[PATCH v11 0/4] fw_cfg: add DMA operations & etc/vmcoreinfo support

[PATCH v11 0/4] fw_cfg: add DMA operations & etc/vmcoreinfo support

[Marc-André Lureau]

Hi,

This series adds DMA operations support to the qemu fw_cfg kernel
module and populates "etc/vmcoreinfo" with vmcoreinfo location
details (since qemu 2.11).

v11:
- add #include <linux/crash_core.h> in last patch,
  fixing kbuild .config test

Marc-André Lureau (4):
  fw_cfg: add DMA register
  fw_cfg: do DMA read operation
  crash: export paddr_vmcoreinfo_note()
  fw_cfg: write vmcoreinfo details

 drivers/firmware/qemu_fw_cfg.c | 268 ++++++++++++++++++++++++++++++++++++-----
 kernel/crash_core.c            |   1 +
 2 files changed, 237 insertions(+), 32 deletions(-)

-- 
2.16.0.rc1.1.gef27df75a1

[PATCH v11 1/4] fw_cfg: add DMA register

[Marc-André Lureau]

Add an optional <dma_off> kernel module (or command line) parameter
using the following syntax:

      [qemu_fw_cfg.]ioport=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]
 or
      [qemu_fw_cfg.]mmio=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]

and initializes the register address using given or default offset.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Gabriel Somlo <somlo@cmu.edu>
---
 drivers/firmware/qemu_fw_cfg.c | 53 ++++++++++++++++++++++++++++++++----------
 1 file changed, 41 insertions(+), 12 deletions(-)

diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
index deb483064f53..740df0df2260 100644
--- a/drivers/firmware/qemu_fw_cfg.c
+++ b/drivers/firmware/qemu_fw_cfg.c
@@ -10,20 +10,21 @@
  * and select subsets of aarch64), a Device Tree node (on arm), or using
  * a kernel module (or command line) parameter with the following syntax:
  *
- *      [qemu_fw_cfg.]ioport=<size>@<base>[:<ctrl_off>:<data_off>]
+ *      [qemu_fw_cfg.]ioport=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]
  * or
- *      [qemu_fw_cfg.]mmio=<size>@<base>[:<ctrl_off>:<data_off>]
+ *      [qemu_fw_cfg.]mmio=<size>@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]
  *
  * where:
  *      <size>     := size of ioport or mmio range
  *      <base>     := physical base address of ioport or mmio range
  *      <ctrl_off> := (optional) offset of control register
  *      <data_off> := (optional) offset of data register
+ *      <dma_off> := (optional) offset of dma register
  *
  * e.g.:
- *      qemu_fw_cfg.ioport=2@0x510:0:1		(the default on x86)
+ *      qemu_fw_cfg.ioport=12@0x510:0:1:4	(the default on x86)
  * or
- *      qemu_fw_cfg.mmio=0xA@0x9020000:8:0	(the default on arm)
+ *      qemu_fw_cfg.mmio=16@0x9020000:8:0:16	(the default on arm)
  */
 
 #include <linux/module.h>
@@ -63,6 +64,7 @@ static resource_size_t fw_cfg_p_size;
 static void __iomem *fw_cfg_dev_base;
 static void __iomem *fw_cfg_reg_ctrl;
 static void __iomem *fw_cfg_reg_data;
+static void __iomem *fw_cfg_reg_dma;
 
 /* atomic access to fw_cfg device (potentially slow i/o, so using mutex) */
 static DEFINE_MUTEX(fw_cfg_dev_lock);
@@ -118,12 +120,14 @@ static void fw_cfg_io_cleanup(void)
 # if (defined(CONFIG_ARM) || defined(CONFIG_ARM64))
 #  define FW_CFG_CTRL_OFF 0x08
 #  define FW_CFG_DATA_OFF 0x00
+#  define FW_CFG_DMA_OFF 0x10
 # elif (defined(CONFIG_PPC_PMAC) || defined(CONFIG_SPARC32)) /* ppc/mac,sun4m */
 #  define FW_CFG_CTRL_OFF 0x00
 #  define FW_CFG_DATA_OFF 0x02
 # elif (defined(CONFIG_X86) || defined(CONFIG_SPARC64)) /* x86, sun4u */
 #  define FW_CFG_CTRL_OFF 0x00
 #  define FW_CFG_DATA_OFF 0x01
+#  define FW_CFG_DMA_OFF 0x04
 # else
 #  error "QEMU FW_CFG not available on this architecture!"
 # endif
@@ -133,7 +137,7 @@ static void fw_cfg_io_cleanup(void)
 static int fw_cfg_do_platform_probe(struct platform_device *pdev)
 {
 	char sig[FW_CFG_SIG_SIZE];
-	struct resource *range, *ctrl, *data;
+	struct resource *range, *ctrl, *data, *dma;
 
 	/* acquire i/o range details */
 	fw_cfg_is_mmio = false;
@@ -170,6 +174,7 @@ static int fw_cfg_do_platform_probe(struct platform_device *pdev)
 	/* were custom register offsets provided (e.g. on the command line)? */
 	ctrl = platform_get_resource_byname(pdev, IORESOURCE_REG, "ctrl");
 	data = platform_get_resource_byname(pdev, IORESOURCE_REG, "data");
+	dma = platform_get_resource_byname(pdev, IORESOURCE_REG, "dma");
 	if (ctrl && data) {
 		fw_cfg_reg_ctrl = fw_cfg_dev_base + ctrl->start;
 		fw_cfg_reg_data = fw_cfg_dev_base + data->start;
@@ -179,6 +184,13 @@ static int fw_cfg_do_platform_probe(struct platform_device *pdev)
 		fw_cfg_reg_data = fw_cfg_dev_base + FW_CFG_DATA_OFF;
 	}
 
