.h +1 -0
index 05a5e1e6f4..31a69f9ce2 100644
@@ -26,6 +26,7 @@ int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state);
 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu);
 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu);
 int kvm_s390_get_hpage_1m(void);
+int kvm_s390_get_protected_dump(void);
 int kvm_s390_get_ri(void);
 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_clock);
 int kvm_s390_get_clock_ext(uint8_t *tod_high, uint64_t *tod_clock);
Signature · 1 line
2.34.1
#30 lore raw

PATCH v3 14/14 s390x: pv: Add dump support

In reply to Janosch Frank

Sometimes dumping a guest from the outside is the only way to get the data that is needed. This can be the case if a dumping mechanism like KDUMP hasn't been configured or data needs to be fetched at a specific point. Dumping a protected guest from the outside without help from fw/hw doesn't yield sufficient data to be useful. Hence we now introduce PV dump support.

The PV dump support works by integrating the firmware into the dump process. New Ultravisor calls are used to initiate the dump process, dump cpu data, dump memory state and lastly complete the dump process. The UV calls are exposed by KVM via the new KVM_PV_DUMP command and its subcommands. The guest's data is fully encrypted and can only be decrypted by the entity that owns the customer communication key for the dumped guest. Also dumping needs to be allowed via a flag in the SE header.

On the QEMU side of things we store the PV dump data in the newly introduced architecture ELF sections (storage state and completion data) and the cpu notes (for cpu dump data).

Users can use the zgetdump tool to convert the encrypted QEMU dump to an unencrypted one.

---
 include/elf.h            |   1 +
 target/s390x/arch_dump.c | 248 ++++++++++++++++++++++++++++++++++-----
 2 files changed, 219 insertions(+), 30 deletions(-)
include/elf.h +1 -0
index 3a4bcb646a..58f76fd5b4 100644
@@ -1649,6 +1649,7 @@ typedef struct elf64_shdr {
 #define NT_TASKSTRUCT	4
 #define NT_AUXV		6
 #define NT_PRXFPREG     0x46e62b7f      /* copied from gdb5.1/include/elf/common.h */
+#define NT_S390_PV_DATA 0x30e           /* s390 protvirt cpu dump data */
 #define NT_S390_GS_CB   0x30b           /* s390 guarded storage registers */
 #define NT_S390_VXRS_HIGH 0x30a         /* s390 vector registers 16-31 */
 #define NT_S390_VXRS_LOW  0x309         /* s390 vector registers 0-15 (lower half) */
target/s390x/arch_dump.c +218 -30

317 lines not on this page · raw message

Signature · 1 line
2.34.1
#28 lore raw

PATCH v3 13/14 s390x: Add KVM PV dump interface

In reply to Janosch Frank

Let's add a few bits of code which hide the new KVM PV dump API from us via new functions.

---
 hw/s390x/pv.c         | 51 +++++++++++++++++++++++++++++++++++++++++++
 include/hw/s390x/pv.h |  8 +++++++
 2 files changed, 59 insertions(+)
hw/s390x/pv.c +51 -0
index a5af4ddf46..48591c387d 100644
@@ -175,6 +175,57 @@ bool kvm_s390_pv_info_basic_valid(void)
     return info_valid;
 }
 
+static int s390_pv_dump_cmd(uint64_t subcmd, uint64_t uaddr, uint64_t gaddr,
+                            uint64_t len)
+{
+    struct kvm_s390_pv_dmp dmp = {
+        .subcmd = subcmd,
+        .buff_addr = uaddr,
+        .buff_len = len,
+        .gaddr = gaddr,
+    };
+    int ret;
+
+    ret = s390_pv_cmd(KVM_PV_DUMP, (void *)&dmp);
+    if (ret) {
+        error_report("KVM DUMP command %ld failed", subcmd);
+    }
+    return ret;
+}
+
+int kvm_s390_dump_cpu(S390CPU *cpu, void *buff)
+{
+    struct kvm_s390_pv_dmp dmp = {
+        .subcmd = KVM_PV_DUMP_CPU,
+        .buff_addr = (uint64_t)buff,
+        .gaddr = 0,
+        .buff_len = info_dump.dump_cpu_buffer_len,
+    };
+    struct kvm_pv_cmd pv = {
+        .cmd = KVM_PV_DUMP,
+        .data = (uint64_t)&dmp,
+    };
+
+    return kvm_vcpu_ioctl(CPU(cpu), KVM_S390_PV_CPU_COMMAND, &pv);
+}
+
+int kvm_s390_dump_init(void)
+{
+    return s390_pv_dump_cmd(KVM_PV_DUMP_INIT, 0, 0, 0);
+}
+
+int kvm_s390_dump_mem(uint64_t gaddr, size_t len, void *dest)
+{
+    return s390_pv_dump_cmd(KVM_PV_DUMP_CONFIG_STATE, (uint64_t)dest,
+                            gaddr, len);
+}
+
+int kvm_s390_dump_finish(void *buff)
+{
+    return s390_pv_dump_cmd(KVM_PV_DUMP_COMPLETE, (uint64_t)buff, 0,
+                            info_dump.dump_config_finalize_len);
+}
+
 #define TYPE_S390_PV_GUEST "s390-pv-guest"
 OBJECT_DECLARE_SIMPLE_TYPE(S390PVGuest, S390_PV_GUEST)
 
include/hw/s390x/pv.h +8 -0
index 6fa55bf70e..f37021e189 100644
@@ -51,6 +51,10 @@ uint64_t kvm_s390_pv_dmp_get_size_cpu(void);
 uint64_t kvm_s390_pv_dmp_get_size_mem(void);
 uint64_t kvm_s390_pv_dmp_get_size_complete(void);
 bool kvm_s390_pv_info_basic_valid(void);
+int kvm_s390_dump_init(void);
+int kvm_s390_dump_cpu(S390CPU *cpu, void *buff);
+int kvm_s390_dump_mem(uint64_t addr, size_t len, void *dest);
+int kvm_s390_dump_finish(void *buff);
 #else /* CONFIG_KVM */
 static inline bool s390_is_pv(void) { return false; }
 static inline int s390_pv_query_info(void) { return 0; }
@@ -66,6 +70,10 @@ static inline uint64_t kvm_s390_pv_dmp_get_size_cpu(void) { return 0; }
 static inline uint64_t kvm_s390_pv_dmp_get_size_mem(void) { return 0; }
 static inline uint64_t kvm_s390_pv_dmp_get_size_complete(void) { return 0; }
 static inline bool kvm_s390_pv_info_basic_valid(void) { return false; }
+static inline int kvm_s390_dump_init(void) { return 0; }
+static inline int kvm_s390_dump_cpu(S390CPU *cpu, void *buff, size_t len) { return 0; }
+static inline int kvm_s390_dump_mem(uint64_t addr, size_t len, void *dest) { return 0; }
+static inline int kvm_s390_dump_finish(void *buff) { return 0; }
 #endif /* CONFIG_KVM */
 
 int s390_pv_kvm_init(ConfidentialGuestSupport *cgs, Error **errp);
Signature · 1 line
2.34.1
#2 lore raw

Re: [PATCH v3 01/14] dump: Introduce GuestPhysBlock offset and length filter functions

In reply to Janosch Frank

Hi

Quoted 45 lines · Janosch Frank

As the code is used in multiple places in dump/dump.c we should introduce functions that calculate the filtered length and offset of a GuestPhysBlock so we can use them in the next patch.

---
 include/sysemu/dump.h | 37 +++++++++++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)
include/sysemu/dump.h +27 -0
index ffc2ea1072..2b39abeeae 100644
@@ -203,4 +203,41 @@ typedef struct DumpState {
 uint16_t cpu_to_dump16(DumpState *s, uint16_t val);
 uint32_t cpu_to_dump32(DumpState *s, uint32_t val);
 uint64_t cpu_to_dump64(DumpState *s, uint64_t val);
+
+static inline int64_t dump_get_memblock_size(GuestPhysBlock *block, int64_t filter_area_start,
+                                             int64_t filter_area_length)
+{
+    int64_t size, left, right;
+
+    /* No filter, return full size */
+    if (!filter_area_length) {
+        return block->target_end - block->target_start;
+    }
+
+    /* calculate the overlapped region. */
+    left = MAX(filter_area_start, block->target_start);
+    right = MIN(filter_area_start + filter_area_length, block->target_end);
+    size = right - left;
+    size = size > 0 ? size : 0;
+
+    return size;
+}
+
+static inline int64_t dump_get_memblock_start(GuestPhysBlock *block, int64_t filter_area_start,
+                                  int64_t filter_area_length)
+{
+    if (filter_area_length) {
+        /* return -1 if the block is not within filter area */
+        if (block->target_start >= filter_area_start + filter_area_length ||
+            block->target_end <= filter_area_start) {

include/sysemu/dump.h typedef struct DumpState { · line 242

Quoted 10 lines
include/sysemu/dump.h +10 -0
+            return -1;
+        }
+
+        if (filter_area_start > block->target_start) {
+            return filter_area_start - block->target_start;
+        }
+    }
+
+    return 0;
+}

Why inline the function and have it in the header? (similar for other inline functions in the series - except obvious empty/stubs functions)

Quoted 4 lines
include/sysemu/dump.h +0 -1
 #endif
--

2.34.1

Reviewed-by on v3 2/14
#4 lore raw

Re: [PATCH v3 02/14] dump: Rename write_elf_loads to write_elf_phdr_loads

In reply to Janosch Frank
Quoted 5 lines · Janosch Frank

Let's make it a bit clearer that we write the program headers of the PT_LOAD type.

Quoted 31 lines
---
 dump/dump.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
dump/dump.c +3 -4
index 4d9658ffa2..0ed7cf9c7b 100644
@@ -490,7 +490,7 @@ static void get_offset_range(hwaddr phys_addr,
     }
 }
 
-static void write_elf_loads(DumpState *s, Error **errp)
+static void write_elf_phdr_loads(DumpState *s, Error **errp)
 {
     ERRP_GUARD();
     hwaddr offset, filesz;
@@ -573,8 +573,8 @@ static void dump_begin(DumpState *s, Error **errp)
         return;
     }
 
-    /* write all PT_LOAD to vmcore */
-    write_elf_loads(s, errp);
+    /* write all PT_LOADs to vmcore */
+    write_elf_phdr_loads(s, errp);
     if (*errp) {
         return;
     }
--

2.34.1

#6 lore raw

Re: [PATCH v3 03/14] dump: Convert GuestPhysBlock iterators and use the filter functions

In reply to Janosch Frank

Hi

Quoted 185 lines · Janosch Frank

The iteration over the memblocks is hard to understand so it's about time to clean it up. Instead of manually grabbing the next memblock we can use QTAILQ_FOREACH to iterate over all memblocks.

