Skip to contentSkip to conversation

pstore: Add boot loader log messages support

PATCH In review Yue Hu 2 participants 3 messages Feb 15, 2019 – May 20, 2019 Pstore filesystem

lore
Share & tools
lore

Yue Hu

author
#0 lore raw

PATCH v2 1/1 pstore: Add boot loader log messages support

From: Yue Hu <[email protected]>

Sometimes we hope to check boot loader log messages (e.g. Android Verified Boot status) when kernel is coming up. Generally it does depend on serial device, but it will be removed for the hardware shipping to market by most of manufacturers. In that case better solder and proper serial cable for different interface (e.g. Type-C or microUSB) are needed. That is inconvenient and even wasting much time on it.

Therefore, let's add a logging support: PSTORE_TYPE_XBL.

---
v2: mention info of interacting with boot loader

 fs/pstore/Kconfig      | 10 +++++++
 fs/pstore/platform.c   | 16 ++++++++++
 fs/pstore/ram.c        | 81 ++++++++++++++++++++++++++++++++++++++++++++++++--
 include/linux/pstore.h | 21 +++++++++----
 4 files changed, 121 insertions(+), 7 deletions(-)
fs/pstore/Kconfig +10 -0
index 0d19d19..ef4a2dc 100644
@@ -137,6 +137,16 @@ config PSTORE_FTRACE
 
 	  If unsure, say N.
 
+config PSTORE_XBL
+	bool "Log bootloader messages"
+	depends on PSTORE
+	help
+	  When the option is enabled, pstore will log boot loader
+	  messages to /sys/fs/pstore/xbl-ramoops-[ID] after reboot.
+	  Boot loader needs to support log buffer reserved.
+
+	  If unsure, say N.
+
 config PSTORE_RAM
 	tristate "Log panic/oops to a RAM buffer"
 	depends on PSTORE
fs/pstore/platform.c +16 -0
index 2d1066e..2e6c3f8f 100644
@@ -65,6 +65,7 @@
 	"mce",
 	"console",
 	"ftrace",
+	"xbl",
 	"rtas",
 	"powerpc-ofw",
 	"powerpc-common",
@@ -530,6 +531,19 @@ static void pstore_register_console(void) {}
 static void pstore_unregister_console(void) {}
 #endif
 
+#ifdef CONFIG_PSTORE_XBL
+static void pstore_register_xbl(void)
+{
+	struct pstore_record record;
+
+	pstore_record_init(&record, psinfo);
+	record.type = PSTORE_TYPE_XBL;
+	psinfo->write(&record);
+}
+#else
+static void pstore_register_xbl(void) {}
+#endif
+
 static int pstore_write_user_compat(struct pstore_record *record,
 				    const char __user *buf)
 {
@@ -616,6 +630,8 @@ int pstore_register(struct pstore_info *psi)
 		pstore_register_ftrace();
 	if (psi->flags & PSTORE_FLAGS_PMSG)
 		pstore_register_pmsg();
+	if (psi->flags & PSTORE_FLAGS_XBL)
+		pstore_register_xbl();
 
 	/* Start watching for new records, if desired. */
 	if (pstore_update_ms >= 0) {
fs/pstore/ram.c +79 -2

195 lines not on this page · raw message

include/linux/pstore.h +16 -5
index b146181..e016bbf 100644
@@ -43,13 +43,14 @@ enum pstore_type_id {
 	PSTORE_TYPE_MCE		= 1,
 	PSTORE_TYPE_CONSOLE	= 2,
 	PSTORE_TYPE_FTRACE	= 3,
+	PSTORE_TYPE_XBL		= 4,
 
 	/* PPC64-specific partition types */
-	PSTORE_TYPE_PPC_RTAS	= 4,
-	PSTORE_TYPE_PPC_OF	= 5,
-	PSTORE_TYPE_PPC_COMMON	= 6,
-	PSTORE_TYPE_PMSG	= 7,
-	PSTORE_TYPE_PPC_OPAL	= 8,
+	PSTORE_TYPE_PPC_RTAS	= 5,
+	PSTORE_TYPE_PPC_OF	= 6,
+	PSTORE_TYPE_PPC_COMMON	= 7,
+	PSTORE_TYPE_PMSG	= 8,
+	PSTORE_TYPE_PPC_OPAL	= 9,
 