+	if (dma)
+		fw_cfg_reg_dma = fw_cfg_dev_base + dma->start;
+#ifdef FW_CFG_DMA_OFF
+	else
+		fw_cfg_reg_dma = fw_cfg_dev_base + FW_CFG_DMA_OFF;
+#endif
+
 	/* verify fw_cfg device signature */
 	fw_cfg_read_blob(FW_CFG_SIGNATURE, sig, 0, FW_CFG_SIG_SIZE);
 	if (memcmp(sig, "QEMU", FW_CFG_SIG_SIZE) != 0) {
@@ -629,6 +641,7 @@ static struct platform_device *fw_cfg_cmdline_dev;
 /* use special scanf/printf modifier for phys_addr_t, resource_size_t */
 #define PH_ADDR_SCAN_FMT "@%" __PHYS_ADDR_PREFIX "i%n" \
 			 ":%" __PHYS_ADDR_PREFIX "i" \
+			 ":%" __PHYS_ADDR_PREFIX "i%n" \
 			 ":%" __PHYS_ADDR_PREFIX "i%n"
 
 #define PH_ADDR_PR_1_FMT "0x%" __PHYS_ADDR_PREFIX "x@" \
@@ -638,12 +651,15 @@ static struct platform_device *fw_cfg_cmdline_dev;
 			 ":%" __PHYS_ADDR_PREFIX "u" \
 			 ":%" __PHYS_ADDR_PREFIX "u"
 
+#define PH_ADDR_PR_4_FMT PH_ADDR_PR_3_FMT \
+			 ":%" __PHYS_ADDR_PREFIX "u"
+
 static int fw_cfg_cmdline_set(const char *arg, const struct kernel_param *kp)
 {
-	struct resource res[3] = {};
+	struct resource res[4] = {};
 	char *str;
 	phys_addr_t base;
-	resource_size_t size, ctrl_off, data_off;
+	resource_size_t size, ctrl_off, data_off, dma_off;
 	int processed, consumed = 0;
 
 	/* only one fw_cfg device can exist system-wide, so if one
@@ -659,19 +675,20 @@ static int fw_cfg_cmdline_set(const char *arg, const struct kernel_param *kp)
 	/* consume "<size>" portion of command line argument */
 	size = memparse(arg, &str);
 
-	/* get "@<base>[:<ctrl_off>:<data_off>]" chunks */
+	/* get "@<base>[:<ctrl_off>:<data_off>[:<dma_off>]]" chunks */
 	processed = sscanf(str, PH_ADDR_SCAN_FMT,
 			   &base, &consumed,
-			   &ctrl_off, &data_off, &consumed);
+			   &ctrl_off, &data_off, &consumed,
+			   &dma_off, &consumed);
 
-	/* sscanf() must process precisely 1 or 3 chunks:
+	/* sscanf() must process precisely 1, 3 or 4 chunks:
 	 * <base> is mandatory, optionally followed by <ctrl_off>
-	 * and <data_off>;
+	 * and <data_off>, and <dma_off>;
 	 * there must be no extra characters after the last chunk,
 	 * so str[consumed] must be '\0'.
 	 */
 	if (str[consumed] ||
-	    (processed != 1 && processed != 3))
+	    (processed != 1 && processed != 3 && processed != 4))
 		return -EINVAL;
 
 	res[0].start = base;
@@ -688,6 +705,11 @@ static int fw_cfg_cmdline_set(const char *arg, const struct kernel_param *kp)
 		res[2].start = data_off;
 		res[2].flags = IORESOURCE_REG;
 	}
+	if (processed > 3) {
+		res[3].name = "dma";
+		res[3].start = dma_off;
+		res[3].flags = IORESOURCE_REG;
+	}
 
 	/* "processed" happens to nicely match the number of resources
 	 * we need to pass in to this platform device.
@@ -722,6 +744,13 @@ static int fw_cfg_cmdline_get(char *buf, const struct kernel_param *kp)
 				fw_cfg_cmdline_dev->resource[0].start,
 				fw_cfg_cmdline_dev->resource[1].start,
 				fw_cfg_cmdline_dev->resource[2].start);
+	case 4:
+		return snprintf(buf, PAGE_SIZE, PH_ADDR_PR_4_FMT,
+				resource_size(&fw_cfg_cmdline_dev->resource[0]),
+				fw_cfg_cmdline_dev->resource[0].start,
+				fw_cfg_cmdline_dev->resource[1].start,
+				fw_cfg_cmdline_dev->resource[2].start,
+				fw_cfg_cmdline_dev->resource[3].start);
 	}
 
 	/* Should never get here */
-- 
2.16.0.rc1.1.gef27df75a1

[PATCH v11 2/4] fw_cfg: do DMA read operation

[Marc-André Lureau]

Modify fw_cfg_read_blob() to use DMA if the device supports it.
Return errors, because the operation may fail.

The DMA operation is expected to run synchronously with today qemu,
but the specification states that it may become async, so we run
"control" field check in a loop for eventual changes.

We may want to switch all the *buf addresses to use only kmalloc'ed
buffers (instead of using stack/image addresses with dma=false).

