Re: [Qemu-devel] [PATCH v4 3/5] fw_cfg: do DMA read operation

[Gabriel L. Somlo]

On Sun, Nov 12, 2017 at 09:36:33AM -0500, Marc-André Lureau wrote:
> 
> 
> ----- Original Message -----
> > On Tue, Oct 31, 2017 at 04:19:36PM +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.
> > > 
> > > To avoid polling with unbound amount of time, the DMA operation is
> > > expected to complete within 200ms, or will return ETIME error.
> > > 
> > > 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 <address@hidden>
> > 
> > A couple of silly questions, inline, below.
> > 
> > Thanks,
> > --G
> > 
> > > ---
> > >  drivers/firmware/qemu_fw_cfg.c | 154
> > >  ++++++++++++++++++++++++++++++++++++-----
> > >  1 file changed, 137 insertions(+), 17 deletions(-)
> > > 
> > > diff --git a/drivers/firmware/qemu_fw_cfg.c
> > > b/drivers/firmware/qemu_fw_cfg.c
> > > index 1f3e8545dab7..8ce5e49b7c62 100644
> > > --- a/drivers/firmware/qemu_fw_cfg.c
> > > +++ b/drivers/firmware/qemu_fw_cfg.c
> > > @@ -33,6 +33,8 @@
> > >  #include <linux/slab.h>
> > >  #include <linux/io.h>
> > >  #include <linux/ioport.h>
> > > +#include <linux/delay.h>
> > > +#include <linux/dma-mapping.h>
> > >  
> > >  MODULE_AUTHOR("Gabriel L. Somlo <address@hidden>");
> > >  MODULE_DESCRIPTION("QEMU fw_cfg sysfs support");
> > > @@ -43,12 +45,26 @@ 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
> > > +#define FW_CFG_DMA_TIMEOUT     200 /* ms */
> > > +
> > >  /* 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
> > >  
> > > +/* platform device for dma mapping */
> > > +static struct device *dev;
> > > +
> > > +/* 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 +73,12 @@ 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 +97,93 @@ 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;
> > > +}
> > > +
> > > +static bool fw_cfg_wait_for_control(struct fw_cfg_dma *d, unsigned long
> > > timeout)
> > > +{
> > > + ktime_t start;
> > > + ktime_t stop;
> > > +
> > > + start = ktime_get();
> > > + stop = ktime_add(start, ms_to_ktime(timeout));
> > > +
> > > + do {
> > > +         if ((be32_to_cpu(d->control) & ~FW_CFG_DMA_CTL_ERROR) == 0)
> > > +                 return true;
> > > +
> > > +         usleep_range(50, 100);
> > > + } while (ktime_before(ktime_get(), stop));
> > > +
> > > + return false;
> > > +}
> > > +
> > > +static ssize_t fw_cfg_dma_transfer(void *address, u32 length, u32 
> > > control)
> > > +{
> > > + dma_addr_t dma_addr = 0;
> > > + struct fw_cfg_dma *d;
> > > + dma_addr_t dma;
> > > + ssize_t ret = length;
> > > + enum dma_data_direction dir =
> > > +         (control & FW_CFG_DMA_CTL_READ ? DMA_FROM_DEVICE : 0);
> > > +
> > > + if (address && length) {
> > > +         dma_addr = dma_map_single(dev, address, length, dir);
> > > +         if (dma_mapping_error(NULL, dma_addr)) {
> > > +                 WARN(1, "%s: failed to map address\n", __func__);
> > > +                 return -EFAULT;
> > > +         }
> > > + }
> > > +
> > > + d = kmalloc(sizeof(*d), GFP_KERNEL | GFP_DMA);
> > > + if (!d) {
> > > +         ret = -ENOMEM;
> > > +         goto end;
> > > + }
> > > +
> > > + dma = dma_map_single(dev, d, sizeof(*d), DMA_BIDIRECTIONAL);
> > > + if (dma_mapping_error(NULL, dma)) {
> > > +         WARN(1, "%s: failed to map fw_cfg_dma\n", __func__);
> > > +         ret = -EFAULT;
> > > +         goto end;
> > > + }
> > > +
> > > + *d = (struct fw_cfg_dma) {
> > > +         .address = cpu_to_be64(dma_addr),
> > > +         .length = cpu_to_be32(length),
> > > +         .control = cpu_to_be32(control)
> > > + };
> > > +
> > > + iowrite32be((u64)dma >> 32, fw_cfg_reg_dma);
> > > + iowrite32be(dma, fw_cfg_reg_dma + 4);
> > > +
> > > + if (!fw_cfg_wait_for_control(d, FW_CFG_DMA_TIMEOUT)) {
> > > +         WARN(1, "%s: timeout", __func__);
> > > +         ret = -ETIME;
> > > + } else if (be32_to_cpu(d->control) & FW_CFG_DMA_CTL_ERROR) {
> > > +         ret = -EIO;
> > > + }
> > > +
> > > + dma_unmap_single(dev, dma, sizeof(*d), DMA_BIDIRECTIONAL);
> > > +
> > > +end:
> > > + kfree(d);
> > > + if (dma_addr)
> > > +         dma_unmap_single(dev, dma_addr, length, dir);
> > > +
> > > + 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(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 +193,36 @@ 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 -EBUSY;
> > 
> > Why EBUSY here (and not, e.g., EINVAL) ?
> 
> no specific reason, I guess EINVAL or ENODEV would be fine too.

