From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:46387) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eH6Yt-0005e9-H9 for qemu-devel@nongnu.org; Tue, 21 Nov 2017 06:17:58 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1eH6Yp-0007eD-CJ for qemu-devel@nongnu.org; Tue, 21 Nov 2017 06:17:55 -0500 Received: from mx1.redhat.com ([209.132.183.28]:48364) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1eH6Yp-0007cs-3R for qemu-devel@nongnu.org; Tue, 21 Nov 2017 06:17:51 -0500 Date: Tue, 21 Nov 2017 06:17:46 -0500 (EST) From: =?utf-8?Q?Marc-Andr=C3=A9?= Lureau Message-ID: <1367386885.43304805.1511263066456.JavaMail.zimbra@redhat.com> In-Reply-To: <20171120232407-mutt-send-email-mst@kernel.org> References: <20171120095519.15214-1-marcandre.lureau@redhat.com> <20171120095519.15214-4-marcandre.lureau@redhat.com> <20171120232407-mutt-send-email-mst@kernel.org> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH v7 3/5] fw_cfg: do DMA read operation List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: "Michael S. Tsirkin" Cc: linux-kernel@vger.kernel.org, qemu-devel@nongnu.org, somlo@cmu.edu, xiaolong ye Hi ----- Original Message ----- > On Mon, Nov 20, 2017 at 10:55:17AM +0100, Marc-Andr=C3=A9 Lureau wrote: > > Modify fw_cfg_read_blob() to use DMA if the device supports it. > > Return errors, because the operation may fail. > >=20 > > 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. > >=20 > > We may want to switch all the *buf addresses to use only kmalloc'ed > > buffers (instead of using stack/image addresses with dma=3Dfalse). > >=20 > > Signed-off-by: Marc-Andr=C3=A9 Lureau > > --- > > drivers/firmware/qemu_fw_cfg.c | 140 > > ++++++++++++++++++++++++++++++++++++----- > > 1 file changed, 123 insertions(+), 17 deletions(-) > >=20 > > diff --git a/drivers/firmware/qemu_fw_cfg.c > > b/drivers/firmware/qemu_fw_cfg.c > > index 740df0df2260..e0fe6ff037c3 100644 > > --- a/drivers/firmware/qemu_fw_cfg.c > > +++ b/drivers/firmware/qemu_fw_cfg.c > > @@ -33,6 +33,8 @@ > > #include > > #include > > #include > > +#include > > +#include > > =20 > > MODULE_AUTHOR("Gabriel L. Somlo "); > > MODULE_DESCRIPTION("QEMU fw_cfg sysfs support"); > > @@ -43,12 +45,25 @@ MODULE_LICENSE("GPL"); > > #define FW_CFG_ID 0x01 > > #define FW_CFG_FILE_DIR 0x19 > > =20 > > +#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 > > =20 > > /* fw_cfg "file name" is up to 56 characters (including terminating nu= l) > > */ > > #define FW_CFG_MAX_FILE_PATH 56 > > =20 > > +/* platform device for dma mapping */ > > +static struct device *dev; > > + > > +/* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level d= ir. > > */ > > +static u32 fw_cfg_rev; > > + > > /* fw_cfg file directory entry type */ > > struct fw_cfg_file { > > =09u32 size; > > @@ -57,6 +72,12 @@ struct fw_cfg_file { > > =09char name[FW_CFG_MAX_FILE_PATH]; > > }; > > =20 > > +struct fw_cfg_dma { > > +=09u32 control; > > +=09u32 length; > > +=09u64 address; > > +} __packed; > > + > > /* fw_cfg device i/o register addresses */ > > static bool fw_cfg_is_mmio; > > static phys_addr_t fw_cfg_p_base; >=20 > Drop __packed please. It causes many gcc versions to do insane things. > Can be a patch on top. Oh? I think __packed should translate to __attribute__((packed)) (include/l= inux/compiler-gcc.h), there would be serious problems if gcc "do insane thi= ngs" with it. >=20 >=20 > > @@ -75,12 +96,79 @@ static inline u16 fw_cfg_sel_endianness(u16 key) > > =09return fw_cfg_is_mmio ? cpu_to_be16(key) : cpu_to_le16(key); > > } > > =20 > > +static inline bool fw_cfg_dma_enabled(void) > > +{ > > +=09return 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, dma_addr_t d= ma) > > +{ > > +=09do { > > +=09=09dma_sync_single_for_cpu(dev, dma, sizeof(*d), DMA_FROM_DEVICE); > > +=09=09if ((be32_to_cpu(d->control) & ~FW_CFG_DMA_CTL_ERROR) =3D=3D 0) > > +=09=09=09return; > > + > > +=09=09usleep_range(50, 100); >=20 > And since in practice we never get to this line, > maybe we should just go back to yield here. Or cond_resched() ? >=20 > > +=09} while (true); > > +} > > + > > +static ssize_t fw_cfg_dma_transfer(void *address, u32 length, u32 cont= rol) > > +{ > > +=09dma_addr_t dma_addr =3D 0; > > +=09static struct fw_cfg_dma d; > > +=09dma_addr_t dma; > > +=09ssize_t ret =3D length; > > +=09enum dma_data_direction dir =3D > > +=09=09(control & FW_CFG_DMA_CTL_READ ? DMA_FROM_DEVICE : 0); > > + > > +=09if (address && length) { > > +=09=09dma_addr =3D dma_map_single(dev, address, length, dir); > > +=09=09if (dma_mapping_error(NULL, dma_addr)) { > > +=09=09=09WARN(1, "%s: failed to map address\n", __func__); > > +=09=09=09return -EFAULT; > > +=09=09} > > +=09} > > + > > +=09d =3D (struct fw_cfg_dma) { > > +=09=09.address =3D cpu_to_be64(dma_addr), > > +=09=09.length =3D cpu_to_be32(length), > > +=09=09.control =3D cpu_to_be32(control) > > +=09}; > > + > > +=09dma =3D dma_map_single(dev, &d, sizeof(d), DMA_BIDIRECTIONAL); > > +=09if (dma_mapping_error(NULL, dma)) { > > +=09=09WARN(1, "%s: failed to map fw_cfg_dma\n", __func__); > > +=09=09ret =3D -EFAULT; > > +=09=09goto end; > > +=09} > > + > > +=09iowrite32be((u64)dma >> 32, fw_cfg_reg_dma); > > +=09iowrite32be(dma, fw_cfg_reg_dma + 4); > > + > > +=09fw_cfg_wait_for_control(&d, dma); > > + > > +=09if (be32_to_cpu(d.control) & FW_CFG_DMA_CTL_ERROR) { > > +=09=09ret =3D -EIO; > > +=09} > > + > > +=09dma_unmap_single(dev, dma, sizeof(d), DMA_BIDIRECTIONAL); > > + > > +end: > > +=09if (dma_addr) > > +=09=09dma_unmap_single(dev, dma_addr, length, dir); > > + > > +=09return ret; > > +} > > + > > /* read chunk of given fw_cfg blob (caller responsible for sanity-chec= k) > > */ > > -static inline void fw_cfg_read_blob(u16 key, > > -=09=09=09=09 void *buf, loff_t pos, size_t count) > > +static ssize_t fw_cfg_read_blob(u16 key, > > +=09=09=09=09void *buf, loff_t pos, size_t count, > > +=09=09=09=09bool dma) > > { > > =09u32 glk =3D -1U; > > =09acpi_status status; > > +=09ssize_t ret =3D count; > > =20 > > =09/* If we have ACPI, ensure mutual exclusion against any potential > > =09 * device access by the firmware, e.g. via AML methods: > > @@ -90,17 +178,36 @@ static inline void fw_cfg_read_blob(u16 key, > > =09=09/* Should never get here */ > > =09=09WARN(1, "fw_cfg_read_blob: Failed to lock ACPI!\n"); > > =09=09memset(buf, 0, count); > > -=09=09return; > > +=09=09return -EINVAL; > > =09} > > =20 > > =09mutex_lock(&fw_cfg_dev_lock); > > -=09iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl); > > -=09while (pos-- > 0) > > -=09=09ioread8(fw_cfg_reg_data); > > -=09ioread8_rep(fw_cfg_reg_data, buf, count); > > +=09if (dma && fw_cfg_dma_enabled()) { > > +=09=09if (pos =3D=3D 0) { > > +=09=09=09ret =3D fw_cfg_dma_transfer(buf, count, key << 16 > > +=09=09=09=09=09=09 | FW_CFG_DMA_CTL_SELECT > > +=09=09=09=09=09=09 | FW_CFG_DMA_CTL_READ); > > +=09=09} else { > > +=09=09=09iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl); > > +=09=09=09ret =3D fw_cfg_dma_transfer(NULL, pos, FW_CFG_DMA_CTL_SKIP); > > +=09=09=09if (ret < 0) > > +=09=09=09=09goto end; > > +=09=09=09ret =3D fw_cfg_dma_transfer(buf, count, > > +=09=09=09=09=09=09 FW_CFG_DMA_CTL_READ); > > +=09=09} > > +=09} else { > > +=09=09iowrite16(fw_cfg_sel_endianness(key), fw_cfg_reg_ctrl); > > +=09=09while (pos-- > 0) > > +=09=09=09ioread8(fw_cfg_reg_data); > > +=09=09ioread8_rep(fw_cfg_reg_data, buf, count); > > +=09} > > + > > +end: > > =09mutex_unlock(&fw_cfg_dev_lock); > > =20 > > =09acpi_release_global_lock(glk); > > + > > +=09return ret; > > } > > =20 > > /* clean up fw_cfg device i/o */ > > @@ -192,7 +299,7 @@ static int fw_cfg_do_platform_probe(struct > > platform_device *pdev) > > #endif > > =20 > > =09/* verify fw_cfg device signature */ > > -=09fw_cfg_read_blob(FW_CFG_SIGNATURE, sig, 0, FW_CFG_SIG_SIZE); > > +=09fw_cfg_read_blob(FW_CFG_SIGNATURE, sig, 0, FW_CFG_SIG_SIZE, false); > > =09if (memcmp(sig, "QEMU", FW_CFG_SIG_SIZE) !