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
---
 drivers/firmware/qemu_fw_cfg.c | 131 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 111 insertions(+), 20 deletions(-)

diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
index 740df0df2260..686f0e839858 100644
--- a/drivers/firmware/qemu_fw_cfg.c
+++ b/drivers/firmware/qemu_fw_cfg.c
@@ -33,6 +33,7 @@
 #include <linux/slab.h>
 #include <linux/io.h>
 #include <linux/ioport.h>
+#include <linux/delay.h>
 
 MODULE_AUTHOR("Gabriel L. Somlo <somlo@cmu.edu>");
 MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
@@ -43,12 +44,22 @@ MODULE_LICENSE("GPL");
 #define FW_CFG_ID         0x01
 #define FW_CFG_FILE_DIR   0x19
 
+#define FW_CFG_VERSION_DMA     0x02
+#define FW_CFG_DMA_CTL_ERROR   0x01
+#define FW_CFG_DMA_CTL_READ    0x02
+#define FW_CFG_DMA_CTL_SKIP    0x04
+#define FW_CFG_DMA_CTL_SELECT  0x08
+#define FW_CFG_DMA_CTL_WRITE   0x10
+
 /* size in bytes of fw_cfg signature */
 #define FW_CFG_SIG_SIZE 4
 
 /* fw_cfg "file name" is up to 56 characters (including terminating nul) */
 #define FW_CFG_MAX_FILE_PATH 56
 
+/* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level dir. */
+static u32 fw_cfg_rev;
+
 /* fw_cfg file directory entry type */
 struct fw_cfg_file {
 	u32 size;
@@ -57,6 +68,12 @@ struct fw_cfg_file {
 	char name[FW_CFG_MAX_FILE_PATH];
 };
 
+struct fw_cfg_dma {
+	u32 control;
+	u32 length;
+	u64 address;
+} __packed;
+
 /* fw_cfg device i/o register addresses */
 static bool fw_cfg_is_mmio;
 static phys_addr_t fw_cfg_p_base;
@@ -75,12 +92,68 @@ static inline u16 fw_cfg_sel_endianness(u16 key)
 	return fw_cfg_is_mmio ? cpu_to_be16(key) : cpu_to_le16(key);
 }
 
+static inline bool fw_cfg_dma_enabled(void)
+{
+	return fw_cfg_rev & FW_CFG_VERSION_DMA && fw_cfg_reg_dma;
+}
+
+/* qemu fw_cfg device is sync today, but spec says it may become async */
+static void fw_cfg_wait_for_control(struct fw_cfg_dma *d)
+{
+	do {
+		u32 ctrl = be32_to_cpu(READ_ONCE(d->control));
+
+		if ((ctrl & ~FW_CFG_DMA_CTL_ERROR) == 0)
+			return;
+
+		usleep_range(50, 100);
+	} while (true);
+}
+
+static ssize_t fw_cfg_dma_transfer(struct device *dev,
+				void *address, u32 length, u32 control)
+{
+	phys_addr_t dma;
+	struct fw_cfg_dma *d = NULL;
+	ssize_t ret = length;
+
+	d = kmalloc(sizeof(*d), GFP_KERNEL);
+	if (!d) {
+		ret = -ENOMEM;
+		goto end;
+	}
+
+	*d = (struct fw_cfg_dma) {
+		.address = cpu_to_be64(virt_to_phys(address)),
+		.length = cpu_to_be32(length),
+		.control = cpu_to_be32(control)
+	};
+
+	dma = virt_to_phys(d);
+
+	iowrite32be((u64)dma >> 32, fw_cfg_reg_dma);
+	iowrite32be(dma, fw_cfg_reg_dma + 4);
+
+	fw_cfg_wait_for_control(d);
+
+	if (be32_to_cpu(READ_ONCE(d->control)) & FW_CFG_DMA_CTL_ERROR) {
+		ret = -EIO;
+	}
+
+end:
+	kfree(d);
+
+	return ret;
+}
+
 /* read chunk of given fw_cfg blob (caller responsible for sanity-check) */
-static inline void fw_cfg_read_blob(u16 key,
-				    void *buf, loff_t pos, size_t count)
+static ssize_t fw_cfg_read_blob(struct device *dev, u16 key,
+				void *buf, loff_t pos, size_t count,
+				bool dma)
 {
 	u32 glk = -1U;
 	acpi_status status;
+	ssize_t ret = count;
 
 	/* If we have ACPI, ensure mutual exclusion against any potential
 	 * device access by the firmware, e.g. via AML methods:
@@ -90,17 +163,36 @@ static inline void fw_cfg_read_blob(u16 key,
 		/* Should never get here */
 		WARN(1, "fw_cfg_read_blob: Failed to lock ACPI!\n");
 		memset(buf, 0, count);
-		return;
+		return -EINVAL;
 	}
 
 	mutex_lock(&fw_cfg_dev_lock);
-	iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl);
-	while (pos-- > 0)
-		ioread8(fw_cfg_reg_data);
-	ioread8_rep(fw_cfg_reg_data, buf, count);
+	if (dma && fw_cfg_dma_enabled()) {
+		if (pos == 0) {
+			ret = fw_cfg_dma_transfer(dev, buf, count, key << 16
+						  | FW_CFG_DMA_CTL_SELECT
+						  | FW_CFG_DMA_CTL_READ);
+		} else {
+			iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl);
+			ret = fw_cfg_dma_transfer(dev, NULL, pos, FW_CFG_DMA_CTL_SKIP);
+			if (ret < 0)
+				goto end;
+			ret = fw_cfg_dma_transfer(dev, buf, count,
+						  FW_CFG_DMA_CTL_READ);
+		}
+	} else {
+		iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl);
+		while (pos-- > 0)
+			ioread8(fw_cfg_reg_data);
+		ioread8_rep(fw_cfg_reg_data, buf, count);
+	}
+
+end:
 	mutex_unlock(&fw_cfg_dev_lock);
 
 	acpi_release_global_lock(glk);
+
+	return ret;
 }
 
 /* clean up fw_cfg device i/o */
@@ -192,7 +284,7 @@ static int fw_cfg_do_platform_probe(struct platform_device *pdev)
 #endif
 