Additionally we move the calculation of the offset and length out by using the dump_get_memblock_*() functions.

---
 dump/dump.c           | 91 +++++++++++--------------------------------
 include/sysemu/dump.h |  9 ++---
 2 files changed, 27 insertions(+), 73 deletions(-)
dump/dump.c +23 -68
index 0ed7cf9c7b..5c9ed25c5a 100644
@@ -591,56 +591,27 @@ static void dump_begin(DumpState *s, Error **errp)
     write_elf_notes(s, errp);
 }
 
-static int get_next_block(DumpState *s, GuestPhysBlock *block)
-{
-    while (1) {
-        block = QTAILQ_NEXT(block, next);
-        if (!block) {
-            /* no more block */
-            return 1;
-        }
-
-        s->start = 0;
-        s->next_block = block;
-        if (s->has_filter) {
-            if (block->target_start >= s->begin + s->length ||
-                block->target_end <= s->begin) {
-                /* This block is out of the range */
-                continue;
-            }
-
-            if (s->begin > block->target_start) {
-                s->start = s->begin - block->target_start;
-            }
-        }
-
-        return 0;
-    }
-}
-
 /* write all memory to vmcore */
 static void dump_iterate(DumpState *s, Error **errp)
 {
     ERRP_GUARD();
     GuestPhysBlock *block;
-    int64_t size;
+    int64_t memblock_size, memblock_start;
 
-    do {
-        block = s->next_block;
-
-        size = block->target_end - block->target_start;
-        if (s->has_filter) {
-            size -= s->start;
-            if (s->begin + s->length < block->target_end) {
-                size -= block->target_end - (s->begin + s->length);
-            }
+    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
+        memblock_start = dump_get_memblock_start(block, s->begin, s->length);
+        if (memblock_start == -1) {
+            continue;
         }
-        write_memory(s, block, s->start, size, errp);
+
+        memblock_size = dump_get_memblock_size(block, s->begin, s->length);
+
+        /* Write the memory to file */
+        write_memory(s, block, memblock_start, memblock_size, errp);
         if (*errp) {
             return;
         }
-
-    } while (!get_next_block(s, block));
+    }
 }
 
 static void create_vmcore(DumpState *s, Error **errp)
@@ -1490,30 +1461,22 @@ static void create_kdump_vmcore(DumpState *s, Error **errp)
     }
 }
 
-static ram_addr_t get_start_block(DumpState *s)
+static int validate_start_block(DumpState *s)
 {
     GuestPhysBlock *block;
 
     if (!s->has_filter) {
-        s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
         return 0;
     }
 
     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
+        /* This block is out of the range */
         if (block->target_start >= s->begin + s->length ||
             block->target_end <= s->begin) {
-            /* This block is out of the range */
             continue;
         }
-
-        s->next_block = block;
-        if (s->begin > block->target_start) {
-            s->start = s->begin - block->target_start;
-        } else {
-            s->start = 0;
-        }
-        return s->start;
-    }
+        return 0;
+   }
 
     return -1;
 }
@@ -1540,25 +1503,17 @@ bool qemu_system_dump_in_progress(void)
     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
 }
 
-/* calculate total size of memory to be dumped (taking filter into
- * acoount.) */
+/*
+ * calculate total size of memory to be dumped (taking filter into
+ * account.)
+ */
 static int64_t dump_calculate_size(DumpState *s)
 {
     GuestPhysBlock *block;
-    int64_t size = 0, total = 0, left = 0, right = 0;
+    int64_t total = 0;
 
     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
-        if (s->has_filter) {
-            /* calculate the overlapped region. */
-            left = MAX(s->begin, block->target_start);
-            right = MIN(s->begin + s->length, block->target_end);
-            size = right - left;
-            size = size > 0 ? size : 0;
-        } else {
-            /* count the whole region in */
-            size = (block->target_end - block->target_start);
-        }
-        total += size;
+        total += dump_get_memblock_size(block, s->begin, s->length);
     }
 
     return total;
@@ -1660,8 +1615,8 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         goto cleanup;
     }
 
-    s->start = get_start_block(s);
-    if (s->start == -1) {
+    /* Is the filter filtering everything? */
+    if (validate_start_block(s) == -1) {
         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
         goto cleanup;
     }
include/sysemu/dump.h +4 -6
index 2b39abeeae..7025e50682 100644
@@ -166,11 +166,10 @@ typedef struct DumpState {
     hwaddr memory_offset;
     int fd;
 
-    GuestPhysBlock *next_block;
-    ram_addr_t start;
-    bool has_filter;
-    int64_t begin;
-    int64_t length;
+    /* Guest memory related data */
+    bool has_filter;           /* Are we dumping parts of the memory? */
+    int64_t begin;             /* Start address of the chunk we want to dump */
+    int64_t length;            /* Length of the dump we want to dump */
 
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
--
Quoted 2 lines

2.34.1

My suggestion in v2 review was to introduce each function & refactoring independently, if possible. And it looks like the validate_start_block() change could be a 3rd patch too.

Reviewed-by on v3 4/14
#9 lore raw

Re: [PATCH v3 04/14] dump: Allocate header

In reply to Janosch Frank
Quoted 6 lines · Janosch Frank

Allocating the header lets us write it at a later time and hence also allows us to change section and segment table offsets until we finally write it.

you could have added from v2

Quoted 195 lines
---
 dump/dump.c           | 127 +++++++++++++++++++++---------------------
 include/sysemu/dump.h |   1 +
 2 files changed, 64 insertions(+), 64 deletions(-)
dump/dump.c +63 -64
index 5c9ed25c5a..2d04e06815 100644
@@ -98,6 +98,7 @@ static int dump_cleanup(DumpState *s)
     memory_mapping_list_free(&s->list);
     close(s->fd);
     g_free(s->guest_note);
+    g_free(s->elf_header);
     s->guest_note = NULL;
     if (s->resume) {
         if (s->detached) {
@@ -126,73 +127,49 @@ static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
     return 0;
 }
 
-static void write_elf64_header(DumpState *s, Error **errp)
+static void prepare_elf64_header(DumpState *s)
 {
-    /*
-     * phnum in the elf header is 16 bit, if we have more segments we
-     * set phnum to PN_XNUM and write the real number of segments to a
-     * special section.
-     */
-    uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
-    Elf64_Ehdr elf_header;
-    int ret;
+    uint16_t phnum = s->phdr_num >= PN_XNUM ? PN_XNUM : s->phdr_num;
+    Elf64_Ehdr *elf_header = s->elf_header;
 
-    memset(&elf_header, 0, sizeof(Elf64_Ehdr));
-    memcpy(&elf_header, ELFMAG, SELFMAG);
-    elf_header.e_ident[EI_CLASS] = ELFCLASS64;
-    elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
-    elf_header.e_ident[EI_VERSION] = EV_CURRENT;
-    elf_header.e_type = cpu_to_dump16(s, ET_CORE);
-    elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
-    elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
-    elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
-    elf_header.e_phoff = cpu_to_dump64(s, s->phdr_offset);
-    elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
-    elf_header.e_phnum = cpu_to_dump16(s, phnum);
+    memcpy(elf_header, ELFMAG, SELFMAG);
+    elf_header->e_ident[EI_CLASS] = ELFCLASS64;
+    elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
+    elf_header->e_ident[EI_VERSION] = EV_CURRENT;
+    elf_header->e_type = cpu_to_dump16(s, ET_CORE);
+    elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
+    elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
+    elf_header->e_ehsize = cpu_to_dump16(s, sizeof(*elf_header));
+    elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
+    elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
+    elf_header->e_phnum = cpu_to_dump16(s, phnum);
     if (s->shdr_num) {
-        elf_header.e_shoff = cpu_to_dump64(s, s->shdr_offset);
-        elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
-        elf_header.e_shnum = cpu_to_dump16(s, s->shdr_num);
-    }
-
-    ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
-    if (ret < 0) {
-        error_setg_errno(errp, -ret, "dump: failed to write elf header");
+        elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
+        elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
+        elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
     }
 }
 
-static void write_elf32_header(DumpState *s, Error **errp)
+static void prepare_elf32_header(DumpState *s)
 {
-    /*
-     * phnum in the elf header is 16 bit, if we have more segments we
-     * set phnum to PN_XNUM and write the real number of segments to a
-     * special section.
-     */
-    uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
-    Elf32_Ehdr elf_header;
-    int ret;
+    uint16_t phnum = s->phdr_num >= PN_XNUM ? PN_XNUM : s->phdr_num;
+    Elf32_Ehdr *elf_header = s->elf_header;
 
-    memset(&elf_header, 0, sizeof(Elf32_Ehdr));
-    memcpy(&elf_header, ELFMAG, SELFMAG);
-    elf_header.e_ident[EI_CLASS] = ELFCLASS32;
-    elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
-    elf_header.e_ident[EI_VERSION] = EV_CURRENT;
-    elf_header.e_type = cpu_to_dump16(s, ET_CORE);
-    elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
-    elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
-    elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
-    elf_header.e_phoff = cpu_to_dump32(s, s->phdr_offset);
-    elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
-    elf_header.e_phnum = cpu_to_dump16(s, phnum);
+    memcpy(elf_header, ELFMAG, SELFMAG);
+    elf_header->e_ident[EI_CLASS] = ELFCLASS32;
+    elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
+    elf_header->e_ident[EI_VERSION] = EV_CURRENT;
+    elf_header->e_type = cpu_to_dump16(s, ET_CORE);
+    elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
+    elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
+    elf_header->e_ehsize = cpu_to_dump16(s, sizeof(*elf_header));
+    elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
+    elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
+    elf_header->e_phnum = cpu_to_dump16(s, phnum);
     if (s->shdr_num) {
-        elf_header.e_shoff = cpu_to_dump32(s, s->shdr_offset);
-        elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
-        elf_header.e_shnum = cpu_to_dump16(s, s->shdr_num);
-    }
-
-    ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
-    if (ret < 0) {
-        error_setg_errno(errp, -ret, "dump: failed to write elf header");
+        elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
+        elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
+        elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
     }
 }
 
@@ -528,6 +505,26 @@ static void write_elf_notes(DumpState *s, Error **errp)
     }
 }
 