 	/* End of the list */
 	PSTORE_TYPE_MAX
@@ -207,10 +208,20 @@ struct pstore_info {
 #define PSTORE_FLAGS_CONSOLE	BIT(1)
 #define PSTORE_FLAGS_FTRACE	BIT(2)
 #define PSTORE_FLAGS_PMSG	BIT(3)
+#define PSTORE_FLAGS_XBL	BIT(4)
 
 extern int pstore_register(struct pstore_info *);
 extern void pstore_unregister(struct pstore_info *);
 
+#define PSTORE_XBL_COOKIE	0x4c425850 /* "PXBL" in ASCII */
+
+struct pstore_xbl_header {
+	uint32_t cookie;
+	uint32_t size_written;
+};
+
+#define PSTORE_XBL_HDR_SIZE	sizeof(struct pstore_xbl_header)
+
 struct pstore_ftrace_record {
 	unsigned long ip;
 	unsigned long parent_ip;
Signature · 1 line
1.9.1
#1 lore raw

Hi!

Thanks for the reminder to review this code. :) Sorry for the delay!

Quoted 10 lines · Yue Hu

From: Yue Hu <[email protected]>

Sometimes we hope to check boot loader log messages (e.g. Android Verified Boot status) when kernel is coming up. Generally it does depend on serial device, but it will be removed for the hardware shipping to market by most of manufacturers. In that case better solder and proper serial cable for different interface (e.g. Type-C or microUSB) are needed. That is inconvenient and even wasting much time on it.

Can you give some examples of how this would be used on a real device? More notes below...

Quoted 49 lines

Therefore, let's add a logging support: PSTORE_TYPE_XBL.

---
v2: mention info of interacting with boot loader

 fs/pstore/Kconfig      | 10 +++++++
 fs/pstore/platform.c   | 16 ++++++++++
 fs/pstore/ram.c        | 81 ++++++++++++++++++++++++++++++++++++++++++++++++--
 include/linux/pstore.h | 21 +++++++++----
 4 files changed, 121 insertions(+), 7 deletions(-)
fs/pstore/Kconfig +10 -0
index 0d19d19..ef4a2dc 100644
@@ -137,6 +137,16 @@ config PSTORE_FTRACE
 
          If unsure, say N.
 
+config PSTORE_XBL
+       bool "Log bootloader messages"
+       depends on PSTORE
+       help
+         When the option is enabled, pstore will log boot loader
+         messages to /sys/fs/pstore/xbl-ramoops-[ID] after reboot.
+         Boot loader needs to support log buffer reserved.
+
+         If unsure, say N.
+
 config PSTORE_RAM
        tristate "Log panic/oops to a RAM buffer"
        depends on PSTORE
fs/pstore/platform.c +1 -0
index 2d1066e..2e6c3f8f 100644
@@ -65,6 +65,7 @@
        "mce",
        "console",
        "ftrace",
+       "xbl",
        "rtas",
        "powerpc-ofw",
        "powerpc-common",
@@ -530,6 +531,19 @@ static void pstore_register_console(void) {}
 static void pstore_unregister_console(void) {}
 #endif

fs/pstore/platform.c static void pstore_register_console(void) {} · line 542

Quoted 10 lines
fs/pstore/platform.c +9 -0
 
+#ifdef CONFIG_PSTORE_XBL
+static void pstore_register_xbl(void)
+{
+       struct pstore_record record;
+
+       pstore_record_init(&record, psinfo);
+       record.type = PSTORE_TYPE_XBL;
+       psinfo->write(&record);
+}

This seems like a very strange way to get the record: this is an "empty" write that has a side-effect of reading the XBL region and copying it into the prz area. I would expect this to all happen in ramoops_pstore_read() instead.