 	/* verify fw_cfg device signature */
-	fw_cfg_read_blob(FW_CFG_SIGNATURE, sig, 0, FW_CFG_SIG_SIZE);
+	fw_cfg_read_blob(&pdev->dev, FW_CFG_SIGNATURE, sig, 0, FW_CFG_SIG_SIZE, false);
 	if (memcmp(sig, "QEMU", FW_CFG_SIG_SIZE) != 0) {
 		fw_cfg_io_cleanup();
 		return -ENODEV;
@@ -201,9 +293,6 @@ static int fw_cfg_do_platform_probe(struct platform_device *pdev)
 	return 0;
 }
 
-/* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level dir. */
-static u32 fw_cfg_rev;
-
 static ssize_t fw_cfg_showrev(struct kobject *k, struct attribute *a, char *buf)
 {
 	return sprintf(buf, "%u\n", fw_cfg_rev);
@@ -222,6 +311,7 @@ struct fw_cfg_sysfs_entry {
 	struct kobject kobj;
 	struct fw_cfg_file f;
 	struct list_head list;
+	struct device *dev;
 };
 
 /* get fw_cfg_sysfs_entry from kobject member */
@@ -344,6 +434,7 @@ static ssize_t fw_cfg_sysfs_read_raw(struct file *filp, struct kobject *kobj,
 				     char *buf, loff_t pos, size_t count)
 {
 	struct fw_cfg_sysfs_entry *entry = to_entry(kobj);
+	struct device *dev = entry->dev;
 
 	if (pos > entry->f.size)
 		return -EINVAL;
@@ -351,8 +442,7 @@ static ssize_t fw_cfg_sysfs_read_raw(struct file *filp, struct kobject *kobj,
 	if (count > entry->f.size - pos)
 		count = entry->f.size - pos;
 
-	fw_cfg_read_blob(entry->f.select, buf, pos, count);
-	return count;
+	return fw_cfg_read_blob(dev, entry->f.select, buf, pos, count, true);
 }
 
 static struct bin_attribute fw_cfg_sysfs_attr_raw = {
@@ -459,7 +549,7 @@ static struct kobject *fw_cfg_sel_ko;
 static struct kset *fw_cfg_fname_kset;
 
 /* register an individual fw_cfg file */
-static int fw_cfg_register_file(const struct fw_cfg_file *f)
+static int fw_cfg_register_file(struct device *dev, const struct fw_cfg_file *f)
 {
 	int err;
 	struct fw_cfg_sysfs_entry *entry;
@@ -471,6 +561,7 @@ static int fw_cfg_register_file(const struct fw_cfg_file *f)
 
 	/* set file entry information */
 	memcpy(&entry->f, f, sizeof(struct fw_cfg_file));
+	entry->dev = dev;
 
 	/* register entry under "/sys/firmware/qemu_fw_cfg/by_key/" */
 	err = kobject_init_and_add(&entry->kobj, &fw_cfg_sysfs_entry_ktype,
@@ -498,14 +589,14 @@ static int fw_cfg_register_file(const struct fw_cfg_file *f)
 }
 
 /* iterate over all fw_cfg directory entries, registering each one */
-static int fw_cfg_register_dir_entries(void)
+static int fw_cfg_register_dir_entries(struct device *dev)
 {
 	int ret = 0;
 	u32 count, i;
 	struct fw_cfg_file *dir;
 	size_t dir_size;
 
-	fw_cfg_read_blob(FW_CFG_FILE_DIR, &count, 0, sizeof(count));
+	fw_cfg_read_blob(dev, FW_CFG_FILE_DIR, &count, 0, sizeof(count), false);
 	count = be32_to_cpu(count);
 	dir_size = count * sizeof(struct fw_cfg_file);
 
@@ -513,12 +604,12 @@ static int fw_cfg_register_dir_entries(void)
 	if (!dir)
 		return -ENOMEM;
 
-	fw_cfg_read_blob(FW_CFG_FILE_DIR, dir, sizeof(count), dir_size);
+	fw_cfg_read_blob(dev, FW_CFG_FILE_DIR, dir, sizeof(count), dir_size, true);
 
 	for (i = 0; i < count; i++) {
 		dir[i].size = be32_to_cpu(dir[i].size);
 		dir[i].select = be16_to_cpu(dir[i].select);
-		ret = fw_cfg_register_file(&dir[i]);
+		ret = fw_cfg_register_file(dev, &dir[i]);
 		if (ret)
 			break;
 	}
@@ -562,14 +653,14 @@ static int fw_cfg_sysfs_probe(struct platform_device *pdev)
 		goto err_probe;
 
 	/* get revision number, add matching top-level attribute */
-	fw_cfg_read_blob(FW_CFG_ID, &fw_cfg_rev, 0, sizeof(fw_cfg_rev));
+	fw_cfg_read_blob(&pdev->dev, FW_CFG_ID, &fw_cfg_rev, 0, sizeof(fw_cfg_rev), false);
 	fw_cfg_rev = le32_to_cpu(fw_cfg_rev);
 	err = sysfs_create_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr);
 	if (err)
 		goto err_rev;
 
 	/* process fw_cfg file directory entry, registering each file */
-	err = fw_cfg_register_dir_entries();
+	err = fw_cfg_register_dir_entries(&pdev->dev);
 	if (err)
 		goto err_dir;
 
-- 
2.16.0.rc1.1.gef27df75a1

[PATCH v11 3/4] crash: export paddr_vmcoreinfo_note()

[Marc-André Lureau]

The following patch is going to use the symbol from the fw_cfg module,
to call the function and write the note location details in the
vmcoreinfo entry, so qemu can produce dumps with the vmcoreinfo note.

