Re: [Qemu-devel] [PATCH v6 5/5] fw_cfg: write vmcoreinfo details

[Marc-André Lureau]

Hi

----- Original Message -----
> On Mon, Nov 13, 2017 at 08:29:58PM +0100, Marc-André Lureau wrote:
> > 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 <address@hidden>
> > Reviewed-by: Gabriel Somlo <address@hidden>
> > ---
> >  drivers/firmware/qemu_fw_cfg.c | 87
> >  +++++++++++++++++++++++++++++++++++++++++-
> >  1 file changed, 86 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/firmware/qemu_fw_cfg.c
> > b/drivers/firmware/qemu_fw_cfg.c
> > index 2ac4cd869fe6..7a70e7a549f6 100644
> > --- a/drivers/firmware/qemu_fw_cfg.c
> > +++ b/drivers/firmware/qemu_fw_cfg.c
> > @@ -35,6 +35,8 @@
> >  #include <linux/ioport.h>
> >  #include <linux/delay.h>
> >  #include <linux/dma-mapping.h>
> > +#include <linux/crash_core.h>
> > +#include <linux/crash_dump.h>
> >  
> >  MODULE_AUTHOR("Gabriel L. Somlo <address@hidden>");
> >  MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
> > @@ -59,6 +61,8 @@ 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
> > +
> >  /* platform device for dma mapping */
> >  static struct device *dev;
> >  
> > @@ -127,7 +131,8 @@ static ssize_t fw_cfg_dma_transfer(void *address, u32
> > length, u32 control)
> >     dma_addr_t dma;
> >     ssize_t ret = length;
> >     enum dma_data_direction dir =
> > -           (control & FW_CFG_DMA_CTL_READ ? DMA_FROM_DEVICE : 0);
> > +           (control & FW_CFG_DMA_CTL_READ ? DMA_FROM_DEVICE : 0) |
> > +           (control & FW_CFG_DMA_CTL_WRITE ? DMA_TO_DEVICE : 0);
> >  
> >     if (address && length) {
> >             dma_addr = dma_map_single(dev, address, length, dir);
> > @@ -225,6 +230,48 @@ static ssize_t fw_cfg_read_blob(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(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__);
> > +           memset(buf, 0, count);
> > +           return -EINVAL;
> > +   }
> > +
> > +   mutex_lock(&fw_cfg_dev_lock);
> > +   if (pos == 0) {
> > +           ret = fw_cfg_dma_transfer(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(NULL, pos, FW_CFG_DMA_CTL_SKIP);
> > +           if (ret < 0)
> > +                   goto end;
> > +           ret = fw_cfg_dma_transfer(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)
> >  {
> > @@ -343,6 +390,37 @@ struct fw_cfg_sysfs_entry {
> >     struct list_head list;
> >  };
> >  
> > +#ifdef CONFIG_CRASH_CORE
> > +static ssize_t write_vmcoreinfo(const struct fw_cfg_file *f)
> > +{
> > +   struct vmci {
> > +           __le16 host_format;
> > +           __le16 guest_format;
> > +           __le32 size;
> > +           __le64 paddr;
> > +   } __packed;
> > +   struct vmci *data;
> > +   ssize_t ret;
> > +
> > +   data = kmalloc(sizeof(struct vmci), GFP_KERNEL | GFP_DMA);
> > +   if (!data)
> > +           return -ENOMEM;
> 
> It's a small bit of data - you can just keep it in a global variable,
> this way failures won't be an issue.

It would still need to be allocated with GFP_DMA. Since it's a one time thing, 
this is just moving the problem isn't it?

> 
> > +
> > +   /* spare ourself reading host format support for now since we
> > +    * don't know what else to format - host may ignore ours
> > +    */
> > +   *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())
> > +   };
> > +   ret = fw_cfg_write_blob(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)
> >  {
> > @@ -582,6 +660,13 @@ static int fw_cfg_register_file(const struct
> > fw_cfg_file *f)
> >     int err;
> >     struct fw_cfg_sysfs_entry *entry;
> >  
> > +#ifdef CONFIG_CRASH_CORE
> > +   if (strcmp(f->name, "etc/vmcoreinfo") == 0 && !is_kdump_kernel()) {
> > +           if (write_vmcoreinfo(f) < 0)
> > +                   pr_warn("fw_cfg: failed to write vmcoreinfo");
> > +   }
> > +#endif
> > +
> >     /* allocate new entry */
> >     entry = kzalloc(sizeof(*entry), GFP_KERNEL);
> >     if (!entry)
> > --
> > 2.15.0.125.g8f49766d64
>