Quoted 21 lines
fs/pstore/platform.c +6 -0
+#else
+static void pstore_register_xbl(void) {}
+#endif
+
 static int pstore_write_user_compat(struct pstore_record *record,
                                    const char __user *buf)
 {
@@ -616,6 +630,8 @@ int pstore_register(struct pstore_info *psi)
                pstore_register_ftrace();
        if (psi->flags & PSTORE_FLAGS_PMSG)
                pstore_register_pmsg();
+       if (psi->flags & PSTORE_FLAGS_XBL)
+               pstore_register_xbl();
 
        /* Start watching for new records, if desired. */
        if (pstore_update_ms >= 0) {
fs/pstore/ram.c +0 -0
index 1adb5e3..b114b1d 100644

fs/pstore/ram.c line 57

Quoted 2 lines
fs/pstore/ram.c +0 -0
@@ -56,6 +56,27 @@
 module_param_named(pmsg_size, ramoops_pmsg_size, ulong, 0400);
 MODULE_PARM_DESC(pmsg_size, "size of user space message log");

How is the base address of the XBL area specified? It looks currently like it's up to the end user to do all the math correctly to line it up with where it's expected?

Quoted 7 lines
fs/pstore/ram.c +6 -0
+/*
+ * interact with boot loader
+ * =========================
+ *
+ * xbl memory layout:
+ * +----+

fs/pstore/ram.c line 74

Quoted 10 lines
fs/pstore/ram.c +10 -0
+ * |dst |
+ * |----|     +------------+
+ * |src ----> |   header   |
+ * +----+     |log messages|
+ *            +------------+
+ *
+ * As above, src memory is used to store header and log messages generated
+ * by boot loader, pstore will copy the log messages to dst memory which
+ * has same size as src. The header in src is to record log messages size
+ * written and make xbl cookie.

Why is such a copy needed? The log is already present in memory; why can't pstore just use what's already there?

Quoted 83 lines
fs/pstore/ram.c +34 -0
+ */
+static ulong ramoops_xbl_size = MIN_MEM_SIZE;
+module_param_named(xbl_size, ramoops_xbl_size, ulong, 0400);
+MODULE_PARM_DESC(xbl_size, "size of boot loader log");
+
 static unsigned long long mem_address;
 module_param_hw(mem_address, ullong, other, 0400);
 MODULE_PARM_DESC(mem_address,
@@ -88,6 +109,7 @@ struct ramoops_context {
        struct persistent_ram_zone *cprz;       /* Console zone */
        struct persistent_ram_zone **fprzs;     /* Ftrace zones */
        struct persistent_ram_zone *mprz;       /* PMSG zone */
+       struct persistent_ram_zone *bprz;       /* XBL zone */
        phys_addr_t phys_addr;
        unsigned long size;
        unsigned int memtype;
@@ -95,6 +117,7 @@ struct ramoops_context {
        size_t console_size;
        size_t ftrace_size;
        size_t pmsg_size;
+       size_t xbl_size;
        int dump_oops;
        u32 flags;
        struct persistent_ram_ecc_info ecc_info;
@@ -106,6 +129,7 @@ struct ramoops_context {
        unsigned int max_ftrace_cnt;
        unsigned int ftrace_read_cnt;
        unsigned int pmsg_read_cnt;
+       unsigned int xbl_read_cnt;
        struct pstore_info pstore;
 };
 
@@ -119,6 +143,7 @@ static int ramoops_pstore_open(struct pstore_info *psi)
        cxt->console_read_cnt = 0;
        cxt->ftrace_read_cnt = 0;
        cxt->pmsg_read_cnt = 0;
+       cxt->xbl_read_cnt = 0;
        return 0;
 }
 
@@ -272,6 +297,10 @@ static ssize_t ramoops_pstore_read(struct pstore_record *record)
        if (!prz_ok(prz) && !cxt->pmsg_read_cnt++)
                prz = ramoops_get_next_prz(&cxt->mprz, 0 /* single */, record);
 
+       if (!prz_ok(prz) && !cxt->xbl_read_cnt++) {
+               prz = ramoops_get_next_prz(&cxt->bprz, 0 /* single */, record);
+       }
+
        /* ftrace is last since it may want to dynamically allocate memory. */
        if (!prz_ok(prz)) {
                if (!(cxt->flags & RAMOOPS_FLAG_FTRACE_PER_CPU) &&
@@ -360,6 +389,26 @@ static size_t ramoops_write_kmsg_hdr(struct persistent_ram_zone *prz,
        return len;
 }
 