CC: Andrew Morton <akpm@linux-foundation.org>
CC: Baoquan He <bhe@redhat.com>
CC: Dave Young <dyoung@redhat.com>
CC: Dave Young <dyoung@redhat.com>
CC: Hari Bathini <hbathini@linux.vnet.ibm.com>
CC: Tony Luck <tony.luck@intel.com>
CC: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Acked-by: Gabriel Somlo <somlo@cmu.edu>
---
 kernel/crash_core.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/kernel/crash_core.c b/kernel/crash_core.c
index 4f63597c824d..a93590cdd9e1 100644
--- a/kernel/crash_core.c
+++ b/kernel/crash_core.c
@@ -376,6 +376,7 @@ phys_addr_t __weak paddr_vmcoreinfo_note(void)
 {
 	return __pa(vmcoreinfo_note);
 }
+EXPORT_SYMBOL(paddr_vmcoreinfo_note);
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
-- 
2.16.0.rc1.1.gef27df75a1

[PATCH v11 4/4] fw_cfg: write vmcoreinfo details

[Marc-André Lureau]

If the "etc/vmcoreinfo" fw_cfg file is present and we are not running
the kdump kernel, write the addr/size of the vmcoreinfo ELF note.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Gabriel Somlo <somlo@cmu.edu>
---
 drivers/firmware/qemu_fw_cfg.c | 84 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 84 insertions(+)

diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
index 686f0e839858..3b3cf6222c97 100644
--- a/drivers/firmware/qemu_fw_cfg.c
+++ b/drivers/firmware/qemu_fw_cfg.c
@@ -34,6 +34,8 @@
 #include <linux/io.h>
 #include <linux/ioport.h>
 #include <linux/delay.h>
+#include <linux/crash_dump.h>
+#include <linux/crash_core.h>
 
 MODULE_AUTHOR("Gabriel L. Somlo <somlo@cmu.edu>");
 MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
@@ -57,6 +59,8 @@ MODULE_LICENSE("GPL");
 /* fw_cfg "file name" is up to 56 characters (including terminating nul) */
 #define FW_CFG_MAX_FILE_PATH 56
 
+#define VMCOREINFO_FORMAT_ELF 0x1
+
 /* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level dir. */
 static u32 fw_cfg_rev;
 
@@ -195,6 +199,47 @@ static ssize_t fw_cfg_read_blob(struct device *dev, u16 key,
 	return ret;
 }
 
+#ifdef CONFIG_CRASH_CORE
+/* write chunk of given fw_cfg blob (caller responsible for sanity-check) */
+static ssize_t fw_cfg_write_blob(struct device *dev, u16 key,
+				 void *buf, loff_t pos, size_t count)
+{
+	u32 glk = -1U;
+	acpi_status status;
+	ssize_t ret = count;
+
+	/* If we have ACPI, ensure mutual exclusion against any potential
+	 * device access by the firmware, e.g. via AML methods:
+	 */
+	status = acpi_acquire_global_lock(ACPI_WAIT_FOREVER, &glk);
+	if (ACPI_FAILURE(status) && status != AE_NOT_CONFIGURED) {
+		/* Should never get here */
+		WARN(1, "%s: Failed to lock ACPI!\n", __func__);
+		return -EINVAL;
+	}
+
+	mutex_lock(&fw_cfg_dev_lock);
+	if (pos == 0) {
+		ret = fw_cfg_dma_transfer(dev, buf, count, key << 16
+					  | FW_CFG_DMA_CTL_SELECT
+					  | FW_CFG_DMA_CTL_WRITE);
+	} else {
+		iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl);
+		ret = fw_cfg_dma_transfer(dev, NULL, pos, FW_CFG_DMA_CTL_SKIP);
+		if (ret < 0)
+			goto end;
+		ret = fw_cfg_dma_transfer(dev, buf, count, FW_CFG_DMA_CTL_WRITE);
+	}
+
+end:
+	mutex_unlock(&fw_cfg_dev_lock);
+
+	acpi_release_global_lock(glk);
+
+	return ret;
+}
+#endif /* CONFIG_CRASH_CORE */
+
 /* clean up fw_cfg device i/o */
 static void fw_cfg_io_cleanup(void)
 {
@@ -314,6 +359,37 @@ struct fw_cfg_sysfs_entry {
 	struct device *dev;
 };
 