+static void prepare_elf_header(DumpState *s)
+{
+    if (dump_is_64bit(s)) {
+        prepare_elf64_header(s);
+    } else {
+        prepare_elf32_header(s);
+    }
+}
+
+static void write_elf_header(DumpState *s, Error **errp)
+{
+    size_t size = dump_is_64bit(s) ? sizeof(Elf64_Ehdr) : sizeof(Elf32_Ehdr);
+    int ret;
+
+    ret = fd_write_vmcore(s->elf_header, size, s);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "dump: failed to write elf header");
+    }
+}
+
 /* write elf header, PT_NOTE and elf note to vmcore. */
 static void dump_begin(DumpState *s, Error **errp)
 {
@@ -557,12 +554,11 @@ static void dump_begin(DumpState *s, Error **errp)
      * vmcore.
      */
 
-    /* write elf header to vmcore */
-    if (dump_is_64bit(s)) {
-        write_elf64_header(s, errp);
-    } else {
-        write_elf32_header(s, errp);
-    }
+    /* Write elf header to buffer */
+    prepare_elf_header(s);
+
+    /* Start to write stuff into file descriptor */
+    write_elf_header(s, errp);
     if (*errp) {
         return;
     }
@@ -1642,6 +1638,9 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         goto cleanup;
     }
 
+    s->elf_header = g_malloc0(dump_is_64bit(s) ?
+                              sizeof(Elf64_Ehdr) : sizeof(Elf32_Ehdr));
+
     /*
      * The goal of this block is to (a) update the previously guessed
      * phys_base, (b) copy the guest note out of the guest.
include/sysemu/dump.h +1 -1
index 7025e50682..58f41bbf45 100644
@@ -171,6 +171,7 @@ typedef struct DumpState {
     int64_t begin;             /* Start address of the chunk we want to dump */
     int64_t length;            /* Length of the dump we want to dump */
 
+    void *elf_header;
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
     uint32_t nr_cpus;           /* number of guest's cpu */
--

2.34.1

#11 lore raw

Re: [PATCH v3 05/14] dump: Split write of section headers and data and add a prepare step

In reply to Janosch Frank
Quoted 61 lines · Janosch Frank

By splitting the writing of the section headers and (future) section data we prepare for the addition of a string table section and architecture sections.

At the same time we move the writing of the section to the end of the dump process. This allows the upcoming architecture section code to add data after all of the common dump data has been written.

---
 dump/dump.c           | 114 ++++++++++++++++++++++++++++++++----------
 include/sysemu/dump.h |   4 ++
 2 files changed, 92 insertions(+), 26 deletions(-)
dump/dump.c +17 -16
index 2d04e06815..980702d476 100644
@@ -342,30 +342,71 @@ static void write_elf_phdr_note(DumpState *s, Error **errp)
     }
 }
 
-static void write_elf_section(DumpState *s, int type, Error **errp)
+static size_t prepare_elf_section_hdr_zero(DumpState *s, void *buff)
 {
-    Elf32_Shdr shdr32;
-    Elf64_Shdr shdr64;
-    int shdr_size;
-    void *shdr;
-    int ret;
+    if (dump_is_64bit(s)) {
+        Elf64_Shdr *shdr64 = buff;
 
-    if (type == 0) {
-        shdr_size = sizeof(Elf32_Shdr);
-        memset(&shdr32, 0, shdr_size);
-        shdr32.sh_info = cpu_to_dump32(s, s->phdr_num);
-        shdr = &shdr32;
+        shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
     } else {
-        shdr_size = sizeof(Elf64_Shdr);
-        memset(&shdr64, 0, shdr_size);
-        shdr64.sh_info = cpu_to_dump32(s, s->phdr_num);
-        shdr = &shdr64;
+        Elf32_Shdr *shdr32 = buff;
+
+        shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
     }
 
-    ret = fd_write_vmcore(shdr, shdr_size, s);
+    return dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+}
+
+static void prepare_elf_section_hdrs(DumpState *s)
+{
+    uint8_t *buff_hdr;
+    size_t len, sizeof_shdr;
+
+    /*
+     * Section ordering:

dump/dump.c static void write_elf_phdr_note(DumpState *s, Error **errp) · line 376

Quoted 10 lines
dump/dump.c +10 -0
+     * - HDR zero (if needed)
+     */
+    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+    len = sizeof_shdr * s->shdr_num;
+    s->elf_section_hdrs = g_malloc0(len);
+    buff_hdr = s->elf_section_hdrs;
+
+    /* Write special section first */
+    if (s->phdr_num == PN_XNUM) {
+            prepare_elf_section_hdr_zero(s, buff_hdr);

From v2 review:

Eventually, drop buff_hdr, and pass only "s" as argument

+ Indentation is off

Quoted 127 lines
dump/dump.c +61 -10
+    }
+}
+
+static void prepare_elf_sections(DumpState *s, Error **errp)
+{
+    if (!s->shdr_num) {
+        return;
+    }
+
+    prepare_elf_section_hdrs(s);
+}
+
+static void write_elf_section_headers(DumpState *s, Error **errp)
+{
+    size_t sizeof_shdr;
+    int ret;
+
+    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+
+    ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
     if (ret < 0) {
-        error_setg_errno(errp, -ret,
-                         "dump: failed to write section header table");
+        error_setg_errno(errp, -ret, "dump: failed to write section headers");
+    }
+}
+
+static void write_elf_sections(DumpState *s, Error **errp)
+{
+    int ret;
+
+    /* Write section zero */
+    ret = fd_write_vmcore(s->elf_section_data, s->elf_section_data_size, s);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "dump: failed to write section data");
     }
 }
 
@@ -557,12 +598,22 @@ static void dump_begin(DumpState *s, Error **errp)
     /* Write elf header to buffer */
     prepare_elf_header(s);
 
+    prepare_elf_sections(s, errp);
+    if (*errp) {
+        return;
+    }
+
     /* Start to write stuff into file descriptor */
     write_elf_header(s, errp);
     if (*errp) {
         return;
     }
 
+    write_elf_section_headers(s, errp);
+    if (*errp) {
+        return;
+    }
+
     /* write PT_NOTE to vmcore */
     write_elf_phdr_note(s, errp);
     if (*errp) {
@@ -575,14 +626,6 @@ static void dump_begin(DumpState *s, Error **errp)
         return;
     }
 
-    /* write section to vmcore */
-    if (s->shdr_num) {
-        write_elf_section(s, 1, errp);
-        if (*errp) {
-            return;
-        }
-    }
-
     /* write notes to vmcore */
     write_elf_notes(s, errp);
 }
@@ -610,6 +653,19 @@ static void dump_iterate(DumpState *s, Error **errp)
     }
 }
 
+static void dump_end(DumpState *s, Error **errp)
+{
+    ERRP_GUARD();
+
+    if (!s->elf_section_data_size) {
+        return;
+    }
+    s->elf_section_data = g_malloc0(s->elf_section_data_size);
+
+    /* write sections to vmcore */
+    write_elf_sections(s, errp);
+}
+
 static void create_vmcore(DumpState *s, Error **errp)
 {
     ERRP_GUARD();
@@ -620,6 +676,12 @@ static void create_vmcore(DumpState *s, Error **errp)
     }
 
     dump_iterate(s, errp);
+    if (*errp) {
+        return;
+    }
+
+    /* Write section data after memory has been dumped */
+    dump_end(s, errp);
 }
 
 static int write_start_flat_header(int fd)
include/sysemu/dump.h +4 -1
index 58f41bbf45..dad10dee0b 100644
@@ -172,6 +172,10 @@ typedef struct DumpState {
     int64_t length;            /* Length of the dump we want to dump */
 
     void *elf_header;
+    void *elf_section_hdrs;
+    uint64_t elf_section_data_size;
+    void *elf_section_data;
+
     uint8_t *note_buf;          /* buffer for notes */
     size_t note_buf_offset;     /* the writing place in note_buf */
     uint32_t nr_cpus;           /* number of guest's cpu */
--

2.34.1

Reviewed-by on v3 6/14
#14 lore raw

Re: [PATCH v3 06/14] dump: Reorder struct DumpState

In reply to Janosch Frank
Quoted 6 lines · Janosch Frank

Let's move ELF related members into one block and guest memory related ones into another to improve readability.

from v2:

Quoted 44 lines
---
 include/sysemu/dump.h | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
include/sysemu/dump.h +10 -8
index dad10dee0b..a94d81303d 100644
@@ -154,15 +154,8 @@ typedef struct DumpState {
     GuestPhysBlockList guest_phys_blocks;
     ArchDumpInfo dump_info;
     MemoryMappingList list;
-    uint32_t phdr_num;
-    uint32_t shdr_num;
     bool resume;
     bool detached;
-    ssize_t note_size;
-    hwaddr shdr_offset;
-    hwaddr phdr_offset;
-    hwaddr section_offset;
-    hwaddr note_offset;
     hwaddr memory_offset;
     int fd;
 
@@ -171,6 +164,16 @@ typedef struct DumpState {
     int64_t begin;             /* Start address of the chunk we want to dump */
     int64_t length;            /* Length of the dump we want to dump */
 
+    /* Elf dump related data */
+    uint32_t phdr_num;
+    uint32_t shdr_num;
+    uint32_t sh_info;
+    ssize_t note_size;
+    hwaddr shdr_offset;
+    hwaddr phdr_offset;
+    hwaddr note_offset;
+    hwaddr section_offset;
+
     void *elf_header;
     void *elf_section_hdrs;
     uint64_t elf_section_data_size;
--

2.34.1

#12 lore raw

Re: [PATCH v3 05/14] dump: Split write of section headers and data and add a prepare step

In reply to Marc-André Lureau
Quoted 75 lines · Marc-André Lureau
Quoted 61 lines · Janosch Frank

By splitting the writing of the section headers and (future) section data we prepare for the addition of a string table section and architecture sections.

At the same time we move the writing of the section to the end of the dump process. This allows the upcoming architecture section code to add data after all of the common dump data has been written.