+static void ramoops_get_xbl_record(struct persistent_ram_zone *prz,
+                                 struct pstore_record *record,
+                                 size_t xbl_size)
+{
+       struct pstore_xbl_header *hdr = NULL;
+       size_t half = xbl_size / 2;
+       size_t max = half - PSTORE_XBL_HDR_SIZE;
+
+       hdr = (struct pstore_xbl_header *)((size_t)prz->buffer + half);
+
+       if (hdr->cookie == PSTORE_XBL_COOKIE) {
+               record->buf = (char *)((size_t)hdr + PSTORE_XBL_HDR_SIZE);
+               record->size = hdr->size_written;
+               if (unlikely(record->size > max))
+                       record->size = max;
+               return;
+       }
+       pr_debug("no valid xbl record (cookie = 0x%08x)\n", hdr->cookie);
+}
+
 static int notrace ramoops_pstore_write(struct pstore_record *record)
 {
        struct ramoops_context *cxt = record->psi->data;
@@ -390,6 +439,16 @@ static int notrace ramoops_pstore_write(struct pstore_record *record)
        } else if (record->type == PSTORE_TYPE_PMSG) {
                pr_warn_ratelimited("PMSG shouldn't call %s\n", __func__);
                return -EINVAL;
+       } else if (record->type == PSTORE_TYPE_XBL) {

fs/pstore/ram.c static int notrace ramoops_pstore_write(struct pstore_record *record) · line 452

Quoted 10 lines
fs/pstore/ram.c +9 -0
+               if (!cxt->bprz)
+                       return -ENOMEM;
+
+               ramoops_get_xbl_record(cxt->bprz, record, cxt->xbl_size);
+               if (record->size == 0)
+                       return -EINVAL;
+
+               persistent_ram_write(cxt->bprz, record->buf, record->size);
+               return 0;
        }

This operates using a very strange side-effect (strange in the sense that none of the other przs work like this). I don't get the sense that XBL is actually a persistent ram zone: these are storage areas to be used for specific front-end purposes (console, oops, ftrace, etc). The XBL seems much more like a read-only backend, but having a read-only backend kind of defeats the purpose of pstore. :) This doesn't seem like it maps to pstore very well (it's not a storage area). You're looking to just expose a memory region somewhere in a kernel filesystem.

I'm not entirely opposed to creating a new type of "thing" for pstore, just so the filesystem infrastructure can be reused (and it's conceptually similar), but it looks like the expected information needed to read XBL are: where in memory to find it, and what is the maximum expected size of the log. Extracting that directly should be a pretty small addition to pstore.

Alternatively, why not inject the XBL into the kernel printk log directly instead?

Also, where can I find a public specification about XBL?

Signature · 1 line
Kees Cook
author
#2 lore raw
Quoted 237 lines · On Wed, 15 May 2019 15:04:09 -0700 Kees Cook <[email protected]> wrote:

Hi!

Thanks for the reminder to review this code. :) Sorry for the delay!

Quoted 10 lines · Yue Hu

From: Yue Hu <[email protected]>

Sometimes we hope to check boot loader log messages (e.g. Android Verified Boot status) when kernel is coming up. Generally it does depend on serial device, but it will be removed for the hardware shipping to market by most of manufacturers. In that case better solder and proper serial cable for different interface (e.g. Type-C or microUSB) are needed. That is inconvenient and even wasting much time on it.

Can you give some examples of how this would be used on a real device? More notes below...

Quoted 49 lines

Therefore, let's add a logging support: PSTORE_TYPE_XBL.

---
v2: mention info of interacting with boot loader

 fs/pstore/Kconfig      | 10 +++++++
 fs/pstore/platform.c   | 16 ++++++++++
 fs/pstore/ram.c        | 81 ++++++++++++++++++++++++++++++++++++++++++++++++--
 include/linux/pstore.h | 21 +++++++++----
 4 files changed, 121 insertions(+), 7 deletions(-)
fs/pstore/Kconfig +10 -0
index 0d19d19..ef4a2dc 100644
@@ -137,6 +137,16 @@ config PSTORE_FTRACE
 
          If unsure, say N.
 