+#ifdef CONFIG_CRASH_CORE
+static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
+{
+	struct vmci {
+		__le16 host_format;
+		__le16 guest_format;
+		__le32 size;
+		__le64 paddr;
+	} __packed;
+	static struct vmci *data;
+	ssize_t ret;
+
+	data = kmalloc(sizeof(struct vmci), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	*data = (struct vmci) {
+		.guest_format = cpu_to_le16(VMCOREINFO_FORMAT_ELF),
+		.size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
+		.paddr = cpu_to_le64(paddr_vmcoreinfo_note())
+	};
+	/* spare ourself reading host format support for now since we
+	 * don't know what else to format - host may ignore ours
+	 */
+	ret = fw_cfg_write_blob(dev, f->select, data, 0, sizeof(struct vmci));
+
+	kfree(data);
+	return ret;
+}
+#endif /* CONFIG_CRASH_CORE */
+
 /* get fw_cfg_sysfs_entry from kobject member */
 static inline struct fw_cfg_sysfs_entry *to_entry(struct kobject *kobj)
 {
@@ -554,6 +630,14 @@ static int fw_cfg_register_file(struct device *dev, const struct fw_cfg_file *f)
 	int err;
 	struct fw_cfg_sysfs_entry *entry;
 
+#ifdef CONFIG_CRASH_CORE
+	if (fw_cfg_dma_enabled() &&
+		strcmp(f->name, "etc/vmcoreinfo") == 0 && !is_kdump_kernel()) {
+		if (write_vmcoreinfo(dev, f) < 0)
+			pr_warn("fw_cfg: failed to write vmcoreinfo");
+	}
+#endif
+
 	/* allocate new entry */
 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 	if (!entry)
-- 
2.16.0.rc1.1.gef27df75a1

Re: [PATCH v11 4/4] fw_cfg: write vmcoreinfo details

[Michael S. Tsirkin]

On Thu, Feb 01, 2018 at 02:03:00PM +0100, Marc-André Lureau wrote:
> @@ -314,6 +359,37 @@ struct fw_cfg_sysfs_entry {
>  	struct device *dev;
>  };
>  
> +#ifdef CONFIG_CRASH_CORE
> +static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
> +{
> +	struct vmci {
> +		__le16 host_format;
> +		__le16 guest_format;
> +		__le32 size;
> +		__le64 paddr;
> +	} __packed;
> +	static struct vmci *data;
> +	ssize_t ret;
> +
> +	data = kmalloc(sizeof(struct vmci), GFP_KERNEL);
> +	if (!data)
> +		return -ENOMEM;
> +
> +	*data = (struct vmci) {
> +		.guest_format = cpu_to_le16(VMCOREINFO_FORMAT_ELF),
> +		.size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
> +		.paddr = cpu_to_le64(paddr_vmcoreinfo_note())
> +	};
> +	/* spare ourself reading host format support for now since we
> +	 * don't know what else to format - host may ignore ours
> +	 */
> +	ret = fw_cfg_write_blob(dev, f->select, data, 0, sizeof(struct vmci));
> +
> +	kfree(data);
> +	return ret;
> +}
> +#endif /* CONFIG_CRASH_CORE */
> +
>  /* get fw_cfg_sysfs_entry from kobject member */
>  static inline struct fw_cfg_sysfs_entry *to_entry(struct kobject *kobj)
>  {

kmalloc during crash is I think a bad idea.
How about preallocating on probe?

diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
index 33e0256..ffd81d1 100644
--- a/drivers/firmware/qemu_fw_cfg.c
+++ b/drivers/firmware/qemu_fw_cfg.c
@@ -363,22 +363,33 @@ struct fw_cfg_sysfs_entry {
 };
 
 #ifdef CONFIG_CRASH_CORE
+
+struct vmci {
+	__le16 host_format;
+	__le16 guest_format;
+	__le32 size;
+	__le64 paddr;
+} *fw_cfg_vmcore_data;
+
+static int fw_cfg_vmcore_init(void)
+{
+	fw_cfg_vmcore_data = kmalloc(sizeof(*fw_cfg_vmcore_data), GFP_KERNEL);
+	if (!fw_cfg_vmcore_data)
+		return -ENOMEM;
+	return 0;
+}
+
+static int fw_cfg_vmcore_cleanup(void)
+{
+	kfree(fw_cfg_vmcore_data);
+}
+
 static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
 {
-	struct vmci {
-		__le16 host_format;
-		__le16 guest_format;
-		__le32 size;
-		__le64 paddr;
-	} __packed;
 	static struct vmci *data;
 	ssize_t ret;
 
-	data = kmalloc(sizeof(struct vmci), GFP_KERNEL);
-	if (!data)
-		return -ENOMEM;
-
-	*data = (struct vmci) {
+	*fw_cfg_vmcore_data = (struct vmci) {
 		.guest_format = cpu_to_le16(VMCOREINFO_FORMAT_ELF),
 		.size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
 		.paddr = cpu_to_le64(paddr_vmcoreinfo_note())
@@ -386,11 +397,23 @@ static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
 	/* spare ourself reading host format support for now since we
 	 * don't know what else to format - host may ignore ours
 	 */
-	ret = fw_cfg_write_blob(dev, f->select, data, 0, sizeof(struct vmci));
+	ret = fw_cfg_write_blob(dev, f->select, fw_cfg_vmcore_data, 0,
+				sizeof(*fw_cfg_vmcore_data));
 
-	kfree(data);
 	return ret;
 }
+
+#else
+
+static int fw_cfg_vmcore_init(void)
+{
+	return 0;
+}
+
+static int fw_cfg_vmcore_cleanup(void)
+{
+}
+
 #endif /* CONFIG_CRASH_CORE */
 
