Skip to contentSkip to conversation

x86: Update cbmem driver

PATCH In review Simon Glass 2 participants 6 messages Aug 13, 2023 – Aug 17, 2023 Char and misc drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: Update cbmem driver

This driver is not actually built since a Kconfig was never created for it.

Add a Kconfig (which is already implied by COREBOOT) and update the implementation to avoid using unnecessary memory. Drop the #ifdef at the top since we can rely on Kconfig to get that right.

To enable it (in addition to serial and video), use:

setenv stdout serial,vidconsole,cbmem

---

 drivers/misc/Kconfig         |  8 ++++++++
 drivers/misc/cbmem_console.c | 26 +++++++++-----------------
 2 files changed, 17 insertions(+), 17 deletions(-)
drivers/misc/Kconfig +8 -0
index b9f5c7a37aed..3df57a5bfbd9 100644
@@ -122,6 +122,14 @@ config VEXPRESS_CONFIG
 	  configuration bus on the Arm Versatile Express boards via
 	  a sysreg driver.
 
+config CBMEM_CONSOLE
+	bool "Write console output to coreboot cbmem"
+	depends on X86
+	help
+	  Enables console output to the cbmem console, which is a memory
+	  region set up by coreboot to hold a record of all console output.
+	  Enable this only if booting from coreboot.
+
 config CMD_CROS_EC
 	bool "Enable crosec command"
 	depends on CROS_EC
drivers/misc/cbmem_console.c +9 -17
index 8bbe33d414da..6ae1389a23e0 100644
@@ -5,27 +5,20 @@
 
 #include <common.h>
 #include <console.h>
-#ifndef CONFIG_SYS_COREBOOT
-#error This driver requires coreboot
-#endif
-
 #include <asm/cb_sysinfo.h>
 
-struct cbmem_console {
-	u32 buffer_size;
-	u32 buffer_cursor;
-	u8  buffer_body[0];
-}  __attribute__ ((__packed__));
-
-static struct cbmem_console *cbmem_console_p;
-
 void cbmemc_putc(struct stdio_dev *dev, char data)
 {
-	int cursor;
+	const struct sysinfo_t *sysinfo = cb_get_sysinfo();
+	struct cbmem_console *cons = sysinfo ? sysinfo->cbmem_cons : NULL;
+	int pos;
+
+	if (!cons)
+		return;
 
-	cursor = cbmem_console_p->buffer_cursor++;
-	if (cursor < cbmem_console_p->buffer_size)
-		cbmem_console_p->buffer_body[cursor] = data;
+	pos = cons->cursor++;
+	if (pos < cons->size)
+		cons->body[pos] = data;
 }
 
 void cbmemc_puts(struct stdio_dev *dev, const char *str)
@@ -40,7 +33,6 @@ int cbmemc_init(void)
 {
 	int rc;
 	struct stdio_dev cons_dev;
-	cbmem_console_p = lib_sysinfo.cbmem_cons;
 
 	memset(&cons_dev, 0, sizeof(cons_dev));
 
Signature · 1 line
2.41.0.640.ga95def55d0-goog
#1 lore raw

+Paul Menzel <[email protected]>

Quoted 90 lines · Simon Glass

This driver is not actually built since a Kconfig was never created for it.

Add a Kconfig (which is already implied by COREBOOT) and update the implementation to avoid using unnecessary memory. Drop the #ifdef at the top since we can rely on Kconfig to get that right.

To enable it (in addition to serial and video), use:

setenv stdout serial,vidconsole,cbmem

---

 drivers/misc/Kconfig         |  8 ++++++++
 drivers/misc/cbmem_console.c | 26 +++++++++-----------------
 2 files changed, 17 insertions(+), 17 deletions(-)
drivers/misc/Kconfig +8 -0
index b9f5c7a37aed..3df57a5bfbd9 100644
@@ -122,6 +122,14 @@ config VEXPRESS_CONFIG
          configuration bus on the Arm Versatile Express boards via
          a sysreg driver.
 