+config PSTORE_XBL
+       bool "Log bootloader messages"
+       depends on PSTORE
+       help
+         When the option is enabled, pstore will log boot loader
+         messages to /sys/fs/pstore/xbl-ramoops-[ID] after reboot.
+         Boot loader needs to support log buffer reserved.
+
+         If unsure, say N.
+
 config PSTORE_RAM
        tristate "Log panic/oops to a RAM buffer"
        depends on PSTORE
fs/pstore/platform.c +1 -0
index 2d1066e..2e6c3f8f 100644
@@ -65,6 +65,7 @@
        "mce",
        "console",
        "ftrace",
+       "xbl",
        "rtas",
        "powerpc-ofw",
        "powerpc-common",
@@ -530,6 +531,19 @@ static void pstore_register_console(void) {}
 static void pstore_unregister_console(void) {}
 #endif

fs/pstore/platform.c static void pstore_register_console(void) {} · line 542

Quoted 10 lines
fs/pstore/platform.c +9 -0
 
+#ifdef CONFIG_PSTORE_XBL
+static void pstore_register_xbl(void)
+{
+       struct pstore_record record;
+
+       pstore_record_init(&record, psinfo);
+       record.type = PSTORE_TYPE_XBL;
+       psinfo->write(&record);
+}  

This seems like a very strange way to get the record: this is an "empty" write that has a side-effect of reading the XBL region and copying it into the prz area. I would expect this to all happen in ramoops_pstore_read() instead.

Quoted 21 lines
fs/pstore/platform.c +6 -0
+#else
+static void pstore_register_xbl(void) {}
+#endif
+
 static int pstore_write_user_compat(struct pstore_record *record,
                                    const char __user *buf)
 {
@@ -616,6 +630,8 @@ int pstore_register(struct pstore_info *psi)
                pstore_register_ftrace();
        if (psi->flags & PSTORE_FLAGS_PMSG)
                pstore_register_pmsg();
+       if (psi->flags & PSTORE_FLAGS_XBL)
+               pstore_register_xbl();
 
        /* Start watching for new records, if desired. */
        if (pstore_update_ms >= 0) {
fs/pstore/ram.c +0 -0
index 1adb5e3..b114b1d 100644

fs/pstore/ram.c line 57

Quoted 2 lines
fs/pstore/ram.c +0 -0
@@ -56,6 +56,27 @@
 module_param_named(pmsg_size, ramoops_pmsg_size, ulong, 0400);
 MODULE_PARM_DESC(pmsg_size, "size of user space message log");  

How is the base address of the XBL area specified? It looks currently like it's up to the end user to do all the math correctly to line it up with where it's expected?

Quoted 7 lines
fs/pstore/ram.c +6 -0
+/*
+ * interact with boot loader
+ * =========================
+ *
+ * xbl memory layout:
+ * +----+

fs/pstore/ram.c line 74

Quoted 10 lines
fs/pstore/ram.c +10 -0
+ * |dst |
+ * |----|     +------------+
+ * |src ----> |   header   |
+ * +----+     |log messages|
+ *            +------------+
+ *
+ * As above, src memory is used to store header and log messages generated
+ * by boot loader, pstore will copy the log messages to dst memory which
+ * has same size as src. The header in src is to record log messages size
+ * written and make xbl cookie.  

Why is such a copy needed? The log is already present in memory; why can't pstore just use what's already there?

Quoted 83 lines
fs/pstore/ram.c +34 -0
+ */
+static ulong ramoops_xbl_size = MIN_MEM_SIZE;
+module_param_named(xbl_size, ramoops_xbl_size, ulong, 0400);
+MODULE_PARM_DESC(xbl_size, "size of boot loader log");
+
 static unsigned long long mem_address;
 module_param_hw(mem_address, ullong, other, 0400);
 MODULE_PARM_DESC(mem_address,
@@ -88,6 +109,7 @@ struct ramoops_context {
        struct persistent_ram_zone *cprz;       /* Console zone */
        struct persistent_ram_zone **fprzs;     /* Ftrace zones */
        struct persistent_ram_zone *mprz;       /* PMSG zone */
+       struct persistent_ram_zone *bprz;       /* XBL zone */
        phys_addr_t phys_addr;
        unsigned long size;
        unsigned int memtype;
@@ -95,6 +117,7 @@ struct ramoops_context {
        size_t console_size;
        size_t ftrace_size;
        size_t pmsg_size;
+       size_t xbl_size;
        int dump_oops;
        u32 flags;
        struct persistent_ram_ecc_info ecc_info;
@@ -106,6 +129,7 @@ struct ramoops_context {
        unsigned int max_ftrace_cnt;
        unsigned int ftrace_read_cnt;
        unsigned int pmsg_read_cnt;
+       unsigned int xbl_read_cnt;
        struct pstore_info pstore;
 };
 