 /* get fw_cfg_sysfs_entry from kobject member */
@@ -725,11 +748,15 @@ static int fw_cfg_sysfs_probe(struct platform_device *pdev)
 	if (fw_cfg_sel_ko)
 		return -EBUSY;
 
+	err = fw_cfg_vmcore_init();
+	if (err)
+		goto err_sel;
+
 	/* create by_key and by_name subdirs of /sys/firmware/qemu_fw_cfg/ */
 	err = -ENOMEM;
 	fw_cfg_sel_ko = kobject_create_and_add("by_key", fw_cfg_top_ko);
 	if (!fw_cfg_sel_ko)
-		goto err_sel;
+		goto err_vmcore;
 	fw_cfg_fname_kset = kset_create_and_add("by_name", NULL, fw_cfg_top_ko);
 	if (!fw_cfg_fname_kset)
 		goto err_name;
@@ -764,6 +791,8 @@ static int fw_cfg_sysfs_probe(struct platform_device *pdev)
 	fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
 err_name:
 	fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
+err_vmcore:
+	fw_cfg_vmcore_cleanup();
 err_sel:
 	return err;
 }
@@ -776,6 +805,7 @@ static int fw_cfg_sysfs_remove(struct platform_device *pdev)
 	fw_cfg_io_cleanup();
 	fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
 	fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
+	fw_cfg_vmcore_cleanup();
 	return 0;
 }
 

Re: [PATCH v11 2/4] fw_cfg: do DMA read operation

[Michael S. Tsirkin]

On Thu, Feb 01, 2018 at 02:02:58PM +0100, Marc-André Lureau wrote:
> Modify fw_cfg_read_blob() to use DMA if the device supports it.
> Return errors, because the operation may fail.
> 
> The DMA operation is expected to run synchronously with today qemu,
> but the specification states that it may become async, so we run
> "control" field check in a loop for eventual changes.
> 
> We may want to switch all the *buf addresses to use only kmalloc'ed
> buffers (instead of using stack/image addresses with dma=false).
> 
> Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

This seems to cause no end of pain. I see no reason to keep
pushing this patch until it's all much more stable.

Something like the below would disable it temporarily.


commit 05054fca8b96900f6b54ab0056458f92b616d74c
Author: Michael S. Tsirkin <mst@redhat.com>
Date:   Fri Feb 2 04:29:59 2018 +0200

    fixup! fw_cfg: do DMA read operation

diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
index d86ad92..33e0256 100644
--- a/drivers/firmware/qemu_fw_cfg.c
+++ b/drivers/firmware/qemu_fw_cfg.c
@@ -41,6 +41,9 @@ MODULE_AUTHOR("Gabriel L. Somlo <somlo@cmu.edu>");
 MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
 MODULE_LICENSE("GPL");
 
+/* enable DMA for all accesses */
+#define FW_CFG_DMA_READ_ENABLE 0
+
 /* selector key values for "well-known" fw_cfg entries */
 #define FW_CFG_SIGNATURE  0x00
 #define FW_CFG_ID         0x01
@@ -171,7 +174,7 @@ static ssize_t fw_cfg_read_blob(struct device *dev, u16 key,
 	}
 
 	mutex_lock(&fw_cfg_dev_lock);
-	if (dma && fw_cfg_dma_enabled()) {
+	if (FW_CFG_DMA_READ_ENABLE && dma && fw_cfg_dma_enabled()) {
 		if (pos == 0) {
 			ret = fw_cfg_dma_transfer(dev, buf, count, key << 16
 						  | FW_CFG_DMA_CTL_SELECT

Re: [PATCH v11 0/4] fw_cfg: add DMA operations & etc/vmcoreinfo support

[Michael S. Tsirkin]

On Thu, Feb 01, 2018 at 02:02:56PM +0100, Marc-André Lureau wrote:
> Hi,
> 
> This series adds DMA operations support to the qemu fw_cfg kernel
> module and populates "etc/vmcoreinfo" with vmcoreinfo location
> details (since qemu 2.11).

At this point you need to scope this down significantly.  I'll give the
vmcore one more chance in this release if you test the hell out of it with 32 and 64 bit
guests and repost tomorrow with just the vmcore enabled and that ASAP,
and without kmalloc on crash which is problematic by itself.

> v11:
> - add #include <linux/crash_core.h> in last patch,
>   fixing kbuild .config test
> 
> Marc-André Lureau (4):
>   fw_cfg: add DMA register
>   fw_cfg: do DMA read operation
>   crash: export paddr_vmcoreinfo_note()
>   fw_cfg: write vmcoreinfo details
> 
>  drivers/firmware/qemu_fw_cfg.c | 268 ++++++++++++++++++++++++++++++++++++-----
>  kernel/crash_core.c            |   1 +
>  2 files changed, 237 insertions(+), 32 deletions(-)
> 
> -- 
> 2.16.0.rc1.1.gef27df75a1

Re: [PATCH v11 4/4] fw_cfg: write vmcoreinfo details

[Marc-Andre Lureau]