Doesn't *really* matter, since we should never reach that code anyway.
But I think EINVAL would cause the casual reader less cause to wonder
what might be going on than any of the others...
 
> > 
> > >   }
> > >  
> > >   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(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(0, pos, FW_CFG_DMA_CTL_SKIP);
> > 
> > s/0/NULL/ ?
> 
> sure

Since the argument is supposed to be a pointer, that'd be following the
"principle of least WTF" :)

With those (in both 3/5 and 5/5),

Reviewed-by: Gabriel Somlo <address@hidden>

Thanks,
--Gabriel

> > 
> > > +                 if (ret < 0)
> > > +                         goto end;
> > > +                 ret = fw_cfg_dma_transfer(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 +314,7 @@ 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(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 +323,6 @@ 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);
> > > @@ -351,8 +470,7 @@ 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(entry->f.select, buf, pos, count, true);
> > >  }
> > >  
> > >  static struct bin_attribute fw_cfg_sysfs_attr_raw = {
> > > @@ -505,7 +623,7 @@ static int fw_cfg_register_dir_entries(void)
> > >   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(FW_CFG_FILE_DIR, &count, 0, sizeof(count), false);
> > >   count = be32_to_cpu(count);
> > >   dir_size = count * sizeof(struct fw_cfg_file);
> > >  
> > > @@ -513,7 +631,7 @@ 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(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);
> > > @@ -544,9 +662,10 @@ static int fw_cfg_sysfs_probe(struct platform_device
> > > *pdev)
> > >    * one fw_cfg device exist system-wide, so if one was already found
> > >    * earlier, we might as well stop here.
> > >    */
> > > - if (fw_cfg_sel_ko)
> > > + if (dev)
> > >           return -EBUSY;
> > >  
> > > + dev = &pdev->dev;
> > >   /* 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);
> > > @@ -562,7 +681,7 @@ 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(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)
> > > @@ -587,6 +706,7 @@ static int fw_cfg_sysfs_probe(struct platform_device
> > > *pdev)
> > >  err_name:
> > >   fw_cfg_kobj_cleanup(fw_cfg_sel_ko);
> > >  err_sel:
> > > + dev = NULL;
> > >   return err;
> > >  }
> > >  
> > > --
> > > 2.15.0.rc0.40.gaefcc5f6f
> > > 
> >