@@ -119,6 +143,7 @@ static int ramoops_pstore_open(struct pstore_info *psi)
        cxt->console_read_cnt = 0;
        cxt->ftrace_read_cnt = 0;
        cxt->pmsg_read_cnt = 0;
+       cxt->xbl_read_cnt = 0;
        return 0;
 }
 
@@ -272,6 +297,10 @@ static ssize_t ramoops_pstore_read(struct pstore_record *record)
        if (!prz_ok(prz) && !cxt->pmsg_read_cnt++)
                prz = ramoops_get_next_prz(&cxt->mprz, 0 /* single */, record);
 
+       if (!prz_ok(prz) && !cxt->xbl_read_cnt++) {
+               prz = ramoops_get_next_prz(&cxt->bprz, 0 /* single */, record);
+       }
+
        /* ftrace is last since it may want to dynamically allocate memory. */
        if (!prz_ok(prz)) {
                if (!(cxt->flags & RAMOOPS_FLAG_FTRACE_PER_CPU) &&
@@ -360,6 +389,26 @@ static size_t ramoops_write_kmsg_hdr(struct persistent_ram_zone *prz,
        return len;
 }
 
+static void ramoops_get_xbl_record(struct persistent_ram_zone *prz,
+                                 struct pstore_record *record,
+                                 size_t xbl_size)
+{
+       struct pstore_xbl_header *hdr = NULL;
+       size_t half = xbl_size / 2;
+       size_t max = half - PSTORE_XBL_HDR_SIZE;
+
+       hdr = (struct pstore_xbl_header *)((size_t)prz->buffer + half);
+
+       if (hdr->cookie == PSTORE_XBL_COOKIE) {
+               record->buf = (char *)((size_t)hdr + PSTORE_XBL_HDR_SIZE);
+               record->size = hdr->size_written;
+               if (unlikely(record->size > max))
+                       record->size = max;
+               return;
+       }
+       pr_debug("no valid xbl record (cookie = 0x%08x)\n", hdr->cookie);
+}
+
 static int notrace ramoops_pstore_write(struct pstore_record *record)
 {
        struct ramoops_context *cxt = record->psi->data;
@@ -390,6 +439,16 @@ static int notrace ramoops_pstore_write(struct pstore_record *record)
        } else if (record->type == PSTORE_TYPE_PMSG) {
                pr_warn_ratelimited("PMSG shouldn't call %s\n", __func__);
                return -EINVAL;
+       } else if (record->type == PSTORE_TYPE_XBL) {

fs/pstore/ram.c static int notrace ramoops_pstore_write(struct pstore_record *record) · line 452

Quoted 10 lines
fs/pstore/ram.c +9 -0
+               if (!cxt->bprz)
+                       return -ENOMEM;
+
+               ramoops_get_xbl_record(cxt->bprz, record, cxt->xbl_size);
+               if (record->size == 0)
+                       return -EINVAL;
+
+               persistent_ram_write(cxt->bprz, record->buf, record->size);
+               return 0;
        }  

This operates using a very strange side-effect (strange in the sense that none of the other przs work like this). I don't get the sense that XBL is actually a persistent ram zone: these are storage areas to be used for specific front-end purposes (console, oops, ftrace, etc). The XBL seems much more like a read-only backend, but having a read-only backend kind of defeats the purpose of pstore. :) This doesn't seem like it maps to pstore very well (it's not a storage area). You're looking to just expose a memory region somewhere in a kernel filesystem.

Quoted 6 lines

I'm not entirely opposed to creating a new type of "thing" for pstore, just so the filesystem infrastructure can be reused (and it's conceptually similar), but it looks like the expected information needed to read XBL are: where in memory to find it, and what is the maximum expected size of the log. Extracting that directly should be a pretty small addition to pstore.

ok, just use it locally in my side :]

Thx.

Quoted 6 lines

Alternatively, why not inject the XBL into the kernel printk log directly instead?

Also, where can I find a public specification about XBL?