Hi

On Fri, Feb 2, 2018 at 3:44 AM, Michael S. Tsirkin <mst@redhat.com> wrote:
> On Thu, Feb 01, 2018 at 02:03:00PM +0100, Marc-André Lureau wrote:
>> @@ -314,6 +359,37 @@ struct fw_cfg_sysfs_entry {
>>       struct device *dev;
>>  };
>>
>> +#ifdef CONFIG_CRASH_CORE
>> +static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
>> +{
>> +     struct vmci {
>> +             __le16 host_format;
>> +             __le16 guest_format;
>> +             __le32 size;
>> +             __le64 paddr;
>> +     } __packed;
>> +     static struct vmci *data;
>> +     ssize_t ret;
>> +
>> +     data = kmalloc(sizeof(struct vmci), GFP_KERNEL);
>> +     if (!data)
>> +             return -ENOMEM;
>> +
>> +     *data = (struct vmci) {
>> +             .guest_format = cpu_to_le16(VMCOREINFO_FORMAT_ELF),
>> +             .size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
>> +             .paddr = cpu_to_le64(paddr_vmcoreinfo_note())
>> +     };
>> +     /* spare ourself reading host format support for now since we
>> +      * don't know what else to format - host may ignore ours
>> +      */
>> +     ret = fw_cfg_write_blob(dev, f->select, data, 0, sizeof(struct vmci));
>> +
>> +     kfree(data);
>> +     return ret;
>> +}
>> +#endif /* CONFIG_CRASH_CORE */
>> +
>>  /* get fw_cfg_sysfs_entry from kobject member */
>>  static inline struct fw_cfg_sysfs_entry *to_entry(struct kobject *kobj)
>>  {
>
> kmalloc during crash is I think a bad idea.
> How about preallocating on probe?

It doesn't kmalloc during crash, it kmalloc during boot. As such, I
don't see a reason to change it.

>
> diff --git a/drivers/firmware/qemu_fw_cfg.c b/drivers/firmware/qemu_fw_cfg.c
> index 33e0256..ffd81d1 100644
> --- a/drivers/firmware/qemu_fw_cfg.c
> +++ b/drivers/firmware/qemu_fw_cfg.c
> @@ -363,22 +363,33 @@ struct fw_cfg_sysfs_entry {
>  };
>
>  #ifdef CONFIG_CRASH_CORE
> +
> +struct vmci {
> +       __le16 host_format;
> +       __le16 guest_format;
> +       __le32 size;
> +       __le64 paddr;
> +} *fw_cfg_vmcore_data;
> +
> +static int fw_cfg_vmcore_init(void)
> +{
> +       fw_cfg_vmcore_data = kmalloc(sizeof(*fw_cfg_vmcore_data), GFP_KERNEL);
> +       if (!fw_cfg_vmcore_data)
> +               return -ENOMEM;
> +       return 0;
> +}
> +
> +static int fw_cfg_vmcore_cleanup(void)
> +{
> +       kfree(fw_cfg_vmcore_data);
> +}
> +
>  static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
>  {
> -       struct vmci {
> -               __le16 host_format;
> -               __le16 guest_format;
> -               __le32 size;
> -               __le64 paddr;
> -       } __packed;
>         static struct vmci *data;
>         ssize_t ret;
>
> -       data = kmalloc(sizeof(struct vmci), GFP_KERNEL);
> -       if (!data)
> -               return -ENOMEM;
> -
> -       *data = (struct vmci) {
> +       *fw_cfg_vmcore_data = (struct vmci) {
>                 .guest_format = cpu_to_le16(VMCOREINFO_FORMAT_ELF),
>                 .size = cpu_to_le32(VMCOREINFO_NOTE_SIZE),
>                 .paddr = cpu_to_le64(paddr_vmcoreinfo_note())
> @@ -386,11 +397,23 @@ static ssize_t write_vmcoreinfo(struct device *dev, const struct fw_cfg_file *f)
>         /* spare ourself reading host format support for now since we
>          * don't know what else to format - host may ignore ours
>          */
> -       ret = fw_cfg_write_blob(dev, f->select, data, 0, sizeof(struct vmci));
> +       ret = fw_cfg_write_blob(dev, f->select, fw_cfg_vmcore_data, 0,
> +                               sizeof(*fw_cfg_vmcore_data));
>
> -       kfree(data);
>         return ret;
>  }
> +
> +#else
> +
> +static int fw_cfg_vmcore_init(void)
> +{
> +       return 0;
> +}
> +
> +static int fw_cfg_vmcore_cleanup(void)
> +{
> +}
> +
>  #endif /* CONFIG_CRASH_CORE */
>
>  /* get fw_cfg_sysfs_entry from kobject member */
> @@ -725,11 +748,15 @@ static int fw_cfg_sysfs_probe(struct platform_device *pdev)
>         if (fw_cfg_sel_ko)
>                 return -EBUSY;
>
> +       err = fw_cfg_vmcore_init();
> +       if (err)
> +               goto err_sel;
> +
>         /* create by_key and by_name subdirs of /sys/firmware/qemu_fw_cfg/ */
>         err = -ENOMEM;
>         fw_cfg_sel_ko = kobject_create_and_add("by_key", fw_cfg_top_ko);
>         if (!fw_cfg_sel_ko)
> -               goto err_sel;
> +               goto err_vmcore;
>         fw_cfg_fname_kset = kset_create_and_add("by_name", NULL, fw_cfg_top_ko);
>         if (!fw_cfg_fname_kset)
>                 goto err_name;
> @@ -764,6 +791,8 @@ static int fw_cfg_sysfs_probe(struct platform_device *pdev)
>         fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
>  err_name:
>         fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
> +err_vmcore:
> +       fw_cfg_vmcore_cleanup();
>  err_sel:
>         return err;
>  }
> @@ -776,6 +805,7 @@ static int fw_cfg_sysfs_remove(struct platform_device *pdev)
>         fw_cfg_io_cleanup();
>         fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset);
>         fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
> +       fw_cfg_vmcore_cleanup();
>         return 0;
>  }
>