---
  dump/dump.c           | 114 ++++++++++++++++++++++++++++++++----------
  include/sysemu/dump.h |   4 ++
  2 files changed, 92 insertions(+), 26 deletions(-)
dump/dump.c +17 -16
index 2d04e06815..980702d476 100644
@@ -342,30 +342,71 @@ static void write_elf_phdr_note(DumpState *s, Error **errp)
      }
  }
 
-static void write_elf_section(DumpState *s, int type, Error **errp)
+static size_t prepare_elf_section_hdr_zero(DumpState *s, void *buff)
  {
-    Elf32_Shdr shdr32;
-    Elf64_Shdr shdr64;
-    int shdr_size;
-    void *shdr;
-    int ret;
+    if (dump_is_64bit(s)) {
+        Elf64_Shdr *shdr64 = buff;
 
-    if (type == 0) {
-        shdr_size = sizeof(Elf32_Shdr);
-        memset(&shdr32, 0, shdr_size);
-        shdr32.sh_info = cpu_to_dump32(s, s->phdr_num);
-        shdr = &shdr32;
+        shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
      } else {
-        shdr_size = sizeof(Elf64_Shdr);
-        memset(&shdr64, 0, shdr_size);
-        shdr64.sh_info = cpu_to_dump32(s, s->phdr_num);
-        shdr = &shdr64;
+        Elf32_Shdr *shdr32 = buff;
+
+        shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
      }
 
-    ret = fd_write_vmcore(shdr, shdr_size, s);
+    return dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+}
+
+static void prepare_elf_section_hdrs(DumpState *s)
+{
+    uint8_t *buff_hdr;
+    size_t len, sizeof_shdr;
+
+    /*
+     * Section ordering:

dump/dump.c static void write_elf_phdr_note(DumpState *s, Error **errp) · line 376

Quoted 10 lines
dump/dump.c +10 -0
+     * - HDR zero (if needed)
+     */
+    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+    len = sizeof_shdr * s->shdr_num;
+    s->elf_section_hdrs = g_malloc0(len);
+    buff_hdr = s->elf_section_hdrs;
+
+    /* Write special section first */
+    if (s->phdr_num == PN_XNUM) {
+            prepare_elf_section_hdr_zero(s, buff_hdr);

From v2 review:

Quoted 1 line

Eventually, drop buff_hdr, and pass only "s" as argument

You need to specify what you mean by "eventually" here.

Quoted 2 lines

+ Indentation is off

I fixed that Tuesday but somehow it didn't end up in this mail... I'll have another go at it.

Quoted 131 lines
Quoted 130 lines
dump/dump.c +61 -10
+    }
+}
+
+static void prepare_elf_sections(DumpState *s, Error **errp)
+{
+    if (!s->shdr_num) {
+        return;
+    }
+
+    prepare_elf_section_hdrs(s);
+}
+
+static void write_elf_section_headers(DumpState *s, Error **errp)
+{
+    size_t sizeof_shdr;
+    int ret;
+
+    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
+
+    ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
      if (ret < 0) {
-        error_setg_errno(errp, -ret,
-                         "dump: failed to write section header table");
+        error_setg_errno(errp, -ret, "dump: failed to write section headers");
+    }
+}
+
+static void write_elf_sections(DumpState *s, Error **errp)
+{
+    int ret;
+
+    /* Write section zero */
+    ret = fd_write_vmcore(s->elf_section_data, s->elf_section_data_size, s);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "dump: failed to write section data");
      }
  }
 
@@ -557,12 +598,22 @@ static void dump_begin(DumpState *s, Error **errp)
      /* Write elf header to buffer */
      prepare_elf_header(s);
 
+    prepare_elf_sections(s, errp);
+    if (*errp) {
+        return;
+    }
+
      /* Start to write stuff into file descriptor */
      write_elf_header(s, errp);
      if (*errp) {
          return;
      }
 
+    write_elf_section_headers(s, errp);
+    if (*errp) {
+        return;
+    }
+
      /* write PT_NOTE to vmcore */
      write_elf_phdr_note(s, errp);
      if (*errp) {
@@ -575,14 +626,6 @@ static void dump_begin(DumpState *s, Error **errp)
          return;
      }
 
-    /* write section to vmcore */
-    if (s->shdr_num) {
-        write_elf_section(s, 1, errp);
-        if (*errp) {
-            return;
-        }
-    }
-
      /* write notes to vmcore */
      write_elf_notes(s, errp);
  }
@@ -610,6 +653,19 @@ static void dump_iterate(DumpState *s, Error **errp)
      }
  }
 
+static void dump_end(DumpState *s, Error **errp)
+{
+    ERRP_GUARD();
+
+    if (!s->elf_section_data_size) {
+        return;
+    }
+    s->elf_section_data = g_malloc0(s->elf_section_data_size);
+
+    /* write sections to vmcore */
+    write_elf_sections(s, errp);
+}
+
  static void create_vmcore(DumpState *s, Error **errp)
  {
      ERRP_GUARD();
@@ -620,6 +676,12 @@ static void create_vmcore(DumpState *s, Error **errp)
      }
 
      dump_iterate(s, errp);
+    if (*errp) {
+        return;
+    }
+
+    /* Write section data after memory has been dumped */
+    dump_end(s, errp);
  }
 
  static int write_start_flat_header(int fd)
include/sysemu/dump.h +4 -1
index 58f41bbf45..dad10dee0b 100644
@@ -172,6 +172,10 @@ typedef struct DumpState {
      int64_t length;            /* Length of the dump we want to dump */
 
      void *elf_header;
+    void *elf_section_hdrs;
+    uint64_t elf_section_data_size;
+    void *elf_section_data;
+
      uint8_t *note_buf;          /* buffer for notes */
      size_t note_buf_offset;     /* the writing place in note_buf */
      uint32_t nr_cpus;           /* number of guest's cpu */
--

2.34.1

#16 lore raw

Re: [PATCH v3 07/14] dump: Swap segment and section header locations

In reply to Janosch Frank

Hi

Quoted 42 lines · Janosch Frank

For the upcoming string table and arch section support we need to modify the elf layout a bit. Instead of the segments, i.e. the guest's memory contents, beeing the last area the section data will live at the end of the file. This will allow us to write the section data after all guest memory has been dumped which is important for the s390 PV dump support.

---
 dump/dump.c | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)
dump/dump.c +7 -5
index 980702d476..6f3274c5af 100644
@@ -590,6 +590,9 @@ static void dump_begin(DumpState *s, Error **errp)
      *   --------------
      *   |  memory     |
      *   --------------
+     *   |  sectn data |
+     *   --------------
+
      *
      * we only know where the memory is saved after we write elf note into
      * vmcore.
@@ -1817,18 +1820,18 @@ static void dump_init(DumpState *s, int fd, bool has_format,
         }
     }
 
+    tmp = (s->phdr_num == PN_XNUM) ? s->sh_info : s->phdr_num;
     if (dump_is_64bit(s)) {
-        s->phdr_offset = sizeof(Elf64_Ehdr);
-        s->shdr_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
-        s->note_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
-        s->memory_offset = s->note_offset + s->note_size;
+        s->shdr_offset = sizeof(Elf64_Ehdr);
+        s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
+        s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * tmp;
     } else {
-

dump/dump.c static void dump_init(DumpState *s, int fd, bool has_format, · line 1834

Quoted 10 lines
dump/dump.c +5 -4
-        s->phdr_offset = sizeof(Elf32_Ehdr);
-        s->shdr_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
-        s->note_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
-        s->memory_offset = s->note_offset + s->note_size;
+        s->shdr_offset = sizeof(Elf32_Ehdr);
+        s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
+        s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * tmp;
     }
+    s->memory_offset = s->note_offset + s->note_size;
+    s->section_offset = s->memory_offset + s->total_size;

Ah! section_offset was actually added from commit e71d353360 ("dump:
Add more offset variables"), but unused until now. It's worth explaining in the commit message.

Quoted 6 lines
dump/dump.c +0 -1
     return;
 
--

2.34.1

#22 lore raw

Re: [PATCH v3 10/14] linux header sync

In reply to Janosch Frank
Quoted 3 lines · Janosch Frank

Add the uapi data for KVM_CAP_S390_PROTECTED_DUMP which I expect to be added with 5.20.

It's going to be overwritten by the next update-linux-headers.sh. I'd mark this patch as WIP or Draft: in the title until it's actually sync and imported from the kernel.

Quoted 93 lines
---
 linux-headers/linux/kvm.h | 54 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 54 insertions(+)
linux-headers/linux/kvm.h +54 -1
index 0d05d02ee4..213216c27c 100644
@@ -1150,6 +1150,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_DISABLE_QUIRKS2 213
 /* #define KVM_CAP_VM_TSC_CONTROL 214 */
 #define KVM_CAP_SYSTEM_EVENT_DATA 215
+#define KVM_CAP_S390_PROTECTED_DUMP 217
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1651,6 +1652,55 @@ struct kvm_s390_pv_unp {
        __u64 tweak;
 };
 
+enum pv_cmd_info_id {
+       KVM_PV_INFO_VM,
+       KVM_PV_INFO_DUMP,
+};
+
+struct kvm_s390_pv_info_dump {
+       __u64 dump_cpu_buffer_len;
+       __u64 dump_config_mem_buffer_per_1m;
+       __u64 dump_config_finalize_len;
+};
+
+struct kvm_s390_pv_info_vm {
+       __u64 inst_calls_list[4];
+       __u64 max_cpus;
+       __u64 max_guests;
+       __u64 max_guest_addr;
+       __u64 feature_indication;
+};
+
+struct kvm_s390_pv_info_header {
+       __u32 id;
+       __u32 len_max;
+       __u32 len_written;
+       __u32 reserved;
+};
+
+struct kvm_s390_pv_info {
+       struct kvm_s390_pv_info_header header;
+       union {
+               struct kvm_s390_pv_info_dump dump;
+               struct kvm_s390_pv_info_vm vm;
+       };
+};
+
+enum pv_cmd_dmp_id {
+        KVM_PV_DUMP_INIT,
+        KVM_PV_DUMP_CONFIG_STATE,
+        KVM_PV_DUMP_COMPLETE,
+        KVM_PV_DUMP_CPU,
+};
+
+struct kvm_s390_pv_dmp {
+        __u64 subcmd;
+        __u64 buff_addr;
+        __u64 buff_len;
+        __u64 gaddr;
+        __u64 reserved[4];
+};
+
 enum pv_cmd_id {
        KVM_PV_ENABLE,
        KVM_PV_DISABLE,
@@ -1659,6 +1709,8 @@ enum pv_cmd_id {
        KVM_PV_VERIFY,
        KVM_PV_PREP_RESET,
        KVM_PV_UNSHARE_ALL,
+        KVM_PV_INFO,
+        KVM_PV_DUMP,
 };
 
 struct kvm_pv_cmd {
@@ -2066,4 +2118,6 @@ struct kvm_stats_desc {
 /* Available with KVM_CAP_XSAVE2 */
 #define KVM_GET_XSAVE2           _IOR(KVMIO,  0xcf, struct kvm_xsave)
 
+#define KVM_S390_PV_CPU_COMMAND _IOWR(KVMIO, 0xd0, struct kvm_pv_cmd)
+
 #endif /* __LINUX_KVM_H */
--

2.34.1

#23 lore raw

Re: [PATCH v3 10/14] linux header sync

Quoted 8 lines · Marc-André Lureau
Quoted 3 lines · Janosch Frank

Add the uapi data for KVM_CAP_S390_PROTECTED_DUMP which I expect to be added with 5.20.