+config CBMEM_CONSOLE
+       bool "Write console output to coreboot cbmem"
+       depends on X86
+       help
+         Enables console output to the cbmem console, which is a memory
+         region set up by coreboot to hold a record of all console output.
+         Enable this only if booting from coreboot.
+
 config CMD_CROS_EC
        bool "Enable crosec command"
        depends on CROS_EC
drivers/misc/cbmem_console.c +9 -18
index 8bbe33d414da..6ae1389a23e0 100644
@@ -5,27 +5,20 @@
 
 #include <common.h>
 #include <console.h>
-#ifndef CONFIG_SYS_COREBOOT
-#error This driver requires coreboot
-#endif
-
 #include <asm/cb_sysinfo.h>
 
-struct cbmem_console {
-       u32 buffer_size;
-       u32 buffer_cursor;
-       u8  buffer_body[0];
-}  __attribute__ ((__packed__));
-
-static struct cbmem_console *cbmem_console_p;
-
 void cbmemc_putc(struct stdio_dev *dev, char data)
 {
-       int cursor;
+       const struct sysinfo_t *sysinfo = cb_get_sysinfo();
+       struct cbmem_console *cons = sysinfo ? sysinfo->cbmem_cons : NULL;
+       int pos;
+
+       if (!cons)
+               return;
 
-       cursor = cbmem_console_p->buffer_cursor++;
-       if (cursor < cbmem_console_p->buffer_size)
-               cbmem_console_p->buffer_body[cursor] = data;
+       pos = cons->cursor++;
+       if (pos < cons->size)
+               cons->body[pos] = data;
 }
 
 void cbmemc_puts(struct stdio_dev *dev, const char *str)
@@ -40,7 +33,6 @@ int cbmemc_init(void)
 {
        int rc;
        struct stdio_dev cons_dev;
-       cbmem_console_p = lib_sysinfo.cbmem_cons;
 
        memset(&cons_dev, 0, sizeof(cons_dev));
 
--

2.41.0.640.ga95def55d0-goog

#2 lore raw
In reply to Simon Glass

Dear Simon and other developers,

drivers/misc/cbmem_console.c line 21

Quoted 6 lines
drivers/misc/cbmem_console.c +3 -3
-	cursor = cbmem_console_p->buffer_cursor++;
-	if (cursor < cbmem_console_p->buffer_size)
-		cbmem_console_p->buffer_body[cursor] = data;
+	pos = cons->cursor++;
+	if (pos < cons->size)
+		cons->body[pos] = data;

While at it, is it OK to increment cons->cursor unconditionally, even when the buffer is full?

It's better to do it after the check, isn't it? E.g.:

	if (cons->cursor < cons->size)
		cons->body[cons->cursor++] = data;

Cheers, Alex.

#3 lore raw

Hi Alex,

Quoted 8 lines · Alex Sadovsky

Dear Simon and other developers,

drivers/misc/cbmem_console.c line 21

Quoted 6 lines
drivers/misc/cbmem_console.c +3 -3
-     cursor = cbmem_console_p->buffer_cursor++;
-     if (cursor < cbmem_console_p->buffer_size)
-             cbmem_console_p->buffer_body[cursor] = data;
+     pos = cons->cursor++;
+     if (pos < cons->size)
+             cons->body[pos] = data;
Quoted 7 lines

While at it, is it OK to increment cons->cursor unconditionally, even when the buffer is full?

It's better to do it after the check, isn't it? E.g.:

        if (cons->cursor < cons->size)
                cons->body[cons->cursor++] = data;

I believe the original intent was to indicate that the buffer had overflowed. But prompted by your review I took a look at the coreboot implementation and it now has an overflow flag.

So I will send a v2 incorporating this.

Thanks,
Simon

#4 lore raw

On 15.08.2023 01:42, Simon Glass:

Quoted 5 lines

I believe the original intent was to indicate that the buffer had overflowed. But prompted by your review I took a look at the coreboot implementation and it now has an overflow flag.

So I will send a v2 incorporating this.

Thank you for this investigation, v2 approach definitely makes sense. Btw, is there any reason to mark 'struct cbmem_console' as packed?

#5 lore raw

Hi Nable,

Quoted 9 lines · Nable

On 15.08.2023 01:42, Simon Glass:

Quoted 5 lines

I believe the original intent was to indicate that the buffer had overflowed. But prompted by your review I took a look at the coreboot implementation and it now has an overflow flag.

So I will send a v2 incorporating this.

Thank you for this investigation, v2 approach definitely makes sense. Btw, is there any reason to mark 'struct cbmem_console' as packed?

Not that I can see. I'll drop that too for v3.

Regards,
Simon