=3D 0) { > > =09=09fw_cfg_io_cleanup(); > > =09=09return -ENODEV; > > @@ -201,9 +308,6 @@ static int fw_cfg_do_platform_probe(struct > > platform_device *pdev) > > =09return 0; > > } > > =20 > > -/* fw_cfg revision attribute, in /sys/firmware/qemu_fw_cfg top-level d= ir. > > */ > > -static u32 fw_cfg_rev; > > - > > static ssize_t fw_cfg_showrev(struct kobject *k, struct attribute *a, = char > > *buf) > > { > > =09return sprintf(buf, "%u\n", fw_cfg_rev); > > @@ -351,8 +455,7 @@ static ssize_t fw_cfg_sysfs_read_raw(struct file *f= ilp, > > struct kobject *kobj, > > =09if (count > entry->f.size - pos) > > =09=09count =3D entry->f.size - pos; > > =20 > > -=09fw_cfg_read_blob(entry->f.select, buf, pos, count); > > -=09return count; > > +=09return fw_cfg_read_blob(entry->f.select, buf, pos, count, true); > > } > > =20 > > static struct bin_attribute fw_cfg_sysfs_attr_raw =3D { > > @@ -505,7 +608,7 @@ static int fw_cfg_register_dir_entries(void) > > =09struct fw_cfg_file *dir; > > =09size_t dir_size; > > =20 > > -=09fw_cfg_read_blob(FW_CFG_FILE_DIR, &count, 0, sizeof(count)); > > +=09fw_cfg_read_blob(FW_CFG_FILE_DIR, &count, 0, sizeof(count), false); > > =09count =3D be32_to_cpu(count); > > =09dir_size =3D count * sizeof(struct fw_cfg_file); > > =20 > > @@ -513,7 +616,7 @@ static int fw_cfg_register_dir_entries(void) > > =09if (!dir) > > =09=09return -ENOMEM; > > =20 > > -=09fw_cfg_read_blob(FW_CFG_FILE_DIR, dir, sizeof(count), dir_size); > > +=09fw_cfg_read_blob(FW_CFG_FILE_DIR, dir, sizeof(count), dir_size, tru= e); > > =20 > > =09for (i =3D 0; i < count; i++) { > > =09=09dir[i].size =3D be32_to_cpu(dir[i].size); > > @@ -544,9 +647,10 @@ static int fw_cfg_sysfs_probe(struct platform_devi= ce > > *pdev) > > =09 * one fw_cfg device exist system-wide, so if one was already found > > =09 * earlier, we might as well stop here. > > =09 */ > > -=09if (fw_cfg_sel_ko) > > +=09if (dev) > > =09=09return -EBUSY; > > =20 > > +=09dev =3D &pdev->dev; > > =09/* create by_key and by_name subdirs of /sys/firmware/qemu_fw_cfg/ = */ > > =09err =3D -ENOMEM; > > =09fw_cfg_sel_ko =3D kobject_create_and_add("by_key", fw_cfg_top_ko); > > @@ -562,7 +666,7 @@ static int fw_cfg_sysfs_probe(struct platform_devic= e > > *pdev) > > =09=09goto err_probe; > > =20 > > =09/* get revision number, add matching top-level attribute */ > > -=09fw_cfg_read_blob(FW_CFG_ID, &fw_cfg_rev, 0, sizeof(fw_cfg_rev)); > > +=09fw_cfg_read_blob(FW_CFG_ID, &fw_cfg_rev, 0, sizeof(fw_cfg_rev), fal= se); > > =09fw_cfg_rev =3D le32_to_cpu(fw_cfg_rev); > > =09err =3D sysfs_create_file(fw_cfg_top_ko, &fw_cfg_rev_attr.attr); > > =09if (err) > > @@ -587,6 +691,7 @@ static int fw_cfg_sysfs_probe(struct platform_devic= e > > *pdev) > > err_name: > > =09fw_cfg_kobj_cleanup(fw_cfg_sel_ko); > > err_sel: > > +=09dev =3D NULL; > > =09return err; > > } > > =20 > > @@ -598,6 +703,7 @@ static int fw_cfg_sysfs_remove(struct platform_devi= ce > > *pdev) > > =09fw_cfg_io_cleanup(); > > =09fw_cfg_kset_unregister_recursive(fw_cfg_fname_kset); > > =09fw_cfg_kobj_cleanup(fw_cfg_sel_ko); > > +=09dev =3D NULL; > > =09return 0; > > } > > =20 > > -- > > 2.15.0.277.ga3d2ad2c43 >=20