It's going to be overwritten by the next update-linux-headers.sh. I'd mark this patch as WIP or Draft: in the title until it's actually sync and imported from the kernel.

Yes, such patches are normally just ignored by maintainers as they provide their own sync patch that's why it doesn't have a proper subject or a WIP prefix. It's needed to be able to compile it until the proper sync has arrived.

Quoted 96 lines
Quoted 95 lines
---
  linux-headers/linux/kvm.h | 54 +++++++++++++++++++++++++++++++++++++++
  1 file changed, 54 insertions(+)
linux-headers/linux/kvm.h +54 -1
index 0d05d02ee4..213216c27c 100644
@@ -1150,6 +1150,7 @@ struct kvm_ppc_resize_hpt {
  #define KVM_CAP_DISABLE_QUIRKS2 213
  /* #define KVM_CAP_VM_TSC_CONTROL 214 */
  #define KVM_CAP_SYSTEM_EVENT_DATA 215
+#define KVM_CAP_S390_PROTECTED_DUMP 217
 
  #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1651,6 +1652,55 @@ struct kvm_s390_pv_unp {
         __u64 tweak;
  };
 
+enum pv_cmd_info_id {
+       KVM_PV_INFO_VM,
+       KVM_PV_INFO_DUMP,
+};
+
+struct kvm_s390_pv_info_dump {
+       __u64 dump_cpu_buffer_len;
+       __u64 dump_config_mem_buffer_per_1m;
+       __u64 dump_config_finalize_len;
+};
+
+struct kvm_s390_pv_info_vm {
+       __u64 inst_calls_list[4];
+       __u64 max_cpus;
+       __u64 max_guests;
+       __u64 max_guest_addr;
+       __u64 feature_indication;
+};
+
+struct kvm_s390_pv_info_header {
+       __u32 id;
+       __u32 len_max;
+       __u32 len_written;
+       __u32 reserved;
+};
+
+struct kvm_s390_pv_info {
+       struct kvm_s390_pv_info_header header;
+       union {
+               struct kvm_s390_pv_info_dump dump;
+               struct kvm_s390_pv_info_vm vm;
+       };
+};
+
+enum pv_cmd_dmp_id {
+        KVM_PV_DUMP_INIT,
+        KVM_PV_DUMP_CONFIG_STATE,
+        KVM_PV_DUMP_COMPLETE,
+        KVM_PV_DUMP_CPU,
+};
+
+struct kvm_s390_pv_dmp {
+        __u64 subcmd;
+        __u64 buff_addr;
+        __u64 buff_len;
+        __u64 gaddr;
+        __u64 reserved[4];
+};
+
  enum pv_cmd_id {
         KVM_PV_ENABLE,
         KVM_PV_DISABLE,
@@ -1659,6 +1709,8 @@ enum pv_cmd_id {
         KVM_PV_VERIFY,
         KVM_PV_PREP_RESET,
         KVM_PV_UNSHARE_ALL,
+        KVM_PV_INFO,
+        KVM_PV_DUMP,
  };
 
  struct kvm_pv_cmd {
@@ -2066,4 +2118,6 @@ struct kvm_stats_desc {
  /* Available with KVM_CAP_XSAVE2 */
  #define KVM_GET_XSAVE2           _IOR(KVMIO,  0xcf, struct kvm_xsave)
 
+#define KVM_S390_PV_CPU_COMMAND _IOWR(KVMIO, 0xd0, struct kvm_pv_cmd)
+
  #endif /* __LINUX_KVM_H */
--

2.34.1

#17 lore raw

Re: [PATCH v3 07/14] dump: Swap segment and section header locations

In reply to Marc-André Lureau
Quoted 51 lines · Marc-André Lureau

Hi

Quoted 48 lines · Janosch Frank

For the upcoming string table and arch section support we need to modify the elf layout a bit. Instead of the segments, i.e. the guest's memory contents, beeing the last area the section data will live at the end of the file. This will allow us to write the section data after all guest memory has been dumped which is important for the s390 PV dump support.

---
  dump/dump.c | 21 ++++++++++++---------
  1 file changed, 12 insertions(+), 9 deletions(-)
dump/dump.c +9 -9
index 980702d476..6f3274c5af 100644
@@ -590,6 +590,9 @@ static void dump_begin(DumpState *s, Error **errp)
       *   --------------
       *   |  memory     |
       *   --------------
+     *   |  sectn data |
+     *   --------------
+
       *
       * we only know where the memory is saved after we write elf note into
       * vmcore.
@@ -1817,18 +1820,18 @@ static void dump_init(DumpState *s, int fd, bool has_format,
          }
      }
 
+    tmp = (s->phdr_num == PN_XNUM) ? s->sh_info : s->phdr_num;
      if (dump_is_64bit(s)) {
-        s->phdr_offset = sizeof(Elf64_Ehdr);
-        s->shdr_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
-        s->note_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
-        s->memory_offset = s->note_offset + s->note_size;
+        s->shdr_offset = sizeof(Elf64_Ehdr);
+        s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
+        s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * tmp;
      } else {
-
-        s->phdr_offset = sizeof(Elf32_Ehdr);
-        s->shdr_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
-        s->note_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
-        s->memory_offset = s->note_offset + s->note_size;
+        s->shdr_offset = sizeof(Elf32_Ehdr);
+        s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
Quoted 8 lines

dump/dump.c static void dump_init(DumpState *s, int fd, bool has_format, · line 1834

Quoted 4 lines
dump/dump.c +3 -0
+        s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * tmp;
      }
+    s->memory_offset = s->note_offset + s->note_size;
+    s->section_offset = s->memory_offset + s->total_size;

Ah! section_offset was actually added from commit e71d353360 ("dump:
Add more offset variables"), but unused until now. It's worth explaining in the commit message.

I think I'll just move it to this patch so the previous one really only does re-ordering and doesn't introduce new struct members.

Quoted 9 lines
Quoted 7 lines
dump/dump.c +0 -1
      return;
 
--

2.34.1

#19 lore raw

Re: [PATCH v3 08/14] dump/dump: Add section string table support

In reply to Janosch Frank
Quoted 76 lines · Janosch Frank

As sections don't have a type like the notes do we need another way to determine their contents. The string table allows us to assign each section an identification string which architectures can then use to tag their sections with.

There will be no string table if the architecture doesn't add custom sections which are introduced in a following patch.

---
  dump/dump.c           | 81 +++++++++++++++++++++++++++++++++++++++++--
  include/sysemu/dump.h |  1 +
  2 files changed, 80 insertions(+), 2 deletions(-)
dump/dump.c +35 -1
index 6f3274c5af..944217349a 100644
@@ -99,6 +99,7 @@ static int dump_cleanup(DumpState *s)
      close(s->fd);
      g_free(s->guest_note);
      g_free(s->elf_header);
+    g_array_unref(s->string_table_buf);
      s->guest_note = NULL;
      if (s->resume) {
          if (s->detached) {
@@ -357,14 +358,47 @@ static size_t prepare_elf_section_hdr_zero(DumpState *s, void *buff)
      return dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
  }
 
+static void write_elf_section_hdr_string(DumpState *s, void *buff)
+{
+    Elf32_Shdr shdr32;
+    Elf64_Shdr shdr64;
+    int shdr_size;
+    void *shdr = buff;
+
+    if (dump_is_64bit(s)) {
+        shdr_size = sizeof(Elf64_Shdr);
+        memset(&shdr64, 0, shdr_size);
+        shdr64.sh_type = SHT_STRTAB;
+        shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
+        shdr64.sh_name = s->string_table_buf->len;
+        g_array_append_vals(s->string_table_buf, ".strtab", sizeof(".strtab"));
+        shdr64.sh_size = s->string_table_buf->len;
+        shdr = &shdr64;
+    } else {
+        shdr_size = sizeof(Elf32_Shdr);
+        memset(&shdr32, 0, shdr_size);
+        shdr32.sh_type = SHT_STRTAB;
+        shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
+        shdr32.sh_name = s->string_table_buf->len;
+        g_array_append_vals(s->string_table_buf, ".strtab", sizeof(".strtab"));
+        shdr32.sh_size = s->string_table_buf->len;
+        shdr = &shdr32;
+    }
+
+    memcpy(buff, shdr, shdr_size);
+}
+
  static void prepare_elf_section_hdrs(DumpState *s)
  {
      uint8_t *buff_hdr;
-    size_t len, sizeof_shdr;
+    size_t len, size = 0, sizeof_shdr;
+    Elf64_Ehdr *hdr64 = s->elf_header;
+    Elf32_Ehdr *hdr32 = s->elf_header;
 
      /*
       * Section ordering:
       * - HDR zero (if needed)
+     * - String table hdr
       */
      sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
      len = sizeof_shdr * s->shdr_num;

dump/dump.c static void prepare_elf_section_hdrs(DumpState *s) · line 412

Quoted 4 lines
dump/dump.c +1 -0
@@ -375,6 +409,22 @@ static void prepare_elf_section_hdrs(DumpState *s)
      if (s->phdr_num == PN_XNUM) {
              prepare_elf_section_hdr_zero(s, buff_hdr);
      }
+    buff_hdr += size;

If I'm not mistaken this line doesn't make sense at all.
It needs to be in the if and add sizeof_shdr so we can account for the zero header.

I.e.:

if (s->phdr_num == PN_XNUM) {
	prepare_elf_section_hdr_zero(s, buff_hdr);
	buff_hdr += sizeof_shdr;
}

Also size isn't really needed, we can directly add dump_arch_sections_write_hdr()'s return to buff_hdr in the next patch.

Quoted 16 lines
dump/dump.c +15 -0
+
+    if (s->shdr_num < 2) {
+        return;
+    }
+
+    /*
+     * String table needs to be last section since strings are added
+     * via arch_sections_write_hdr().
+     */
+    write_elf_section_hdr_string(s, buff_hdr);
+    if (dump_is_64bit(s)) {
+        hdr64->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
+    } else {
+        hdr32->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
+    }
  }
#7 lore raw

Re: [PATCH v3 03/14] dump: Convert GuestPhysBlock iterators and use the filter functions

In reply to Marc-André Lureau

On 7/21/22 16:36, Marc-André Lureau wrote:
[..]

Quoted 18 lines
Quoted 18 lines
include/sysemu/dump.h +4 -5
index 2b39abeeae..7025e50682 100644
@@ -166,11 +166,10 @@ typedef struct DumpState {
      hwaddr memory_offset;
      int fd;
 
-    GuestPhysBlock *next_block;
-    ram_addr_t start;
-    bool has_filter;
-    int64_t begin;
-    int64_t length;
+    /* Guest memory related data */
+    bool has_filter;           /* Are we dumping parts of the memory? */
+    int64_t begin;             /* Start address of the chunk we want to dump */
+    int64_t length;            /* Length of the dump we want to dump */
 
Quoted 11 lines
Quoted 5 lines
include/sysemu/dump.h +0 -1
      uint8_t *note_buf;          /* buffer for notes */
      size_t note_buf_offset;     /* the writing place in note_buf */
--

2.34.1

My suggestion in v2 review was to introduce each function & refactoring independently, if possible. And it looks like the validate_start_block() change could be a 3rd patch too.

Alright, I just squashed and split this into 5 patches:
* Introducing the 2 new functions
* Converting dump_iterate and removing get_next_block
* get_start_block -> validate_start_block
* Removal of next_block and start from DumpState, last user was
get_start_block
* Re-work of dump_calculate_size

I don't think we can easily adapt to dump_get_memblock_size() and dump_get_memblock_start() independently.

I'll also move the DumpState comment hunk to the removal of start and next_block tomorrow.

Reviewed-by on v3 12/14
#27 lore raw

Re: [PATCH v3 12/14] s390x: Introduce PV query interface

In reply to Janosch Frank
Quoted 3 lines · Janosch Frank

Introduce an interface over which we can get information about UV data.

Quoted 145 lines
---
  hw/s390x/pv.c              | 61 ++++++++++++++++++++++++++++++++++++++
  hw/s390x/s390-virtio-ccw.c |  5 ++++
  include/hw/s390x/pv.h      | 10 +++++++
  3 files changed, 76 insertions(+)
hw/s390x/pv.c +61 -0
index 401b63d6cb..a5af4ddf46 100644
@@ -20,6 +20,11 @@
  #include "exec/confidential-guest-support.h"
  #include "hw/s390x/ipl.h"
  #include "hw/s390x/pv.h"
+#include "target/s390x/kvm/kvm_s390x.h"
+
+static bool info_valid;
+static struct kvm_s390_pv_info_vm info_vm;
+static struct kvm_s390_pv_info_dump info_dump;
 
  static int __s390_pv_cmd(uint32_t cmd, const char *cmdname, void *data)
  {
@@ -56,6 +61,42 @@ static int __s390_pv_cmd(uint32_t cmd, const char *cmdname, void *data)
      }                                  \
  }
 
+int s390_pv_query_info(void)
+{
+    struct kvm_s390_pv_info info = {
+        .header.id = KVM_PV_INFO_VM,
+        .header.len_max = sizeof(info.header) + sizeof(info.vm),
+    };
+    int rc;
+
+    /* Info API's first user is dump so they are bundled */
+    if (!kvm_s390_get_protected_dump()) {
+        return 0;
+    }
+
+    rc = s390_pv_cmd(KVM_PV_INFO, &info);
+    if (rc) {
+        error_report("KVM PV INFO cmd %x failed: %s",
+                     info.header.id, strerror(rc));
+        return rc;
+    }
+    memcpy(&info_vm, &info.vm, sizeof(info.vm));
+
+    info.header.id = KVM_PV_INFO_DUMP;
+    info.header.len_max = sizeof(info.header) + sizeof(info.dump);
+    rc = s390_pv_cmd(KVM_PV_INFO, &info);
+    if (rc) {
+        error_report("KVM PV INFO cmd %x failed: %s",
+                     info.header.id, strerror(rc));
+        return rc;
+    }
+
+    memcpy(&info_dump, &info.dump, sizeof(info.dump));
+    info_valid = true;
+
+    return rc;
+}
+
  int s390_pv_vm_enable(void)
  {
      return s390_pv_cmd(KVM_PV_ENABLE, NULL);
@@ -114,6 +155,26 @@ void s390_pv_inject_reset_error(CPUState *cs)
      env->regs[r1 + 1] = DIAG_308_RC_INVAL_FOR_PV;
  }
 
+uint64_t kvm_s390_pv_dmp_get_size_cpu(void)
+{
+    return info_dump.dump_cpu_buffer_len;
+}
+
+uint64_t kvm_s390_pv_dmp_get_size_complete(void)
+{
+    return info_dump.dump_config_finalize_len;
+}
+
+uint64_t kvm_s390_pv_dmp_get_size_mem(void)
+{
+    return info_dump.dump_config_mem_buffer_per_1m;
+}
+
+bool kvm_s390_pv_info_basic_valid(void)
+{
+    return info_valid;
+}
+
  #define TYPE_S390_PV_GUEST "s390-pv-guest"
  OBJECT_DECLARE_SIMPLE_TYPE(S390PVGuest, S390_PV_GUEST)
 
hw/s390x/s390-virtio-ccw.c +5 -0
index cc3097bfee..f9401e392b 100644
@@ -366,6 +366,11 @@ static int s390_machine_protect(S390CcwMachineState *ms)
 
      ms->pv = true;
 
+    rc = s390_pv_query_info();
+    if (rc) {
+        goto out_err;
+    }
+
      /* Set SE header and unpack */
      rc = s390_ipl_prepare_pv_header();
      if (rc) {
include/hw/s390x/pv.h +10 -0
index 1f1f545bfc..6fa55bf70e 100644
@@ -38,6 +38,7 @@ static inline bool s390_is_pv(void)
      return ccw->pv;
  }
 
+int s390_pv_query_info(void);
  int s390_pv_vm_enable(void);
  void s390_pv_vm_disable(void);
  int s390_pv_set_sec_parms(uint64_t origin, uint64_t length);
@@ -46,8 +47,13 @@ void s390_pv_prep_reset(void);
  int s390_pv_verify(void);
  void s390_pv_unshare(void);
  void s390_pv_inject_reset_error(CPUState *cs);
+uint64_t kvm_s390_pv_dmp_get_size_cpu(void);
+uint64_t kvm_s390_pv_dmp_get_size_mem(void);
+uint64_t kvm_s390_pv_dmp_get_size_complete(void);
+bool kvm_s390_pv_info_basic_valid(void);
  #else /* CONFIG_KVM */
  static inline bool s390_is_pv(void) { return false; }
+static inline int s390_pv_query_info(void) { return 0; }
  static inline int s390_pv_vm_enable(void) { return 0; }
  static inline void s390_pv_vm_disable(void) {}
  static inline int s390_pv_set_sec_parms(uint64_t origin, uint64_t length) { return 0; }
@@ -56,6 +62,10 @@ static inline void s390_pv_prep_reset(void) {}
  static inline int s390_pv_verify(void) { return 0; }
  static inline void s390_pv_unshare(void) {}
  static inline void s390_pv_inject_reset_error(CPUState *cs) {};
+static inline uint64_t kvm_s390_pv_dmp_get_size_cpu(void) { return 0; }
+static inline uint64_t kvm_s390_pv_dmp_get_size_mem(void) { return 0; }
+static inline uint64_t kvm_s390_pv_dmp_get_size_complete(void) { return 0; }
+static inline bool kvm_s390_pv_info_basic_valid(void) { return false; }
  #endif /* CONFIG_KVM */
 
  int s390_pv_kvm_init(ConfidentialGuestSupport *cgs, Error **errp);
#29 lore raw

Re: [PATCH v3 13/14] s390x: Add KVM PV dump interface

In reply to Janosch Frank

Hi Janosch,

looks good to me.
Have a look on my comments.

Quoted 51 lines · Janosch Frank

Let's add a few bits of code which hide the new KVM PV dump API from us via new functions.

---
  hw/s390x/pv.c         | 51 +++++++++++++++++++++++++++++++++++++++++++
  include/hw/s390x/pv.h |  8 +++++++
  2 files changed, 59 insertions(+)
hw/s390x/pv.c +34 -0
index a5af4ddf46..48591c387d 100644
@@ -175,6 +175,57 @@ bool kvm_s390_pv_info_basic_valid(void)
      return info_valid;
  }
 
+static int s390_pv_dump_cmd(uint64_t subcmd, uint64_t uaddr, uint64_t gaddr,
+                            uint64_t len)
+{
+    struct kvm_s390_pv_dmp dmp = {
+        .subcmd = subcmd,
+        .buff_addr = uaddr,
+        .buff_len = len,
+        .gaddr = gaddr,
+    };
+    int ret;
+
+    ret = s390_pv_cmd(KVM_PV_DUMP, (void *)&dmp);
+    if (ret) {
+        error_report("KVM DUMP command %ld failed", subcmd);
+    }
+    return ret;
+}
+
+int kvm_s390_dump_cpu(S390CPU *cpu, void *buff)
+{
+    struct kvm_s390_pv_dmp dmp = {
+        .subcmd = KVM_PV_DUMP_CPU,
+        .buff_addr = (uint64_t)buff,
+        .gaddr = 0,
+        .buff_len = info_dump.dump_cpu_buffer_len,
+    };
+    struct kvm_pv_cmd pv = {
+        .cmd = KVM_PV_DUMP,
+        .data = (uint64_t)&dmp,
+    };
+
+    return kvm_vcpu_ioctl(CPU(cpu), KVM_S390_PV_CPU_COMMAND, &pv);
+}
+

hw/s390x/pv.c bool kvm_s390_pv_info_basic_valid(void) · line 221

Quoted 10 lines
hw/s390x/pv.c +10 -0
+int kvm_s390_dump_init(void)
+{
+    return s390_pv_dump_cmd(KVM_PV_DUMP_INIT, 0, 0, 0);
+}
+
+int kvm_s390_dump_mem(uint64_t gaddr, size_t len, void *dest)
+{
+    return s390_pv_dump_cmd(KVM_PV_DUMP_CONFIG_STATE, (uint64_t)dest,
+                            gaddr, len);
+}

Wouldn't be kvm_s390_dump_mem_state() a more precise name? Or kvm_s390_dump_mem_meta, as the corresponding section in the dump has that name (pv_mem_meta)

The current name may lead to the conclusion that this function dumps the guests memory, which it does not.

hw/s390x/pv.c bool kvm_s390_pv_info_basic_valid(void) · line 227

Quoted 6 lines
hw/s390x/pv.c +6 -0
+
+int kvm_s390_dump_finish(void *buff)
+{
+    return s390_pv_dump_cmd(KVM_PV_DUMP_COMPLETE, (uint64_t)buff, 0,
+                            info_dump.dump_config_finalize_len);
+}

IIRC this is the only place were you call "complete-dump" "finish". In the next patch you call that function in "get_data_complete()". This is the only reference to that function.

Why not simply call it kvm_s390_dump_complete() to reduce confusion?

Quoted 30 lines
hw/s390x/pv.c +1 -0
+
  #define TYPE_S390_PV_GUEST "s390-pv-guest"
  OBJECT_DECLARE_SIMPLE_TYPE(S390PVGuest, S390_PV_GUEST)
 
include/hw/s390x/pv.h +8 -0
index 6fa55bf70e..f37021e189 100644
@@ -51,6 +51,10 @@ uint64_t kvm_s390_pv_dmp_get_size_cpu(void);
  uint64_t kvm_s390_pv_dmp_get_size_mem(void);
  uint64_t kvm_s390_pv_dmp_get_size_complete(void);
  bool kvm_s390_pv_info_basic_valid(void);
+int kvm_s390_dump_init(void);
+int kvm_s390_dump_cpu(S390CPU *cpu, void *buff);
+int kvm_s390_dump_mem(uint64_t addr, size_t len, void *dest);
+int kvm_s390_dump_finish(void *buff);
  #else /* CONFIG_KVM */
  static inline bool s390_is_pv(void) { return false; }
  static inline int s390_pv_query_info(void) { return 0; }
@@ -66,6 +70,10 @@ static inline uint64_t kvm_s390_pv_dmp_get_size_cpu(void) { return 0; }
  static inline uint64_t kvm_s390_pv_dmp_get_size_mem(void) { return 0; }
  static inline uint64_t kvm_s390_pv_dmp_get_size_complete(void) { return 0; }
  static inline bool kvm_s390_pv_info_basic_valid(void) { return false; }
+static inline int kvm_s390_dump_init(void) { return 0; }
+static inline int kvm_s390_dump_cpu(S390CPU *cpu, void *buff, size_t len) { return 0; }
+static inline int kvm_s390_dump_mem(uint64_t addr, size_t len, void *dest) { return 0; }
+static inline int kvm_s390_dump_finish(void *buff) { return 0; }
  #endif /* CONFIG_KVM */
 
  int s390_pv_kvm_init(ConfidentialGuestSupport *cgs, Error **errp);
Reviewed-by on v3 11/14
#25 lore raw

Re: [PATCH v3 11/14] s390x: Add protected dump cap

In reply to Janosch Frank
Quoted 3 lines · Janosch Frank

Add a protected dump capability for later feature checking.

Quoted 49 lines
---
  target/s390x/kvm/kvm.c       | 7 +++++++
  target/s390x/kvm/kvm_s390x.h | 1 +
  2 files changed, 8 insertions(+)
target/s390x/kvm/kvm.c +7 -0
index 7bd8db0e7b..cbd8c91424 100644
@@ -157,6 +157,7 @@ static int cap_ri;
  static int cap_hpage_1m;
  static int cap_vcpu_resets;
  static int cap_protected;
+static int cap_protected_dump;
 
  static bool mem_op_storage_key_support;
 
@@ -362,6 +363,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
      cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
      cap_vcpu_resets = kvm_check_extension(s, KVM_CAP_S390_VCPU_RESETS);
      cap_protected = kvm_check_extension(s, KVM_CAP_S390_PROTECTED);
+    cap_protected_dump = kvm_check_extension(s, KVM_CAP_S390_PROTECTED_DUMP);
 
      kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
      kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
@@ -2043,6 +2045,11 @@ int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
      return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
  }
 
+int kvm_s390_get_protected_dump(void)
+{
+    return cap_protected_dump;
+}
+
  int kvm_s390_get_ri(void)
  {
      return cap_ri;
target/s390x/kvm/kvm_s390x.h +1 -0
index 05a5e1e6f4..31a69f9ce2 100644
@@ -26,6 +26,7 @@ int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state);
  void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu);
  int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu);
  int kvm_s390_get_hpage_1m(void);
+int kvm_s390_get_protected_dump(void);
  int kvm_s390_get_ri(void);
  int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_clock);
  int kvm_s390_get_clock_ext(uint8_t *tod_high, uint64_t *tod_clock);
#31 lore raw

Re: [PATCH v3 14/14] s390x: pv: Add dump support

In reply to Janosch Frank

Hi Janosch,

looks good to me.
Have a look on my comments.

Quoted 181 lines · Janosch Frank

Sometimes dumping a guest from the outside is the only way to get the data that is needed. This can be the case if a dumping mechanism like KDUMP hasn't been configured or data needs to be fetched at a specific point. Dumping a protected guest from the outside without help from fw/hw doesn't yield sufficient data to be useful. Hence we now introduce PV dump support.

The PV dump support works by integrating the firmware into the dump process. New Ultravisor calls are used to initiate the dump process, dump cpu data, dump memory state and lastly complete the dump process. The UV calls are exposed by KVM via the new KVM_PV_DUMP command and its subcommands. The guest's data is fully encrypted and can only be decrypted by the entity that owns the customer communication key for the dumped guest. Also dumping needs to be allowed via a flag in the SE header.

On the QEMU side of things we store the PV dump data in the newly introduced architecture ELF sections (storage state and completion data) and the cpu notes (for cpu dump data).

Users can use the zgetdump tool to convert the encrypted QEMU dump to an unencrypted one.

---
  include/elf.h            |   1 +
  target/s390x/arch_dump.c | 248 ++++++++++++++++++++++++++++++++++-----
  2 files changed, 219 insertions(+), 30 deletions(-)
include/elf.h +1 -0
index 3a4bcb646a..58f76fd5b4 100644
@@ -1649,6 +1649,7 @@ typedef struct elf64_shdr {
  #define NT_TASKSTRUCT	4
  #define NT_AUXV		6
  #define NT_PRXFPREG     0x46e62b7f      /* copied from gdb5.1/include/elf/common.h */
+#define NT_S390_PV_DATA 0x30e           /* s390 protvirt cpu dump data */
  #define NT_S390_GS_CB   0x30b           /* s390 guarded storage registers */
  #define NT_S390_VXRS_HIGH 0x30a         /* s390 vector registers 16-31 */
  #define NT_S390_VXRS_LOW  0x309         /* s390 vector registers 0-15 (lower half) */
target/s390x/arch_dump.c +64 -26
index 08daf93ae1..e081aa9483 100644
@@ -16,7 +16,8 @@
  #include "s390x-internal.h"
  #include "elf.h"
  #include "sysemu/dump.h"
-
+#include "hw/s390x/pv.h"
+#include "kvm/kvm_s390x.h"
 
  struct S390xUserRegsStruct {
      uint64_t psw[2];
@@ -76,9 +77,16 @@ typedef struct noteStruct {
          uint64_t todcmp;
          uint32_t todpreg;
          uint64_t ctrs[16];
+        uint8_t dynamic[1];  /*
+                              * Would be a flexible array member, if
+                              * that was legal inside a union. Real
+                              * size comes from PV info interface.
+                              */
      } contents;
  } QEMU_PACKED Note;
 
+static bool pv_dump_initialized;
+
  static void s390x_write_elf64_prstatus(Note *note, S390CPU *cpu, int id)
  {
      int i;
@@ -177,52 +185,82 @@ static void s390x_write_elf64_prefix(Note *note, S390CPU *cpu, int id)
      note->contents.prefix = cpu_to_be32((uint32_t)(cpu->env.psa));
  }
 
+static void s390x_write_elf64_pv(Note *note, S390CPU *cpu, int id)
+{
+    note->hdr.n_type = cpu_to_be32(NT_S390_PV_DATA);
+    if (!pv_dump_initialized) {
+        return;
+    }
+    kvm_s390_dump_cpu(cpu, &note->contents.dynamic);
+}
 
  typedef struct NoteFuncDescStruct {
      int contents_size;
+    uint64_t (*note_size_func)(void); /* NULL for non-dynamic sized contents */
      void (*note_contents_func)(Note *note, S390CPU *cpu, int id);
+    bool pvonly;
  } NoteFuncDesc;
 
  static const NoteFuncDesc note_core[] = {
-    {sizeof_field(Note, contents.prstatus), s390x_write_elf64_prstatus},
-    {sizeof_field(Note, contents.fpregset), s390x_write_elf64_fpregset},
-    { 0, NULL}
+    {sizeof_field(Note, contents.prstatus), NULL, s390x_write_elf64_prstatus, false},
+    {sizeof_field(Note, contents.fpregset), NULL, s390x_write_elf64_fpregset, false},
+    { 0, NULL, NULL}
  };
 
  static const NoteFuncDesc note_linux[] = {
-    {sizeof_field(Note, contents.prefix),   s390x_write_elf64_prefix},
-    {sizeof_field(Note, contents.ctrs),     s390x_write_elf64_ctrs},
-    {sizeof_field(Note, contents.timer),    s390x_write_elf64_timer},
-    {sizeof_field(Note, contents.todcmp),   s390x_write_elf64_todcmp},
-    {sizeof_field(Note, contents.todpreg),  s390x_write_elf64_todpreg},
-    {sizeof_field(Note, contents.vregslo),  s390x_write_elf64_vregslo},
-    {sizeof_field(Note, contents.vregshi),  s390x_write_elf64_vregshi},
-    {sizeof_field(Note, contents.gscb),     s390x_write_elf64_gscb},
-    { 0, NULL}
+    {sizeof_field(Note, contents.prefix),   NULL, s390x_write_elf64_prefix,  false},
+    {sizeof_field(Note, contents.ctrs),     NULL, s390x_write_elf64_ctrs,    false},
+    {sizeof_field(Note, contents.timer),    NULL, s390x_write_elf64_timer,   false},
+    {sizeof_field(Note, contents.todcmp),   NULL, s390x_write_elf64_todcmp,  false},
+    {sizeof_field(Note, contents.todpreg),  NULL, s390x_write_elf64_todpreg, false},
+    {sizeof_field(Note, contents.vregslo),  NULL, s390x_write_elf64_vregslo, false},
+    {sizeof_field(Note, contents.vregshi),  NULL, s390x_write_elf64_vregshi, false},
+    {sizeof_field(Note, contents.gscb),     NULL, s390x_write_elf64_gscb,    false},
+    {0, kvm_s390_pv_dmp_get_size_cpu,       s390x_write_elf64_pv, true},
+    { 0, NULL, NULL}
  };
 
  static int s390x_write_elf64_notes(const char *note_name,
-                                       WriteCoreDumpFunction f,
-                                       S390CPU *cpu, int id,
-                                       void *opaque,
-                                       const NoteFuncDesc *funcs)
+                                   WriteCoreDumpFunction f,
+                                   S390CPU *cpu, int id,
+                                   void *opaque,
+                                   const NoteFuncDesc *funcs)
  {
-    Note note;
+    Note note, *notep;
      const NoteFuncDesc *nf;
-    int note_size;
+    int note_size, content_size;
      int ret = -1;
 
      assert(strlen(note_name) < sizeof(note.name));
 
      for (nf = funcs; nf->note_contents_func; nf++) {
-        memset(&note, 0, sizeof(note));
-        note.hdr.n_namesz = cpu_to_be32(strlen(note_name) + 1);
-        note.hdr.n_descsz = cpu_to_be32(nf->contents_size);
-        g_strlcpy(note.name, note_name, sizeof(note.name));
-        (*nf->note_contents_func)(&note, cpu, id);
+        notep = &note;
+        if (nf->pvonly && !s390_is_pv()) {
+            continue;
+        }
 
-        note_size = sizeof(note) - sizeof(note.contents) + nf->contents_size;
-        ret = f(&note, note_size, opaque);
+        content_size = nf->contents_size ? nf->contents_size : nf->note_size_func();
+        note_size = sizeof(note) - sizeof(notep->contents) + content_size;
+
+        /* Notes with dynamic sizes need to allocate a note */
+        if (nf->note_size_func) {
+            notep = g_malloc0(note_size);
+        }
+
+        memset(notep, 0, sizeof(note));
+
+        /* Setup note header data */
+        notep->hdr.n_descsz = cpu_to_be32(content_size);
+        notep->hdr.n_namesz = cpu_to_be32(strlen(note_name) + 1);
+        g_strlcpy(notep->name, note_name, sizeof(notep->name));
+
+        /* Get contents and write them out */
+        (*nf->note_contents_func)(notep, cpu, id);
+        ret = f(notep, note_size, opaque);
+
+        if (nf->note_size_func) {
+            g_free(notep);
+        }
 
          if (ret < 0) {
              return -1;

target/s390x/arch_dump.c int s390_cpu_write_elf64_note(WriteCoreDumpFunction f, CPUState *cs, · line 292

Quoted 8 lines
target/s390x/arch_dump.c +5 -0
@@ -247,12 +285,159 @@ int s390_cpu_write_elf64_note(WriteCoreDumpFunction f, CPUState *cs,
      return s390x_write_elf64_notes("LINUX", f, cpu, cpuid, opaque, note_linux);
  }
 
+/* PV dump section size functions */
+static uint64_t get_dump_mem_size_from_len(uint64_t len)
+{
+    return (len / (1 << 20)) * kvm_s390_pv_dmp_get_size_mem();
+}

similar to my comment of the last patch rearding memory state. IIRC you are not calculating the size of the memory dump but the size of the memory_state data.

target/s390x/arch_dump.c int s390_cpu_write_elf64_note(WriteCoreDumpFunction f, CPUState *cs, · line 297

Quoted 5 lines
target/s390x/arch_dump.c +5 -0
+
+static uint64_t get_size_mem(DumpState *s)
+{
+    return get_dump_mem_size_from_len(s->total_size);
+}

get_size_mem_state ?

Quoted 12 lines
target/s390x/arch_dump.c +12 -0
+
+static uint64_t get_size_complete(DumpState *s)
+{
+    return kvm_s390_pv_dmp_get_size_complete();
+}
+
+/* PV dump section data functions*/
+static int get_data_complete(DumpState *s, uint8_t *buff)
+{
+    int rc;
+
+    if (!pv_dump_initialized) {

target/s390x/arch_dump.c int s390_cpu_write_elf64_note(WriteCoreDumpFunction f, CPUState *cs, · line 319

Quoted 10 lines
target/s390x/arch_dump.c +10 -0
+        return 0;
+    }
+    rc = kvm_s390_dump_finish(buff);
+    if (!rc) {
+            pv_dump_initialized = false;
+    }
+    return rc;
+}
+
+static int dump_mem(DumpState *s, uint64_t gaddr, uint8_t *buff, uint64_t buff_len)

dump_mem_state ?

target/s390x/arch_dump.c int s390_cpu_write_elf64_note(WriteCoreDumpFunction f, CPUState *cs, · line 328

Quoted 8 lines
target/s390x/arch_dump.c +8 -0
+{
+    /* We need the gaddr + len and something to write to */
+    if (!pv_dump_initialized) {
+        return 0;
+    }
+    return kvm_s390_dump_mem(gaddr, buff_len, buff);
+} > +
+static int get_data_mem(DumpState *s, uint8_t *buff)

get_data_mem_state ?

Quoted 143 lines
target/s390x/arch_dump.c +113 -4
+{
+    int64_t memblock_size, memblock_start;
+    GuestPhysBlock *block;
+    uint64_t off;
+
+    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
+        memblock_start = dump_get_memblock_start(block, s->begin, s->length);
+        if (memblock_start == -1) {
+            continue;
+        }
+
+        memblock_size = dump_get_memblock_size(block, s->begin, s->length);
+
+        off = get_dump_mem_size_from_len(block->target_start);
+        dump_mem(s, block->target_start, buff + off,
+                 get_dump_mem_size_from_len(memblock_size));
+    }
+
+    return 0;
+}
+
+struct sections {
+    uint64_t (*sections_size_func)(DumpState *s);
+    int (*sections_contents_func)(DumpState *s, uint8_t *buff);
+    char sctn_str[12];
+} sections[] = {
+    { get_size_mem, get_data_mem, "pv_mem_meta"},
+    { get_size_complete, get_data_complete, "pv_compl"},
+    {NULL , NULL, ""}
+};
+
+static uint64_t arch_sections_write_hdr(void *opaque, uint8_t *buff)
+{
+    DumpState *s = opaque;
+    Elf64_Shdr *shdr = (void *)buff;
+    struct sections *sctn = sections;
+    uint64_t off = s->section_offset;
+
+    if (!s390_is_pv()) {
+        return 0;
+    }
+
+    for (; sctn->sections_size_func; off += shdr->sh_size, sctn++, shdr++) {
+        memset(shdr, 0, sizeof(*shdr));
+        shdr->sh_type = SHT_PROGBITS;
+        shdr->sh_offset = off;
+        shdr->sh_size = sctn->sections_size_func(s);
+        shdr->sh_name = s->string_table_buf->len;
+        g_array_append_vals(s->string_table_buf, sctn->sctn_str, sizeof(sctn->sctn_str));
+    }
+
+    return (uintptr_t)shdr - (uintptr_t)buff;
+}
+
+
+/* Add arch specific number of sections and their respective sizes */
+static void arch_sections_add(void *opaque)
+{
+    DumpState *s = opaque;
+    struct sections *sctn = sections;
+
+    /*
+     * We only do a PV dump if we are running a PV guest, KVM supports
+     * the dump API and we got valid dump length information.
+     */
+    if (!s390_is_pv() || !kvm_s390_get_protected_dump() ||
+        !kvm_s390_pv_info_basic_valid()) {
+        return;
+    }
+
+    /*
+     * Start the UV dump process by doing the initialize dump call via
+     * KVM as the proxy.
+     */
+    if (!kvm_s390_dump_init()) {
+            pv_dump_initialized = true;
+    }
+
+    for (; sctn->sections_size_func; sctn++) {
+        s->shdr_num += 1;
+        s->elf_section_data_size += sctn->sections_size_func(s);
+    }
+}
+
+/*
+ * After the PV dump has been initialized, the CPU data has been
+ * fetched and memory has been dumped, we need to grab the tweak data
+ * and the completion data.
+ */
+static void arch_sections_write(void *opaque, uint8_t *buff)
+{
+    DumpState *s = opaque;
+    struct sections *sctn = sections;
+
+    /* shdr_num should only have been set > 1 if we are protected */
+    assert(s390_is_pv());
+
+    for (; sctn->sections_size_func; sctn++) {
+        sctn->sections_contents_func(s, buff);
+        buff += sctn->sections_size_func(s);
+    }
+}
+
  int cpu_get_dump_info(ArchDumpInfo *info,
                        const struct GuestPhysBlockList *guest_phys_blocks)
  {
      info->d_machine = EM_S390;
      info->d_endian = ELFDATA2MSB;
      info->d_class = ELFCLASS64;
+    info->arch_sections_add_fn = *arch_sections_add;
+    info->arch_sections_write_hdr_fn = *arch_sections_write_hdr;
+    info->arch_sections_write_fn = *arch_sections_write;
 
      return 0;
  }
@@ -261,7 +446,7 @@ ssize_t cpu_get_note_size(int class, int machine, int nr_cpus)
  {
      int name_size = 8; /* "LINUX" or "CORE" + pad */
      size_t elf_note_size = 0;
-    int note_head_size;
+    int note_head_size, content_size;
      const NoteFuncDesc *nf;
 
      assert(class == ELFCLASS64);
@@ -270,12 +455,15 @@ ssize_t cpu_get_note_size(int class, int machine, int nr_cpus)
      note_head_size = sizeof(Elf64_Nhdr);
 
      for (nf = note_core; nf->note_contents_func; nf++) {
-        elf_note_size = elf_note_size + note_head_size + name_size +
-                        nf->contents_size;
+        elf_note_size = elf_note_size + note_head_size + name_size + nf->contents_size;
      }
      for (nf = note_linux; nf->note_contents_func; nf++) {
+        if (nf->pvonly && !s390_is_pv()) {
+            continue;
+        }
+        content_size = nf->contents_size ? nf->contents_size : nf->note_size_func();
          elf_note_size = elf_note_size + note_head_size + name_size +
-                        nf->contents_size;
+                        content_size;
      }
 
      return (elf_note_size) * nr_cpus;